Rapid Damage-Free Robotic Harvesting of Tomatoes (Springer Tracts in Mechanical Engineering) 9811612838, 9789811612831

Citation preview

Springer Tracts in Mechanical Engineering

Jizhan Liu · Zhiguo Li · Pingping Li

Rapid Damage-Free Robotic Harvesting of Tomatoes

Springer Tracts in Mechanical Engineering Series Editors Seung-Bok Choi, College of Engineering, Inha University, Incheon, Korea (Republic of) Haibin Duan, Beijing University of Aeronautics and Astronautics, Beijing, China Yili Fu, Harbin Institute of Technology, Harbin, China Carlos Guardiola, CMT-Motores Termicos, Polytechnic University of Valencia, Valencia, Spain Jian-Qiao Sun, University of California, Merced, CA, USA Young W. Kwon, Naval Postgraduate School, Monterey, CA, USA Francisco Cavas-Martínez, Departamento de Estructuras, Universidad Politécnica de Cartagena, Cartagena, Murcia, Spain Fakher Chaari, National School of Engineers of Sfax, Sfax, Tunisia Francesca di Mare, Institute of Energy Technology, Ruhr-Universität Bochum, Bochum, Nordrhein-Westfalen, Germany

Springer Tracts in Mechanical Engineering (STME) publishes the latest developments in Mechanical Engineering - quickly, informally and with high quality. The intent is to cover all the main branches of mechanical engineering, both theoretical and applied, including: • • • • • • • • • • • • • • • • •

Engineering Design Machinery and Machine Elements Mechanical Structures and Stress Analysis Automotive Engineering Engine Technology Aerospace Technology and Astronautics Nanotechnology and Microengineering Control, Robotics, Mechatronics MEMS Theoretical and Applied Mechanics Dynamical Systems, Control Fluids Mechanics Engineering Thermodynamics, Heat and Mass Transfer Manufacturing Precision Engineering, Instrumentation, Measurement Materials Engineering Tribology and Surface Technology

Within the scope of the series are monographs, professional books or graduate textbooks, edited volumes as well as outstanding PhD theses and books purposely devoted to support education in mechanical engineering at graduate and post-graduate levels. Indexed by SCOPUS, zbMATH, SCImago. Please check our Lecture Notes in Mechanical Engineering at http://www.springer. com/series/11236 if you are interested in conference proceedings. To submit a proposal or for further inquiries, please contact the Springer Editor in your region : Ms. Ella Zhang (China) Email: [email protected] Priya Vyas (India) Email: [email protected] Dr. Leontina Di Cecco (All other countries) Email: [email protected] All books published in the series are submitted for consideration in Web of Science.

More information about this series at http://www.springer.com/series/11693

Jizhan Liu · Zhiguo Li · Pingping Li

Rapid Damage-Free Robotic Harvesting of Tomatoes

Jizhan Liu College of Agricultural Engineering Jiangsu University Zhenjiang, Jiangsu, China Pingping Li College of Forest Resources and Environment Nanjing Forestry University Nanjing, Jiangsu, China

Zhiguo Li College of Mechanical and Electronic Engineering North West Agriculture and Forestry University Yangling, Shaanxi, China

ISSN 2195-9862 ISSN 2195-9870 (electronic) Springer Tracts in Mechanical Engineering ISBN 978-981-16-1283-1 ISBN 978-981-16-1284-8 (eBook) https://doi.org/10.1007/978-981-16-1284-8 Jointly published with Science Press, Beijing, China The print edition is not for sale in China (Mainland). Customers from China (Mainland) please order the print book from: Science Press, Beijing, China. © Science Press, Beijing and Springer Nature Singapore Pte Ltd. 2021 This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part of the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation, broadcasting, reproduction on microfilms or in any other physical way, and transmission or information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology now known or hereafter developed. The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication does not imply, even in the absence of a specific statement, that such names are exempt from the relevant protective laws and regulations and therefore free for general use. The publishers, the authors, and the editors are safe to assume that the advice and information in this book are believed to be true and accurate at the date of publication. Neither the publishers nor the authors or the editors give a warranty, express or implied, with respect to the material contained herein or for any errors or omissions that may have been made. The publishers remain neutral with regard to jurisdictional claims in published maps and institutional affiliations. This Springer imprint is published by the registered company Springer Nature Singapore Pte Ltd. The registered company address is: 152 Beach Road, #21-01/04 Gateway East, Singapore 189721, Singapore

Foreword

The rapid development of robotics has led to profound changes in human production and life. In agriculture, especially in the fruit and vegetable industry, which relies heavily on labor, the development and application of agricultural robots for fruit harvesting, seedling transplanting, pesticide spraying, weeding, and transportation will bring about a tremendous revolution in the industry. Due to the unstructured environment and the significant difference in individual fruit or vegetable targets, the robotic harvesting is regarded as one of the most challenging robotic technologies. Its research needs to be highly promoted by the integration of biomechanics, optimization design, advanced perception and intelligent control, and so on. The author is a young scientist who started the earlier in this research field and has an important influence. He has a deep insight into the development trend and progress in this field. He has devoted himself to the theoretical research and prototype development for over a dozen years and has achieved the fruitful achievements. In particular, the author first put forward the topics of dynamic grip collision and speedy damage-free harvesting. Around this topic and with the support of the National Natural Science Foundation of China and the other projects, the systematic research technical system, including mechanical properties and interaction rules, modeling and simulation, optimum design method, prototype development, and control optimization, was formed. This book reported the research results on this topic. The book systematically introduced the global progress of the research on robotic harvesting and discovered the intrinsic link between the development characteristics of robotic harvesting and the respective social/economic conditions and agricultural business patterns in different countries (regions) around the globe. It also introduced the related achievements and findings of the robotic speedy damage-free harvesting, which reflected the multi-disciplinary fusion characteristics of the research. The research of gripping collision of a viscoelastic object, laser cutting of plant material, plant-fruit response to vacuum sucking and pulling, and performance probability distribution based on individual differences are all very distinctive and valuable. The book has a wide content and outstanding innovation, reflecting the latest research progress in the field of intelligent agricultural equipment in China. It also attaches great importance to the integration of readers’ universality and academic depth. It is worthy of reading v

vi

Foreword

by scientists, engineers, and students who are engaged in research and development on robotics. The author has visited the Robotics and Automation Laboratory of Michigan State University. As the chief of the lab when he visited, I have many years of academic exchanges and technical cooperation with him. Thanks for the invitation of the author during the visit to Jiangsu University, I was very pleased to preface the book and recommend it to readers. Ning Xi Visiting Professor of the University of Hong Kong, President of the IEEE Robotics and Automation Society (2018–2019)

Preface

I was born in the countryside in North China. When I was a little child, I have become an important labor around my parents. Bending and waving a sickle to harvest wheat earlier from 3 am or waving a hoe to weed in the hot summer midday, callus and sweat have accompanied my entire childhood. At the age of 18, the old-fashioned green train brought me to the regions south of the Yangtze River to study engineering. It seemed to be fated that I would combine engineering with agriculture into a lifelong career. The decade I have been committed to the research of agricultural equipment is just a period of fast loss, aging of rural labor force and a period of rapid agricultural mechanization in China. Bending to wave a sickle has long been a memory of history, and mechanizations are promoting the flourishing of Chinese agriculture. However, in sharp contrast to the rapid mechanization of land-intensive grain and oil production, labor-intensive fruit and vegetable production are facing the dilemma of lack of machinery. Traditional combine harvesters are far from sending fresh fruit and vegetable products into consumers’ hands, and there is no doubt that more intelligent equipment is the only choice. Tomatoes are not only the favorite of Chinese consumers, but also the most demanding vegetables in the global. However, the unmanned harvesting of fresh tomato fruit is one of the most significant challenges. Associated with Prof. Peter P. Ling of Ohio State University and my junior fellow apprentice Jizhang Wang, I have gone to a 10-hectare large tomato production greenhouse in the United States. Under the highly mechanized operation, it was found still necessary to use dozens of labors to complete the annual harvesting task. Targeting the actual need, the research on tomato harvesting robot started in Japan and the United States, and China also has great development in this field. Although the prospect of unmanned harvesting is so beautiful, there is still a long way! Since the beginning of 2006, Li Zhiguo and I have started the development of the first prototype, and systematically carried out the theoretical and engineering research on robotic harvesting of tomato. With the cooperation and participation of several graduate students, the complete technology chain has been established, which consists of the macroscopic–microscopic structure and viscoelastic properties of fruit, robot–fruit interaction mechanisms, mathematical modeling and virtual vii

viii

Preface

simulation of grip-deformation damage, prototype development, fruit identification and location, design method and control optimization, etc. In particular, we found the outstanding contradiction between the requirements of “damage-free” and “highefficiency” in robot harvesting operations. Around “speedy damage-free harvesting”, we have put forward and carried out the researches on the grip collision of viscoelastic targets, the laser cutting of stem, the plant-fruit response to vacuum sucked pulling, the performance probability distribution based on individual differences of tomato, and so on. These researches received attention both at home and abroad. An old Chinese saying goes “Taking ten years to sharpen a sword”. After 10 years of hard work, and with the support of our supervisor Prof. Pingping Li, the research on the robotic harvesting of tomato has achieved little success. I received a Ph.D. degree, and then became a professor. After I got Ph.D. degree, Zhiguo went to the United Kingdom to continue his research with the support of the Mary Curie Scholarship. Now, he also grew into an outstanding young scientist. This book is not only the integration of research work around speedy damage-free harvesting of tomato but also the record of our days of youth. The book reports on the research of me, Zhiguo Li, and students including Fengyun Wang, Xinxin Bai, Xiuqiong Xu, Jun Ni, Qi Ni, and Yang Hu. It introduced the global progress of the research on robotic harvesting and prospected its future development. We do our best to reflect the multi-disciplinary high-level fusion characteristics of research on robotic harvesting and to construct a clear and rigorous logic system. We also try to balance the breadth of readers with the depth of academics, both for peer reviewers and robotics enthusiasts. I would like to thank the National Natural Science Foundation of China for its continued support. I am grateful to the School of Agricultural Engineering of Jiangsu University, the Key Laboratory of Modern Agricultural Equipment and Technology of the Ministry of Education, and the support of the Priority Academic Program Development of Jiangsu Higher Education Institutions (PAPD). I also thank for the kindly help of Prof. Hanping Mao, Prof. Jianjun Yin, and Prof. Xinzhong Wang. Finally, I need to thank my two Ph.D. students, Muhammad Faheem and Yu Peng, for their contribution to the proofreading of this book. The road ahead is long, and we need to seize every minute. Adhering to the ambition, we will dedicate ourselves to work hard with our colleagues for a beautiful prospect of robotic harvesting! Zhenjiang, China

Jizhan Liu

Contents

1 History and Present Situations of Robotic Harvesting Technology: A Review . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.1 An Industry of Fresh-Eat Fruits and Vegetables and Its Labor-Cost Harvesting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.2 The History and Current Situation of the Development of Robotic Harvesting Equipment in Global . . . . . . . . . . . . . . . . . . . . 1.2.1 Tomato Harvesting Robots . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.2.2 Fruit Harvesting Robot for Orchards . . . . . . . . . . . . . . . . . . . . 1.2.3 Harvesting Robots for Fruits and Vegetables . . . . . . . . . . . . . 1.2.4 Other Fruit Harvesting Robots . . . . . . . . . . . . . . . . . . . . . . . . . 1.2.5 Other Harvesting Robots . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.3 Summary and Prospect . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.3.1 The Continuous Progress of Robotic Harvesting Technology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.3.2 Technical Keys to the Development of Harvesting Robot Technology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.3.3 The Historical Characteristics of the Technology Development of the Harvesting Robots . . . . . . . . . . . . . . . . . . 1.3.4 The Breakthrough Points of the Technology Development of Harvesting Robots . . . . . . . . . . . . . . . . . . . . . 1.3.5 Key Fields of Technology Development of Harvesting Robots . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 Damage and Damage-Free Harvesting in Robotic Operation . . . . . . . 2.1 Cause of Fruit Damage in Robot Harvesting . . . . . . . . . . . . . . . . . . . . 2.2 Passive Compliant Structure in Robotic Harvesting . . . . . . . . . . . . . . 2.2.1 Elastic Surface Material . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2.2 Under-Actuated End-Effectors . . . . . . . . . . . . . . . . . . . . . . . . . 2.2.3 Elastic-Medium Fingers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.3 Active Compliance Control in Robotic Harvesting . . . . . . . . . . . . . . . 2.4 The Robotic Speedy Damage-Free Harvesting . . . . . . . . . . . . . . . . . .

1 1 2 2 15 38 65 74 88 88 89 90 93 95 95 107 107 108 108 110 112 114 118 ix

x

Contents

2.4.1 The Significance and Particularity of Robotic Speedy Damage-Free Harvesting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.4.2 The Particularity of the Collision in Robotic Speedy Gripping of Fruit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.4.3 The Research System of Speedy Damage-Free Harvesting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 The Physical and Mechanical Properties of Tomato Fruit and Stem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.1 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.1.1 Research Significance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.1.2 Content and Innovation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2 The Physical/Mechanical Properties Index System of Tomato Fruit-Stem Related to Robot’s Harvesting . . . . . . . . . . . . . . . . . . . . . . 3.3 Physical Properties of Tomato Fruit and Stem . . . . . . . . . . . . . . . . . . 3.3.1 Structure of Tomato Fruit and Stem . . . . . . . . . . . . . . . . . . . . . 3.3.2 Physical Property of Tomato Fruit and Stem [3–5] . . . . . . . . 3.4 Mechanical Properties of Tomato Fruit Components [3] . . . . . . . . . . 3.4.1 Material, Equipment, and Method . . . . . . . . . . . . . . . . . . . . . . 3.4.2 Results and Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.5 Compressive Mechanical Properties of the Whole Tomato . . . . . . . . 3.5.1 The Compression Force–Deformation Properties [4, 5] . . . . 3.5.2 Creep Properties [33] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.5.3 Stress Relaxation Properties [33] . . . . . . . . . . . . . . . . . . . . . . . 3.5.4 Load–Unload Properties [33] . . . . . . . . . . . . . . . . . . . . . . . . . . 3.6 Frictional Mechanical Properties of Tomato Fruits [3] . . . . . . . . . . . 3.6.1 Static and Sliding Friction Coefficients . . . . . . . . . . . . . . . . . . 3.6.2 Measurement of Rolling Resistance Coefficient . . . . . . . . . . 3.7 Mechanical Structure Model of the Whole Tomato Fruit . . . . . . . . . . 3.7.1 The Wheel-like Simplification Mechanical Structure of Fruit [4, 46] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.7.2 Mechanical Properties of Tomatoes with Different Numbers of Locules [3] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.8 Mechanical Damage in Tomato Fruits [3] . . . . . . . . . . . . . . . . . . . . . . 3.8.1 Mechanical Damage Mechanism of Tomato Fruit . . . . . . . . . 3.8.2 Physiological Change of Tomatoes After Being Compress . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.9 The Properties of Tomato Stem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.9.1 Stem System [5, 81] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.9.2 Mechanical Properties of Tomato Fruit System [4, 5] . . . . . . 3.9.3 Results [4, 5] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

118 120 121 123 127 127 127 127 128 129 129 131 134 134 143 148 148 153 155 157 160 160 163 164 164 166 176 176 176 184 184 186 190 192

Contents

4 Development of Damage-Free Hand–Arm System for Tomato Harvesting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.1 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.1.1 Research Significance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.1.2 Content and Innovation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.2 Development of Damage-Free Harvesting End-Effector . . . . . . . . . . 4.2.1 System Scheme Design of Damage-Free Harvesting End-Effector . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.3 Motion Configuration Scheme . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.4 System Components of the End-Effector . . . . . . . . . . . . . . . . . . . . . . . 4.4.1 Mechanism Design of End-Effector [81] . . . . . . . . . . . . . . . . 4.4.2 Design of the Sensing System [81] . . . . . . . . . . . . . . . . . . . . . 4.4.3 Design of Control System [81] . . . . . . . . . . . . . . . . . . . . . . . . . 4.4.4 Design of Power Supply System [81] . . . . . . . . . . . . . . . . . . . 4.4.5 Structure Design of the End-Effector [81] . . . . . . . . . . . . . . . 4.4.6 Prototype and Its Performance Indicators [81] . . . . . . . . . . . . 4.4.7 Upper Lower Type End-Effector . . . . . . . . . . . . . . . . . . . . . . . 4.4.8 Passive–active Coupled Compliant End-Effector for Robot Tomato Harvesting [95] . . . . . . . . . . . . . . . . . . . . . . 4.5 Damage-Free Harvesting Hand–arm System Based on Commercial Manipulator [96] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.5.1 Background and Needs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.5.2 The Control System Structure of Commercial Manipulator [31] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.5.3 Control System Integration Between the Manipulator and the End-Effector [31, 34] . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 Mathematical Modeling of Speedy Damage-Free Gripping of Fruit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.1 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.1.1 Research Significance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.1.2 Content and Innovation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.2 Experiment of Speedy Fruit Gripping and Special Collision Characteristics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.2.1 Experiment of Speedy Fruit Gripping [1, 2] . . . . . . . . . . . . . . 5.2.2 Collision Characteristics of Speedy Fruit Gripping . . . . . . . . 5.3 The Special Collision Issue of Speedy Fruit Gripping . . . . . . . . . . . . 5.4 Dynamic Characteristics in Different Phases of Speedy Fruit Gripping [1] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.5 Fruit Compression Model [1, 3] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.5.1 The Viscoelastic Properties of Fruit and the Characterization of Constitutive Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

xi

197 197 197 197 198 198 199 213 214 223 225 228 230 231 233 233 236 236 237 239 241 247 247 247 247 248 248 248 250 250 252

252

xii

Contents

5.5.2 Burger’s Modified Model for Characterization of Creep Properties of Whole Fruit . . . . . . . . . . . . . . . . . . . . . 5.6 Complex Collision Model in Speedy Gripping of Fruit [1] . . . . . . . . 5.6.1 Phase of Constant-Speed Loading and Phase of Stress Relaxing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.6.2 Phase of Collision Decelerating . . . . . . . . . . . . . . . . . . . . . . . . 5.7 The Basic Law of Collision in Robotic Gripping of Fruit [1] . . . . . . 5.7.1 The Law of Collision Force in Robotic Gripping of Fruit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.7.2 The Influence of Initial Gripping Speed and Fruit Ripeness on Gripping Collision Time . . . . . . . . . . . . . . . . . . . 5.7.3 The Influence of Initial Gripping Speed and Fruit Ripeness on Gripping Collision Deformation . . . . . . . . . . . . . 5.7.4 The Influence of Initial Gripping Speed and Fruit Ripeness on Peak Collision Force . . . . . . . . . . . . . . . . . . . . . . 5.8 The Theoretical Calculation of the Time Consumption of Gripping [2] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.8.1 The Stroke Composition of the Finger Gripping Process . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.8.2 Dimension Relation of Fruit Gripping with Robotic Fingers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.8.3 The Time Consumption Composition of the Finger Gripping Process . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.8.4 Selection of Damage-Free Control Mode . . . . . . . . . . . . . . . . 5.8.5 Time Calculation of Damage-Free Gripping . . . . . . . . . . . . . 5.9 Collision Stage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6 Simulation of Damage-Free Robotic Gripping of Fruit . . . . . . . . . . . . . 6.1 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.1.1 Research Significance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.1.2 Content and Innovation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.2 Finite Element Model of Fruit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.2.1 Viscoelastic Finite Element Model of the Whole Tomato Fruit [1] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.2.2 Nonlinear Multi-component Finite Element Model of Tomato Fruit [3] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.3 Simulation of Static Gripping Process [3] . . . . . . . . . . . . . . . . . . . . . . 6.3.1 Geometry Model Finger-Fruit Contacting Process . . . . . . . . 6.3.2 Creating Contact Pair . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.3.3 Model Verification Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.3.4 Prediction Method of Gripping Damage . . . . . . . . . . . . . . . . . 6.3.5 The Component Stress Simulation of Different Loading Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.4 Dynamic Simulation of Gripping Process [1] . . . . . . . . . . . . . . . . . . .

256 263 263 264 265 265 266 267 268 270 270 271 272 272 273 274 274 277 277 277 277 278 278 288 289 289 290 291 292 298 314

Contents

6.4.1 The Software Implementation of Dynamic Gripping Simulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.4.2 The Establishment of System Virtual Prototype for Gripping . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.4.3 Simulation Analysis of Tomato Fruit Gripping with the End-Effector . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7 Modeling of the Vacuum Sucked Pulling of Tomato Fruit . . . . . . . . . . 7.1 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.1.1 Function of Vacuum Sucked Pulling in Robotic Harvesting [1] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.1.2 Research Significance [1, 18] . . . . . . . . . . . . . . . . . . . . . . . . . . 7.1.3 Content and Innovation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.2 Modeling of Mechanical Behavior for Sucking with Suction Pad . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.2.1 Mechanical Relation Between Suction Pad and Spherical Surface [1] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.2.2 Experiment on Influence Factors of Suction Force . . . . . . . . 7.2.3 The Effect of Fruit Surface Contour on Pull-off Force . . . . . 7.3 Mechanical Model of Vacuum Sucked Pulling . . . . . . . . . . . . . . . . . . 7.3.1 Kinematic and Force Balance Analyses of Pulling of On-plant Fruit with Suction Pad . . . . . . . . . . . . . . . . . . . . . 7.3.2 Static Analysis of Pulling of On-plant Fruit with Suction Pad . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.3.3 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.4 Probability Model of Sucked Pulling of On-plant Tomato Fruit [1] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.4.1 Rate of Interference and Success of Fruit Gripping . . . . . . . . 7.4.2 The Proportion of Fruit Number Per Cluster for Different Harvesting Rounds . . . . . . . . . . . . . . . . . . . . . . . 7.4.3 The Required Sucked Pulling Distance and Its Probability for Different Fruit Number in Each Cluster . . . . 7.4.4 Theoretical Influence of Required Sucked Pulling Distance on the Rate of Gripping Interference . . . . . . . . . . . . 7.4.5 Determination of Sucked Pulling Distance . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8 Fruit Detaching Methods for Robotic Damage-Free Tomato Harvesting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8.1 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8.1.1 Research Significance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8.1.2 Content and Innovation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8.2 Theoretical and Experimental Comparison of Non-tool Fruit Detaching Methods [1, 2] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8.2.1 Non-tool Fruit Detaching Methods . . . . . . . . . . . . . . . . . . . . .

xiii

314 315 318 322 323 323 323 324 325 326 326 329 332 334 334 335 338 345 345 346 349 360 361 363 365 365 365 365 366 366

xiv

Contents

8.2.2 Experiments of Non-tool Detaching of Tomato Fruit . . . . . . 8.2.3 Theory of Strength and Detachment of Abscission Layers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8.2.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8.3 Experimental Exploration of Laser Cutting of Stems [36, 37] . . . . . 8.3.1 Put Forward Laser Cutting of Stems . . . . . . . . . . . . . . . . . . . . 8.3.2 The Principle and Advantages of Laser Cutting of Biomaterials [36, 43] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8.3.3 Particularity and Feasibility of Laser Cutting of Stem [36, 43] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8.3.4 Experiments on Laser Drilling and Cutting of Tomato Stems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8.3.5 Results and Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8.3.6 Realization of Laser Cutting of Peduncles [43] . . . . . . . . . . . 8.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9 Control Optimization and Test Study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9.1 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9.1.1 Research Significance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9.1.2 Content and Innovation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9.2 Parameter Optimization of Speedy Flexible Gripping [1] . . . . . . . . . 9.2.1 PID Parameter Adjustment of the Motion Control System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9.2.2 Energy Consumption Analysis of Acceleration and Deceleration Stage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9.2.3 Speed Optimization of Speedy Flexible Gripping . . . . . . . . . 9.3 Control Optimization of Vacuum Sucked Pulling [7] . . . . . . . . . . . . . 9.3.1 The Relationship Between Maximum Pulling Speed and Displacement in Acceleration Stage . . . . . . . . . . . . . . . . . 9.3.2 The Relationship Between the Dynamic Pulling Force and the Threshold of Vacuum Degree . . . . . . . . . . . . . . . . . . . 9.3.3 Optimization of Displacement/Position Parameters for Sucked Pulling of Fruit . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9.3.4 Optimization of Control Mode for Motion Coordination . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9.4 Hand–Arm Coordination Control for Speedy Flexible Harvesting [8] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9.4.1 Hand–Arm Coordinative Control Modes . . . . . . . . . . . . . . . . 9.4.2 Hand–Arm Coordinated Harvesting Experiments . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

367 372 374 379 379 379 382 382 386 393 395 397 403 403 403 403 404 404 415 426 432 432 437 438 442 445 445 447 452

Nomenclature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 453

Chapter 1

History and Present Situations of Robotic Harvesting Technology: A Review

1.1 An Industry of Fresh-Eat Fruits and Vegetables and Its Labor-Cost Harvesting Fruits and vegetables are both daily necessities, and also they are important economic crops. According to the statistics, the global production of total fruits and vegetables in 2019 reached 8.83 × 108 t and 11.30 × 108 t, respectively. Globally, the rate between fresh-eat and processed fruits and vegetables is about 7:3. China’s vegetable and fruit planting area and output all rank the first in the world, but the proportion of processing fruits and vegetables is only about 5%. Usually, it is not necessary for the harvesting of processing fruits and vegetables to distinguish the ripeness, and also a certain damage is tolerant in the harvest. For example, tomato fruit can be whole-plant harvested and apple fruit can be harvested mechanically by vibratory excitation. In developed countries, the non-selective mechanized harvesting of processed fruits and vegetables has been gradually popularized. But for the larger proportion of fresh-eat fruits and vegetables, the non-selective mechanized harvesting method cannot adapt to both the individual difference of the fruit maturity and the harsh demand of non-destructive harvest. So far, it is still dependent on human labor for selective harvesting. With the gradual mechanization of the production of fruit and vegetable cultivation, harvesting has become the last link to break through the whole process of mechanical operation. According to the investigation, the labor consumption of strawberry production in Japan reaches 20,000 h/ha [1], and the harvest takes up about 40% of the total labor amount [1, 2]. Meanwhile, the shortage of agricultural labor and the rising cost of labor have seriously affected the development of the fruit and vegetable industry. In China, in recent years, the labor force, especially the young and middle-aged labor force, has also been rapidly transferred to other industries. In the busy farming season, the labor shortage has begun to appear in the vast rural areas. The labor intensity of the elderly and women in rural areas has greatly increased, and the production efficiency has decreased obviously. © Science Press, Beijing and Springer Nature Singapore Pte Ltd. 2021 J. Liu et al., Rapid Damage-Free Robotic Harvesting of Tomatoes, Springer Tracts in Mechanical Engineering, https://doi.org/10.1007/978-981-16-1284-8_1

1

2

1 History and Present Situations of Robotic Harvesting Technology: A Review

The contradiction between the rapid development of fruit and vegetable production, the shortage of agricultural labor, and the excessive intensity of labor is becoming more and more obvious, and the replacement of complex manual selective harvesting can only be realized through the in-depth study of the technology of the harvesting robot. The research and development of fruit and vegetable harvesting robot are of great significance for reducing the labor intensity of agricultural practitioners, liberating the agricultural labor force and improving the intensive production level of fruits and vegetables.

1.2 The History and Current Situation of the Development of Robotic Harvesting Equipment in Global A typical harvesting robot for fruits or vegetables is usually composed of mobile platform, manipulator, end-effector, vision system, and control system. Since fruit and vegetable species and varieties, and cultivation patterns are all numerous and complicated, various kinds of harvesting robots and their end-effectors have been developed at home and abroad. The action principle, structure form, complexity, operation effect, and performance also have a very big difference.

1.2.1 Tomato Harvesting Robots 1.

Fresh-eat tomato and its robotic harvesting problem

As a favorite fresh-eat vegetable, its robotic harvesting has been paid much attention by researchers worldwide. Concerned research has been carried out continuously for many years, and a series of achievements have been produced. At the same time, the tomato is also one of the fruits and vegetables that are most difficult to be harvested by robots. At present, in the face of fresh food, common tomato fruits are usually picked as single fruit one by one, while cherry-tomato fruit is usually picked in clusters. Compared with cucumber, eggplant, apple, and other fruits and vegetables, there is usually 3–5 tomato fruit in one cluster. They grow densely and touch each other, and the difference of fruit-stem posture is more significant (Fig. 1.1). The great difference of growth posture and distribution poses a greater challenge to the implementation of intelligent robotic harvesting: (1)

Recognition of the target fruit

The close and occlusion of the fruit are more serious. For the vision system of the harvesting robot, although the color difference between mature tomato fruit and leaves is distinct, it is difficult to identify and locate the target fruit since multiple fruit images are integrated into one or even completely overlapped to be difficult to be segmented [3, 4].

1.2 The History and Current Situation of the Development …

(a) Tomato

(b) Cucumber

3

(c) Eggplant

Fig. 1.1 Difference of growth posture and distribution among the fruit of tomato, cucumber, and eggplant

(2)

Implementation of the picking action

Due to the in-cluster growth and mutual touch of tomato fruit, there is limited gripping space for the target fruit, which may lead to the failure of gripping or bruise of adjacent fruit. Meanwhile, the large difference of posture of the fruit stem and length of the stem will increase the difficulty of mechanical cutting and difference of detachment of bending, pulling, or twisting. As a result, the success rate is limited. 2.

Tomato harvesting robots in Japan

It is easy to understand that planting patterns have great influence on the performance of the robotic harvesting. For the traditional cup-type planting, the fruit is very scattered, and the robot needs a much larger workspace. At the same time, the spatial distribution of the branches makes the harvesting operation very difficult. In Japan, the fresh-eat tomato usually adopts single-truss cultivation mode, and the plant is cultivated vertically with the support of the pillar or the rope. Several fruits are suspended in a cluster. Due to the short petiole, the fruit recognition is greatly simplified, and the performance of the picking operation is guaranteed [5]. In the early 80s of the last century, the study of the tomato harvesting robot started in Japan. For decades, the Kyoto University, Okayama University, Shimane University, Kanagawa Institute of Technology, Osaka State University, and other universities as well as Department of Protected Cultivation, Aichi have all developed different prototypes of the tomato harvesting robot, and experts such as Kondo N. and Monta M. have led the research upsurge of technology of tomato harvesting robot. A variety of prototypes mainly use an electrically wheeled or railed vehicles for application in greenhouses, and a few of them are crawler type for outdoor cultivation [6]. For regular cultivation pattern, due to the complexity of the canopy and the randomness of the distribution of the fruit, degrees of developed joint-type

4

1 History and Present Situations of Robotic Harvesting Technology: A Review

manipulators have increased from 3 to 6–7. While for single-truss cultivation, since all tomato fruits grow inversely, the difficulty of harvesting is greatly reduced, and the Cartesian coordinate manipulator was applied by Kondo [7]. (1)

Robotic harvesting of individual common tomato fruit

Kawamura, N, et al. from Kyoto University carried out the development of tomato harvesting robot earlier [8, 9] (Fig. 1.2). The robot consisted of 0.52/0.25 m/s doublespeed electric wheeled vehicle and 5-DOF (degree of freedom) manipulator, and a fixed camera moving was used to realize the fruit location by two-position detection. Both its whole machine structure and the target detection technology are relatively simple, but it has become an important exploration for robot harvesting technology at the early stage. In view of the needs of the elderly or the disabled, Takahashi Y., et al., from Kanagawa Institute of Technology, proposed a manual-operated Cartesian-type tomato harvesting robot [10] (Fig. 1.3), which was remote operated through a human interface. The end-effector is a scissor to cut off the stem directly and cannot hold the fruit. The fruit falls down or falls into the replaced box. It has a simple structure and function, which has certain applicability to more sparse plant canopy with larger admissible motion space and fruit with a certain impact resistance. But for most fruits and vegetables, it cannot meet the requirements of robot harvesting. The complex canopy space makes the fruit vulnerable to bruise and the falling position is unpredictable, which affects the recovery of fruits. Fig. 1.2 Tomato harvesting robot developed in Kyoto University [8, 9]

1.2 The History and Current Situation of the Development …

5

Fig. 1.3 Tomato harvesting robot with a human interface developed in Kanagawa Institute of Technology [10]

Kondo N., the most famous scholar in the field of agricultural robot research in the world, developed the prototype of tomato harvesting robot. It adopted the wheeled vehicle and the 7-DOF redundant manipulator. The manipulator has five rotational joints and two straight joints to realize both up-and-down and forward-and-backward movements so that the working space and posture diversity of the manipulator can effectively meet the obstacle avoidance and arrival requirements in fruit harvesting. The two-finger and flexible four-finger end-effectors were developed, respectively [11–13] (Fig. 1.4). Meanwhile, a vacuum suction system was installed and the similar action principle was adopted, which first isolates the target fruit from the adjacent fruit in the same cluster by sucking and pulling back, and then holds the fruit to detach it by twisting or bending.

Fig. 1.4 Tomato harvesting robots developed by Kondo et al. [13]

6

1 History and Present Situations of Robotic Harvesting Technology: A Review

Kondo N. and Monta M., et al. put forward different robot structures for the inverted single-truss tomato cultivation mode [7] (Fig. 1.5). In this mode, the roots of tomato are in the upper part of the hydroponic culture tank, while the fruit clusters hang down. The robot can find and reach the fruit cluster more easily after pruning. Since the hydroponic trough can move, the mobile vehicle is not needed for the robot. At the same time, the drooping growth of the fruit cluster makes complex manipulator unnecessary. Therefore, a rectangular coordinate manipulator was adopted, and the same end-effector as the single-truss tomato cultivation mode was adopted. The test results showed that the total success rate of the end-effector was 78%. Ripe fruit was harvested successfully with no damage, but some mostly soft, over-mature fruit could not be harvested because the fingers slipped over the very soft fruit and failed to create a bending moment on the peduncles [7]. Second, even if the end-effector grasped over-mature fruits, it was difficult to detach them because their peduncles became woody [7]. Finally, the fingers could not bend off peduncles of immature fruits because their abscission layers were not grown enough [7]. The tomato harvesting robot developed by Hayashi S., et al. in Department of Protected Cultivation, Aichi, uses the binocular vision to realize fruit identification and location and uses a 2900 mm*1400 mm crawler vehicle and a Mitsubishi 5-DOF multi-joint manipulator [6] (Fig. 1.6). Its end-effector sucks the target fruit with a vacuum suction pad and pulls it back for certain distance. The two fingers driven by the DC motor grip the fruit vertically and cut it off by rotation. During the operation of the end-effector, the vacuum negative pressure is detected by the pressure switch to determine whether the sucker is sucked successfully or not. The vacuum sucker sucked and pulled the fruit back to 30 mm. Test results showed that when the stem was too short, the pulling force of the suction pad would exceed the vacuum suction force, resulting in the failure. The average harvesting period was 41 s, and the recognition

Fig. 1.5 Tomato harvesting robot for the inverted single-truss tomato cultivation mode [7]

1.2 The History and Current Situation of the Development …

7

Fig. 1.6 Tomato harvesting robot developed in Department of Protected Cultivation, Aichi [6]

time of mature fruit and success rate were 7 s and 92.5%, respectively. After the successful recognition, 83.8% could be picked up, but one-third was damaged. As a result, the total success rate was only 56.8%. A prototype of a simple track-type tomato harvesting robot was developed by Yasukawa S. et al. in Kyushu University of Technology [14] (Fig. 1.7). It includes a commercial 6-DOF series manipulator and an end-effector, and the recognition of target fruit was realized by the fusion of color and infrared information of the Kinect V2 somatosensory camera. The prototype still needs indoor and field test. Tomato harvesting robot developed by Yaguchi H., et al. in the University of Tokyo is composed of an omnidirectional vehicle, a Universal UR5 manipulator, a PS4

Fig. 1.7 Tomato harvesting robot developed in Kyushu University of Technology [14]

8

1 History and Present Situations of Robotic Harvesting Technology: A Review

binocular stereo camera of Sony, and a rotational plucking gripper [15] (Fig. 1.8a). The robot could work in direct sunlight. The robot motion was improved and the harvesting speed was up from 85 to 23 s/fruit. However, the gripper may grasp multiple fruits in case of very cluttered cluster and the calyx also may be broken when the stem angle is deep from the rotation axis [15]. This research team also developed a humanoid dual-arm tomato harvesting robot [16] (Fig. 1.8b). The robot is equipped with an omni-directional vehicle, which installed an Xtion somatosensory camera on the head and a Carmine somatosensory camera on the wrist. Each manipulator had 7-DOF and a grip-and-cut integrated endeffector. The robot has completed the indoor harvesting test of the tomato clusters. But for now in the experiment, humans need to send motion orders to the robot because the system is not completely autonomous yet, and it requires humans to

(a) Wheeled

(b) Humanoid dual-arm Fig. 1.8 Tomato harvesting robot developed in the University of Tokyo [15, 16]

1.2 The History and Current Situation of the Development …

9

judge whether one harvesting behavior is successful or not since the system has not equipped the corresponding functional modules [16]. It has proved the feasibility of fruit harvesting with the humanoid robot, but both the identification and the operation still need to be improved. (2)

Robotic harvesting of individual cherry-tomato fruit

The cherry-tomato harvesting robot developed by Kondo N., et al. also used an electric four-wheel vehicle and a 7-DOF redundant manipulator that is the same as that used for common tomato fruit harvesting [17–19]. However, a new type end-effector for cherry tomato was developed, which pulled a fruit into its tube opening pneumatically by suction and nipped its peduncle near the joint by nipper closing actuated by two springs and a solenoid [17–19]. The detached fruit would be transported to a holding container through the tube by suction [17–19] (Fig. 1.9). As the fruit was transported and dropped into the container by the tube, the end-effector was usually used only for the smaller fruit of cherry tomatoes, strawberries, and so on, and the hose must be carefully designed to avoid the damage to the fruit [18]. A color camera mounted on the vehicle was moved horizontally and vertically for acquiring the two images needed for stereo vision, so as to achieve the target fruit location. It was found that the success rate of harvesting was 70%, and harvest failures were mostly caused by fruit disturbance by the end-effector while harvesting a target fruit or a neighboring fruit. Moreover, some fruits with short and hard peduncles were still difficult to harvest. At the same time, the robot has a good harvest for single-truss fruit cluster (one long stem in truss), and for multiple-truss (more than two long stems in truss), the first-attempt success rate was merely 23% due to the error of stem positioning [17]. Tanigaki K. of Osaka Prefecture University thought that the training mode of the plants has great influence on the performance of robotic harvesting. If plants are

Fig. 1.9 Cherry-tomato harvesting robot developed by Kondo et al. [17]

10

1 History and Present Situations of Robotic Harvesting Technology: A Review

trained using the method of vase form training in traditional training mode, fruits will be distributed widely in the large cherry crown, which makes the harvesting work difficult for the robot [20]. Therefore, a harvesting robot was designed for application on single trunk training. By this cultivation, cherry trees are trained vertically with their branches pruned and fruits are located around the trunk [20]. Since the leaf stalk is pruned short, fruits can be easily recognized by machine vision scanning around the trunk [20]. To harvest the fruit with its peduncle, a special end-effector was designed which consisted of a fruit sucking device, an open-close mechanism, a back-and-forth mechanism, and a pair of fingers [20]. After the fingers grasp the peduncle, the end-effector is lifted up to remove the peduncle from the tree [20]. The results of the observation showed that the time required to harvest each fruit was about 14 s. But for fruits close to each other, since the adjacent fruit may enter between the fingers, the peduncle of the target fruit was not held well, resulting in incorrect harvesting. Meanwhile, if the strength of the peduncle at its root was large compared with other fruits, the peduncle may break at its midpoint when the end-effector was raised after having gripped the peduncle [20] (Fig. 1.10).

Fig. 1.10 Cherry-tomato harvesting robot developed in Osaka Prefecture University [20]

1.2 The History and Current Situation of the Development …

11

Fig. 1.11 Cherry-tomato harvesting robot developed by Fujiura et al. [22]

Fujiura T., et al. of Shimane University and Osaka Prefecture University developed a prototype of cherry-tomato harvesting robot [21–25] (Fig. 1.11). The electric fourwheel drive vehicle was used with the 4-DOF Cartesian-type manipulator. The nearinfrared stereo vision sensor was installed in the front of the end-effector. In the early development, the fruit was sucked into the tube, and then it was changed to suck and swing to cut it off and return it into the fruit box through an open bag. The experiment in a greenhouse of Southern Osaka showed that the success rate of 135 fruits was 85%, and the intact calyx rate was 92%, and the total harvest time of 22 fruit was 252 s. In the experiment carried out in its cultivation facilities of Osaka Prefecture University, of total 129 fruits, 81% were successfully harvested while 98% of the harvested fruits had an intact calyx. (3)

Robotic cluster harvesting of cherry-tomato fruit

Kondo N. thought that a key reason why present tomato harvesting robot prototype has not been commercialized yet was that robot’s operation speed was same or below than human operator’s speed [26, 27]. As cluster harvesting was increasingly common in European countries, USA, and Japan, an end-effector that can harvest a whole tomato cluster with 4–6 fruits was manufactured and tested in a greenhouse using selective compliance assembly robot arm (SCARA) [26–28] (Fig. 1.12). The robot arm approaches a fruit cluster, and both upper and lower fingers close and surround the main stem when the limit switch detects the main stem. And then, the upper finger grasps and cuts the peduncle. Finally, a pushing device moves forward to touch the fruit cluster to keep it from swinging and then transports the fruit cluster to a location above a tray [26–28]. Experimental results indicated that the success rate of harvesting tomato clusters was only 50% since for the high-density plants, many fruit clusters, peduncles, and main stems are hidden by leaves and node lengths were too short for the end-effector to extend into the plant and grasp the main stem near the tomato cluster. The remaining failure rate was due to insufficient air pressure to grasp the peduncles repeatedly and the loss of a fruit from the harvested cluster [27]. The cluster harvesting cannot adapt to the difference of fruit mature period in

12

1 History and Present Situations of Robotic Harvesting Technology: A Review

Fig. 1.12 Cherry-tomato harvesting robot developed by Kondo [26–28]

the same cluster, and its applicability relies on the development of new variety and new cultivation mode of cherry tomatoes and specific market demand. 3.

Tomato harvesting robots in other countries

The tomato harvesting robot technology has also been launched in the US, Chinese Mainland, and Chinese Taiwan. The prototype developed by Ling P., et al. of Ohio State University adopted a flexible under-actuated four-finger end-effector [29] (Fig. 1.13) with a hydraulic vehicle and a commercial Moto man 6-DOF manipulator. The target fruit was pulled away from the fruit cluster with the suction of a vacuum suction, and then it was detached by pulling with enveloped fingers. But no further development and test reports were reported.

Fig. 1.13 Tomato harvesting robot developed at Ohio State University [29]

1.2 The History and Current Situation of the Development …

13

In the development of tomato harvesting robot, Chiu Y., et al. of National Ilan University, Taiwan, China, combined the Mitsubishi RV-M1 5-DOF joint-type manipulator with a three-axis carrier [30–33] (Fig. 1.13). The end-effector was similar under-actuated four-finger structure with a suction pad. In fruit harvesting, four solenoids pulled the wires causing the four fingers to bend and clamp simultaneously [30]. A CCD camera is constructed in a mobile platform for the establishment of binocular vision. The dimension of the prototype was 1650 × 700 × 1350 mm, and the total weight was 219 kgf. Test results showed that the successful harvesting rate was 73.3% for pot-cultivated tomatoes. The average harvesting time needs about 74.6 s/sample. It was found that two of the failures occurred during the suction process since the suction pad was positioned near the edge of the fruit, and the angles for six of the fruits were not aligned correctly [30]. Additionally, peduncles could be twisted effectively, which prevented the fruits from being detached from the stalk [30] (Fig. 1.14). The research of tomato harvesting robot in China started later, but the present investment of sci-tech funds and researchers and achievements has been at the forefront of the world. The prototype developed by Ji C. and Li W. of China Agricultural University was based on the commercial crawler flat bottom [34, 35] (Fig. 1.15) combined with a 4-DOF joint-type manipulator and a grip-and-cut integrated twofinger pneumatic end-effector, and a binocular vision system was fixed on the vehicle. Experiment results showed that the average harvesting time need about 28 s/sample, and the success rate of harvesting was 86%. In the experiment, the shadow, bright spot, and occlusion affected the recognition result, and the manipulator might scrape the stem and leaf and cause the fruit offset in dense canopy. The end-effector might not grip the stem and thick stem could not be cut off, and also the falling of fruits during pulling might happen. Feng Q. of National Research Center of Intelligent Equipment for Agriculture, Wang X. of Hebei University of Technology, et al. developed a harvesting robot

Fig. 1.14 Tomato harvesting robot developed in National Ilan University, Taiwan., China [30–33]

14

1 History and Present Situations of Robotic Harvesting Technology: A Review

Fig. 1.15 Tomato harvesting robot developed in China Agricultural University [34, 35]

prototype for high-wire cultivation tomatoes [36, 37] (Fig. 1.16), and it used wheeltype lifting platform and 4-DOF joint-type manipulator. A sleeve-shaped end-effector was designed which consisted of a sleeve, six air floats, and a rotation mechanism. The telescopic cylinder makes the sleeve extend to insert the target fruit into the sleeve, and the air floats are inflated and expanded rapidly to hold the fruit tightly but softly. Finally, the sleeve with the fruit is driven to rotate back and forth to detach the fruit. Meanwhile, a visual unit that was composed of the CCD camera, the linearstructured laser generator, and the sliding driver were used to identify and locate the target fruit. It was found in experiments that the execution time of a single harvest cycle was about 24 s, and the success rates for harvesting tomatoes in bright light and weak light conditions were 83.9% and 79.4%, respectively.

Fig. 1.16 Tomato harvesting robot developed in National Research Center of Intelligent Equipment for Agriculture [36, 37]

1.2 The History and Current Situation of the Development …

15

Fig. 1.17 Tomato harvesting Robot of Shanghai Jiaotong University [38]

Zhao Y., et al. from Shanghai Jiao Tong University has developed a dual-arm tomato harvesting robot to improve the operation efficiency in an unstructured environment [38] (Fig. 1.17). The platform vehicle designed as a carrier was driven on the heating pipes. Two 3-DOF Cartesian-type manipulators were adopted and two different kinds of end-effectors such as saw cutting type and pneumatic-type gripper were installed, respectively. A color stereo camera was used as the vision system. Due to the complexity of the working environment, an artificial recognition approach conducted by the operator through marking the tomato object on the graphic user interface was used for tomato recognition and localization [38]. The varying illumination is the main factor to influence the performance of image match. On the other hand, the time consumption of 3D image reconstructing is high which directly affected the picking cycle time. In addition, Jiangsu University, Zhejiang University, Northeast Agricultural University, China Jiliang University, and other organizations have also carried out a great deal of research on the identification of tomato fruit, the design and analysis of manipulators, and even the design of night lighting system matching with the machine vision system [39–41]. Timeline of the development of tomato harvesting robots is shown in Table 1.1.

1.2.2 Fruit Harvesting Robot for Orchards 1.

Citrus harvesting robots

Citrus is a widely welcomed fruit. In addition to the mechanized harvesting for the production of fruit juice, the robotic selective harvesting for fresh-eat citruses has also been paid attention too. As the main citrus-producing country, the United States and Italy have carried out a lot of research on the robotic harvest of citrus. Relative research has also been carried out in Japan, Britain, China’s Mainland, and Taiwan.

16

1 History and Present Situations of Robotic Harvesting Technology: A Review

Table 1.1 Timeline of the development of tomato harvesting robots Country Organization

Crop

Prototype

Vehicle

(Region)

Common tomato

Manipulator

End effector

DOF

DOF

Japan

Kyoto University

Wheeled

3

2

Japan

Okayama University

Wheeled

7

2

Japan

Okayama University

Wheeled

7 (Commercial)

1

Crawler-type

5 (Commercial)

1

-

5 (Cartesian)

-

3 (Cartesian)

-

5

Method of recognition & localization

Research phase

CCD camera

Indoor

moving

experiment

-

Field test

Common Tomato

Cherry tomato

Common tomato

CCD camera moving

Experiment

Department of Japan

Protected

Binocular vision

Experiment

-

Experiment

1

CCD camera

-

-

3D vison sensor

2

Lipstick Camera

Cultivation, Aichi

Common Japan

Okayama University

1(Flexible

Tomato

Common Japan Tomato

Common Japan Tomato

Common USA Tomato

Kanagawa Institute of Technology

Osaka Prefecture University

Ohio State University

-

6

Commercial)

Partial prototype

Prototype

Common Japan

Okayama University

Wheeled

7

2

Field test

Tomato

(continued)

1.2 The History and Current Situation of the Development …

17

Table 1.1 (continued)

Common China Tomato

Cherry tomato

Japan

China Agricultural University

Osaka Prefecture University

-

-

1

4

3

3D vison sensor

Experiment

Binocular vision Tomato cluster

Japan

Kyoto University

-

4 (SCARA)

3

+4 lighting

Experiment

devices +polarizing filters

Shimane University, Cherry tomato

Japan

Osaka Prefecture

Wheeled

4 (Cartesian)

1

University

Common Taiwan, China Tomato

Common and cherry tomato

Japan

National Ilan University

Rail-type

China Agricultural

Magnetic

University

guidance

8 (Joint arm 5+ carrier 3)

4

Infrared stereo vision

CCD camera up-and -down

Binocular vision

Field test

Experiment

4

1

Experiment

Rail-type

4

3

Rail-type

3 (Dual-arm)

2

Binocular vision

Field test

Omni directional

6

2

PS4 camera

Experiment

National Research Common tomato

China

Center of Intelligent Equipment for

CCD camera+ Laser range finder

Field test

Agriculture

Common tomato

Common tomato

China

Japan

Shanghai Jiaotong University

University of Tokyo

(continued)

18

1 History and Present Situations of Robotic Harvesting Technology: A Review

Table 1.1 (continued)

Common tomato

Common tomato

(1)

RGB-D Japan

University of Tokyo

Omni directional

7 (Dual-arm)

1

cameras(Xtion & Carmine)

Japan

Kyushu University of Technology

Indoor experiment

RGB-D camera Rail-type

Prototype

6 (Kinect)

Citrus harvesting robots developed in Italy

Italy is one of the major citrus-producing countries in the world and in Europe, and its proportion of fresh citrus is much higher than that of processed citrus. Research on the robot citrus harvesting technology has been carried out earlier and more in Italy. Blandini G. of the University of Catania and Levi P. of Robotic and Division A found the mechanical harvesting with air blowers, or mechanical shakers can only be used for processing citrus in which it is difficult to detach the citrus fruit, in order to avoid damage to fruits and trees [42]. The citrus harvesting robot they developed was composed of wheeled vehicle, 3-DOF Cartesian-type manipulator, and rotary cutting end-effector [42, 43] (Fig. 1.18). The average harvest time was 10 s, 70% were identified and only 65% of them were successfully harvested. The citrus harvesting robot developed by Muscato G. and Fortuna L. of University of Catania [43–45] (Fig. 1.19) uses a unique structure to install a large main hydraulic arm on the tank-like crawler vehicle, and a 5-DOF hydraulically driven base that supports two 3-DOF electrically driven telescoping arms [45]. The two electrical arms carry a small high-resolution wide-angle color camera in the center of the gripper to realize the target identifying and 3D localization [45]. After the main arm

Fig. 1.18 Citrus harvesting robots developed in the University of Catania [42, 43]

1.2 The History and Current Situation of the Development …

19

Fig. 1.19 The citrus harvesting robot with master–slave manipulators developed in University of Catania [43, 45]

and platform have been positioned near the canopy of the tree using manual controls, a fully automatic harvesting system is invoked [44]. When the robot is tested in real conditions, using its own real-time hardware, it lacks the external computing power that allows detailed monitoring and debugging of the intermediate processes [45]. Then, the tests conducted on this prototype by Recc M. of the University College London and Plebe A. of the University of Catania in Italy found that the difficult cases are ones in which the oranges are partly occluded or in which the segmentation algorithm finds a large leaf with sufficient orange color to be considered an orange. In each image, 86% of the fruit was located, but 16% of the fruits were positioned many times, and further processing was needed to identify the correct center. The two environmental parameters which mostly affect the harvesting are the lighting conditions and the wind speed [44]. The average harvesting time per fruit is between 5–10 s and [44]. To simplify the structure in the development of the new citrus harvesting robot, Muscato G. of the University of Catania adopts a Cartesian type. To enhance the performance, the robot is equipped with two arms [43]. These are identical, but the lower one does not have a pneumatic extension, because the low-growing fruit is always on the outside branches of the tree [43]. The height of the picking system is more than 3.5 m and the weight is over 2 tons, and the robot has to be housed on a caterpillar vehicle [43] (Fig. 1.20). In addition, different Italy organizations have done a lot of work on the development of various principles of Citrus harvesting end-effectors. The Allotta B. of the ARTS laboratory has developed a pneumatic flexural three-finger end-effector [43, 46] (Fig. 1.21). Once the orange had been grasped, the arm retreated, thus tautening the stalk. After three fingers grasping the fruit, the manipulator retreats and tightens the fruit, and the control system identified the position of the stalk by means of a 6-axis force sensor mounted on the wrist [43, 46]. At this point, the wrist rotated on itself to bring the stalk into the cutting position and the stalk was cut by a circular micro-saw. The prototype was a success and the results obtained were extraordinary, but the implementation costs were comparable to those estimated for the whole robot [43].

20

1 History and Present Situations of Robotic Harvesting Technology: A Review

Fig. 1.20 The Cartesian-type citrus harvesting robot developed in University of Catania [43]

Fig. 1.21 The pneumatic flexural three-finger end-effector developed in ARTS laboratory [43]

Subsequently, Muscato G. et al. improved this type of end-effector (Fig. 1.22). The shape of the teeth is such that, when the jaw lifts, the stalk of the fruit in front of the grasping device is trapped and guided between the blades of the clippers. For the further improved end-effector, besides the two actuators to control the jaw and clippers, a third one was added to command a sliding tray placed at the bottom of the pincers, which slides out before the stalk is cut [43]. The presence of this tray along with the lengthening of the lower jaw allows the orange to remain trapped inside [43]. However, the success rate was not tested. The citrus harvesting robot developed by Raparelli T. et al. of the University of L’aquila is based on a multiple picking devices, which is in the shape of an oversized hairbrush [47] (Fig. 1.23). Each tooth is formed essentially by a moving part and a fixed frame, and the moving part is made by a circular tube which is installed with cutting tools joined to a square plate in the rear part [47]. In this front part of the tube is mounted the cutting knife, in the upper side, and the moving plate, in the internal part [47]. The moving plate is studied to fix a proximity sensor used to start the cutting knife and after it is used to rotate the moving plate to send the orange in the

1.2 The History and Current Situation of the Development …

21

Fig. 1.22 The jaw-type citrus harvesting end-effector developed in University of Catania [43]

Fig. 1.23 The tube-knife-type citrus harvesting end-effector developed in University of L’Aquila [47]

rear part of the tube [47]. The tube has an opening in the lower side to discharge the orange [47]. Experimental tests indicated that the cutting operation does not show any problem, but the machine fails in cutting a branch of 8 mm, and in this case, it needs 2–3 repetitive operations to cut the branch [47]. Furthermore, the moving plate showed some problems in capturing the orange after cutting, in the case of tick foliage [47]. (2)

Citrus harvesting robots developed in USA

The United States is another major country to carry out the technical research of citrus harvesting robot. The citrus production in Florida, the world famous “Orange State”, accounts for 75% of the total output of the United States. The University of Florida has carried out the constant research for decades from the citrus mechanized harvesting to the robotic harvesting of citrus, which is now leading the world. Harrell R. at the University of Florida developed a prototype of citrus harvesting robot in 1987 [48–51]. The hydraulically driven spherical manipulator was housed at the back of a trailer [43]. The end-effector had a color camera mounted at the tip, an ultrasound sensor to estimate the distance of the fruit, and a rotating ring to grasp

22

1 History and Present Situations of Robotic Harvesting Technology: A Review

Fig. 1.24 Earlier prototype of citrus harvesting robot developed in University of Florida [49–51]

the fruit and cut its stalk [43]. This end-effector did not provide any system for collection and storage. It was found in experiments that the hydraulic driving system was not sufficiently accurate and the speed of response by the system as a whole was lower, thus preventing the machine from giving the hoped-for results [43]. The greatest problem lay in the vision system, which is difficult to solve a problem of segmentation in a highly unstructured and variable environment under the natural light [43] (Fig. 1.24). In recent years, a 7-DOF manipulator was adopted by Burks T. et al. in the development of citrus harvesting robot. On the basis of the commercial pneumatic gripper, a pair of ultrasonic sensors were installed to detect the range information, and a CCD camera is located for an “in hand view” of the harvesting environment and the visual servo control [52] (Fig. 1.25). And then a three-finger end-effector was redesigned. It is driven by a single motor and equipped with a camera, an infrared sensor, and an ultrasonic sensor [53]. The related research is still being carried out. In addition, the citrus harvesting robot developed by Lee B. of UC Davis consisted of a hydraulically powered compressing frame attached to a forklift and an electropneumatic fruit detachment system [54, 55] (Fig. 1.26). The frame had two hydraulic cylinders. The electro-pneumatic system included a pneumatically actuated cutting device and had two pneumatic cylinders to provide two horizontal sliding motions to drive the cutting device inside the canopy [55]. Because its position and posture are

Fig. 1.25 New prototype of citrus harvesting robot developed in University of Florida [52, 53]

1.2 The History and Current Situation of the Development …

23

Fig. 1.26 The fork-type citrus harvesting robot developed in UC Davis [54, 55]

limited in view of the horizontal displacement in the direction of X and Y, and the cutting direction of the scissors cannot be adjusted according to the stem position. Furthermore, a more compact actuator mechanism should be redesigned to drive the cutting device, which may reduce the contact surface area between the canopy and the harvester, thus facilitating the canopy penetration and reducing accidental fruit removal [54]. The implementation of sensing techniques for citrus fruit detection and a fruit-collecting mechanism is also needed [54]. The Vision Robotics Corporation, San Diego built a Scout prototype capable of thoroughly scanning small orange trees [57] (Fig. 1.27). The platform is a fourwheeled trailer that may move down rows via an integral winch. The new 8 arm consisting of shoulder and elbow joints, which move two 4 links in a plane perpendicular to the Scout’s direction of travel, and two wrist joints, which enable the cameras to look up-and-down and side-to-side [57]. The camera head uses a scouting hand with four sets of cameras each looking at a different directions, whose “cross-eyed” configuration has an almost 180º field of view [57]. The Scout will take all eight pictures simultaneously, ultimately enabling a continuous scan, and the data for all

Fig. 1.27 The citrus Scout prototype developed by Vision Robotics Corporation, San Diego [56]

24

1 History and Present Situations of Robotic Harvesting Technology: A Review

Fig. 1.28 Citrus harvesting robot developed by Energid Technologies Corporation [57]

the fruit is combined to generate a 3D model of the fruit on the tree [57]. All hardware is complete for the prototype, and preliminary detection and control are also complete. The company plans to demonstrate the full system scouting in production of orange groves. The Energid Technologies Corporation also developed a new generation imageguided citrus fruit picker [57] (Fig. 1.28). It adopts a goat truck with a modified boom arm and multiple frog-tongue modules. Each frog-tongue module consists of an aiming gimbal, a high-speed double-acting pneumatic cylinder, a 2-DOF linear positioning unit, a pneumatic valve, and motor controllers. Each module is mounted on a common frame with a regular array of outward facing cameras. It is found from field tests that at least 98% of the fruit on the tree is visible using a moving grid of cameras and the frog-tongue system can mechanically remove over 98% of the fruit observed. (3)

Citrus harvesting robots developed in other countries and regions

In 1990, the citrus harvesting robot, which was developed by Fujiura T. of Shimane University in Japan, adopted a crawler chassis and a hydraulic manipulator and developed an end-effector with pneumatic flexible fingers. After the fruit was gripped, the stem was cut by the scissor [58]. However, it was found in the experiment that when there is occlusion of the branches and leaves, the flexure fingers will be bent before reaching the fruit, resulting in the failure of gripping [58] (Fig. 1.29). In 1988, the citrus picking robot developed by Kubota Co., Japan, installed a lifting cantilever on the mobile platform, and the cantilever front end installed a 3-DOF vertical joint manipulator [5]. The end-effector sucks the fruit by a vacuum suction with a vacuum pump and pulls it back, while its sparse cover moves forward to make the fruit enter the cage and separate from the other fruits. Then the stem is cut off by a barber-shaped cutter to complete the harvesting (Fig. 1.30). The endeffector is equipped with a proximity sensor, and a stroboscopic light source and a mini camera are installed for fruit detection. However, the end-effector had the poor adaptability to fruit size difference and slower movement speed, and the harvest success rate was only 30% [5].

1.2 The History and Current Situation of the Development …

25

Fig. 1.29 The citrus harvesting robot developed by Shimane University, Japan [58]

Fig. 1.30 Citrus harvesting robot developed by Kubota Co., Japan [5]

Lee F. of National Chung Hsing University, Taiwan, China, developed a robotic citrus harvesting system, which consists of a 3-DOF joint manipulator with threefinger end-effector and binocular vision system [59]. The end-effector is driven by a DC motor, which moves backward at the same time when the three fingers are close to grasp the fruit, and cuts the stem with a cutter driven by a slide block (Fig. 1.31). The test results show that the positioning error of the visual servo manipulator is larger.

Fig. 1.31 Citrus harvesting system developed in National Chung Hsing University, Taiwan, China [59]

26

1 History and Present Situations of Robotic Harvesting Technology: A Review

Fig. 1.32 Citrus harvesting system developed in Southeast University, China [60]

Although the end-effector has good adaptability to the single fruit, for the multiple fruit, it may touch fruits other than the target fruit, which will lead to the moving of the target fruit with other fruits and the failure of fruit gripping [59]. In addition, due to the complex structure of the end-effector, sometimes it will be twinned by the branches during fruit picking. Therefore, the author thinks that further improvement is needed [59]. In Chinese Mainland, relevant research on citrus harvesting robots is also carried out. For example, the automatic harvesting system developed by Lu W., Southeast University, including the 5-DOF manipulator and an end-effector composed of a sucker, nut ring, disk sawing knife, and collection capsule [60] (Fig. 1.32). Yao J., Zhejiang A&F University, made a simple picking manipulator model [61] (Fig. 1.33), adopting a three-joint plane folding arm. When the stem is cut off, the fruit may fall into a tube to realize the collection. The citrus harvesting end-effector, developed by Zhang S., Zhejiang University of Technology, consists of a pneumatic joint threefinger gripper and a disk sawing cutter [62] (Fig. 1.34). But generally speaking, the research on citrus harvesting robots is relatively weak in China, and it is urgent to promote the development and test of the whole prototypes. Timeline of the development of citrus harvesting robots is shown in Table 1.2.

Fig. 1.33 Citrus harvesting manipulator developed in Zhejiang A&F University, China [61]

1.2 The History and Current Situation of the Development …

27

Fig. 1.34 Citrus harvesting end-effector developed in Zhejiang University of Technology [62]

2.

Apple Harvesting robots

Apple is one of the most important fruits in the world. It is widely planted in Asia, Europe, America, and Oceania. China is the largest producer of apples, whose output accounts for 46% of the world’s total. Apple’s automatic harvesting technology has also received widespread attention in major countries and regions of the world. (1)

Apple harvesting robots in USA

Peterson D. of USDA Agricultural Research Service, Appalachian Fruit Research Station has developed a robotic bulk harvesting system for apples grown on Yshaped narrow inclined trellises (Fig. 1.35). A three-wheel, all-wheel-drive power unit was used to support the robotic bulk harvester, a hydraulic rapid displacement actuator was supported by a frame which was attached to the rod end of an extending cylinder [64]. A mechanical conveyance system used rollers, tracks, and a support structure to position the RDA along the row (x-axis) and parallel to the leader (y-axis, 52.5° to horizontal) [63]. The vision system was composed of a Sony digital video camera and a Hot Connect 8920 frame-grabber, and a catching surface was used to catch and immobilize the fruit in the area of expected detachment [63]. To conduct the harvesting, the power unit was positioned along the row, and either processed automatically to determine RDA actuation position or the position was manually selected from the computer image using the mouse, and the harvest cycle continued [64]. Test results indicated that the image-based automatic positioning is difficult, but the human operator-controlled “mouse selection” technique performed particularly well [64]. Fruit removal averaged 95% and detached fruit graded 99% [63]. Washington State University, Cameron University, Massachusetts Institute of Technology, and other universities developed an apple harvesting robot prototype for V-trellis fruiting wall architecture cooperatively [64–68] (Fig. 1.36). It adopted a

28

1 History and Present Situations of Robotic Harvesting Technology: A Review

Table 1.2 Timeline of the development of citrus harvesting robots Country Organization

Prototype

Method of recognition &

Research

DOF

DOF

localization

phase

Wheeled

3

1

CCD camera

Field test

Trailer

3

1

Vehicle

(Region)

Italy

USA

Japan

University of Catania

University of Florida

Shimane University

Manipulator End effector

2 Ultrasonic Sensors+CCD camera+Light

Field test

source+Self-focusing lens

Crawler 3(hydraulic)

2

Color TV camera

Field test

-type

13(5DOF

Italy

University of Catania

Crawler

main arm +dual

-type

-

3DOF slave

Wide-angle lens +Dual CCD camera in hand

Filed test

arms)

Taiwan, China

Italy

USA

National Chung Hsing

-

3

1

Binocular vision

3

-

-

7

1

University

University of Catania

University of Florida

Indoor test

Crawler -

-type

CCD camera in hand+ -

Field test Ultrasonic sensor

(continued)

1.2 The History and Current Situation of the Development …

29

Table 1.2 (continued)

2(Cartesian, USA

UC Davis

Forklift

hydraulic)

Field test

Vision USA

Robotics

4-wheeled

Corporation,

trailer

4

-

8 CCD cameras

Partial prototype

San Diego

Energid USA

Technologies

Truck

-

-

3

An array of CCD cameras in hand

Corporation

Partial test

Southeast China

University Jiangsu

1

-

Partial prototype

University

Fig. 1.35 The robotic bulk harvesting system developed in Appalachian Fruit Research Station [63]

John Deere Gator utility vehicle, a custom-designed and 7-DOF serial link manipulator [69]. The end-effector is an under-actuated, tendon-driven multi-finger gripper, and the machine vision system consisted of a single CCD color camera mounted on top of a time-of-flight-based 3D camera [68]. The picking efficiency for the apples in field test was 84.6%, and the total cycle time required to harvest a single apple was approximately 7.6 s [68]. It was found that horticultural practices are critical to the performance of robotic tree fruit harvesting systems. The picking process did

30

1 History and Present Situations of Robotic Harvesting Technology: A Review

Fig. 1.36 The apple harvesting robot developed by Washington State University and other universities [68]

cause vibration of the tree and remaining fruit, and additional sensing of force, fruit orientation, and stem location is required for improved harvesting efficiencies and overall robustness [68]. (2)

Apple harvesting robots in Europe

The robotic apple harvester developed by the Katholieke Hogeschool Limburg, Belgium, adopts the Panasonic 6-DOF industrial manipulator with the external vertical axis mounted on a tractor and develops a silicone funnel with a diameter of 10.5 cm with a camera mounted inside [70] (Fig. 1.37). A touch panel PC with human–machine interface with the PLC is integrated into the controller. A threepoint suspension consists of two hydraulic feet and one turn over cylinder is added between the manipulator and the tractor, which ensures a stable positioning of the robot platform during the picking cycle, controlled by two-level sensors [69]. Field experiments demonstrate that about 80% of the apples are detected and harvested. However, a more firm connection of the stems to the limbs results in the detaching failure relying on suction, and apples not harvested are either not detectable by the vision system or not reachable for the robot manipulator [69]. The overall cycle time

Fig. 1.37 The apple harvesting robot developed in Katholieke Hogeschool Limburg, Belgium [69]

1.2 The History and Current Situation of the Development …

31

to pick one apple is 8 to 10 s. It is found the communication bottleneck between the vision system and the robot through the central controller has an adverse effect on the overall picking cycle time [69]. By improving the bandwidth of this connection and on optimizing the image processing, the picking cycle period is expected to be reduced to less than 5 s [70]. Furthermore, special attention should go to improving the actual apple detach movement by taking into account the relative pose of the apple with respect to the branch [69]. The apple picking robot developed by Technical University of Munich, Germany [70] (Fig. 1.38), used a tractor and developed a complex 9-DOF uplift-type redundant manipulator with a complex sensing system consisting of a 3D camera, a RGB camera, and a ripeness sensor. At present, only the 9-DOF manipulator is verified by experiment. (3)

Apple harvesting robots in China

The apple harvesting robot developed by Luo H. of Beijing Forestry University is operated in a master–slave manner [71] (Fig. 1.39). The manipulator of the robot is from an ordinary SCARA robot with a third rotation joint R3 added to its distal, which is further installed on a vertical translational joint. The end-effector is installed in the manipulator with two passive joints to adjust its gesture, and the cutters and graspers are driven by microcylinders together to hold the apple steadily and to cut the apple stem. Two CCD cameras are vertically set at the bottom center of the endeffector frame and horizontally set at the frame side of the end-effector, respectively, to switch the field of view in close range. During the operation, the operator observed the fruit through the monitor and controls the manipulator and the end-effector by the joystick to complete the harvesting. The apple harvesting robot developed by Zhao D., Jiangsu University and Zhang X., China Agricultural Machinery Academy of China [72–75] (Fig. 1.40) uses a GPS navigating crawler-type mobile platform and a 5-DOF PRRRP-structure manipulator (1-DOF uplifting, 3-DOF rotation, and 1-DOF wrist expansion). The spoon-shaped

Fig. 1.38 The apple harvesting robot developed in Technical University of Munich, Germany [70]

32

1 History and Present Situations of Robotic Harvesting Technology: A Review

Fig. 1.39 The apple harvesting robot developed in Beijing Forestry University [71]

Fig. 1.40 The apple harvesting robot developed by Jiangsu University and China Agricultural Machinery Academy of China [72–75]

end-effector contains a pneumatic-driven gripper and an electric cutting device. When the fruit was grasped, the DC motors transmitted power by flexible wire to drive the cutter rotating around the gripper, cutting off the stalk in front of end-effector at any position [73]. In the laboratory test, the success rate was 86%, the average time of each fruit picking was 14.3 s, while the success rate and the average time of the field test were 77% and 15.4 s, respectively. Gu B. et al. Nanjing Agricultural University, installed the Motoman 6-DOF industry manipulator on the developed light crawler intelligent mobile platform through a lateral sliding mechanism to expand the working space of harvesting. The force sensor and the slip sensor were installed on the industrial gripper, and the binocular vision system was adopted [76, 77] (Fig. 1.41). Orchard test showed that the success rate of fruit recognition was between 60 and 88%, and the effect of backlight on a sunny day was great. In recognized fruits, 37–47% could be successfully picked. The success rate of harvesting was affected by the sliding in gripping, the occlusion of branches and leaves, and the fruit offset and swing caused by collision and wind, respectively.

1.2 The History and Current Situation of the Development …

33

Fig. 1.41 The apple harvesting robot developed in Nanjing Agricultural University [76, 77]

(4)

Apple harvesting robots in Japan and Australia

The apple harvesting robot [78–80] (Fig. 1.42), developed by Bulanon D., Hokkaido University, Japan, has a 3-DOF Cartesian-type manipulator mounted on a mobile elevator to effectively expand the operating space. The end-effector is equipped with a laser ranging sensor and a color CCD camera. Once the camera detects the largest single fruit in the image and the camera axis is positioned in line with the fruit center, the laser ranging system could easily measure the fruit’s distance. Then the DC motor drives the two fingers to hold the peduncle and the step motor rotates the wrist to bend the fruit off. The end-effector adopts the same structure as that put forward by Kataoka T., Iwate University [81] (Fig. 1.43). To prevent the fruit from falling off during the peduncle gripping, a slanting support bar was added at the bottom in the improved end-effector. Results of the field test showed that the robot successfully harvested about 89% of the apples, and the position of the peduncle, size of the peduncle, and difficulty in the fruit recognition were the main factors that led to failure [80]. Although the success rate is considerably high, the response time of the robot is still long [80].

Fig. 1.42 The apple harvesting robot developed in Hokkaido University, Japan [78]

34

1 History and Present Situations of Robotic Harvesting Technology: A Review

Fig. 1.43 The apple harvesting end-effector developed in Iwate University, Japan [81]

Setiawan A., et al., University of New South Wales, Australia, used John Deere 4210 Tractor as the mobile platform, and a DENSO VS-E series 6-DOF manipulator mounted on a linear slider attached to the front loader of the tractor to construct the prototype of apple harvesting robot [82] (Fig. 1.44). A low-cost cup-type end-effector was designed, which consists of five main parts: one main cup, two inner rings, and two end caps [82]. As a clumping mechanism, two rubber bladders connected to a pneumatic tube are attached to the inner surface of the cup [82]. For this prototype, only a verification experiment was carried out in the laboratory for string-hanged apples, and there is no report to the performance and test of the whole machine. Timeline of the development of apple harvesting robots is shown in Table 1.3. 3.

Kiwifruit harvesting robots

The kiwifruit’s epidermis is densely covered with dense villi, it has a soft texture, and it has sweet and sour taste and rich nutrition. It is widely welcomed by consumers. The kiwifruit produced in New Zealand is popular in the world, and shares about 70% of the world market, ranking the highest in the world. At present, research on robotic kiwifruit harvesting technology is mainly carried out in New Zealand and China. (1)

Kiwifruit harvesting robots in New Zealand

The kiwifruit harvesting robot developed by Scarfe A. et al., Massey University (Fig. 1.45), adopts the autonomous four-wheel drive vehicle with GPS and machine vision navigation [83, 84] (Fig. 1.45). A pneumatic thumb-finger end-effector was designed to bend the fruit off. Eight Webcams are used to look up into the canopy

1.2 The History and Current Situation of the Development …

35

Fig. 1.44 The apple harvesting robot developed in University of New South Wales, Australia [82]

to identify fruit and to perform stereopsis in order to determine the three position coordinates of each fruit [83, 84]. No test result of the whole machine has been reported until now. (2)

Kiwifruit harvesting robots in China

The Northwest A&F University has carried out a centralized study on the field of robotic kiwifruit harvesting. The prototype consists of the electric four-wheel independent steering platform, the 3-DOF Cartesian-type manipulator, the end-effector, and the fruit recognition and location system that includes a CCD camera and a Kinect sensor. Based on the mechanical properties of the kiwifruit, a 3-DOF endeffector was developed [85, 86] (Fig. 1.46), it picks the fruit by gripping with two fingers and twisting of the wrist. The infrared limit switch and the finger pressure sensor were used to detect the fruit position and measure the gripping force. The success rate of the fruit picking test in the laboratory was 90% [86, 87]. And then, since there are larger space and little shelter at the bottom, and the fruits grow in clusters, a new end-effector was developed. It approaches the fruit from the bottom,

36

1 History and Present Situations of Robotic Harvesting Technology: A Review

Table 1.3 Timeline of the development of apple harvesting robots Country Organization

Prototype

End-effector

DOF

DOF

3

1

Tractor

6

1

-

Mobile lifter

3

2

+Laser ranging

(Region)

USA

Method of

Manipulator

Vehicle

Appalachian

Three-wheel

Fruit Research

(gasoline

Station

engine)

recognition & localization

Digital video camera

Research phase

Field test

University of Australia

New South

-

Wales

Japan

Hokkaido University

CCD camera

Katholieke Belgium

Hogeschool

Tractor

6(Commercial)

1

7

1

5

2

9

-

Limburg

Nanjing China

China

Jiangsu University

University of Munich

hand

Field test

Binocular vision Indoor test

-type

Crawler -type

Technical Germany

CCD camera in

Crawler

Agricultural University

Field test

sensor

CCD camera in hand

Field test

3D camera+CCD Tractor

camera ripeness

Prototype

sensor

(continued)

1.2 The History and Current Situation of the Development …

37

Table 1.3 (continued)

China

Beijing Forestry University

-

4

3

Wheeled

7

actuated,

Washington USA

State University and others

3(under

multi-finger)

Dual CCD

Partial

camera in hand

prototype

CCD camera+ Depth camera

Field test

Fig. 1.45 The kiwifruit harvesting robot developed in Massey University [83]

Fig. 1.46 The kiwifruit harvesting end-effectors developed in Northwest A&F University [86, 87]

38

1 History and Present Situations of Robotic Harvesting Technology: A Review

Fig. 1.47 The kiwifruit harvesting robot developed in Northwest A&F University [88, 89]

and envelops and grips the fruit from two sides, and then rotates up to separate the fruit from stem [88] (Fig. 1.47). The end-effector is still equipped with an infrared limit switch and a finger pressure sensor. The success rate in the field test is 89%, and the average picking time is 22 s [88, 89].

1.2.3 Harvesting Robots for Fruits and Vegetables 1.

Strawberry harvesting robots (1)

Strawberry harvesting robots in Japan Japan is the main country in the production and sale of strawberry. In recent decades, its annual output has been in the forefront of the world, and its research on strawberry harvesting robots is also far ahead, and many kinds of the prototype have been put forward. According to the different patterns of ground cultivation, ridge-top cultivation, and tabletop cultivation of strawberry, the principle and structure of recognition and picking have great differences for different robots. Different researchers have developed various kinds of robotic equipments. (1) Strawberry harvesting robots for ridge-top culture

In Japan, traditional ground cultivation has been gradually replaced by ridge cultivation (hilltop cultivation) and hang bench cultivation (tabletop cultivation, elevated cultivation). Ridge cultivation is a cultivation mode for planting rows into 20–30 cm high ridges and planting crops on ridges. By raising the cultivation rows, it can

1.2 The History and Current Situation of the Development …

39

enhance ventilation, save water, expand the surface area of soil, and prevent pollution of fruit, thus effectively improving the yield and quality. Meanwhile, the aisles left by the high ridges greatly facilitate the implementation of field management operations. The ridge cultivation in Japan is divided into ridge side and ridge-top cultivation, and their fruits are grown on a horizontal plane of the ridge and on both sidewalls of the ridge, respectively. The ridge-top cultivation is more suitable for robotic harvesting than a training system on sidewalls, since the fruits can be easily detected by a visual sensor and can be approached by a manipulator with no obstacles [90]. At present, various types of harvesting robots are mainly developed for ridge-top cultivation, and vertical downward picking operations are adopted for the flat growth of stems and fruits. The first generation of strawberry harvesting robot prototype developed by Kondo N. et al. for ridge culture installs 3-DOF Cartesian-type manipulator on the gantry-type mobile platform and develops a suction-rotary-cutting end-effector [91] (Fig. 1.48). The end-effector was developed to suck fruit into an inner cylinder connected to a vacuum device and to cut peduncle by the inner cylinder rotation [91]. During harvesting operation, an ultrasonic sensor first measured distance to ridge top and an image was acquired by a color CCD camera, and then the manipulator moved to a position above a target fruit and the end-effector moved downward to suck the fruit until it was detected by three pairs of photo-interrupters; finally, the inner cylinder rotated and the fruit’s peduncle was cut [91]. The experiment showed that the fruit sucking method was very effective for harvesting small-sized fruit [91]. However, 34% of fruits could not be harvested because some fruits were not sucked and some peduncles were not cut successfully [91].

Fig. 1.48 The first generation of strawberry harvesting robot prototype developed by Kondo N. et al. for ridge culture [91]

40

1 History and Present Situations of Robotic Harvesting Technology: A Review

Arima S. and Kondo N., et al., then developed a multi-functional robot [90]. The robotic harvesting unit picking still used the gantry-type mobile platform and color CCD camera and adopted a 5-DOF (three prismatic and two rotary) Cartesian-type manipulator. As the sucking-type end-effector would suck some adjacent immature fruits with the target fruit [90], a new hook-cut end-effector was designed. A hook was installed to approach and hang the peduncle of target fruit, then the target fruit was separated from other fruit by the hook raising, and finally the peduncle was grasped by fingers and cut [90]. However, the performance of the end-effector has not been verified experimentally (Fig. 1.49). Kondo N. et al. then developed the prototype of the second-generation harvesting robot, which mainly improved the end-effector structure on the basis of the first generation [92, 93] (Fig. 1.50). It sucked the fruit into the sucking head relying on

Fig. 1.49 The improved strawberry harvesting robot prototype by Kondo N., et al. for ridge culture [90]

Fig. 1.50 The second generation of strawberry harvesting robot prototype developed by Kondo N., et al. for ridge culture

1.2 The History and Current Situation of the Development …

41

Fig. 1.51 The strawberry harvesting robot for ridge culture developed in Miyazaki University [95–97]

the vacuum device, and then the manipulator moved upward and internal cylinder rotated to cut its peduncle to cut off the stem [93–95]. Even if the peduncle was not cut, an added open–close section was closed and detached the fruit from the peduncle [92, 93]. This kind of end-effector can be considered as the combination of the abovementioned two kinds of end-effectors, which can effectively improve the success rate of picking, but the complexity of the corresponding structure is also increased. For delicate fruit with soft long peduncles, such as strawberry, sucking and hooking may be more feasible than the way of gripping. The harvesting robot for ridge culture developed by the Cui Y. et al. of Miyazaki University in Japan used a 3-DOF Cartesian-type manipulator to take a pneumatic grip-cut integrated end-effector to move downward and cut off the peduncle [95– 97] (Fig. 1.51). The robot is equipped with a global camera installed on the top of robot frame with a local camera installed on the end-effector to identify and locate fruits on the background of white or black thin sheets [96–98]. Under this condition, the difficulty of recognition and location and picking is significantly reduced. The experiment shows that the average time of recognition and picking is 1 s and 6 s, respectively. It can achieve 93% accuracy of strawberry’s stem detection, and the detection failure is due to the occlusion caused by leaves or if the strawberries are so close. The success rate of harvesting on black thin sheets reached 96%, and there was no false harvesting but a small amount was missed. Since the white plastic sheet reflected more light than the black sheet and caused some bright spots appearing on the fruit surface, the fruits with low maturity cannot be picked up, but the success rate is also over 90%. (2)

Strawberry harvesting robots for tabletop culture

Tabletop culture is a new type of cultivation mode developed in recent years. Through vertical support or hanging, the culture bed leaves the ground and the fruit grows on both sides of the culture bed. Because of its advantages of labor saving, cleanliness,

42

1 History and Present Situations of Robotic Harvesting Technology: A Review

Fig. 1.52 The early strawberry harvesting robot for tabletop culture developed by Arima et al. [98]

and high yield, it has been rapidly developed and promoted. Tabletop culture makes fruits hang down from the table with long peduncle and simple posture, so there are little obstacles and occlusion. As a result, the mature fruits are easier to detect and harvest, which is extremely beneficial to robotic harvesting. The large-scale popularization of tabletop culture has provided favorable conditions for the application of harvesting robots [5], and many achievements have also been made in related research. Arima S., et al. developed a strawberry harvesting robot for hanged culture bed [98] (Fig. 1.52). A four-wheel independent driving electric chassis was adopted. A polar coordinate-type manipulator was also adopted because it was not necessary to avoid obstacles and control of the manipulator was not complicated [99]. Assumed that the fruit is in the same vertical plane of the bedside, two-dimensional information was obtained by a single CCD camera. A pneumatic horizontal suck-twist-cut endeffector equipped with two pairs of photo-interrupters was developed. Experiments show that the robot could harvest all target fruits without damage, but some immature fruits located around target fruits were also picked up. The average harvesting was 14–20 s. It was considered that the mechanisms of robotic components and their software algorithms could be simple because the strawberry on tabletop culture had a suitable condition for robotic harvesting [99]. Based on the above research, Arima S., Kondo N. et al. simplified the structure of the robot. A Cartesian-type manipulator with three prismatic joints and one rotary joint suspended under cultivation bed [99]. And the same suck-twist-cut end-effector was used. The target fruit was pneumatically sucked into the end-effector by a blower, and then the end-effector moved toward a target fruit until it was detected by three pairs of photo-interrupters on sucking head [100]. Finally, the robot’s wrist joint rolled and its peduncle was introduced into a curved hole and cutoff [100] (Fig. 1.53). The robot still relied on a single CCD camera for fruit recognition and location. The experiment showed that the fruits could be removed without damage, but there was

1.2 The History and Current Situation of the Development …

43

Fig. 1.53 The strawberry harvesting robot for tabletop culture improved by Kondo N. et al

still a possibility of nearby immature fruits being harvested together with the target fruits. Due to the simplification of the structure and control algorithm, the average harvesting time is shortened to about 10 s [93, 99–102]. Kondo N. and Hayashi S. of Institute of Agricultural Machinery, BRAIN, NARO, et al. developed the second generation of harvesting robot prototype (Fig. 1.54). A 3-DOF Cartesian-type manipulator (gantry, up–down movement, and horizontal rotation) is installed on the railed moving platform, and a grip-cut integrated endeffector is used. The machine vision unit was composed of five light sources with 120 light-emitting diode (LED) chips for each source, and three aligned color charge coupled device (CCD) cameras [1]. The center camera was used to calculate the inclination of the peduncle, whereas the two cameras mounted on both sides of the center camera were used in a stereo vision system to determine the 3D position of the fruit [1]. Further, a suck-grip-cut end-effector was designed by adding a vacuum suction device on the grip-cut integrated end-effector [1, 93, 103, 104]. In suction

Fig. 1.54 The second generation of strawberry harvesting robot prototype developed by Kondo et al. [1]

44

1 History and Present Situations of Robotic Harvesting Technology: A Review

picking, the suction device was used to hold the fruit before cutting to compensate for the three-dimensional (3D) position error caused by the variations in the size and shape of the fruit and the conditions around the fruit [1]. The experimental results showed that the machine vision unit correctly detected a peduncle of the target fruit at a rate of 60% [1]. In harvesting tests conducted throughout the harvest season on target fruits with a maturity of 80% or more, the successful harvesting rate of the system was 41.3% when fruits were picked using a suction device before cutting the peduncle, while the rate was 34.9% when fruits were picked without suction [1]. The execution time for the successful harvest of a single fruit, including the time taken to transfer the harvested fruit to a tray, was 11.5 s [1]. The main cause of unsuccessful harvesting was that the machine vision unit did not detect the fruit in mismatching the left and right camera images. It was also found that the suction has little effect on picking success rate, and only helps to reduce the swinging of the fruit during the transfer to the tray. Furthermore, mistaken harvesting of immature fruits was observed, resulting in a decrease in yield [1]. Based on the improvement of the second-generation prototype, Hayashi S. has developed the third-generation and fourth-generation prototypes (Figs. 1.55 and 1.56). They both still used the railed mobile platform and the gantry-type cylindrical manipulator. However, the suction device was canceled on the end-effector according to the experimental results of the second-generation prototype, and a transmissiontype photoelectric sensor was added for confirming the presence of the picked fruit [105]. The third-generation prototype still used the vision system structure of the twogeneration prototype. The fourth prototype was designed with the aim of reducing its weight while maintaining the basic structure of the railed mobile platform and the gantry-type cylindrical manipulator of the third prototype [105]. A rectangular LED source was used, and the transmission-type photoelectric sensor was replaced with a reflection-type sensor. The weight of the whole machine has been reduced from

Fig. 1.55 The third generation of strawberry harvesting robot prototype developed by BRAIN, NARO

1.2 The History and Current Situation of the Development …

45

Fig. 1.56 The fourth generation of strawberry harvesting robot prototype developed by BRAIN, NARO

345 kg of the third-generation prototype to 245 kg [105]. The third prototype showed a harvesting success rate of 60.0–65.6% and a picking execution time of 8.8 s. The fourth prototype showed a picking execution time of 6.3 s by removing ineffective waiting time, although the success rate was lower, at 52.6% [106]. Most of the failures came from the undetection of target fruit because of the occlusion by other fruit, and some came from the failure of peduncle cutting due to the detection errors [105]. Specifically, the robot could not detect some matured red fruits and ignored other fruits [105]. One of the reasons for the lower harvesting success rate may be the uneven illumination provided by the new LED light source, and it further needed improvement on the machine vision algorithm and layout of the LED lights [105]. Yamamoto S. and Hayashi S. et al. have also developed a stationary robotic strawberry harvester for the railed moveable cultivated bed system [106–108] (Fig. 1.57). The robot used a method to detect and harvest fruit from bottom to up. It used 7DOF industrial manipulator. And a suck-blow-grip-pull end-effector was developed, which first approaches the fruit from the bottom and sucks, and then blows out the adjacent fruit and leaves by the nozzle and grips the fruit with two fingers, and finally inclines a certain angle to pull the fruit off. It is found in the experiment that the success rate of harvesting mature fruit was 67.1%, and the mistake rate when adjoining fruit was on the outer or aisle side of the target fruit is somewhat higher. Furthermore, fruit damage, the unwilling harvest of adjacent fruits, and immature fruits might happen. The average harvesting time was 31.5 s. Yamashita T. of Maekawa Seisakusho, et al. developed the M-3 strawberry harvesting robot for hanging cultivated bed [109, 110] (Fig. 1.58). It adopts the railed mobile platform and the manipulator with three rotating joints, an in-hand camera, and a to-hand binocular stereo vision system are applied. A double-side grip-cut integrated end-effector was developed to realize the inside-approach harvesting of strawberry on both sides. This robot can travel beneath a hanging bench and picks

46

1 History and Present Situations of Robotic Harvesting Technology: A Review

Fig. 1.57 The stationary robotic strawberry harvester for the railed moveable cultivated bed developed by Brain [106–108]

Fig. 1.58 The strawberry harvesting robot for hanging bench developed by Maekawa Seisakusho [109, 110]

fruits on both sides using a single manipulator [109]. Test results showed that the success rates for detection and for picking were 68.7 and 50.6% for var. Benihoppe, and 75.0% and 63.0% for var. Amaotome [110]. The average harvesting time was more than 37 s. (2)

Strawberry harvesting robots in China

China has also started early in the field of strawberry harvesting, and it has been developing continuously in many organizations. The Zhang T. team of China Agricultural University first carried out the research on the strawberry harvesting robot. They put forward prototypes of strawberry harvesting robot for both ridge culture and tabletop culture. The robotic strawberry harvesting system for ridge culture is configured by a Cartesian-type manipulator

1.2 The History and Current Situation of the Development …

47

(three prismatic joints) with a grip-cut integrated end-effector, and two CCD cameras are installed on the frame and arm, respectively, to form a vision system [111] (Fig. 1.59a). The prototype of “Harvesting Child No.1” for tabletop culture used a microcrawler chassis, a Cartesian-type manipulator (three prismatic joints), and a grip-cut integrated end-effector [112] (Fig. 1.59b). A camera is installed at the bottom of the end-effector to detect the fruit and judge the position deviation, and an optical fiber sensor is installed on each fingertip to judge the existence of the peduncle. In the experiment, the success rate reached 88%, and the average harvesting time was 18.54 s. This prototype of “Harvesting Child No.1” is demonstrated at the 7th World Strawberry Conference and on TV program of CCTV (China Central Television). The Li W. team of China Agricultural University used the same robot hardware structure for tomato and strawberry harvesting. The success rate and the average harvesting time showed in paper were 86% and 28 s, respectively, but the different working performance to tomato and strawberry was not introduced in detail [34] (Fig. 1.60). A strawberry harvesting robot for tabletop culture was developed by Feng Q., et al., National Engineering Research Center for Information Technology in Agriculture [113, 114] (Fig. 1.61). The robot adopted sonar-navigation four-wheel-drive vehicle, 6-DOF industrial joint-type manipulator and binocular vision system. The non-destructive end-effector contains two fingers forming a pneumatic gripper for peduncle grasping, a suction cup for fruit holding, and an electrical heating knife for peduncle cutting [114]. The total weight is 105 kg and the dimension is 1500 × 700 × 1600 mm. The field experiment showed that the 100 targets were all detected by the vision unit, but 86 fruits were successfully harvested after harvesting attempt for 121 times because of the operation faults [114]. So one harvesting attempt needs

Fig. 1.59 The strawberry harvesting robots developed by Zhang T. Z. team of China Agricultural University [111, 112]

48

1 History and Present Situations of Robotic Harvesting Technology: A Review

Fig. 1.60 The strawberry harvesting robots developed by Li W. team of China Agricultural University [34]

Fig. 1.61 The strawberry harvesting robot for tabletop culture developed in National Engineering Research Center for Information Technology in Agriculture [113, 114]

averagely 22.3 s, and one successful attempt costs 31.3 s [114]. Among them, picking failure mainly stems from sucking failure and post-harvest drop. Among the 14 fruits unsuccessfully harvested, 10 fruits could not be held by suction cup because of the greater error or the smaller fruit size, and 4 fruits dropped down before transported to the container [114]. The prototype of the robot was also exhibited in many kinds of exhibitions (Table 1.4).

1.2 The History and Current Situation of the Development …

49

Table 1.4 Timeline of the development of strawberry harvesting robots Country Organization

Prototype

Vehicle

(Region)

Japan

Japan

Ehime University

Okayama University

Wheel

Manipulator End-effector DOF

5 (Polar coordinate)

DOF

Method of recognition & localization

Research phase

1

CCD camera

Field test

Gantry

3(Cartesian)

1

CCD camera

Field test

Gantry

3(Cartesian)

3

CCD camera

Field test

Hanging

4(Cartesian)

1

CCD camera

Field test

Gantry

3(Cartesian)

2

CCD camera

Prototype

Department of Japan

Technology Development, Ishii Industry Co., Ltd.

Okayama Japan

University, Ehime University

Japan

Ehime University

CCD camera on China

China Agricultural University

base+ CCD -

3(Cartesian)

2

camera on arm

Railed

3(Cartesian)

2

source+3 CCD

Partial prototype

LED light Japan

BRAIN, NARO

Field test

camera

(continued)

50

1 History and Present Situations of Robotic Harvesting Technology: A Review

Table 1.4 (continued)

Japan

Miyazaki University

CCD camera in -

3(Cartesian)

1

hand+ CCD camera to hand

Indoor test

Binocular vision Japan

BRAIN, NARO

Railed

7(Joint-type)

3

+RGW LED+3D

Field test

vison sensor

CCD camera in Japan

Maekawa Seisakusho

hand + Railed

3(Rotation)

Field test

1 Binocular vision to hand

National Engineering Research Center China

for Information

Wheeled

6(Joint-type)

2

Binocular vision Field test

Crawler-type

3(Cartesian)

1

CCD camera

Railed

3(Cartesian)

1

source+3CCD

Technology in Agriculture

China

China Agricultural University

Field test

LED light Japan

BRAIN, NARO

Field test

camera

China

China Agricultural

Magnetic-guide

University

crawler-type

4

1

Binocular vision Field test

1.2 The History and Current Situation of the Development …

51

Fig. 1.62 The early cucumber harvesting end-effector developed in University of Tokyo [115]

2.

Cucumber harvesting robots

Cucumber is one of the most popular and most productive fruit types in the world. The output of cucumber in is 84.3% of the world’s total output, and the output of Europe is 9.6% of the world’s output, which is the most important producing area in the world. At present, the research of cucumber harvesting robots is mainly focused on and carried out in Holland, Japan, and China. (1)

Cucumber harvesting robots in Japan

The University of Tokyo, Japan, developed a prototype of the cucumber harvesting end-effector early in the 80 s of last century [115] (Fig. 1.62). The gripper serves to grasp and hold a fruit, then the sensor and cutter travel up following the shape of the grasped fruit [115]. The sensor detects where the peduncle is by measuring the diameter of the fruit continuously until the diameter is sensed to be less than 8 (mm), at which point, it is judged to be the peduncle [115]. The cutter then rotates and cuts the peduncle to release the fruit from the canopy [115]. Motion tests have been made with real cucumber fruits, which were placed in the hand manually, in the laboratory [115]. It was found that the hand could follow the shape of most fruits except for those that were very curved, and the sensor could detect all peduncles except for those that were heavily curved [115]. A prototype of cucumber harvesting robot for the inclined trellis training system was developed by Kondo N. et al. in Japan [116–119] (Fig. 1.63). It is composed of a four-wheel electric platform, a manipulator made up of an inclined prismatic joint capable of sliding along the trellis, and five rotational joints, an end-effector, and a dual-wavelength vision sensor to identify and locate the cucumber fruit under the green background. Based on the principle of end-effector developed by University of Tokyo [115], the end-effector is replaced by a potentiometer to detect the peduncle, and the mechanical structure is optimized.

52

1 History and Present Situations of Robotic Harvesting Technology: A Review

Fig. 1.63 The cucumber harvesting robot for inclined trellis training system developed by Kondo et al. [116–119]

Van Hente E. also reported the research and development of the cucumber harvesting robot for inclined trellis system in Ehime University in Japan [120] (Fig. 1.64). The machine consists of a mobile platform running on the hot water heating pipes of the greenhouse [120]. Due to this growing system, instead of using camera technology for detection and localization of the fruits, the robot uses laser range sensors and ultrasound sensors [120]. But the report did not describe more information about the robot. (2)

Cucumber harvesting robots in Netherland

Van Henten E. of Institute of Agricultural and Environmental Engineering and Wageningen University in Netherlands has carried out long-term and in-depth research on robot harvesting of cucumber [121, 122] (Fig. 1.65). The cucumber harvesting robot for high-wire cultivation system consists of an autonomous vehicle, a 7-DOF manipulator, an end-effector, two-camera vision systems, and miscellaneous electronic and pneumatic hardware. The autonomous vehicle uses the heating pipes mounted on the ground as a rail for guidance and support [121]. The Mitsubishi RV-E2 commercial 6-DOF manipulator is mounted on a linear slide. The end-effector is modified based on Mitsubishi motor gripper 1E-HM01, a suction cup and a thermal cutting device are added to grasp the fruit and to separate the fruit from the plant, respectively [122]. The vision system includes two-camera systems. One-camera system is mounted on the vehicle on a rail for the detection of the fruit, and determination of the ripeness, quality, and 3D localization of the fruit [122]. This camera system uses two synchronized CCD cameras mounted onto one wide-angle optical system, and the detection of the fruit is achieved by using different filters on each of the two cameras [121]. Additionally, a lightweight camera is mounted on top of the end-effector for stereo imaging in the neighborhood of the cucumber during the final approach of the cucumber with the gripper [121].

1.2 The History and Current Situation of the Development …

53

Fig. 1.64 The cucumber harvesting robot for inclined trellis training system developed in Ehime University, Japan [120]

Test results showed that the average success rate of harvesting was 74.4%, and a cycle time of 124 s per harvested cucumber was measured under practical circumstances [122]. The main source of failure had to do with the inability of the robotic system to position the end-effector with sufficient accuracy at the stalk of the fruit, and the occlusion of the branches and leaves also has an important effect on the harvesting failure [122]. In the overall execution time of a single harvest cycle, the cutting, transportation of fruits, identification and positioning, the motion planning, and the execution of the multi-DOF manipulator are more than 25, 20, and 12 s, respectively, resulting in low work efficiency. (3)

Cucumber harvesting robots in China

Tang X. and Zhang T. of China Agricultural University also developed the robotic cucumber harvesting hand–arm system for inclined trellis training system [123]. It

54

1 History and Present Situations of Robotic Harvesting Technology: A Review

Fig. 1.65 The cucumber harvesting robot for high-wire cultivation system developed in Institute of Agricultural and Environmental Engineering, Wageningen, Netherlands [121, 122]

adopts a manipulator, composed of four rotation (swinging) joints, two vertical and horizontal prismatic joints, and a grip-cut integrated end-effector (Fig. 1.66). The China Agricultural University and Zhejiang University of Technology have developed a cucumber harvesting robot [34, 124–126] (Fig. 1.67). It used a crawler chassis with visual navigation and a 4-DOF joint manipulator. The vision system is composed of the binocular cameras, the halogen lamp, and the near-infrared filter. The fruit recognition is realized based on the near-infrared image. The end-effector is equipped with a mini camera for close range positioning. When the end-effector is close to the cucumber, the feedback of micro-switch controls the two pneumatic flexure joints holding the cucumber. After the pressure feedback, the blade rotates 180 degrees to cut the peduncle to complete the harvesting. The success rate of harvesting after idealized treatment of branches and leaves was 85%, and the average harvesting time reached 45 s. Liu C. and Jin L. of Shanghai Jiao Tong University developed a cucumber robot harvesting system based on binocular vision. The cucumber fruit is identified and located based on the color and texture features. A 3-DOF cylindrical-coordinate manipulator (one rotation joints, two prismatic joints) was developed, and a 2-DOF end-effector was configured to grip the fruit and cut the peduncle off. The work is still in the early stage of [127] (Fig. 1.68) (Table 1.5). 3.

Eggplant harvesting robot

Eggplant originated in the tropics of Asia. At present, the planting and consumption of eggplant is concentrated in Asia, with the output of more than 90% of the world. The research of robotic eggplant harvesting technology is also carried out mainly in Japan, China, and other countries in Asia. Hayashi S. et al. of Department of Fruit Vegetables, National Institute of Vegetable and Tea Science developed a harvesting robot for V-shaped training eggplant [128,

1.2 The History and Current Situation of the Development …

55

Fig. 1.66 The robotic cucumber harvesting hand–arm system for inclined trellis training system developed in China Agricultural University [123]

129] (Fig. 1.69). The robot applies a crawler chassis with an industrial 5-DOF manipulator. Its end-effector is composed of a fruit-grasping mechanism, a size-judging mechanism, and a peduncle-cutting mechanism [128, 129]. The CCD camera was attached to the center of the end-effector. The machine vision algorithm for detecting the eggplant fruit was based on color characteristics and morphological features, and the guidance of the manipulator and the end-effector is realized by the increment of the pixel area proportion of the detected fruit with the movement of the in-hand camera. A vacuum suction system is installed on the end-effector, and the four fingers are used to realize the flexible grasping of eggplant. In the laboratory, a harvesting test without chassis movement was carried out after removing the leaves in front of the fruit manually. The success rate was 62.5%. and the failures were mainly due to the unsuccessful detection of the fruit base [128, 129]. Additionally, the average cutting position of the peduncle was 9.3 mm higher from the fruit base in the samples harvested successfully [129, 130]. The robotic harvesting system required 64.1 s to

56

1 History and Present Situations of Robotic Harvesting Technology: A Review

Fig. 1.67 The cucumber harvesting robot developed in China Agricultural University and Zhejiang University of Technology [34, 124–126]

Fig. 1.68 The cucumber harvesting robot system developed in Shanghai Jiao Tong University, China [127]

1.2 The History and Current Situation of the Development …

57

Table 1.5 Timeline of the development of cucumber harvesting robots Country Organization

Prototype

Vehicle

(Region)

Japan

University of Tokyo

-

Manipulator

End-effector

DOF

DOF

-

3

Method of recognition & localization

-

Research phase

Partial prototype

ISEKI Company Wheeled

Japan Okayama

6(5 rotation+ 1prismatic)

3

Dual-wavelength vision sensor

Field test

University

To-hand Camera+ Netherlands

Wageningen University

In-hand Railed

7(Commercial)

3

camera+Flash

Field test

tube filters

China China

Agricultural

-

University

6(4 rotation+ 2prismatic)

1

-

Partial prototype

China Agricultural

Near-infrared

University China

Crawler-type

4(4 rotation)

2(flexible)

Zhejiang

binocular

Field test

+In-hand camera

University of Technology

China

Japan

Shanghai Jiao Tong University

Ehime University

-

Railed

3(1 rotation+ 2 prismatic)

-

2

-

Binocular vision

Laser range sensors+ Ultra sound sensors

Partial prototype

Field test

58

1 History and Present Situations of Robotic Harvesting Technology: A Review

Fig. 1.69 The eggplant harvesting robot for V-shaped training developed in Department of Fruit Vegetables, National Institute of Vegetable and Tea Science, Japan [128, 129]

harvest one eggplant, and the execution time for judging the fruit size, the approach, and the fruit apex detection was 46.1 s and accounted for major contribution of the total time span [129, 130]. Hayashi S. et al. modified the prototype based on the abovementioned [130, 131] (Fig. 1.70). The end-effector is changed to a grip-cut integrated structure, and the ultrasonic distance sensor is added while retaining the in-hand camera. The DOF of the manipulator is increased to 7. The greenhouse field test showed that the successful harvesting rate averaged 29.1% and the undersized-fruit harvesting rate averaged 3.5% [130, 131]. The main cause of the low success rate was that leaves occluded the marketable-sized fruit. The harvesting execution time averaged 43.2 s per fruit, and the total harvesting time in the experimental row ranged from 16 to 53 min [130, 131].

Fig. 1.70 The modified eggplant harvesting robot for V-shaped training by Hayashi et al. [130, 131]

1.2 The History and Current Situation of the Development …

59

Fig. 1.71 The robotic eggplant harvesting hand–arm system developed in Weifang University and China Agricultural University, China [132–134]

In addition, Song J. of Weifang University and Zhang T., Liu C. of China Agricultural University also introduced the design of the opening eggplant harvesting robot system [132–134] (Fig. 1.71). It uses the 4-DOF joint manipulator, and a single motor-driven grip-cut integrated end-effector with a camera in hand was developed. The success rate and average time consumption in harvesting experiments were found 89% and 37.4%, respectively, but no specific test conditions and methods were introduced (Table 1.6). 4.

Sweet pepper harvesting robots

At present, the world’s total output of dry and fresh peppers has exceeded 60 million ton, becoming the third largest vegetable crop in the world after bean and tomato. Sweet pepper, as an important fresh vegetable, has been widely welcomed in Europe, America, Asia, and Oceania. The research on the robotic harvesting of sweet pepper has gained more impetus in Japan and the European Union. Kochi University of Technology of Japan has carried out continuous research on the robotic harvesting of sweet pepper. The overall size of the first developed prototype by Kitamura S. et al. is 1000 × 550 × 1400 mm (Fig. 1.72). It uses four-wheel electric chassis and 3-DOF Cartesian manipulator to move the binocular vision system and the scissor to move synchronously [135–139]. Then the motor drives the scissor to rotate horizontally, and the fruit is cut and falls. The LED light is installed in the CCD camera and a distinction method for the fruit of sweet pepper

60

1 History and Present Situations of Robotic Harvesting Technology: A Review

Table 1.6 Timeline of the development of eggplant harvesting robots Country Organization

Prototype

Vehicle

(Region)

Manipulator End-effector DOF

DOF

5

2

Method of recognition & localization

Research phase

Department of Fruit Vegetables, National Japan

Institute of Vegetable and

Crawler-type

Tea Science

CCD camera in

Partial

hand

test

Ehime University Department of Fruit In-hand CCD

Vegetables, National Japan

Institute of Vegetable and

Crawler-type

7

1

camera+

Field test

Tea Science distance sensor Ehime University

Weifang University, China

China Agricultural

-

4

1

CCD camera in hand

-

University

Fig. 1.72 The Cartesian-type sweet pepper harvesting robot developed in Kochi University of Technology, Japan [135–139]

using the reflection of LED light is proposed [137]. Experiments showed that both the recognition and cutting were good without leaves, but the occlusion of branches and leaves would disturb the operation of recognition and cutting. Bachche S. of Kochi University of Technology, Japan then developed a jointarm sweet pepper harvesting robot for inclined trellis training system [140–142]

1.2 The History and Current Situation of the Development …

61

(Fig. 1.73). In the crawling chassis equipped with line tracing system, the 2-DOF manipulator is installed, and the binocular vision system with LED light is still used to identify and locate the fruit. The robot first configured a single-DOF grip-cut integrated end-effector and then developed a new type of thermal cutting system, which has two electrodes connected by Nichrome wire (Fig. 1.74). When the vision system detects the sweet peppers to be harvested, 12 V DC input was supplied to the device which produces 300 V AC with 50 Hz frequency at 20 mA current as output [140–142]. At the same time, servomotor drives the notch plate with electrodes and gripper bars to move inward together, until the stem of fruit contacts with electrodes. Then the high voltage causes vaporization of water from fruit stem and results in the detachment. The system still needs to be tested for the whole machine.

Fig. 1.73 The sweet pepper harvesting robot for inclined trellis training system developed in Kochi University of Technology, Japan [140–142]

Fig. 1.74 The thermal cutting end-effector

62

1 History and Present Situations of Robotic Harvesting Technology: A Review

The European Union launched a large European FP7 CROPS project in 2010– 2014, and 14 organizations from different member countries jointly develop advanced crop robots. The Wageningen University and Research Center, Applied Plant Research (WUR) led the development of the sweet pepper harvesting robot. The platform can move in between the crop rows on the greenhouse rail system. A 9DOF redundant and modular manipulator is designed which consists of a vertical moving slide rail and eight rotational joints. Two different types of end-effectors for detaching sweet pepper fruits from the plant were realized. The first Fin-ray gripper features a combined grip-and-cut mechanism, while the second lip-type end-effector first stabilizes the fruit using a suction cup after which two rings with a lip each enclose the fruit and cut the fruit peduncle [143]. Both types of end-effectors have integrated LED illumination as well as two cameras: a miniature remote head RGB color camera and a small-size 3D TOF (time of flight) camera to realize the recognition and location of fruit [143]. The system was tested under simplified laboratory conditions with unconcluded single fruits, and it was found 97% of all fruits could be detected, while 86% could be reached and 79% picked [144]. The robot developed was able to harvest sweet peppers in a commercial greenhouse, but at limited success rates: harvest success was 6% when the fin ray end-effector was mounted, and 2% when the lip-type end-effector was mounted [145]. However, the lip-type end-effector causes smaller stem damage. The cycle time per fruit was commonly 94 s, which is too long compared with an economically feasible time of 6 s. Motion planning and execution were especially time-consuming because of a low speed and because of the waypoints required [144] (Fig. 1.75). Lehnert C. et al. of Queensland University of Technology, Australia, has developed two prototypes of sweet pepper harvesting robot. The first prototype of the QUT harvester consists of a 6-DOF robotic arm from Universal Robots and a customized harvesting tool mounted on its end-effector (Fig. 1.76a). The robot arm is mounted to a scissor lift giving an extra two passive DOF. The modified prototype consists of a custom wheeled differential drive mobile base with a custom end-effector and

Fig. 1.75 The sweet pepper harvesting robot developed in CROPS project, European Union [143, 144]

1.2 The History and Current Situation of the Development …

63

Fig. 1.76 The sweet pepper harvesting robot developed in Queensland University of Technology, Australia [145, 146]

7-DOF manipulator (6-DOF articulated arm + lift joint) [145, 146] (Fig. 1.76b). The end-effector is able to grip sweet peppers with a suction cup and then cut them free from the plant using an oscillating blade [146]. The end-effector also contains a real sense RGB-D camera for perceiving the crop and a micro-switch for checking whether the suction cup is coupled with the cutting blade [146]. Initial field trials in protected cropping environments demonstrate the efficacy of this approach achieving a 46% success rate for an unmodified crop, and 58% for modified crop [146]. Vitzrabin E., from the Ben-Gurion University of the Negev, Israel, introduced the study of the sweet pepper harvesting robot [147] (Fig. 1.77). The robot is based on the fusion of color and depth information of the Kinect RGB-D camera to realize

Fig. 1.77 The sweet pepper harvesting robot developed in Ben-Gurion University of the Negev, Israel [147]

64

1 History and Present Situations of Robotic Harvesting Technology: A Review

the recognition and location of the fruit, and the fruit is picked by the 6-DOF Cyton gamma multi-joint robot. However, the details of the robot and operation test are not introduced (Table 1.7). Table 1.7 Timeline of the development of sweet pepper harvesting robots Country Organization

Prototype

Vehicle

Manipulator DOF

(Region)

End-effector DOF

University of

localization

Wheeled

3(Cartesian)

Crawler-type

2

Railed

1

+LED light source

Kochi University of

Union

Binocular vision

Partial

cutting

+LED light source

test

9

2

TOF+CCD

Field test

Wheeled

7(6articulated+1lift)

2

RGB-D

Partial

camera(RealSense)

test

Wheeled

7(6articulated+1lift)

2

-

-

-

Wageningen University, et al

Queensland Australia

University of Technology

Queensland Australia

University of Technology

Ben-Gurion Israel

Partial

1+Thermal

Technology

European

phase

test

Technology

Japan

Research

Binocular vision

Kochi Japan

Method of recognition &

University of the Negev

RGB-D camera(RealSense)

Field test

RGB-D

Partial

camera(Kinect)

test

1.2 The History and Current Situation of the Development …

65

1.2.4 Other Fruit Harvesting Robots Researchers from various countries have also developed other kinds of harvesting robots for other fruits or vegetables, laying an important foundation for promoting the progress of the technology. 1.

Fruit harvesting robots for orchard

In Japan, Kondo N. et al. carried out the research and development of grape harvesting robot as early as the 90s of last century [148–150] (Fig. 1.78). The prototype for trellis training system adopts a crawler chassis and designs a 5-DOF polar coordinate manipulator (including one prismatic joint and four rotational joints). The wrist is equipped with a complex 3-DOF end-effector to grasp and cut a rachis, and the added pushing device which moved back and forth could make the end-effector grasp also too short rachis at harvesting time, could reduce bunch to swing at carrying time, and could orient the bunch at placing time [149]. In order to distinguish between the green fruit and the branch and leaf background, the vision system consists of several image sensors and optical filters, which can be used to distinguish the fruit by the reflected near-infrared images. It was observed in the experiment that the end-effector was able to grasp and cut rachis successfully, since the structure of the end-effector permitted the reasonable error, although detecting error by the vision sensor was about 20 mm. The pushing device also worked effectively except for too short rachis [148–150]. However, it took a long time for visual feedback control to recognize objects, and the CCD camera needs automatic exposure in the field [149]. In the 90 s of last century, Japan launched the research of watermelon harvesting robot. Lida M. et al., Kyoto University, developed a hydraulic-driven watermelon harvesting robot (Fig. 1.79a). A large five-joint manipulator and a four-finger envelope end-effector are installed on the, and the success rate of harvesting reached 65%

Fig. 1.78 The grape harvesting robot developed by Kondo et al. [148–150]

66

1 History and Present Situations of Robotic Harvesting Technology: A Review

Fig. 1.79 The watermelon harvesting robot developed in Kyoto University [150, 154]

[151–153]. Then the team also developed the “STORK” robot (Fig. 1.79a), and a four–link-type manipulator with a sucking end-effector was installed on the wheeled tractor. Two small CCD cameras are installed on the arm to form a stereoscopic vision system for identification and positioning. The test results show that the success rate of harvesting is 66.7% [154]. Hwang H. et al. of Sungkyunkwan University, Korea also developed a multifunctional tele-operative modular robotic system for greenhouse watermelon [155] (Fig. 1.80). It adopts the gantry-type mobile unit driven on the guided rail [155]. Four different kinds of end tools such as cutting tool for branch and fruit/flower pruning, compliant vacuum pad for watermelon turning, nozzle device for watering and pesticide application, and compliant harvest vacuum pad with stem-cutting

Fig. 1.80 The watermelon harvesting robot developed in Sungkyunkwan University, Korea [155]

1.2 The History and Current Situation of the Development …

67

device and load cell were developed [155]. The CCD color camera and 4-DOF Cartesian robot manipulator were attached to the front part of the gantry to ensure the proper workspace, and the stem-cutting scissor for harvesting was composed of pruning scissor and vacuum pad for watermelon turning [155]. According to the results of field test, at the stage of task execution without human intervention, the developed system showed an average 15 s to harvest and load each watermelon [155]. Xia H. of South China Agricultural University, Wang H. of National Research Center of Intelligent Equipment for Agriculture, and Zhang R. of Zhongkai University of Agriculture and Engineering have carried out the design or development of the automatic pineapple picking machine or robotic end-effector, respectively, but the complete prototype and experiment have not been formed [156–158]. Coconut is an important tropical woody oil crop in Southeast Asia. Its plant is high up to 15–30 m, so the fruit harvesting can only rely on manual climbing, and there is great difficulty and danger. Therefore, the research of coconut harvesting robot has been emphasized and carried out. WAN I. et al. of University Putra Malaysia developed a semi-automatic coconut harvester [159, 160] (Fig. 1.81a). A Kubota L3010 hydrostatic drive tractor is modified to realize unmanned tractor control, and large hydraulic manipulator was installed. Various end-effectors or robot fingers were designed and fabricated, such as hydraulic two-finger gripper, two-finger gripper with a cutting tool, and V-shaped cutter is driven by a hydraulic motor. The tractor, manipulator with different end-effectors, and binocular vision system constitute a coconut picking robot (Fig. 1.81b), and the research on single CCD recognition and spatial location based on webcam has also been carried out. Dates are rich in Western Asia and North Africa, with a height of up to 35 m. Saudi Arabia is at the forefront of the world production of dates, producing about one

Fig. 1.81 The semi-automatic coconut harvester and coconut harvesting robot developed in University Putra Malaysia [159, 160]

68

1 History and Present Situations of Robotic Harvesting Technology: A Review

Fig. 1.82 The date palm harvesting robot developed in King Saud University, Saudi Arabia [161]

million tons of dates annually [161]. The unique nature of the date palm tree requires highly labor-intensive orchard operation under hazardous condition [161]. Aljanobi A. of King Saud University developed a Mobile Robotic Unit for Date Harvesting (MRUDH). A Caterpillar Telehandlers TH62 lifter (maximum lift heights 7.2 m) is assembled on the Caterpillar Telehandlers forks, and its special platform carries the robotic arm of Neuronics and control box (Fig. 1.82). However, both the detection and control methods are still not known to form the literature. Razzaghi E. et al. of Iran Tehran University carried out the research and development of the lifting harvesting robot for date palm [162] (Fig. 1.83). The U-shaped frame is set as a fixed rail, and the installation of the three-prismatic-joint manipulator on the rail-type lifting platform is done, so as to pick up the high date palm. But the specific methods of detection and control are unknown. Shokripour H. of University Putra Malaysia [163] (Fig. 1.84a), Wibowo T. of Politeknik Elektronika Negeri Surabaya in Indonesia [164] (Fig. 1.84b), Mani A. of Anna University in India [165] (Fig. 1.85a), Srinivas A. of PSG College of Technology in India [166] (Fig. 1.85b), and Dubey A. of VIT University in India [167] (Fig. 1.86), respectively, developed different coconut tree climbing and harvesting robots. Shokripour H. uses a wheeled design instead of legs for climbing and installs a rack-and-pinion reciprocating saw blade on the climbing robot. Senthilkumar S. directly installed the motor-driven disk sawing knife on the tree climbing robot, and Mani A. installed a 3-DOF manipulator and a disk sawing knife on the tree climbing robot. Wibowo T. installed the motor-driven vertical slide bar on the tree climbing robot, a specifically designed saw and wireless camera are attached at the top of the bar. Dube A.P. uses a hollow, cylindrical steel pipe driven through bevel gear transmission to realize the vertical motion, and a disk sawing knife is mounted on top of the

1.2 The History and Current Situation of the Development …

69

Fig. 1.83 The date palm harvesting robot developed in Iran Tehran University [162]

Fig. 1.84 The coconut tree climbing and harvesting robots developed in University Putra Malaysia and Politeknik Elektronika Negeri Surabaya, Indonesia [163, 164]

cylindrical part. Megalingam R. et al. of Amrita Vishwa Vidyapeetham University, India also proposed to install the two-joint manipulator and the disk sawing knife on the climbing robot, and to achieve the harvesting of coconut by remote control [168]. Abraham A. et al. of East Point College of Engineering and Technology, India, have proposed the idea of a new insect crawling robot [169, 170] (Fig. 1.87). The robot is equipped with double cameras on the front of the head as eyes. It has six legs on the body, of which four are used to crawl up the trunk, and the two front legs are

70

1 History and Present Situations of Robotic Harvesting Technology: A Review

Fig. 1.85 The coconut tree climbing and harvesting robots developed in Anna University and PSG College of Technology, India [165, 166]

Fig. 1.86 The coconut tree climbing and harvesting robots developed in VIT University, India [167]

used for harvesting. One front leg has three human-like fingers, on a rotating spindle, which helps the coconut bunch to be pushed away and clear any obstacle like a dry leaf in the robot’s path, during climbing and harvesting [169, 170]. The other front leg has a sharp circular high-speed rotating blade for harvesting the desired coconuts [169, 170].

1.2 The History and Current Situation of the Development …

71

Fig. 1.87 The “Insect” coconut tree climbing and harvesting robots developed in East Point College of Engineering and Technology, India [169, 170]

2.

Harvesting robots for vegetables

A selective asparagus harvester was developed by Clary C. of Washington State University [171] (Fig. 1.88). Using the upper pulse laser to detect the height of asparagus, the position of asparagus was determined with the lower pulse laser, and the selective harvest of the single shoot was carried out. A single harvester head is

Fig. 1.88 The selective asparagus harvester developed in Washington State University [171]

72

1 History and Present Situations of Robotic Harvesting Technology: A Review

Fig. 1.89 The asparagus harvesting robot of Industrial Technology Center of Nagasaki, Japan [172]

mounted on the front of a motorized carrier unit and consists of four components: spear detection, spear cutting, spear capture, and harvester head height control [171]. The upper laser is mounted at 23 cm to detect spears ready for harvest, and the lower laser is used to determine the location of the spear at ground level in order to time the cut. It was concluded that with further improvements to the harvester, it would be successful at achieving an efficiency of 70% to 80% compared to hand harvesting [171]. Damage to the spears was not significant. Irie N. of the Industrial Technology Center of Nagasaki, Japan, developed a railed asparagus harvesting robot [172] (Fig. 1.89). Its 3D vision system, composed of two sets of slit laser projectors and one TV camera, recognizes the target asparagus to have desired length and measures the 3D position of the target asparagus. The robotic arm has 4 DOF related to the position, and the end-effector has 2 DOF for grasping and cutting asparagus. The experiment showed that the time required to harvest one asparagus was 11.9 s without considering the time taken for identification and location. Two generation robotic cabbage harvesters have been developed by the Murakami N. of the National Agricultural Research Center of Japan [173–175] (Fig. 1.90). The second-generation prototype adopts the hydraulic motor-driven crawler chassis of BRAIN Company, and the added encoder carries out the traveling distance measurement. Two hydraulic pumps are installed to drive the arm and the end-effector, respectively. The manipulator has 3 DOF, whose expansion is driven by hydraulic motor and ball screw. In the four-finger end-effector designed, two fingers are used to hold the cabbage, and the other two fingers are used to cut the stem. The prototype realizes the detection by the CCD camera on the top, and the action of the end-effector is activated by touching the leaves or the ground through the limit switch on the fingertip. The experiment shows that the success rate of harvesting is 39.2%, and the average time is 55 s.

1.2 The History and Current Situation of the Development …

73

Fig. 1.90 Two generation robotic cabbage harvesters developed in the National Agricultural Research Center, Japan [173–175]

Cho S. et al. of Seoul National University, South Korea, put forward a potted lettuce harvesting robot system in a plant factory [176] (Fig. 1.91). The system uses a lettucetransfer device to feed the plants into harvesting position, and uses the machine vision device to detect the leaf area and geometrical shape of lettuce plant and six photoelectric sensors attached to a vertical bar to detect the height. And then, the lettuce is picked by grip-cut integrated end-effector, and 3-DOF cylindrical-coordinate system manipulator moves the cut lettuce on the packing conveyor. However, no more details of the prototype were reported.

Fig. 1.91 The potted lettuce harvesting robot developed in Seoul National University, South Korea [176]

74

1 History and Present Situations of Robotic Harvesting Technology: A Review

Fig. 1.92 The lettuce harvesting robot developed in Shimane University [177]

Chung S. et al. of Shimane University, Japan developed a harvesting robot for ground-cultured lettuce [177] (Fig. 1.92). The robot includes wheeled chassis, stereoscopic vision system, end-effector, and transportation system and a collection box. The end-effector relies on two upper U-type fingers to grip the lettuce and lower V-shaped finger to realize the position judgment by force feedback to the ground. Finally, the lower blade cut the root off.

1.2.5 Other Harvesting Robots In addition to the harvesting of various kinds of fruits and vegetables, the research on the harvesting of tea, flowers, and other special crops has been carried out in succession. But in general, compared with the research of fruit and vegetable harvesting robots, the research results of other special crop harvesting robots are limited, and the continuation and depth of the development are still inadequate. Present harvesting robot prototypes for other special crops are still far from a complete system and operation application. 1.

Tea harvesting robots

Tea is a drink that is drank directly with boiled water. Tea, coffee, and cocoa, also known as the world’s three largest beverage, are widely consumed worldwide. At present, mechanized harvesting is only suitable for low-grade tea, but the harvesting of shoots and leaves of high-grade tea depends entirely on manual work. Because of the traditional consumption habits, people enjoy drinking the bagged broken tea in European and American, while Chinese has the tradition of drinking the original leaf tea. Therefore, the research oriented to the robotic harvesting of high-quality tea is mainly carried out in China. Chen J. et al. of Hefei University of Technology built the tea harvesting robot by installing the two-finger gripper to a parallel robot, and the color features obtained by the CCD camera are used to identify the leaf buds and the edge projections are

1.2 The History and Current Situation of the Development …

75

Fig. 1.93 The tea harvesting robot developed in Hefei University of Technology [178]

used to realize 3D reconstruction and measurement of the leaf buds [178] (Fig. 1.93). This structure of the parallel harvesting robot is simulated and analyzed by Gao F. et al. of Nanjing Forestry University [179, 180]. Li H., Lu X., and Xu L. of China Agricultural University developed the whole system of fresh tea harvesting robot [181] (Fig. 1.94). The crawler-type tractor is used as the chassis, and a CCD camera was installed on the top. A 4-DOF Cartesian manipulator and the gear-driven double-finger gripper were installed, and the tea is cut off by the fingertip. The team also carried out an improved design of a double arm tea picking robot. The left and right arms were arranged with 3-DOF Cartesian-type manipulator, and an end-effector was installed [182]. Qin G. et al. of Nanjing Research Institute of Agricultural Mechanization, Ministry of Agriculture, developed the 4CZ-12 self-walking intelligent tea harvesting robot [183] (Fig. 1.95). It uses a crawler chassis, and a 3-DOF Cartesian manipulator is lifted to the top of the tea canopy by a hydraulic cylinder. A single camera and a grating projector are configured to obtain the contour surface of the tender shoots of the tea tree. With the same structure of the fresh tea harvesting robot in China Agricultural University, the harvested end-effector was used to collect the tea in real time by the negative-pressure tea collection tube installed on the upper part of the shear mouth.

76

1 History and Present Situations of Robotic Harvesting Technology: A Review

Fig. 1.94 The fresh tea harvesting robot developed in China Agricultural University [181]

Fig. 1.95 The intelligent tea harvesting robot developed in Nanjing Research Institute of Agricultural Mechanization, Ministry of Agriculture [183]

1.2 The History and Current Situation of the Development …

2.

77

Flower harvesting robots

With the rapid rise of high-efficiency flower industry in the international range, the scale of its planting and output value is growing rapidly. The research on flower harvesting robot has begun to be paid more attention, although it started late. Noordam J. et al. of Agro technology and Food Innovations of Holland reported the development of the rose harvesting robot. The flowers are transported by the transportation line, and the ripeness judgment and positioning are realized by the binocular vision and the moving sheet of laser light, respectively. The vertical prismatic manipulator drives the end-effector to cut off the stem and then pulls out the flowers. The system has achieved a high level of 3D detection/location and automatic harvesting of tall and complex roses [184] (Fig. 1.96).

Fig. 1.96 The rose harvesting robot developed in Agrotechnology and Food Innovations of Netherlands [184]

78

1 History and Present Situations of Robotic Harvesting Technology: A Review

Fig. 1.97 The Gerbera Jamesonii harvesting robot developed in the Leibniz University of Hannover, Germany [185]

Thomas R., Leibniz University of Hannover, Germany, has developed a robotic harvesting system for the potted Gerbera Jamesonii [185] (Fig. 1.97). The bottom line of the main plant was attached to the upper edge of the transporting pallet, on which the plant was transported [185]. Images of plants were taken with a stereo camera system, which consisted of two high-resolution CCD cameras with near-infrared filters [185]. The harvesting work was realized by the industrial robot RV-E3NLM manufactured by Mitsubishi [185]. This robot had 6 DOF and was mounted onto an additional vertical linear axis for the expansion of its work space [185]. A pneumatic harvest grabber was developed, which harvested the pedicels by cutting them off [185]. In the harvest experiments, 80% of all pedicels could be harvested, but the average duration to harvest one plant reached about 10 min [185]. Antonelli M., University of L’Aquila, Italy, developed a mobile saffron harvesting robot [186] (Fig. 1.98). A new harvesting module for saffron flower was installed on Zaffy, the Agrirobot developed by Elettronica R. [186]. The new module is composed of a gripper, a vision system for the timing of the gripping step, a pneumatic system to move away from the leaves, and a suction system for the harvesting of the flower 187] . A photodetector and a scan plane compose the vision system [186]. The gripper is mounted on a vertical cylindrical pipe, and two fingers with a half-moon shape are joined at the endings by two hinges [186]. A vacuum cleaner powered by a battery on board of Zaffy chassis feeds the pneumatic system for leaves shifting and for the suction system [186]. The design of the machine is ingenious, and the success rate of automatic mobile harvesting of saffron reaches 60%. Kohan A. et al. of Islamic Azad University, Iran developed an arm–hand system for the harvesting of Rosa Damascena [187]. The 4-DOF manipulator is used, and a curved form clipper was designed, which has two appendages covered with soft

1.2 The History and Current Situation of the Development …

79

Fig. 1.98 The mobile saffron harvesting robot developed in the University of L’Aquila, Italy [186]

rubber to keep the flower after picking [187]. The vision system includes two CCD cameras forming a stereo vision system that can move matching, once an image has been segmented, the image along the horizontal axis relatively [187] (Fig. 1.99). Gürel C. of Atılım University, Turkey has carried out a conceptual design on rose harvesting robot. It is considered that the robot should have four main functions of “analyzing the rose”, “cutting the stem”, “move pot”, and “operational control”. A test platform was set up [188] (Fig. 1.100), but the development of the actual prototype and further research have not yet been carried out.

Fig. 1.99 The arm–hand system for the harvesting of Rosa Damascena developed in Kohan A. of Islamic Azad University, Iran [187]

80

1 History and Present Situations of Robotic Harvesting Technology: A Review

Fig. 1.100 The test platform of harvesting robot developed in Atılım University, Turkey [188]

3.

General-purpose, multi-functional, and harvesting robots

In addition to the vigorous development of all kinds of special harvesting robots, there are a number of researches on general-purpose and multi-functional picking robots. (1)

General-purpose harvesting robots

The robot structure, detection, and control method can be applied to the harvesting operation of multiple varieties of fruit, which will greatly improve the versatility of the robot equipment. Even if many harvesting robots are designed for special fruit such as apple and tomato at present, the overall or key part of the equipment structure and detection control scheme will be effectively extended to the similar category of crops. The versatility of robots is of great importance to farmers and equipment manufacturers. The Agribot, developed by Ceres R., Institute of Automatic Industrial, Spain, is suitable for harvesting apples, oranges, tomatoes, and other spherical fruits [189, 190] (Fig. 1.101). The operator drives the vehicle which carries two picking arms

1.2 The History and Current Situation of the Development …

81

Fig. 1.101 The spherical fruit harvesting robot developed in Institute of Automatic Industrial, Spain [189, 190]

and detects every piece of fruit pointing to it with a laser spot [190]. The fruit localization module which consists of a point laser range-finder and the tilt/pan mechanism is directed by a human operator [190]. The gripper has been specifically designed. Firstly, a local approximation is solved by a passive V-shaped centering device acting in a horizontal higher plane relative to the fruit, guiding the stem to the V vertex where the cutting tool is placed [190]. Then, a suction pneumatic cup is used to prevent from moving during cutting operation [190]. In order to actually detect whether the fruit has been grasped, a pressure sensor is used [190]. Finally, once the fruit is detached, it is foreseen to be released and guided by a flexible pipe collector to the vehicle where it is stored [190]. The general harvesting arm–hand system of kiwi and apple developed by Kahya E., Namık Kemal University, Turkey [191] (Fig. 1.102), uses a 4-DOF manipulator and a pneumatic scissor with a vacuum sucker. A 2D camera and an ultrasonic sensor are applied to realize the recognition and location of the fruit, and the preliminary test is completed.

82

1 History and Present Situations of Robotic Harvesting Technology: A Review

Fig. 1.102 The general-purpose harvesting arm–hand system developed in Namık Kemal University, Turkey [191]

Xiong J., South China Agricultural University, has developed a multi-type fruit harvesting robot [192] (Fig. 1.103). This robot can achieve the harvesting of either single fruit or cluster of fruits. The robot consists of a GPS navigation car, a 4DOF manipulator, a grip-cut end-effector, and a binocular stereo vision system. The experiment of robotic harvesting of litchi clusters and single citrus fruits was carried out, and the success rate is 80% for litchi and 85% for citrus, respectively. In addition, Kawamura N. and Yukawa N. et al. of Kyoto University, Japan put forward the structure of fruit harvesting robot [9, 193] (Fig. 1.104); Fujiura T. et al. of Shimane University proposed the structure of harvesting robot for orchard [194– 196] (Fig. 1.105); Kondo N. et al. of Okayama University proposed the 8-DOF manipulator structure [197] (Fig. 1.106); He B. of China Agricultural University proposed auto-recognition of navigation path in orchard for harvest robot based on machine vision [198] (Fig. 1.107); and Font D. of Lleida University of Spain

Fig. 1.103 The multi-type fruit harvesting robot developed in South China Agricultural University, China [192]

1.2 The History and Current Situation of the Development …

83

Fig. 1.104 Two types of fruit harvesting robots developed in the Faculty of Agriculture, Kyoto University, Japan [9, 193]

Fig. 1.105 The harvesting robot for orchard developed in Shimane University, Japan [194–196]

Fig. 1.106 The 8-DOF manipulator for harvesting developed in Okayama University, Japan [197]

proposed an automatic fruit harvesting by combining a low-cost stereovision camera and a robotic arm [199] (Fig. 1.108). All the above researches do not specifically target specific fruit and vegetable objects but have obtained certain research results for the general technology of harvesting robots.

84

1 History and Present Situations of Robotic Harvesting Technology: A Review

Fig. 1.107 The harvest robot and auto-recognition of navigation path for orchard developed in China Agricultural University [198]

Fig. 1.108 The low-cost automatic fruit harvesting system developed in Lleida University, Spain [199]

(2)

Multi-functional robots

Different working parts, including harvesting, can be changed or integrated into the same robot body to meet the needs of various functions. This kind of robot can better adapt to the practical situation of multiple links and strong seasonality in an agricultural operation, and effectively improve the equipment utilization and market competitiveness. However, undoubtedly its technical difficulty has also significantly increased. Kondo N. et al. of Okayama University, Japan developed a multipurpose agricultural robot working in the vineyard [150] (Fig. 1.109). Four end-effectors for harvesting, berry thinning, spraying, and bagging were developed based on the unified crawler chassis, 5-DOF manipulator, and vision system, respectively [150]. By replacing the end-effector, it meets the needs of multi-link operation of grape planting management. Hwang H. of Sungkyunkwan University, South Korea, developed a multifunctional tele-operative modular robotic system for greenhouse watermelon [155] (Fig. 1.110). On the gantry system equipped with 4-DOF Cartesian-type robotic

1.2 The History and Current Situation of the Development …

85

Fig. 1.109 The multipurpose agricultural robot working in vineyard developed in Okayama University, Japan [150]

Fig. 1.110 The multi-functional tele-operative modular robotic system for greenhouse watermelon developed in Sungkyunkwan University, South Korea [155]

manipulator, pruning, watering, pesticide application, and harvest operation are realized with exchangeable modular-type end-effectors [155]. Among them, the combination of the pruning scissors and sucker turning parts can achieve the harvesting. This design idea makes the device design more efficient in integration. Arima S. et al. of Ehime University in Japan developed a multi-operation robot for ridge-top strawberry [90] (Fig. 1.111). On the gantry four-wheeled platform, the spray nozzles and harvesting arm–hand system can be mounted, respectively, to realize spraying and harvesting operation. The robot can coordinate with a mobile grading robot which follows just behind to realize immediate grading of harvested fruits by the harvesting robot. Fujiura T. of Osaka Prefecture University, Japan developed a multi-functional robot for mobile tomato cultivation facility [200] (Fig. 1.112). Tomato plants cultivated in the beds also move with the bed movement and tomato stems are trained along the poles [200]. It consisted mainly of a 5-DOF manipulator, a 3D vision sensor,

86

1 History and Present Situations of Robotic Harvesting Technology: A Review

Fig. 1.111 The multi-operation strawberry robot developed in Ehime University, Japan [90]

Fig. 1.112 The multi-functional robot for mobile tomato cultivation facility developed in Osaka Prefecture University, Japan [200]

plant training end-effector, defoliation end-effector, and harvesting end-effector. Comprehensive experiment of the robot system is necessary for the actual work. Dohi M. of the Hyogo Prefectural Institute of Agriculture, Japan developed a transplanting and harvesting dual-purpose robot [201] (Fig. 1.113). The detection and height judgment of the seedlings were carried out by optical fiber sensors. The transplanting of the cabbage and spinach seedlings was realized by installing seedling box and transplanting end-effector on the body of the vehicle. Also, the harvesting of leaf vegetable was realized by installing harvesting end-effector.

1.2 The History and Current Situation of the Development …

87

Fig. 1.113 The transplanting and harvesting dual-purpose robot developed in Hyogo Prefectural Institute of Agriculture, Japan [201]

Dohi M., which moved to Hyogo Prefectural Institute of Agriculture, then further increased the weeding function based on the above robot [202] (Fig. 1.114). The machine uses an electric four-wheel chassis, and two weeding end-effectors were designed. One weeding end-effector was attached to a weed knife with a spiral shape, and the other was attached to three weeding knives. To configure a vertical rotating spiral weeding knife. The color images of weeds are obtained by a color camera, and a stereo camera is used to locate the weeds.

Fig. 1.114 The multipurpose robot for vegetable production developed in Hyogo Prefectural Institute of Agriculture, Japan [202]

88

1 History and Present Situations of Robotic Harvesting Technology: A Review

Fig. 1.115 The 9-DOF redundant modular multipurpose agricultural manipulator developed in Technische Universitat Munchen, Germany [203]

Baur J. et al. of Technische Universitat Munchen, Germany, developed a 9-DOF redundant modular multipurpose agricultural manipulator [203] (Fig. 1.115). The manipulator is designed as an agricultural robot to fulfill several tasks like harvesting, spraying, or pruning for different types of fruits or plants to achieve dexterous manipulation and to avoid obstacles.

1.3 Summary and Prospect 1.3.1 The Continuous Progress of Robotic Harvesting Technology In developed countries, because of limit of the labor number, cost, and the needs of high-quality fresh food and vegetable and fresh-cut flowers, the robotic harvesting technology has been carried out earlier. Particularly in Japan, compared with Europe and the United States, the robotic harvesting technology is highly valued in view of its small-scale cultivation and high-quality agricultural products. The research organizations involve universities, research institutions, and agricultural machinery enterprises, with a large number of researchers and research achievements, which greatly promoted the development level of this technology. In the United States, the advantages of agricultural resources are prominent, so the development of large-scale, high-yield, and low-cost agriculture is vigorously promoted. Therefore, robot harvesting technology has long been dominated by space station operations supported by NASA. But the mechanical and robotic harvesting of citrus in Florida (famous “Citrus State”) has lasted for decades, and the distinctive features of the robotic citrus harvesting technology have been formed.

1.3 Summary and Prospect

89

Fig. 1.116 Changing trend of patent numbers in agricultural robots in Chinese Mainland and abroad

Europe has a high contribution in robotic harvesting. On the basis of scattered studies such as Holland, Spain, Italy, and on behalf of the CROPS Project of the European Union, cooperation on the robotic harvesting of sweet peppers and other characteristic fruits and vegetables is carried out. Its technical research and development have reached a high level. The research on harvesting robot technology started late in China. However, with the disappearance of the “demographic dividend”, the shortage of labor force has quickly become a bottleneck restricting the development of agricultural development, especially the development of labor-intensive fruit and vegetable industry. Since the beginning of this century, the input and achievements of agricultural robot technology, most are about harvesting robot technology, in China have become a geometric growth trend (Fig. 1.116). China Agricultural University, Jiangsu University, Zhejiang University, South China Agricultural University, Chinese Academy of Agricultural Mechanization Sciences, and National Research Center of Intelligent Equipment for Agriculture have become the leading unit to promote the research of the harvesting robot technology. The research and development level of China’s harvesting robot has basically achieved synchronization with the international arena and has been leading in a few sub-fields.

1.3.2 Technical Keys to the Development of Harvesting Robot Technology 1.

Adaptability to complex and unstructured environment

Highly complex and unknown plant environment has brought great challenges to the sensing and operation of the harvesting robot. Although scholars have carried out

90

1 History and Present Situations of Robotic Harvesting Technology: A Review

many types of researches on such topics as the path search and navigation in greenhouse [204–208], the fruit identification under the overlap or occlusion [209–214], the path planning and obstacle avoidance in the complex canopy environment [209– 214], and the harvesting under large difference of peduncle size and posture [1, 17, 215], the operation of the harvesting robot in the field is still difficult to achieve the ideal experimental results after the manual environmental simplification [73, 89, 96, 129]. The sensing and adaptability of robots to complex and unstructured environments still remain a bottleneck problem affecting the harvesting robot technology’s maturity and application. 2.

Fusion of complex robot system

The harvesting robot is a complex system composed of the moving chassis, manipulator, end-effector, and navigation and detection unit. The performance of the harvesting robot depends not only on each unit but also on the deep fusion of the whole system. At present, some studies have been carried out in the arm–hand coordination [29, 216], body–hand coordination [217], hand–eye coordination [218], arm–eye coordination [219–223], body–eye coordination [224], etc., but there is still a certain distance from the actual application performance. For example, the visual servo control is still based on the open loop and the “looking then moving” model which relies on the precise modeling. Furthermore, the fusion between vehicle navigation and fruit detection; the precision position/posture compensation; and the action parallel interaction among hand, arm, and vehicle still need to be deeply studied and gradually overcome.

1.3.3 The Historical Characteristics of the Technology Development of the Harvesting Robots 1.

The determinant of the cultivation pattern to the technical characteristics

Crop cultivation mode has a decisive influence on the robot structure scheme, technical difficulty, and performance. Arima S. and Hayashi S. et al. have discussed the requirements for the cultivation mode of robot harvesting [118, 225, 226]. In the traditional cup-type planting, the tomato fruit is very scattered, and the robot needs much larger workspace. At the same time, the spatial distribution of the branches makes the harvesting work very difficult. While for the single trunk cultivation and the high-wire cultivation mode [20, 31, 36], fruits are cultivated in the vertical direction through the rod or rope support, several fruits are suspended in clusters, so the recognition and harvesting of fruit are greatly simplified, and the performance of harvesting has been effectively guaranteed [5] (Fig. 1.117). For the different cultivation patterns of strawberry, such as the ground, ridge, and elevated cultivation, because of the significant difference in the distribution

1.3 Summary and Prospect

(a)

91

(b)

Fig. 1.117 Single trunk and high-wire cultivation. a single trunk cultivation [20, 31], b high-wire cultivation [36]

of strawberry, the posture of fruit stalk, the relationship between fruit and leaf, the detection background, and so on, the scheme of the harvesting robot is very different. In different mode, the elevated cultivation is not only welcomed as a new laborsaving mode, but also because of its prominent leaf separation, two-dimensional fruit distribution, the vertical orientation of fruit stalk, and the simplification of the background [5]; not only the process of fruit detection is greatly simplified [96, 99], but also the harvesting can be realized through simpler Cartesian manipulator and horizontal grip-cut structure [1, 93, 99, 103, 105, 112]. As a result, the success rate and efficiency of robotic harvesting are greatly improved. At the same time, in view of different mobile cultivation facilities, the stationary harvesting robot and harvesting operation can be realized [7, 23, 107, 108, 227], which is quite different from most harvesting robots which need the help of the chassis movement. Thus, both the complexity of the robot system and the difficulty of manipulation are effectively simplified, which has become an important direction for its development (Fig. 1.118). 2.

The boost of mechanical and electrical technology revolution to technological breakthrough of harvesting robots

The high complexity of the harvesting robot technology leads to its long development cycle, high cost, poor performance, and reliability, which greatly affects the progress of technology and the process of application and popularization. It is gratifying that the rapid development of mechanical and electronic technology has greatly accelerated the research of harvesting robot technology, and the development cycle is shortened and the structure of the robot is also more refined. In recent years, some key modules, such as the manipulator and the chassis, are rapidly mature,

92

1 History and Present Situations of Robotic Harvesting Technology: A Review

Fig. 1.118 Moveable cultivation unit. a Mobile cultivation pot of cherry tomato [23], b Mobile cultivation bed of strawberry [108]

customized, and commercialized. The sensors, such as distance, vision, and force, and motor systems are miniaturized continuously. Also, the 3D light field camera [228, 229] and the RealSense depth camera [145] are used to achieve the low-cost real-time reliable fruit recognition it is possible to locate. RFID [230, 231], laser cutting [232, 233], and other new technologies are rapidly introduced into robot navigation and fruit detaching. The maturating and modularization of algorithms and tools such as image processing and robot motion solving simplify the implementation of detection and control. All these mechanical and electrical technology revolution will enable the harvesting robot technology to be vigorously promoted and usher in a new breakthrough (Fig. 1.119).

Fig. 1.119 Advanced sensors. a 3D light field camera [228, 229], b RealSense [146, 147]

1.3 Summary and Prospect

3.

93

The urgency of the social and economic conditions on the application of harvesting robot technology

The large-scale popularization of a technology depends on both the maturity of its technology and the maturity of external conditions. When the economic development level and labor cost make the application of the harvesting robot technology have the demand urgency and the function irreplaceable, and highlight the higher return rate compared with the manpower input, the harvesting robot technology will usher in the spring of development.

1.3.4 The Breakthrough Points of the Technology Development of Harvesting Robots 1.

The industrialization of the agriculture environment

Compared with industrial robots, the unstructured environment of agriculture and the differentiation of individual objects are the key bottlenecks that restrict the performance and application of agricultural robots, which also makes the design of agricultural robots very difficult and of high cost. The practice of the development of agricultural robots also shows that, in the field of agriculture which is more closer to the structured and standardized industry, such as milking [234], grafting [235], grading [236], livestock and poultry cleaning, and other fields, the technical requirements are greatly reduced, and the actual application of robot technology has been realized. The development of harvesting robot technology must also be integrated with the improvement of planting mode. Closer to industrial production, more effective reduction of individual differences and canopy complexity will greatly promote the development and popularization of harvesting robot technology. On the premise of large-scale production, it should be combined effectively from variety selection, standardization of seedling nursing, industrialization of planting environment, and standardization of planting and management. Furthermore, to replace the frequent replacement of many kinds of fruits and vegetables with more specific and efficient production, so as to eliminate the key constraints and obstacles for the realization of robot harvesting operation. 2.

Standardization of robot structure

Tractor, precision seeder, combine harvester, and other agricultural machinery, all of which have realized the standardization of the whole machine structure and components, so as to achieve mass production and large-scale application in the world. The development of harvesting robot technology is still in the stage of competitive

94

1 History and Present Situations of Robotic Harvesting Technology: A Review

exploration with different principles and different structures. Its maturity and application also objectively require the deep cooperation of the academic and the industry, and the deep integration of agricultural equipment and agronomy to promote the gradual formation of the robot structure standard according to the cultivation pattern and crop variety. Further, the demonstration of the robotic production mode will promote the matching and standardization of agricultural production. At present, the types and production environment of fruits and vegetables are complicated, and the principle and structure of harvesting robots are quite different, but their application is so limited. The standardization of a harvesting robot structure is the necessity of development, and it is believed that it will help to effectively break the current dilemma. 3.

Multi-robot collaboration

When the robotic harvesting is applied to actual production, it is necessary not only to complete the different actions of recognition, detaching, and placing in harvesting stage, but also to complete the logistics and placing of the fruits and vegetables, and even further completing the sorting, cleaning, and packing in the factory production. The production efficiency of harvesting, logistics, and placing by a single robot is bound to be very low, even without considering the post-harvest processing of agricultural products. Therefore, considering the parallel harvesting operation of multi-hand [237], multi-arm [38, 238, 239], multiple-robot [240], and the collaborative operation between the harvesting robot and the logistics robot or conveyor [241, 242] will be inevitable for the development of robot harvesting technology. It will greatly promote the development and application of harvesting robot technology. 4.

Human–robot cooperation

In addition to identifying and harvesting action, there are many processes such as fruit placing, navigation, turning around, or line transfer logistics. Furthermore, some auxiliary processes, such as charging and maintenance of equipment, are also necessary. However, the human–robot cooperation technologies, such as the whole-process remote control [10, 243] or motion sensing manipulation [244, 245], are difficult to meet the needs of high efficiency, labor saving, and easy harvesting operation. To build the reasonable system, human–robot cooperation mode is essential. As a basic principle, we may hand over most of the working process composed of high-intensity and accurate operation tasks, such as positioning and harvesting, to robots. However, a very small amount of tasks that need complex decision-making but is easy to perform manually may be realized with the manual intervention [246]. The realization of reasonable human–robot cooperation will significantly reduce the complexity of the autonomous operation and accelerate its practical application.

1.3 Summary and Prospect

95

1.3.5 Key Fields of Technology Development of Harvesting Robots The author believes that meeting the three characteristics of “large amount”, “urgent need”, and “high efficiency” should become the key areas of the development of harvesting robot technology. 1.

Large-amount fruits and vegetables

The large fresh fruit and vegetable products, such as tomatoes, cucumbers, and sweet peppers, as the basic needs of residents, must be guaranteed effective supply. With the rapid development of society, the labor shortage and high cost will impact on the stable supply and price of fresh food and vegetable products. Therefore, the robotic operation will become the rigid demand for fruit and vegetable industry and social development. 2.

Hilly forest fruits

The harvesting of forest fruit from tall trees in a hilly area is difficult, challenging, and dangerous. The harvesting of fresh forest fruits must be free from the dependence on manual operations, and the harvesting of dangerous fruits will also become the focus of development. 3.

High value-added cash crops

Because of the high complexity of technology and high cost, in the current stage, the harvesting robot technology will receive economic returns only in the fields of harvesting of fresh-cut flowers, harvesting and binding of saplings, harvesting of cuttage seedlings, and harvesting of sapling seeds. The harvesting robot technology will also be welcomed and popularized by agricultural operators in the above fields.

References 1. Hayashi S, Shigematsu K, Yamamoto S et al (2010) Evaluation of a strawberry-harvesting robot in a field test [J]. Biosys Eng 105(2):160–171 2. Sanders K (2005) Orange harvesting systems review [J]. Biosyst Eng 90(2):115–125 3. Yin J, Mao H, Wang X et al (2006) Automatic Segement method for multi-tomato images under various growth conditions [J]. Trans CSAE 22(10):149–153 4. Yin H, Chai Y, Yang SX et al (2009) Ripe tomato recognition and localization for a tomato harvesting robotic system [C]. Proceedings of the international conference of soft computing and pattern recognition 5. Kondo N, Monta M, Noguchi N (2011) Agricultural robots-mechanisms and practice [M]. Kyoto University Press 6. Hayashi S, Sakaue O (1997) Basic operation of tomato harvesting system using robot: manufacture of two-finger harvesting hand with auxiliary cutting device and basic experiment for harvest [J]. Bull Nat Res Inst Veg Ornam Plant Tea, 133–142

96

1 History and Present Situations of Robotic Harvesting Technology: A Review

7. Monta M, Kondo N, Ting K et al (1998) Harvesting end effector for inverted single truss tomato production systems [J]. J Jpn Soc Agric Mach 60(6):97–104 8. Kawamura N, Fujiura T, Ura M et al (1985) Fruit harvesting robot [J]. J Japn Soc Agric Mach 47(2):237–241 9. Kawamura N, Namikawa K, Fujiura T et al (1984) Study on agricultural robot (Part 1)— microcomputer-controlled ManipulatorSystem for fruit harvesting [J]. J Jpn Soc Agric Mach 46(3):353–358 10. Takahashi Y, Ogawa J, Saeki K (2001) Automatic tomato picking robot system with human interface using image processing [C]. Proceedings of the 27th annual conference of the ieee of industrial electronics society 11. Monta M, Kondo N, Ting K (1998) End-effectors for tomato harvesting robot [J]. Artif Intell Rev 12(1):11–25 12. Kondo N, Ting K (1998) Robotics for Bioproduction systems [M]. Amer Soc Agric Eng (ASAE) 13. Krikke J (22005) Robotics research exploits opportunities for growth [J]. Pervas Comput IEEE 4(3):7–10 14. Yasukawa S, Li B, Sonoda T et al (2017) Development of a tomato harvesting robot [C]. Proceedings of the international conference on artificial life and robotics, 19–22 15. Yaguchi H, Nagahama K, Hasegawa T et al (2016) Development of an autonomous tomato harvesting robot with rotational plucking gripper [C]. Proceedings of the intelligent robots and systems (IROS) 16. Chen X, Chaudhary K, Tanaka Y et al (2015) Reasoning-based vision recognition for agricultural humanoid robot toward tomato harvesting [C]. Proceedings of the international conference on intelligent robots and systems 17. Kondo N, Nishitsuji Y, Ling P et al (1996) Visual feedback guided robotic cherry tomato harvesting [J]. Trans ASAE 39(6):2331–2338 18. Kondo N, Ting K (1998) Robotics for plant production [J]. Artif Intell Rev 12(1–3):227–243 19. Kondo N, Nishitsuji Y, Monta M et al (1996) position detection algorithm for robotic harveting small fruits in truss—comparison with human Being’s Eyesight [J]. J Jpn Soc Agric Mach 58(Supplement):489–492 20. Tanigaki K, Fujiura T, Akase A et al (2008) Cherry-harvesting robot [J]. Comput Electron Agric 63(1):65–72 21. Han L, Fujipura K, Katsuhiko M et al (1999) Mini tomato harvesting robot with 3D vision sensor [J]. J Jpn Soc Agric Mach 61(Supplement):389–390 22. Fujipura K (2010) Tomato harvesting robot with 3D vision sensor ( forefront of system technology in agriculture) [J]. Syst Control inform 54(4):149–154 23. Fujiura T, Wada T, Nishiura Y et al (2010) Robot for harvesting cherry tomatoes in mobile cultivation facilities (Part 2)-harvesting experiment [J]. J Jpn Soc Agric Mach 72(2):152–159 24. Fujiura T, Wada T, Nishiura Y et al (2010) Robot for harvesting cherry tomatoes in mobile cultivation facilities, 1: outline of the robot and 3-D image recognition [J]. J Soc Agric Mach 72(2):143–151 25. Fujiura T, Wada T, Nishiura Y et al (2010) Robot for harvesting cherry tomatoes in mobile cultivation facilities, 2: harvesting experiment [J]. J Soc Agric Mach 72(2):152–159 26. Kondo N, Yamamoto K, Shimizu H et al (2009) A machine vision system for tomato cluster harvesting robot [J]. Eng Agric Environ Food 2(2):60–65 27. Kondo N, Yata K, Iida M et al (2010) Development of an end-effector for a tomato cluster harvesting robot [J]. Eng Agric Environ Food 3(1):20–24 28. Kondo N, Taniwaki S, Tanihara K et al (2007) An end-effector and manipulator control for tomato cluster harvesting robot [C]. Proceedings of ASABE annual international meeting 29. Ling P, Ehsani R, Ting K et al (2004) Sensing and end-effector for a robotic tomato harvester [C]. Proceedings of the ASAE annual meeting 30. Chiu Y, Chen S, Lin J (2013) Study of an autonomous fruit picking robot system in greenhouses [J]. Eng Agric Environ Food 6(3):92–98

References

97

31. Chiu Y, Yang P, Chen S (2013) Development of the end-effector of a picking robot for greenhouse-grown tomatoes [J]. Appl Eng Agric 29(6):1001–1009 32. Chiu Y, Chen S, Lin J (2016) Study of an autonomous picking robot system for greenhouse grown tomatoes [C]. Proceedings of the 6th international symposium on machinery and mechatronics for agriculture and biosystems engineering (ISMAB) 33. Chiu Y, Chen S, Yang P et al (2012) Integrated test of an autonomous tomato picking robot [C]. Proceedings of the 7th CIGR Section VI international technical symposium “Innovating the Food Value Chain” Postharvest technology and Agrifood processing2012 34. Ji C (2014) Vision information acquisition for fruit harvesting robot and development of robot prototype system [D]. China Agricultural University 35. Ji C, Zhang Z, Yu J, et al (2013) Research on key technology of truss tomato picking robot in greenhouse [J]. Sciencepaper Online 36. Feng Q, Wang X, Wang G, et al (2015) Design and test of tomatoes harvesting robot [C]. Proceedings of the IEEE international conference on information and automation 37. Wang X, Wu P, Feng Q et al (2016) Design and test of tomatoes harvesting robot [J]. J Agric Mech Res 4:94–98 38. Zhao Y, Gong L, Liu C et al (2016) Dual-arm robot design and testing for harvesting tomato in greenhouse [J]. IFAC-PapersOnLine 49(16):161–165 39. Xiang R, Duan P (2016) Design and experiment of night lighting system for tomato picking robot [J]. Trans Chinese Soc Agric Mach 47(7):8–14 40. Wang X (2012) Studies on information aquisition and path planning of greenhouse tomato harvesting robot with selective harvesting operation [D]. Jiangsu University 41. Liang X, Miao X, Cui S et al (2005) Experiments of optimization and simulation on kinematics of a tomato harvesting manipulator [J]. Trans Chinese Soc Agric Mach 36(7):96–100 42. Blandini G, Levi P (1989) First approaches to robot utilization for automatic citrus harvesting [J]. Land Water Use, 1903–1907 43. Armada MA, Muscato G, Prestifilippo M et al (2005) A prototype of an orange picking robot: past history, the new robot and experimental results [J]. Ind Robot: Int J 32(2):128–138 44. Recce M, Taylor J, Plebe A et al (1996) Vision and neural control for an orange harvesting robot [C]. Proceedings of the international workshop on neural networks for identification, control, robotics, and signal/image processing 45. Plebe A, Grasso G (2001) Localization of spherical fruits for robotic harvesting [J]. Mach Vis Appl 13(2):70–79 46. Allotta B, Buttazzo G, Dario P et al (1996) A force/torque sensor-based technique for robot harvesting of fruits and vegetables [C]. Proceedings of the IEEE international workshop on intelligent robots and systems’ 90.’ Towards a New Frontier of Applications’ 47. Raparelli T (2005) Development of a picking device of an orange harvesting machine [C]. Proceedings of 6th international conference on fluid power, transmission and control, 335–339 48. HARRELL, R. Economic Analysis of Robotic Citrus Harvesting in Florida [J]. Transactions of the ASAE, 1987, 30(2), 298–304 49. Burks T, Villegas F, Hannan M et al (2005) Engineering and horticultural aspects of robotic fruit harvesting: opportunities and constraints [J]. HortTechnology 15(1):79–87 50. Pool T, Harrell R (1991) An end-effector for robotic removal of citrus from the tree [J]. Transactions of the ASAE 34(2):373–378 51. Harrell R, Slaughter D, Adsit P (1989) A Fruit-tracking system for robotic harvesting [J]. Mach Vis Appl 2(2):69–80 52. Hannan M, Burks T (2004) Current developments in automated citrus harvesting [C]. Proceedings of the american society of agricultural and biological engineers, No. 043087 53. Flood S (2006) Design of a robotic citrus harvesting end effector and force control model using physical properties and harvesting motion tests [D]. University of Florida 54. Lee B, Rosa U (2006) Development of a canopy volume reduction technique for easy assessment and harvesting of valencia citrus fruits [J]. Trans ASABE 49(6):1695–1703 55. Lee B, Rosa U, Cheet-Ancheri K (2006) End-effector for automated citrus harvesting [C]. Proceedings of the ASABE annual international meeting), No. 061143

98

1 History and Present Situations of Robotic Harvesting Technology: A Review

56. Macconnell T (2008) Robotic mechanical harvester for fresh market citrus for fresh market citrus [R]. Postharvest Annual Report 57. Aloisio C, Mishra R, Chang C et al (2012) Next generation image guided citrus fruit picker [C]. Proceedings of the IEEE international conference on technologies for practical robot applications (TePRA) 58. Fujiura T, Ura M, Kawamura N et al (2010) Fruit harvesting robot for orchard [J]. J Jpn Soc Agric Mach 52:35–42 59. Lee F, Wang J (1999) Development of robotic arm for citrus fruit harvesting [J]. J Agric Mach 8(3):1–8 60. Lu W, Song A, Cai J et al (2011) Structural design and kinematics algorithm research for orange harvesting robot [J]. J South Univ (Natural Science Edition) 41(1):95–100 61. Yao J, Xiong W, Yang L (2012) Design and optimization of robotic end-effector for citrus harvesting [J]. Sci Technol Inform 4:1–1 62. Zhang S (2011) Research om end-effector of citrus harvesting robot [D]. Zhejiang University of Technology 63. Peterson D, Bennedsen B, Anger W et al (1999) A systems approach to robotic bulk harvesting of apples [J]. Trans ASAE 42(4):871–876 64. Davidson J, Mo C (2015) Mechanical design and initial performance testing of an applepicking end-effector. [C]. Proc ASME IMECE2015 65. Davidson J, MO C (2014) Conceptual design of an end-effector for an apple harvesting robot [C]. Proceedings of the international conf. on automation technology for off-road equipment 66. Davidson JM, Silwal A et al (2015) Human-machine collaboration for the robotic harvesting of fresh market apples [C]. Proceedings of IEEE ICRA workshop on robotics in agriculture 67. Davidson J, Silwal A, Hohimer C et al (2016) Proof-of-concept of a robotic apple harvester [C]. Proceedings of the IEEE/RSJ international conference on intelligent robots and systems 68. Silwal A, Davidson J, Karkee M et al (2017) Design, integration, and field evaluation of a robotic apple harvester [J]. J Field Robot 34(6):1140–1159 69. Baeten J, Donné K, Boedrij S et al (2008) Autonomous fruit picking machine: a robotic apple harvester [C]. Proceedings of the field and service robotics 70. Nguyen T, Kayacan E, de Baedemaeker J et al (2013) Task and motion planning for apple harvesting robot [J]. IFAC Proc 46(18):247–252 71. Luo H, Wei G (2015) System design and implementation of a novel robot for apple harvest [J]. INMATEH-Agric Eng 46(2):85–94 72. Guo J, Zhao D, Ji W et al (2010) Design and control of the open apple-picking-robot manipulator [C]. Proceedings of the 3rd IEEE international conference on computer science and information technology (ICCSIT) 73. Guo J, Zhao D, Lv J et al (2011) Design and control of an apple harvesting robot [J]. Biosys Eng 110(2):112–122 74. Zhao D, Ji W, Chen Y et al (2014) Development and design of fruit harvesting robot [J]. Robot Techn Appl 5:16–20 75. Ma Q (2012) Research on the key technology of apple picking-robot [D]. Chinese academy of agricultural mechanization sciences 76. Gu B, Ji C, Wang H et al (2012) Design and experiment of intelligent mobile fruit picking robot [J]. Trans Chinese Soc Agric Mach 43(6):153–160 77. Su Y, Yang L, Song X et al (2016) Design and experiment of intelligent mobile apple picking robot [J]. J Agric Mech Res 1:159–162 78. Bulanon DM, Kataoka T, Okamoto H et al (2005) Feedback control of manipulator using machine vision for robotic apple harvesting. Proceedings of ASAE, No. 053114 79. Bulanonet D, Kataoka T, Okamoto H (2004) Determing the 3-D location of the apple fruit during harvest [C]. Proceedings of the ASAE conference 80. Bulanon D, Kataoka T (2010) Fruit detection system and an end effector for robotic harvesting of fuji apples [J]. Agric Eng Int: CIGR J 12(1) 81. Kataoka T, Ishikawa Y, Hiroma T et al (2010) Hand mechanism for apple harvesting robot [J]. J Jpn Soc Agric Mach 61(1):131–139

References

99

82. Setiawan A, Furukawa T, Preston A (2004) A low-cost gripper for an apple picking robot [C]. Proceedings of the IEEE international conference on robotics and automation 83. Scarfe A (2012) Development of an autonomous kiwifruit harvester [D]. Massey University 84. Scarfe A, Flemmer R, Bakker H et al (2009) Development of an autonomous kiwifruit picking robot [C]. Proceedings of the 4th international conference on autonomous robots and agents 85. Zhang F, Li Z, Wang B et al (2014) Study on Kiwifruit physical parameters and damage factors [J]. J Agric Mech Res 11:141–145 86. Chen J, Wang H, Jiang H et al (2012) Design of end-effector for kiwifruit harvesting robot [J]. Trans Chinese Soc Agric Mach 43(10):151–154 87. Gao H, Wang H, Chen J (2013) Research and design of kiwi fruit harvesting robot [J]. J Agric Mech Res 35(2):73–76 88. Fu L, Zhang F, Yoshinori G et al (2015) Development and experiment of end-effector for kiwifruit harvesting robot [J]. Trans Chinese Soc Agric Mach 46(3):1–8 89. Zhang F (2014) Research and design on the nondestructive end-effector of kiwifruit harvesting robot [D]. Northwest A & F University 90. Arima S, Shibusawa S, Kondo N et al (2003) Traceability based on multi-operation robot; information from spraying, harvesting and grading operation robot [C]. Proceedings of the IEEE/ASME international conference on advanced intelligent mechatronics, AIM 91. Kondo N, Hisaeda K, Hatou K et al (2000) Harvesting robot for strawberry grown on annual hill top (Part 1) manufacture of the first prototype robot and fundamental harvesting experiment [J]. J Soc High Technol Agric 12(1):23–29 92. Kondo N, Monta M, Hisaeda K (2001) Harvesting robot for strawberry grown on annual hill top, 2: manufacture of the second prototype robot and fundamental harvesting experiment [J]. J Soc High Technol Agric (Japan) 93. Kondo N, Ninomiya K, Hayashi S et al (2005) A new challenge of robot for harvesting strawberry grown on table top culture [C]. Proceedings of the ASABE annual international meeting 94. Cui Y, Nagata M, Gejima Y et al (2007) Study on strawberry harvesting robot using machine vision for strawberry grown on annual hill top (Part 2) [J]. J Jpn Soc Agric Mach 69(2):60–68 95. Feng G, Qixin C, Masateru N (2008) Fruit detachment and classification method for strawberry harvesting robot. Int J Adv Rob Syst 5(1):41–48 96. Cui Y, Kobayashi T, Nagata M (2005) Basic study on ultrasonic sensor for harvesting robot of strawberry [R]. Bull Faculty Agri,c Miyazaki University 51(1):9–16 97. Cui Y, Nagata M, Gejima Y et al (2006) Study on strawberry harvesting robot using machine vision for strawberry grown on annual hill top (part 1)—development of robot and recognition of fruit{J]. J Jpn Soc Agric Mach 68(6):59–67 98. Arima S, Kondo N, Yagi Y et al (2001) Harvesting robot for strawberry grown on table top culture, 1: harvesting robot using 5 Dof manipulator [J]. J Soc High Technol Agric (Japan) 13(3):159–166 99. Arima S, Kondo N, Monta M (2004) Strawberry harvesting robot on table-top culture [C]. Proceedings of the ASAE annual meeting 100. Nagasaki Y, Hayashi S, Nakamoto Y et al (2013) Development of a table-top cultivation system for robot strawberry harvesting [J]. Jpn Agric Res Quart 47(2):165–169 101. Arima S, Kondo N, Yagi Y et al (2001) Harvesting robot for strawberry grown on table top culture (part 1)—harvesting robot using 5 DOF manipulator [J]. J Soc High Technol Agric 13(3):159–166 102. Arima S, Monta M, Namba K et al (2003) harvesting robot for strawberry grown on table top culture (part 2)—harvesting robot with a suspended manipulator under cultivation bed [J]. J Soc High Technol Agric 15(3):162–168 103. Shiigi T, Kurita M, Kondo N et al (2008) Strawberry harvesting robot for fruits grown on table top culture [C]. Proceedings of the ASABE annual international meeting 104. Shigematsu K, Hayashi S, Yamamoto S et al (2009) Study on the annual utilization of a harvesting robot for forcing culture in strawberries [J]. J Jpn Soc Agric Mach 71(6):106–114

100

1 History and Present Situations of Robotic Harvesting Technology: A Review

105. Hayashi S, Yamamoto S, Sarito S et al (2012) Development of a movable strawberryharvesting robot using a travelling platform [C]. Proceedings of the international conference of agricultural engineering 106. Yamamoto S, Hayashi S, Yoshida H et al (2009) Development of a stationary robotic strawberry harvester with picking mechanism that approaches target fruit from below (part 1)—development of the end-effector [J]. J Jpn Soc Agric Mach 71(6):71–78 107. Yamamoto S, Hayashi S, Yoshida H et al (2010) Development of a stationary harvester with picking mechanism that approaches target fruit form below(partv2)—construction of the machine’s optical system [J]. J Jpn Soc Agric Mach 72(2):133–142 108. Yamamoto S, Hayashi S, Yoshida H et al (2010) Development of a stationary harvester with picking mechanism that approachestarget fruit form below(Part v3)—performance test with a movable bench system [J]. J Jpn Soc Agric Mach 72(5):479–486 109. Yamashita T, Tanaka M, Yamamoto S, Yamashita T, Tanaka M, Yamamoto S et al (2012) Development of RT-components for the M-3 strawberry harvesting robot [J]. Trans Soc Instr Control Eng 48(1):51–59 110. Hayashi S, Yamamoto S, Saito S et al (2012) Study on an Inside-approach Harversting Robot for Picking Strawberries [J]. J Jpn Soc Agric Mach 74(4):325–333 111. Chen L (2005) Study on picking system for strawberry harvesting robots [D]. China Agricultural University 112. Zhang K, Yang L, Wang L et al (2012) Design and experiment of elevated substrate culture strawberry picking robot [J]. Trans Chinese Soc Agric Mach 43(9):165–172 113. Feng Q, Zheng W, Jiang K et al (2012) Design of strawberry harvesting robot on table—top culture [J]. J Agric Mech Res 7:122–126 114. Feng Q, Wang X, Zheng W et al (2012) New strawberry harvesting robot for elevated-trough culture [J]. Int J Agric Biol Eng 5(2):1–8 115. Amaha K, Takakura T (1989) Development of a robotic hand for harvesting cucumber fruits [J]. J Agri Met 45(2):93–97 116. Arima S, Kondo N, Shibano Y et al (1994) Studies on cucumber harvesting robot(part 2)— manufacture of hand based on physical properties of cucumber plant and basic experiment to harvest [J]. J Jpn Soc Agric Mach 56(6):69–76 117. Arima S (1996) Fruit vegetable harvesting robot and cultivation style [J]. J Jpn Soc Agric Mach 58(6):158–164 118. Kondo N, Monta M, Noguchi N (2011) Agricultural robots: mechanisms and practice [M]. Apollo Books 119. Arima S, Kondo N, Nakamura H (1996) Development of robotic system for cucumber harvesting [J]. Jpn Agric Res Quart 30:233–238 120. van Henten E, Bac CW, Hemming J et al (2013) Robotics in protected cultivation [J]. IFAC Proc Volumes 46(18):170–177 121. van Henten E, Hemming J, van Tuijl B et al (2002) An autonomous robot for harvesting cucumbers in greenhouses [J]. Auton Robots 13(3):241–258 122. van Henten E, van Tuijl B, Hemming J et al (2003) Field Test of an Autonomous Cucumber Picking Robot [J]. Biosys Eng 86(3):305–313 123. Tang X, Zhang T, Liu L et al (2009) A New robot system for harvesting cucumber [C]. Proceedings of the ASABE Anual meeting, No. 096463 124. Ji C, Feng Q, Yuan T et al (2011) Development and performance analisis on cucumber harvesting robot system in greenhouse [J]. Robot 33(6):726–730 125. Qian S, Yang Q, Wang Z et al (2010) Research on holding characteristics of cucumber and end-effector of cucumber picking [J]. Trans CSAE 26(7):107–112 126. Wang Y (2010) Research on motion planning and control system of the cucumber picking robot [D]. Zhejiang University of Technology 127. Jin L (2013) Key technology research of cucumbers harvest robot based on machine vision [D]. Shanghai Jiaotong University 128. Hayashi S, Ganno K, Ishii Y et al (2002) Robotic harvesting system for eggplants [J]. Jpn Agric Res Quart 36(3):163–168

References

101

129. Hayashi S, Ganno K, Ishii Y et al (2001) Development of a harvesting end-effector for eggplants [J]. J Soc High Technol Agric 13(2):97–103 130. Hayashi S, Ganno K, Kurosaki H et al (2003) Robotic harvesting system for eggplants trained in v-shape (Part 2) [J]. J Soc High Technol Agric 15(4):211–216 131. Hayashi S, Ota T, Kubota K et al (2005) Robotic harvesting technology for fruit vegetables in protected horticultural production [R]. Information and technology for sustainable fruit and vegetable production, 227–236 132. Liu C, Zhang T, Yang L (2008) Design of end-effector of eggplant harvesting robot [J]. J Agric Mech Res (12):62–64 133. Song J, Sun X, Zhang T et al (2009) Design and development of transmission of picking robot for eggplant [J]. J Mech Trans 33(5):36–38 134. Song J, Sun X, Zhang T et al (2009) Design and experiment of opening picking robot for eggplant [J]. Trans Chinese Soc Agric Mach 40(1):143–147 135. Oka K, Kitamura S, Sweet Pepper Picking Robot in Greenhouse Horticulture [J]. http://www. lab.kochi-tech.ac.jp/oka-lab/society/ron_pdf/okalab20051010a.pdf 136. Kitamura S, Oka K (2005) recognition and cutting system of sweet pepper for picking robot in greenhouse horticulture [C]. Proceedings of the IEEE international conference on mechatronics and automation 137. Kitamura S, Oka K, Ikutomo K et al (2008) A distinction method for fruit of sweet pepper using reflection of led light [C]. Proceedings of the SICE annual conference 138. Kitamura S, Oka K (2006) improvement of the ability to recognize sweet peppers for picking robot in greenhouse horticulture [C]. Proceedings of the international joint conference on SICE-ICASE 139. Eizentals P, Oka K (2016) 3d pose estimation of green pepper fruit for automated harvesting [J]. Comput Electron Agric 128:127–140 140. Bachche S, Oka K (2013) Performance testing of thermal cutting systems for sweet pepper harvesting robot in greenhouse horticulture [J]. J Syst Design Dyn 7(1):36–51 141. Bachche S (2013) Automatic harvesting for sweet peppers in greenhouse horticulture [D]. Kochi University of Technology 142. Bachche S, Oka K, Sakamoto H (2012) Development of thermal cutting system for sweet pepper harvesting robot in greenhouse horticulture [C]. Proceedings of the JSME conference on robotics and mechatronics, 27–29 143. Hemming J, Bac C, Van Tuijl B et al (2014) A robot for harvesting sweet-pepper in greenhouses [C]. Proceedings of the international conference of agricultural engineering [C]. Proceedings of the JSME Conference on Robotics and Mechatronics, 27–29 144. Bac C (2015) Improving obstacle awareness for robotic harvesting of sweet-pepper [M]. Wageningen University 145. Lehnert C, Sa I, Mccool C et al (2016) Sweet pepper pose detection and grasping for automated crop harvesting [C]. Proceedings of the IEEE international conference on robotics and automation2016 146. Lehnert C, English A, McCool C et al (2017) Autonomous sweet pepper harvesting for protected cropping systems [J]. IEEE Robot Autom Lett 2(2):872–879 147. Vitzrabin E, Edan Y (2016) Changing task objectives for improved sweet pepper detection for robotic harvesting [J]. IEEE Robot Autom Lett 1(1):578–584 148. Kondo N (1995) Harvesting robot based on physical properties of grapevine [J]. Jpn Agric Res Quart 29:171–171 149. Monta M, Kondo N, Shibano Y (1995) Agricultural robot in grape production system [C]. Proceedings of the IEEE international conference on robotics and automation 150. Kondo N, Shibano Y, Mohri K et al (1993) Basic Studies on robot to work in Vinyard(Part 1)—manipulator and harvesting hand [J]. J Jpn Soc Agric Mach 55(6):85–94 151. Michihisa I, Umeda M, Namikawa K (1996) Studies on agricultural hydraulic manipulator (Part 3) [J]. J Jpn Soc Agric Mach 58(4):19–27 152. Michihisa I, Furube K, Namikawa K et al (1996) Development of watermelon harvesting gripper [J]. J Jpn Soc Agric Mach 58(3):19–26

102

1 History and Present Situations of Robotic Harvesting Technology: A Review

153. Sakai S, Osuka K, Fukushima H et al (2002) Watermelon harvesting experiment of a heavy material handling agricultural robot with LQ control [C]. Proceedings of the IEEE/RSJ international conference on intelligent robots and systems 154. Umeda M, Kubota S, Iida M (1999) Development of “Stork”, a Watermelon-Harvesting Robot [J]. Artif Life Robot 3(3):143–147 155. Hwang H, K S (2003) Development of multi-functional tele-operative modular robotic system for greenhouse watermelon [C]. Proceedings of the IEEE/ASME international conference on advanced intelligent mechatronics, AIM 156. Xia H, Wang H, Zhen W (2013) Design and kinematics analysis of pineapple picking hybrid mechanism [J]. J Mach Design 30(9):28–32 157. Wang H, Li B, Liu G et al (2012) Design and test of pineapple harvesting manipulator [J]. Trans CSAE 28(z2):42–46 158. Zhang R, Shi J, Zhang R (2011) Structure design of automatic pineapple picking machine [J]. J Anhui Agri Sci 39(16):9861–9863 159. Razali M, Ismail W, Ramili A (2007) Integrated recognition strategies for vision system (Oil Palm Harvesting) [C]. Proceedings of the 2nd national intelligent systems and information technology symposium 160. Ismail W, Ishak W (2010) research and development of oil palm harvester robot at Universiti Putra Malaysia [J]. Int J Eng Technol 7(2):87–94 161. Aljanobi A, Al-Hamed S, Al-Suhaibani S (2010) A setup of mobile robotic unit for fruit harvesting [C]. Proceedings of the IEEE international workshop on robotics in Alpe-AdriaDanube region 162. Razzaghi E, Massah J, Vakilian K (2015) Mechanical analysis of a robotic date harvesting manipulator [J]. Russian Agric Sci 41(1):80–85 163. Shokripour H, Ishak W, Shokripour R et al (2012) Development of an automatic cutting system for harvesting oil palm fresh fruit bunch (FFB) [J]. Afr J Agric Res 7(17):2683–2688 164. Wibowo T, Sulistijono I, Risnumawan A (2016) End-to-end coconut harvesting robot [C]. Proceedings of the international electronics symposium (IES) 165. Mani A, Jothilingam A (2014) Design and fabrication of coconut harvesting robot: cocobot [J]. Int Jo Innov Res Sci Eng Technol 3(s3):1351–1354 166. Srinivas A, Kuriachan M, Veerabhadhran et al (2015) Development of automated coconut harvester prototype [J]. Int J Innov Res Sci 4(8):7134–7140 167. Dubey A, Pattnaik S, Banerjee A et al (2016) Autonomous control and implementation of coconut tree climbing and harvesting robot [J]. Procedia Comput Sci 85:755–766 168. Megalingam R, Pathmakumar T, Venugopal T et al (2015) DTMF based robotic arm design and control for robotic coconut tree climber [C]. Proceedings of the international conference on computer, communication and control 169. Abraham A, Girish M, Vitala H et al (2014) Design of harvesting mechanism for advanced remote-controlled coconut harvesting robot (A.R.C.H-1) [J]. Indian J Sci Technol 7(10):1465– 1470 170. Abraham A, Vysakh A, Philip J et al, Fabrication of advanced remote-controlled coconut harvesting robot (A.R.C.H-1) [J]. Indian J Sci Technol 7(10):1465–1470 171. Clary C, Ball T, Ward E et al (2007) Performance and economic analysis of a selective asparagus harvester [J]. Appl Eng Agric 23(5):571–577 172. Irie N, Taguchi N, H T et al (2009) Asparagus harvesting robot coordinated with 3-D vision sensor [C]. Proceedings of the IEEE international conference on industrial technology2009 173. Murakami N, Otsuka K, Inoue K et al (1994) Robotic cabbage harvester [J]. JSAM 56(4):67– 74 174. Murakami N, Otsuka K, Inoue K et al (1999) Development of robotic cabbage harvester. (Part 1). Operational speed of the designed robot: operational speed of the designed robot [J]. Journal of JSAM, 61:85–92 175. Murakami N, Otsuka K, Inoue K et al (2013) Development of robotic cabbage harvester. (Part 2). Field test by developed hand: field test by developed hand [J]. Mater Manuf Process 28(4):436–442

References

103

176. Cho S, Chang S, Kim Y et al (2002) Development of a three-degrees-of-freedom robot for harvesting lettuce using machine vision and fuzzy logic control [J]. Biosys Eng 82(2):143–149 177. Chung S, Fujiura T, Dohi M et al (1999) Selective harvesting for crisp head vegetables (Part 3)—recognition of lettuce form 3-D image [J]. J Jpn Soc Agric Mach 61(5):101–107 178. Chen J, Chen Y, Jin X et al (2015) Research on a parallel robot for tea flushes plucking [C]. Proceedings of the international conference on education, management, information and medicine 179. Gao F (2013) Structural design and simulation of a high-quality tea parallel plucking robot [D]. Nanjing Forestry University 180. Gao F, Chen Y (2015) Structure design and workspace analysis of a parallel high-quality tea plucking robot [J]. Mach Tool Hydra (3):12–15 181. Li H, Li C, Xu L et al (2013) fresh tea picking robot based on DSP [C]. Proceedings of the international conference on computer and computing technologies in agriculture 182. Lu X, Li H, Xu L et al (2015) The improve design of tea picking robot with two arms [J]. J Agric Mech Res 2:101–106 183. Qin G, Zhao Y, Xiao H et al (2014) 4CZ-12 intelligent tea harvest robot design and field experiment [J]. J Chinese Agric Mech 35(1):152–156, 169 184. Noordam J, Hemming J, van Heerde C et al (2005) Automated rose cutting in greenhouses with 3d vision and robotics: analysis of 3d vision techniques for stem detection [C]. Int Conf Sustain Greenhouse Syst-Greensys 691:885–892 185. Thomas R, Marco K (2009) Robotic harvesting of gerbera Jamesonii based on detection and three-dimensional modeling of cut flower pedicels [J]. Comput Electron Agric 66(1):85–92 186. Antonelli M, Auriti L, Zobel P et al (2011) Development of a new harvesting module for saffron flower detachment [J]. Romanian Rev Precision Mech Opt Mechatron 39:163–168 187. Kohan A, Borghaee A, Yazdi M et al (2011) robotic harvesting of rosa damascena using stereoscopic machine vision [J]. World Appl Sci J 12(2):231–237 188. Gürel C, Erden A (2013) conceptual design of a rose harvesting robot for greenhouses [C]. Proceedings of The 20th international conference on mechatronics and machine vision in practice-M2ViP, 18–20 189. Ceres R, Pons J, Jiménez A et al (1998) Design and implementation of an aided fruit-harvesting robot (Agribot) [J]. Industr Robot 25(5):337–346 190. Ceres R, Pons J, Jiménez A et al (1998) Agribot: a robot for aided fruit harvesting [J]. Industr Robot 25(5):337–346 191. Kahya E, Arim S (2014) Research on robotics application in fruit harvesting [J]. J Agric Sci Technol b(5):386–392 192. Xiong J, Ye M, Zou X et al (2013) System design and performance analysis on multi-type fruit harvesting robot [J]. Trans Chinese Soc Agric Mach 44(s1):230–235 193. Yukawa N, Namikawa K, Fujiura T (1990) Visual feedback control of fruit harvesting robot (Part 1)—by use of processing of combined binary and gray images by light Emittance [J]. J Jpn Soc Agric Mach 52(3):53–59 194. Fujiura T, Ura M, Kawamura N et al (1990) Fruit harvesting robot for orchard [J]. J Jpn Soc Agric Mach 52(2):35–42 195. Fujiura T, Kawamura N, Ura M et al (1988) Fruit harvesting robot for orchard (I) [J]. J Jpn Soc Agric Mach 50(Supplement):131 196. Ura M, Kawamura N, Fujiura T et al (1988) Fruit harvesting robot for orchard (II) [J]. J Jpn Soc Agric Macnh 50(Supplement):132 197. Kondo N, Shibano Y (1991) Study of agricultural robots with 8 degrees of freedom—use of redundancy for avoiding obstacles [J]. J Jpn Soc Agric Macnh 53(6):49–57 198. He B, Liu G, Ji Y et al (2011) Auto recognition of navigation path for harvest robot based on machine vision [J]. Comput Comput Technol Agric IV, 138–148 199. Font D, Pallejà T, Tresanchez M et al (2014) A proposal for automatic fruit harvesting by combining a low cost stereovision camera and a robotic arm [J]. Sensors 14(7):11557–11579 200. Fujiura T, Otomune T, Ishida T et al (20045) Multifunctional robot for tomato production [C]. The 2nd international symposium on machinery and mechatronics for agriculture and bio-systems engineering, 125–128

104

1 History and Present Situations of Robotic Harvesting Technology: A Review

201. Dohi M, Fujiura T, Nakao S et al (1994) Multi-purpose robot for vegetable production (part 2) application to vegetable transplanting and harvesting: application to vegetable transplanting and harvesting [J]. J Jpn Soc Agric Macnh 56:101–108 202. Dohi M (1996) Development of multipurpose robot for vegetable production [J]. Jpn Agric Res Quart 30(4):227–232 203. Baur J, Pfaff J, Ulbrich H et al (2012) Design and development of a redundant modular multipurpose agricultural manipulator [C]. IEEE/ASME international conference on advanced intelligent mechatronics (AIM). IEEE, 823–830 204. Qiu Q (2015) A new algorithm for greenhouse corridor edge detection with RGB-D data [C]. Proceedings of the IEEE international conference on cyber technology in automation, control, and intelligent systems (CYBER), 1074–1079 205. Nissimov S, Goldberger J, Alchanatis V (2015) Obstacle detection in a greenhouse environment using the Kinect sensor [J]. Comput Electron Agric 113(C):104–115 206. González R, Rodríguez F, Sánchezhermosilla J et al (2009) Navigation techniques for mobile robots in greenhouses [J]. Appl Eng Agric 25(25):153–165 207. Yuan T, Ren Y, Li W et al (2012) Navigation information acquisition based on illumination chromaticity stability analysisi for greenhouse robot [J]. Trans Chinese Soc Agric Mach 43(10):161–166 208. Wang X, H X, Mao H et al (2012) Navigation line detection of tomato gidges in greenhouse based on least square method [J]. Trans Chinese Soc Agric Mach 43(6):161–166 209. Yamamoto K, Guo W, Yoshioka Y et al (2014) On plant detection of intact tomato fruits using image analysis and machine learning methods [J]. Sensors 14(7):12191–12206 210. Wei X, Jia K, Lan J et al (2014) Automatic method of fruit object extraction under complex agricultural background for vision system of fruit picking robot [J]. Optik—Int J Light Electron Optics 125(19):5684–5689 211. Puttemans S, Vanbrabant Y, Tits L et al (2016) automated visual fruit detection for harvest estimation and robotic harvesting [C]. Proceedings of the international conference on image processing theory, TOOLS and applications 212. Yin H, Chai Y, Yang SX et al (2011) Ripe tomato detection for robotic vision harvesting systems in greenhouses [J]. J Reprod Med 57(54):1539–1546 213. Miao Z, Shen Y, Wang X et al (2016) Imaged recognition algorithm and experiment of overlapped fruits in natural environment [J]. Trans Chinese Soc Agric Mach 47(6):21–26 214. Feng Q, Chen W, Yang Q et al (2015) Identification and localization of overlapping tomatoes based on linear structured vision system [J]. J China Agric Univ 20(4):100–106 215. Liu J (2010) Analysis and optimal control of vacuum suction system for tomato harvesting robot [D]. Jiangsu University 216. Liu J, Li Z, Wang F et al (2013) Hand-arm coordination for a tomato harvesting robot based on commercial manipulator [C]. Proceedings of the IEEE international conference on robotics and Biomimetics 217. Mann M, Rubinstein D, Shmulevich I et al (2014) Motion planning of a mobile cartesian manipulator for optimal harvesting of 2-D crops [J]. Trans ASABE 57(1):283–295 218. Han L, Fujiura T, Yamada H et al (2010) Cherry tomato harvesting robot with 3-d sensor on its end effector (part 2): image recognition and harvest experiment [J]. J Jpn Soc Agric Macnh 62:127–136 219. Barth R, Hemming J, Henten E (2016) Design of an eye-in-hand sensing and servo control framework for harvesting robotics in dense vegetation [J]. Biosys Eng 146:71–84 220. García-Luna F, Morales-Díaz A (2016) Towards an artificial vision-robotic system for tomato identification [J]. IFAC-PapersOnLine 49(16):365–370 221. Liu B, Gong L, Chen Q et al (2014) A stereo visual interface for manipulating the grasping operation of a fruit-harvesting robot [C]. International conference on intelligent robotics and applications, 239–244 222. Ejvan H, Hemming J, Bajvan T et al (2003) Collision-free motion planning for a cucumber picking robot. [J] Biosystems Eng 86(2):135–144

References

105

223. Carducci G, Foglia M, Gentile A et al (2005) Pneumatic robotic arm controlled by on-off valves for automatic harvesting based on vision localisation [C]. Proc IEEE Int Conf Industr Technol 2:1017–1022 224. Hayashi S, Takeshita D, Yamamoto S et al (2013) Collision-free control of a strawberryharvesting robot by recognition of immature fruits [J]. Shokubutsu Kankyo Kogaku 25(1):29– 37 225. Arima S, Kondo N, Shibano Y et al (1993) Studies on cucumber harvesting robot(part 1)— investigation of cultivation training and mechanism of manipulator [J]. J Jpn Soc Agric Mach 56(1):55–64 226. Hayashi S, Ota T, Ikuta K et al (2004) Questionnaire survey on labor saving of harvesting fruit picking work of strawberry [J]. J Jpn Soc Agric Macnh 66(Supplement):111–112 227. Kondo N, Monta M, Ting K et al (1996) Harvesting robot system for single truss upside down tomato production (1): single truss upside down tomato system and constitution of robot system [J]. J JSAM 58:467–470 228. Polder G, Hofstee J (2014) Phenotyping large tomato plants in the greenhouse using a 3d light-field camera [C]. Proceedings of the ASABE-CSBE/ASABE joint meeting, No. 1882255 229. Polder G, Lensink D, Veldhuisen B (2013) Phenobot—a robot System for Phenotyping large tomato plants in the greenhouse using a 3D light field camera [R]. Wageningen Ur Glastuinbouw/enza Zaden 230. Li X, Li L, Li N (2017) Autonomous positioning and navigation design of picking robot based on RFID and WSN [J]. J Agric Mech Res 39(9):215–218 231. Xiong Q (2017) Autonomous positioning and navigation design of fruit and vegetable picking robot-based on information fusion of RFID and WSN [J]. J Agric Mech Res 39(10):223–227 232. Liu J, Hu Y, Xu X et al (2011) Feasibility and influencing factors of laser cutting of tomato peduncles for robotic harvesting [J]. Afr J Biotech 10(69):15552–15563 233. Liu J, Xu X, Li P (2014) Analysis and experiment on laser cutting of fruit peduncles [J]. Trans Chinese Soc Agric Mach 45(01):59–64 234. John A, Clark C, Freeman M et al (2016) Review: milking robot utilization, a successful precision livestock farming evolution [J]. Animal 10(9):1–9 235. Zhang K, Chu J, Zhang T et al (2017) Development status and analysis of autonomous grafting technology for vegetables [J]. Trans Chinese Soc Agric Mach 48(3):1–13 236. Chopde S, Patil M, Shaikh A et al (2017) Developments in computer vision system, focusing on its applications in quality inspection of fruits and vegetables-a review [J]. Agric Rev 38(2):94–102 237. Li G, Ji C, Gu B et al (2016) Kinematics analysis and experiment of apple harvesting robot manipulator with multiple end-effectors [J]. Trans Chinese Soc Agric Mach 47(12):14–21 238. Zion B, Mann M, Levin D et al (2014) Harvest-Order planning for a Multiarm robotic harvester [J]. Comput Electron Agric 103(2):75–81 239. Mann M, Zion B, Shmulevich I et al (2016) Combinatorial optimization and performance analysis of a multi-arm cartesian robotic fruit harvester—extensions of graph coloring [J]. J Intell Rob Syst 82(3–4):399–411 240. Tang H, Zheng B (2017) Reseach on remote formation control of picking robot based on Singalr and web [J]. J Agric Mech Res 39(3) 241. George V, Yiannis S, David S (2012) Dispatching and routing of robotic crop-transport aids for fruit pickers using mixed integer programming [C]. Proceedings of the ASABE annual meeting, No. 1338247 242. Liu J, Wang J, Chen Y (2015) Cooperative operation system and control method of greenhouse fruit and vegetable harvesting and transportation by robot [P]. ZL201210530180.5. 04-22 243. Fu Z, Wang L (2012) Design of a remote picking robot based on ATmega32 [J]. Electron Design Eng 20(4):151–154 244. Quan L, Li C, Feng Z et al (2017) Algorithm of works’ decision for three arms borot in greenhouse based on control with motion sensing thechnology [J]. Trans Chinese Soc Agric Mach 48(3):14–23

106

1 History and Present Situations of Robotic Harvesting Technology: A Review

245. Xu C, Wang Q, Chen H et al (2017) Design and simulation of artificial limb picking robot based on somatosensory interaction [J]. Trans CSAE 33(s1):49–55 246. Liu J, Peng H, Wang J et al, Substrate Auto-Paver for Table-Top Culture [P]. US 10334790B2. 2019-07-22

Chapter 2

Damage and Damage-Free Harvesting in Robotic Operation

2.1 Cause of Fruit Damage in Robot Harvesting Because of the tenderness of fruits and vegetables, how to effectively reduce the damage of fruits and vegetables in the harvesting and postharvest transportation and sorting has always been a hot research spot in the field of agricultural products. The main causes of fruit damage in robot harvesting are: (1)

Gripping damage

The collision rupture or bruise in gripping is the most common form of damage in robot harvesting. Especially the collision between the fingers and the fruit under rapid operation will greatly increase the probability of damage to the fruit. (2)

Interference collision damage

In various processes, such as gripping, detaching, and postharvest transportation, the fruit may interfere with the tool, the robot body, or the working environment, and the robot may interfere in the adjacent fruit during the harvesting process and lead to damage. (3)

Loss of fruit falling

In the harvesting, transportation and, placing of single fruit and fruit clusters, the fruit and fruit berries may fall off due to the unreliable or vibration, which can lead to the loss of the yield and quality of fruit. (4)

Placing damage

For different objects, such as spherical fruit, slender fruit, and fruit clusters, in the process of fruit placing after harvesting, the fall of fruit and the collision among different fruit will cause damage of fruit or fall of berries.

© Science Press, Beijing and Springer Nature Singapore Pte Ltd. 2021 J. Liu et al., Rapid Damage-Free Robotic Harvesting of Tomatoes, Springer Tracts in Mechanical Engineering, https://doi.org/10.1007/978-981-16-1284-8_2

107

108

2 Damage and Damage-Free Harvesting in Robotic Operation

For most harvesting, it is necessary to grip the fruit by applying force from the endeffector to the fruit, thus avoiding the gripping damage, especially in the high-speed operation, becomes the key problem to realize robotic damage-free harvesting.

2.2 Passive Compliant Structure in Robotic Harvesting Compliance control is an important research field in robot technology. It is through a passive compliant structure or an active compliance control to make the robot adapts to the interaction between the objects and adapts to the shape, size, and posture change of the object and reduces the damage and failure of the operation. The ways of compliance include passive and active methods, in which a passive compliance is to give the robot certain flexibility and compensation ability by flexibility of contact parts and joints, while active compliance is to achieve the purpose of active adaptation and reliable damage-free gripping by the servo control of force and sliding sense.

2.2.1 Elastic Surface Material On the basis of rigid structure, most robotic harvesting end-effectors add elastic materials, such as rubber layer, on the inside surface of the finger to increase the buffer to reduce the possibility of gripping damage to the fruit. For example, the two-finger end-effector for tomato harvesting developed by Kondo N. et al. in Japan [1] (Fig. 2.1a), the three-finger citrus harvesting end-effector developed by Hannan M. of the University of Florida in the USA [2] (Fig. 2.1b), the cherry-tomato harvesting end-effector developed by Tanigaki K. in Osaka Prefecture University of Japan [3] (Fig. 2.1c), the two-finger kiwifruit harvesting end-effector developed by Fu L. et al. in Northwest A&F University of China [4] (Fig. 2.1d), all adopted the method of adding elastic materials onto the finger surface. Dimeas F., University of Patras, Greece, developed a strawberry end-effector for harvesting by investigating the hand motion of a skilled worker [5] (Fig. 2.2). The end-effector grasps the fruit and then rotates its wrist around the pitch axis to bend the stem [5]. An experimental prototype is produced on a domestic 3D printer with fused deposition modeling (FDM) and by using polylactic acid (PLA) thermoplastic material [5]. The fingers provide a wide surface to attach polyurethane foam and their formation enables the grasping with many contact points to distribute the force and prevent damage to the fruit [5].

2.2 Passive Compliant Structure in Robotic Harvesting

109

(a)

(b)

(c)

(d)

Fig. 2.1 End-effector with elastic surface material for robotic harvesting. a Tomato [1], b Citrus [2], c Cherry tomato [3], d Kiwi fruit [4]

Fig. 2.2 End-effector with polyurethane foam surface for robotic strawberry harvesting [5]

110

2 Damage and Damage-Free Harvesting in Robotic Operation

2.2.2 Under-Actuated End-Effectors Kondo, Monta et al., Japan, developed an end-effector with four flexible fingers and a suction pad [6–8] (Fig. 2.3). Each finger consists of four tubes connected in a series and a cable that is attached inside the tubes and connects to a fingertip [6]. The finger whose materials are nylon bends inward when a cable is pulled because the finger tends to contract toward the non-supported side [6]. Davidson of Washington State University, USA, developed an end-effector for robotic apple harvesting [9, 10] (Fig. 2.4). This end-effector includes three identical fingers arranged symmetrically around a circular palm with a soft rubber insert [9, 10]. Under-actuation between the fingers is provided by a disc differential that is a variant of the seesaw mechanism [9, 10]. Should a finger contact the object before the other two, the differential will rotate and enable further displacement of the Support Cable

Tubes

Pull

Fig. 2.3 The under-actuated end-effector for robotic tomato harvesting developed by Kondo N. et al., Japan [6–8]

Fig. 2.4 The under-actuated end-effector for robotic apple harvesting developed in Washington State University [9, 10]

2.2 Passive Compliant Structure in Robotic Harvesting

111

remaining two tendons [9, 10]. The under-actuated design using flexure joints with passive compliance is helpful to increase robustness to position error [9, 10]. An under-actuated end-effector for robotic tomato harvesting, similar to the previous end-effector by Monta et al. [8], has been developed by Chiu et al. in National Ilan University, Taiwan, China [11, 12] (Fig. 2.5). The end-effector is a four-fingered gripper comprising fingers, a spring plate, a fruit suction device, and finger bending mechanisms [12]. Each finger has four sections. One spring plate is inserted into each of the four fingers, with a wire connecting the tip to the solenoid [12]. When the solenoid is magnetized, it pulls the steel wire on the inside of the finger, drawing the tip in and forcing the spring plate to bend [12]. Ling et al., Ohio State University, USA, developed a four-finger end-effector for robotic tomato harvesting [13] (Fig. 2.6). The fingers are similar to the previous end-effector by Monta et al. [8] in relative shape and actuation. Both systems are highly under-actuated, having only one power input for all four fingers. Like the previous model, the fingers depend upon moments produced by force applied to the

Fig. 2.5 The under-actuated end-effector for robotic tomato harvesting developed in National Ilan University, Taiwan, China [11, 12]

Fig. 2.6 The under-actuated end-effector for robotic tomato harvesting developed in Ohio State University, USA [13]

112

2 Damage and Damage-Free Harvesting in Robotic Operation

Fig. 2.7 The under-actuated multi-fingered end-effector for fruit gripping [14]

cables [13]. The fingers used rigid and lightweight square and rectangular acetyl butyl styrene (ABS) tubing instead of nylon to restrict lateral movement in gripping. Yin et al. of Jiangsu University, China, also developed the under-actuated multifingered end-effector [14] (Fig. 2.7). There are three fingers in each finger, each finger has three knuckles and is composed of eight two link rods and one slider, which are driven by a linear stepping motor to realize the opening of the fingers and the gripping of the fruit.

2.2.3 Elastic-Medium Fingers 1.

Gas medium

Yang et al. of Zhejiang University of Technology, China, developed a pneumatic flexible end-effector [15–17] (Fig. 2.8). It took three pneumatic flexible bending joints as its fingers and one flexible pneumatic torsion joint as its wrist. When certain compressed gas is inflated into the bending and torsion joints, the fingers are driven to rotate and bend to achieve the flexible gripping of objects such as apple fruit. Allotta et al., ARTS laboratory in Italy, has also developed a pneumatic flexural three-finger end-effector [18, 19] (Fig. 2.9). Setiawan of the University of New South Wales in Australia developed an apple harvesting robot. Its end-effector consists of five main parts: one main cup, two inner rings, and two end-caps [20] (Fig. 2.10). As a clumping mechanism, two rubber bladders connected to a pneumatic tube are attached to the inner surface of the cup [20]. When the fruit enters the main cup, the two rubber bladders are inflated to hold the fruit to avoid potential damage. The tomato harvest end-effector, which was developed by Wang X., Hebei University of Technology, China, also adopted a similar principle but more complex in structure. Firstly, the vacuum suction sucked the fruit and pulled it into the round

2.2 Passive Compliant Structure in Robotic Harvesting

113

Fig. 2.8 The pneumatic flexible end-effector developed in Zhejiang University of Technology, China [15–17]

Fig. 2.9 The pneumatic flexural three-finger end-effector developed in ARTS laboratory, Italy [18, 19]

cup, and then the fruit was gripped with three inflatable airbags, and then the tomato fruit was detached by twisting [21] (Fig. 2.11). 2.

Liquid medium

Pettersson et al. of the Swedish Institute for Food and Biotechnology, designed a magnetorheological robot gripper for handling delicate food products with varying shapes [22] (Fig. 2.12). Magnetorheological fluid is a suspension of micron-sized polarizable particles where the rheological properties change when a magnetic field is applied [22]. Using a magnetic field to control the pressure of the magnetorheological

114

2 Damage and Damage-Free Harvesting in Robotic Operation

Fig. 2.10 The cup-type end-effector with inner bladders developed in the University of New South Wales, Australia [20]

Fig. 2.11 The cup-type end-effector with inner bladders developed in Hebei University of Technology, China [21]

fluid, small gripping force can be used to achieve flexible gripping of different shapes of fruits.

2.3 Active Compliance Control in Robotic Harvesting All kinds of passive compliant structures can effectively make up for the shortage of rigid gripping and are conducive to the implementation of reliable damage-free gripping. However, the specific passive compliance structure is limited to specific

2.3 Active Compliance Control in Robotic Harvesting

115

Fig. 2.12 The magnetorheological robot gripper developed in Swedish Institute for Food and Biotechnology, Sweden [22]

harvesting objects. Especially for larger size and shape differences, the performance will be significantly affected. Active compliance control is an active intervention for gripping damage and slippage by sensing and controlling the interaction force between the end-effector and the object. Active compliance is based on the interaction of force, and it has stronger adaptability and wider application to different objects. In order to realize the compliance control of the gripping process, for various end-effectors of harvesting robots developed at home and abroad, different forces or sliding sensors are installed on finger surface and fingertip, or force sensor is installed on the wrist. A double-finger apple harvesting end-effector (Fig. 2.13), was developed by Ji et al., Jiangsu University, China. Force-sensitive resistances are installed on the inner surface of the in-finger surface, and a generalized proportional integral (GPI) output feedback control scheme is proposed. The rate of damage-free gripping for both apple and pear fruit reached nearly 90%. The time from contact with fruit to complete the

Fig. 2.13 The apple harvesting end-effector with gripping force feedback developed in Jiangsu University, China [23, 24]

116

2 Damage and Damage-Free Harvesting in Robotic Operation

Fig. 2.14 The end-effector with pressure feedback for robotic kiwifruit harvesting developed in Northwest A&F University, China [25]

gripping is about 3 s. However, it is necessary to close the two fingers firstly at low voltage and low speed, which will lead to too long time consumption of the total gripping process [23, 24]. The 3 DOF end-effector (Fig. 2.14) for robotic kiwifruit harvesting developed by Zhang F. et al., Northwest A&F University, China, is equipped with an FSR402 pressure sensor on the inner surface of the arc surface to detect the clamping pressure [25]. For the pneumatic flexural three-finger end-effector developed by Allotta et al., ARTS laboratory in Italy [18, 19] (Fig. 2.9), a wrist-mounted six-axis F/T (Force/Torque) sensor was used to obtain force information during the fruit harvesting. After the three fingers grasping the fruit, the manipulator pulls the fruit away, and the wrist six-axis F/T detects the force and torque information to determine the position of the stem, and then the wrist is rotated cut off the stem by a circular saw (Fig. 2.15). For the three-finger grip-bend end-effector developed by Dimeas et al., University of Patras, Greece, the PPS (pressure profile sensor) arrays are placed on each finger that can sense the distribution of the force along with the finger. A hierarchical control scheme is proposed based on accurate grasping criteria that detect misplaced strawberries on the gripper or uneven distribution of forces and a fuzzy controller for the force regulation of the gripper [26] (Fig. 2.16). The pneumatic two-finger end-effector developed by the University of Cassino in Italy, installed only a force cell in a suitable fingertip to control the flow pressure in the actuator of the gripping mechanism. However, only the changing rule of gripping

2.3 Active Compliance Control in Robotic Harvesting

117

Fig. 2.15 The end-effector with wrist six-axis F/T sensor developed in ARTS laboratory, Italy [18, 19]

Fig. 2.16 The end-effector with PPS arrays developed at the University of Patras, Greece [26]

force for tomato fruit under different working conditions has been measured, but the study of gripping force control has not been carried out [27] (Fig. 2.17). Fig. 2.17 The end-effector with a force cell developed at the University of Cassino, Italy [27]

118

2 Damage and Damage-Free Harvesting in Robotic Operation

Wang X. et al., Nanjing Agricultural University, China, carried out the research of fruit and vegetable grasping technology-based external force control loop. The force sensor was embedded in the finger surface, and the gripping force was controlled by the proportional and incremental proportional–integral algorithm according to the set value. But the gripping time in the test was more than 1 min, which was difficult to adapt to the need for speedy operation [28]. Then the team’s Feng et al. carried out a study on the control algorithm of fruit gripping with PVDF piezoelectric film based on sliding and external force control loop, but the description of gripping time in the test results was not found [29].

2.4 The Robotic Speedy Damage-Free Harvesting 2.4.1 The Significance and Particularity of Robotic Speedy Damage-Free Harvesting 1.

The characteristics of present flexible harvesting research

The tender fleshy fruits are very susceptible to mechanical damage, which has attracted the attention of the technical research and equipment developers of the harvesting robots. In order to achieve the purpose of damage-free harvesting, different attempts of adding flexible materials or structures, adding sensors, and carrying out the control of damage-free harvesting have been made. Passive compliance plays an essential role in robotic harvesting to reduce the damage, and further active compliance control can be classified into two main categories. (1)

(2)

2.

Based on the slip detection and feedback, the gripping force can be actively adjusted to realize stable gripping. By the slip avoidance control with the minimum gripping force input, the damage of the object may be avoided. This slip avoidance control is based on the unknown reliable grip force, and its advantages are the active adaptability to different objects. However, it must be based on the emergence of the sliding trend and the control has to be carried out after the detection, and the ultimate goal is also to be reliably gripping instead of avoiding the damage. By gripping force control in the whole process from contact object to stable gripping based on the direct force feedback, to avoid the fruit and vegetable damage caused by the excessive gripping force. Although various control algorithms are proposed, it is mainly based on the quasi-static rules of the contact force and control strategy. It is limited to contact with the fruit and achieve stable gripping at a much lower speed. The problem of gripping collision in speedy harvesting operation

2.4 The Robotic Speedy Damage-Free Harvesting

119

Fig. 2.18 The fruit gripping experiment of harvesting robot

In actual operation, the gripper starts closing from a certain degree of opening to contact the fruit first and then continues to move until grips the fruit stably. A complete gripping process includes two major stages before and after contact. In the present study, the stage of gripping after contact is only concerned. However, if the fingers have to move much slower to contact the fruit due to the need for damage-free gripping, it is no doubt that the time consuming before contact will become too long and determine the efficiency of the gripping operation. It is found in the experiment (Fig. 2.18) that the low-speed clamping is a quasistatic loading process to the fruit, the collision between the finger and the fruit during speedy gripping may resulting in extremely high peak gripping force instantaneously. The peak gripping force is obviously larger than that resulted from the static force balance relation. This will cause the failure of control criteria under quasi-static conditions. At the same time, it is found that the probability of internal tissue damage caused by the collision is much greater than that of skin rupture. The dynamic collision phenomenon in fast gripping causes the probability of fruit damage to be greatly increased, which seriously affects the success rate of fruit and vegetable robot harvesting, and causes great economic loss. The collision in speedy gripping has become a new and key issue in the rapid damage-free harvesting of fruits and vegetables. In the process of manual harvesting of fruits, the time needed to grip a fruit stably is usually not more than 1 s. However, the low efficiency of present various kinds of robot systems to avoid fruit damage has become one of the great obstacles to the practical application of the harvesting robot technology. Therefore, the study on the collision model and the damage mechanism of the speedy harvesting of fruits is of great theoretical value, and it is of great significance to the practical application of the harvesting robot technology.

120

2 Damage and Damage-Free Harvesting in Robotic Operation

2.4.2 The Particularity of the Collision in Robotic Speedy Gripping of Fruit 1.

The characteristics of existing collision research

The contact and collision between the manipulator and the object is the hot spot of international research. Kim of McGill University in Canada studied the collision of the hand of space manipulator with the object [30], Brogliato of Domaine University of France studied the collision of the single DOF. robot with the object [31], Yoshida of the Tohoku University Japan carried out the dynamic modeling of the contact between the long flexible manipulator and the object [32], and Hariharesan of Texas Technology University studied the dynamic modeling of the collision of the flexible manipulator with the object [33]. However, the above researches only aim at the elastic collision in unidirectional contact between the end-effector and the object. The collision is the decisive factor in the mechanical damage of fruits in the harvesting, transportation, and sorting of fruits and vegetables, and has always been the focus of fruit and vegetable research. Lu et al. [34], Idah et al. [35], Stropek et al. [36], Albaloushi et al. [37], Fluck et al. [38], Chen et al. [39], Gołacki et al. [40], Li et al. [41], and Lien [42], respectively, studied the collision acceleration, collision force, and collision energy of fruit drop. Lu [43, 44] and Groves [45] modeled the collision force and deformation in fruit drop based on the viscoelastic model of fruit. However, the existing research has only focused on the problem of fruit drop and unidirectional contact collision between fruits during storage and transportation. 2.

Bi-directional constraints of collision in robotic speedy gripping of fruit

Compared with the free collision between the end-effector and the object of the manipulator, the collision in fruit gripping is characterized by the bilateral restrictive collision between the end-effector and the fruit and the transition process from the collision to the reliability constraint. And compared with the free collision of oneway contact between the fruit and the ground during transportation, the collision in fruit gripping has two characteristics of continuous energy input and constraint, which are very different in collision mechanism and form. 3.

The flexible feature of the object of collision in robotic speedy gripping of fruit

Compared with the rigid collision between the end-effector and the workpiece, the modeling of the constraint collision process is more complex because of the viscoelastic and rheological properties of the fruit and vegetable objects. The interaction force of the collision process presents a special change rule. The individual difference of the viscoelastic and rheological properties of fruits makes the modelbased fast compliant gripping control cannot be carried out with accurate uniform parameters, but the large scope of application of control must be sought on the basis of the statistical laws of viscoelastic and rheological properties.

2.4 The Robotic Speedy Damage-Free Harvesting

121

2.4.3 The Research System of Speedy Damage-Free Harvesting Take tomato fruit as the object, the research system of speedy damage-free harvesting includes the follows: 1.

Systematic measurement and modeling of macro-mesomechanical properties of fruits

To carry out the experimental measurement and analysis of the mechanical properties of whole fruit, tissue, and microstructures with a different maturity of tomato fruit, thus to provide the basis for discovering the mechanism of gripping damage and the control of compliance: (1) (2)

(3)

(4) (5) 2.

To know well the biological and physical characteristics of tomato fruit related to robotic harvesting; To know well the viscoelastic and rheological properties of tomato fruit, and then to build the constitutive model of static compression properties of tomato fruit; To find out the structure of each mesoscopic tissue of tomato fruit, and then to obtain the mechanical properties of each tissue and reveal the mesoscopic mechanism of mechanical damage of tomato fruit To know well the anisotropy of the mechanical properties of tomato fruit and the effects on the weightlessness and moisture content of compressed tomato; To establish a combined judgment method based on both apparent and inner damage. Experimental analysis of fruit damage under multi-mode and multi-parameter gripping with robotic end-effector system

To develop an experimental platform of multi-sensing and integrated control of the harvesting robot, thus to complete the multi-mode and multi-parameter gripping experiments of tomato fruit and provide the basis and verification basis for the theoretical modeling of the fruit gripping collision and the realizing of optimal control of fast compliant gripping: (1)

(2) (3) 3.

To establish the statistical relationship among reliable gripping control mode, motor input, peak gripping force, and fruit deformation under low-speed (quasistatic) conditions; To establish the statistical relationship among gripping speed, gripping collision force, and collision deformation under different motor inputs; To find out the influence law of mechanical anisotropy on gripping collision force. Virtual simulation of gripping collision integrated finite element model with virtual prototyping

On the basis of the modeling of mechanical properties of fruit and the experimental analysis of gripping collision, to establish a virtual simulation method based on the

122

2 Damage and Damage-Free Harvesting in Robotic Operation

combination of finite element model and virtual prototyping, thus to realize the virtual simulation of the collision process of the static gripping of the fruit of the harvesting robot, which provides the model support for the realization of the speedy damage-free gripping of the fruit: (1) (2)

(3)

(4)

4.

To establish a multi-component viscoelastoplastic finite element model of the whole tomato fruit; To establish the dynamic model and virtual prototype of the harvesting robot’s end-effector system, and to complete the motion simulation of the gripping process; To establish a virtual simulation method of gripping and collision based on the combination of the finite element model of tomato fruit and the virtual prototyping of end-effector; To realize the prediction of fruit damage in gripping and the determination of the relationship between the motor control parameters and fruit damage through dynamic simulation of static and speedy dynamic gripping collision. A comparative study on the rule and effect of different stem detaching methods

The detaching method of fruiting has an important influence on the loading of the fruit in the harvesting. In order to achieve the high success rate of flexible harvesting, the comparison of different detaching methods is necessary, which may provide technical support for the realization of the speedy harvest operation. (1)

(2)

5.

To carry out the mechanical analysis and experimental comparison of different wrist motions for fruit detaching, thus to find out the best non-tool detaching method by a comprehensive consideration of multiple indicators for robotic harvesting; To carry out the feasibility study of non-contact laser cutting technology of stem, and to establish the relationship between the cutting efficiency and different influencing factors and put forward the optimal control mode so as to provide strong support for the realization of the speedy harvesting operation. Optimal control of speedy flexible gripping for fruit harvesting robot

To propose the optimal control mode and parameters of speedy flexible gripper for harvesting robot, thus to realize the coordinated control of the hand–arm movement of harvesting robot. (1)

(2) (3)

By gripping collision experiment and simulation, to effectively reveal the inelastic contact collision phenomena and the law between powered mechanism and viscoelastic object; To propose the optimal control mode and parameters of speedy flexible gripping for harvesting robot; To establish an integrated mode of the passive compliant structure and the active compliance control to realize flexible harvesting of fruit.

2.4 The Robotic Speedy Damage-Free Harvesting

(4)

123

To establish the hand–arm system of harvesting robot by integrating the general-purpose industrial manipulator with a special flexible end-effector, thus to realize the coordinated motion control of the hand–arm of harvesting robot.

References 1. Monta M, Kondo N, Ting K (1998) End-effectors for tomato harvesting robot. Artif Intell Rev 12(1):11–25 2. Hannan M, Burks T (2004) Current developments in automated citrus harvesting. In: Proceedings of the American society of agricultural and biological engineers, p 043087 3. Tanigaki K, Fujiura T, Akase A et al (2008) Cherry-Harvesting Robot. Comput Electron Agric 63(1):65–72 4. Fu L, Zhang F, Yoshinori G et al (2015) Development and experiment of end-effector for kiwifruit harvesting robot. Trans Chin Soc Agric Mach 46(3):1–8 5. Dimeas F, Sako D, Moulianitis V et al (2013) Towards designing a robot gripper for efficient strawberry harvesting. In: Proceedings of 22nd international workshop on robotics in AlpeAdria-Danube region, pp 220–226 6. Kondo N, Ting K (1998) Robotics for bioproduction systems. American Society of Agricultural Engineers (ASAE) 7. Krikke J (2005) Robotics research exploits opportunities for growth. Pervasive Comput IEEE 4(3):7–10 8. Monta M, Kondo N, Ting K et al (1998) Harvesting end effector for inverted single truss tomato production systems. J Jpn Soc Agric Mach 60(6):97–104 9. Davidson J, Mo C (2015) Mechanical design and initial performance testing of an apple-picking end-effector. In: Proceedings of the ASME international mechanical engineering congress and exposition, No. 57397 10. Davidson J, Mo C, Silwal A et al (2015) Human-machine collaboration for the robotic harvesting of fresh market apples. https://doi.org/10.13140/rg.2.1.3457.2880 11. Chiu Y, Chen S, Yang P et al (2012) Integrated test of an autonomous tomato picking robot. In: Proceedings of the 7th CIGR section VI international technical symposium “Innovating the Food Value Chain” Postharvest Technology and AgriFood Processing 12. Chiu Y, Yang P, Chen S (2013) Development of the end-effector of a picking robot for greenhouse-grown tomatoes. Appl Eng Agric 29(6):1001–1009 13. Ling P, Ehsani R, Ting K et al (2004) Sensing and end-effector for a robotic tomato harvester. In: Proceedings of the ASAE annual meeting paper 14. Wang T, Yin J (2006) A type of underactuated finger mechanism design and statics analysis for picking fruit. J Agric Mech Res 3:112–114 15. Jin Y (2010) Research on end-effector of apple picking based on flexible pneumatic actuator FPA. Zhejiang University of Technology 16. Bao G, Gao F, Xun Y et al (2009) Flexible end-effector based on flexible pneumatic actuator and its grasping model. Trans CSAE 25(10):121–126 17. Yang Q, Jin Y, Qian S et al (2010) Research on end-effector of apple picking based on new flexible pneumatic actuator. Trans Chin Soc Agric Mach 41(9):154–158 18. Allotta B, Buttazzo G, Dario P et al (1990) A force/torque sensor-based technique for robot harvesting of fruits and vegetables. In: Proceedings of the IEEE international workshop on intelligent robots and systems’ 90.’ Towards a new frontier of applications’

124

2 Damage and Damage-Free Harvesting in Robotic Operation

19. Armada MA, Muscato G, Prestifilippo M et al (2005) A prototype of an orange picking robot: past history, the new robot and experimental results. Ind Robot Int J 32(2):128–138 20. Setiawan AI, Furukawa T, Preston A (2004) A low-cost gripper for an apple picking robot. In: Proceedings of the IEEE international conference on robotics and automation, vol 5, pp 4448–4453 21. Wang X, Wu H, Feng Q et al (2016) Design and test of tomatoes harvesting robot. J Agric Mech Res 4:94–98 22. Pettersson A, Davis S, Gray J et al (2010) Design of a magnetorheological robot gripper for handling of delicate food products with varying shapes. J Food Eng 98(3):332–338 23. Guo J, Zhao D, Ji W et al (2010) Design and control of the open apple-picking-robot manipulator. In: Proceedings of the 3rd IEEE international conference on computer science and information technology (ICCSIT) 24. Ji W, Luo D, Li J et al (2014) Compliance grasp force control for end-effector of fruit-vegetable picking robot. Trans CSAE 30(9):19–26 25. Zhang F (2014) Research and design on the nondestructive end-effector of kiwifruit harvesting robot. Northwest A & F University 26. Dimeas F, Sako D (2015) Design and fuzzy control of a robotic gripper for efficient strawberry harvesting. Robotica 33(5):1085–1098 27. Ceccarelli M, Figliolini G et al (2000) Designing a robotic gripper for harvesting horticulture products. Robotica 18(1):105–111 28. Wang X, Ji C, Zhou J et al (2009) Technology of grasp fruit and vegetable based on external force control loops. Acta Agric Zhejiangensis 21(6):627–632 29. Feng L (2010) Study on control algorithm for fruit capturing based on slippery feeling and strength outer ring regulation. Nanjing Agricultural University 30. Kim S, Misra A, Modi V (2001) Contact dynamics and force control of space manipulator systems. Philos Trans Math Phys Eng Sci 359(1788):2271–2286 31. Brogliato B, Orhant P (1998) Contact stability analysis of a one degree-of-freedom robot. Dyn Control 8(1):37–53 32. Yoshida K, Mavroidis C, Dubowsky S (1996) Impact dynamics of space long reach manipulators. In: Proceedings of the IEEE international conference on robotics and automation 33. Hariharesan S (1998) Modeling, simulation and experimental verification of contact/impact dynamics in flexible articulated structures. Texas Tech University 34. Lu F, Ishikawa Y, Kitazawa H et al (2010) Measurement of impact pressure and bruising of apple fruit using pressure-sensitive film technique. J Food Eng 96(4):614–620 35. Idah P, Ajisegiri E, Yisa M (2013) An assessment of impact damage to fresh tomato fruits. AU J Technol 10(4):271–275 36. Stropek Z (2007) Determining apple mass on the basis of rebound energy during impact. In: Polish academy of sciences branch in Lublin. TEKA. Commission of motorization and power industry in agriculture, vol 7, pp 100–105 37. Albaloushi N, Azam M, Eissa A (2012) Mechanical properties of tomato fruits under storage conditions. J Appl Sci Res 8:3053–3064 38. Fluck R, Halsey L (1973) Impact forces and tomato bruising. Fla Agric Exp Stn J Ser 5109:239– 242 39. Chen P, Ruiz-Altisent M, Barreiro P (1996) Effect of impact mass on firmness sensing fruits. Trans ASAE 39(3):1019–1023 40. Gołacki K, Bobin G, Stropek Z (2009) Bruise resistance of apples (Melrose variety). Teka Komisji Motoryzacji i Energetyki Rolnictwa 9:40–47 41. Li, X, Wang W (1996) The research on mathematical model of apple impact response. Trans CSAE, 12(4):204–207 42. Lien C, Hurng H (2007) The study of bruise for apple. J Agric Mach 16(1):49–59

References

125

43. Lu L (2008) Cushioning dropping modeling of a falling apple on the corrugated board. Trans CSAE 24(9):276–280 44. Lu L (2009) Non-linear viscoelastic modeling of the fruits under dropping impact. Eng Mech 26(4):228–233 45. Groves J (1985) Predicting physical properties of tomatoes with impact force analysis. The Ohio State University

Chapter 3

The Physical and Mechanical Properties of Tomato Fruit and Stem

3.1 Summary 3.1.1 Research Significance Usually, the robotic one-by-one harvesting action on tomato fruits consists of two basic motion elements—gripping and detaching. To discover the physical and mechanical properties of tomato fruit and stem are the basic premise for achieving damage-free harvesting. The mechanical properties of the fruit are the external expression of its internal macro-microstructure and mechanical structure. The structure of tomato and apple, pear, and other fruits is obviously different. The analysis of its internal macro–micro structure and bearing capacity is helpful to understand the behavior of its external mechanical properties and to further reveal the mechanism of local damage under its overall loading.

3.1.2 Content and Innovation In this chapter, property tests and analysis of tomato fruit, stem, and fruit-stem joint were carried out for the needs of robotic damage-free harvesting operation. The main innovations are as follows: 1.

Construction of property index system

A complete fruit-stem physical/mechanical property index system as related to robot’s harvesting was established, and the methods of measurement and analysis of several key indices were put forward. Then the physical and mechanical properties of tomato fruit, stem, and fruit-stem joint were systematically measured and analyzed.

© Science Press, Beijing and Springer Nature Singapore Pte Ltd. 2021 J. Liu et al., Rapid Damage-Free Robotic Harvesting of Tomatoes, Springer Tracts in Mechanical Engineering, https://doi.org/10.1007/978-981-16-1284-8_3

127

128

2. (1)

3 The Physical and Mechanical Properties of Tomato Fruit and Stem

Macro–micro combined analysis of tomato fruit Macroscopic viscoelastic properties

The creep, stress relaxation, load–unload experiments of tomato fruits at different ripeness stages were carried out. Furthermore, a viscous-elastic constitutive model was established, whose parameters were decided for different ripeness levels. It provides a basis for explaining the dynamic response of the whole process of fingerfruit collision in speedy gripping. (2)

Micromechanical properties

The tensile, compressive, bending, and shear properties of tomato epidermis, pericarp, and gel tissue were systematically analyzed, and the viscoelastic properties of different components of tomato fruit under different loads were discovered. (3)

Macro–micro combined bearing structure

The simplified wheel-like load-bearing structure of tomato fruit was proposed, which can effectively explain the anisotropy, the change with ripeness level, and the position difference of the mechanical properties of tomato fruit. (4)

Damage evaluation

According to the biological theory of fruit damage, the multi-physiological changes and their biological mechanisms of gripping damage were proposed for tomato fruit, and the shelf life was determined as the evaluation index of internal damage of tomato fruit. (5)

Gripping and mechanical damage

According to the experiment results, the relationship was successfully established among the mechanical properties of tomato fruit, the action parameters of gripping, and the mechanical damage. Also the probability curve of fruit gripping damage under different conditions was obtained in the following chapters. As a result, the gap among the gripping operation, the fruit properties, the fruit damage evaluation, and fruit damage probability was broken, which provides a basis for achieving flexible harvesting control.

3.2 The Physical/Mechanical Properties Index System of Tomato Fruit-Stem Related to Robot’s Harvesting In view of the previous studies on fruit physical/mechanical properties, which the main focus was on fruit storage and mechanical damage prevention and a few studies focused on mechanized harvesting of fruits, while the measurement and analysis

3.2 The Physical/Mechanical Properties Index System …

129

Table 3.1 Fruit-stem physical/mechanical property index system related to robotic harvesting Index object

Fruit

Stem (Abscission layer)

Physical properties

Structure and geometry

Structure of stem systema

Height/Diameter/Mass/Density/Porosity Pedicel length Morphological characteristics (Sphericity/Fruit shape index)

Pedicel diameter

Static/Dynamic sliding friction coefficient

Stem length

Rolling resistance coefficient Mechanical properties Compression force/deformation Compression rupture force/deformation

Stem diameter Bend-off momenta Bending elastic coefficienta

Creep

Pull-off force/deformation

Stress relaxation

Twist-off torque/deformation

Load–unload a The

measurement methods were first proposed by the authors

indicators for robotic selective harvesting were incomplete, and the lack of analytical analysis methods, we es lished a fruit-stem physical/mechanical property index system related to robot’s harvesting. Furthermore, an indirect measurement and analysis method for the fracture moment of fruit stems and for the bending elastic coefficient of fruit stem while with a method for measuring fruit-stem breaking force/deformation and measuring torque/deformation of twisting were proposed and constructed. The measurement and data observation records of each characteristic index shown in the above table were completed, which lay a foundation for further theoretical modeling analysis and control implementation (Table 3.1).

3.3 Physical Properties of Tomato Fruit and Stem 3.3.1 Structure of Tomato Fruit and Stem 1.

Morphological structure of tomato fruit and stem

The morphological structure of tomato fruit and fruit stem is shown in Fig. 3.1. The fruit is approximately an ellipsoid, and the radial section is approximately circular. According to the fruit shape index (diameter/height), it can be divided into the flat circular shape, the elliptical shape, the nearly circular shape, and the long circular shape. A pedicel is a stem that attaches a single fruit to the cluster, while a peduncle is the stem or branch from the main stem of the cluster that holds a group of pedicels. The abscission layer is a relatively simple structure located about halfway through the peduncle and consists of a band of anatomically distinct cells in which cell separation occurs for the shedding of fruits [1]. The presence of the abscission layer

130

3 The Physical and Mechanical Properties of Tomato Fruit and Stem

Peduncle

Axial Abscission layer

Pedicel

Calyx

Radial

Fig. 3.1 Morphology of tomato fruit and stem

is an important trait for manual and robotic harvesting as it makes fruit detachment easier while retaining both the calyx and the distal end of the pedicel [1]. 2.

Structure of tomato fruit

The structure of tomato fruit is shown in Fig. 3.2. Tomato fruit is a true fruit developed from the ovary [2]. Peel is developed from the ovary wall, which can be divided into three parts: exocarp, mesocarp, and endocarp. The exocarp, also known as the epidermis, is only a thin layer of cells, which is the outermost part of the fruit. The mesocarp is thicker and consists of several parenchyma cells and vascular bundles. The endocarp is from the inside epidermis of the carpel, which is a single layer of tissue. The gel part of the fruit is the soft tissue of the placenta that is fertilized and becomes hypertrophy to enclose the seed. Convenient for research, we name the tomato exocarp, the mesocarp, the seeds with the colloid tissue around as the

(a) Top view

(b) Longitudinal section

(c) Cross section

Fig. 3.2 Tomato fruit and its sectional view. 1. Fruit shoulder 2. Stalk pit 3. Fruit waist 4. Exocarp 5. Mesocarp 6. Locular gel(seeds with colloid tissue around) 7. Core 8. Radial wall 9. Locule 10. Fruit navel

3.3 Physical Properties of Tomato Fruit and Stem

(a) Exocarp

(b) Mesocarp

131

(c) Locular gel

Fig. 3.3 The tissue ultrastructure of tomato exocarp, mesocarp, and locular gel

exocarp, mesocarp, and locular gel. The ultra microstructure of exocarp, mesocarp, and locular gel of tomato fruit is shown in Fig. 3.3, which are scanned with a Leica Z16 APO scanning electron microscope (Magnification: 2.5×). In the same visible range, the cells of exocarp are smallest but they are in largest number, while the cells of mesocarp and locular gel tissue are larger but their arrangement is sparse and loose. The chemical constituents of the cell wall of fruit tissue are polysaccharides and proteins, and polysaccharides include cellulose, hemicellulose, and pectin. While the main chemical components of cell membranes are lipids and proteins, in addition to small amounts of glycoproteins, glycolipids, and trace amounts of nucleic acids. 3.

The number of tomato locules

The selected materials were two varieties of tomato, Fenguan906, and Jinguang28. The number of tomato locules is mainly determined by genes. Each Fenguan906 tomato fruit usually has 3–4 locules, while the locules number of Jinguang28 is normally 5–6. We define that tomato fruit with 3, 4, 5, and 6 locules are three-locular, four-locular, five-locular, and six-locular tomato, respectively. Fenguan906 tomato fruits are mainly three-locular and four-locular, and most of Jinguang28 tomato fruits are five-locular and six-locular (Fig. 3.4).

3.3.2 Physical Property of Tomato Fruit and Stem [3–5] 1.

Geometric properties of tomato fruit and stem

The study used Jinpeng1, Jinpeng5, and Fenguan906 which are widely cultivated in Zhenjiang City. From the Guantangqiao Vegetable Base, the Nanjing Military Region Vegetable Base, and the Zhenjiang Vegetable Research Institute, 50 tomatoes after the “Breakers” stage were taken. Their size and weight were measured using a vernier caliper and an electronic balance (Table 3.2).

132

3 The Physical and Mechanical Properties of Tomato Fruit and Stem

(a) Three-locular

(b) Four-locular

Fig. 3.4 Different locule number of tomato fruit

Table 3.2 The measurement result of physical properties of tomato fruit and stem Index

Maximum

Mean

Minimum

Fruit diameter (mm)

104.7

71.3

45.9

Fruit shape index (diameter/height) Pedicel length (mm) Pedicel diameter (mm) Stem length (mm) Stem diameter (mm)

1.09 15.8 7.0 141.6 8.97

0.91 11.5 3.77 66.2 5.47

0.75 8.0 2.3 15.3 3.39

Observation and experiment showed that the ripeness level of tomato fruit and the size, shape, and weight of ripe tomato fruit on the same plant were quite different. The pedicel was shorter, and the attachment points of peduncles of several tomato fruits in same cluster were close to each other, causing the fruits to interact with each other with different postures. It was found that the physical properties of each variety were close. More than 80% of the tomato fruits were between 60 and 80 mm in diameter, and over 95% were between 50 and 90 mm, which was close to the normal distribution (Fig. 3.5). The stem diameter was between 3.3 mm and 4.8 mm, but the length of the pedicels and the peduncles was very different. The high diversity of either tomato fruit distribution or size of tomato fruit and stem posed a challenge to robotic one-by-one selective harvesting. 2. 1)

Properties of fruit mass and density Measurement of density and porosity

Density is the ratio of the mass of a material to its volume. The porosity is the ratio of the pore volume within the material to the total volume of the material. The mass of mesocarp and gel slices of tomato fruit samples was measured by electronic scale

3.3 Physical Properties of Tomato Fruit and Stem 120

Probability density /%

6 Probability density Cumulative density

5

100

4

80

3

60

2

40

1

20

0 0

20

40

60

80

100

Cumulative density /%

Fig. 3.5 Normal distribution of tomato fruit diameter

133

0 120

Diameter /mm

and the volume was measured by water displacement method. The bulk density ρb of tomato, the density ρp , ρg of mesocarp, and gel were calculated by Eq. (3.1), and the porosity e of tomato was calculated by Eq. (3.2) [6–8] ρ= e=

V − (V p +

1 (M ρg

V

m V − ρ p V p ))

(3.1)

× 100 %

(3.2)

where m—the mass of material (g); M—the total mass of tomato (g); V —the total mass of tomato (cm3 ); V p —the volume of mesocarp (cm3 ); ρ p —the density of mesocarp (g/cm3 ); ρ g —he density of the gel (g/cm3 ). 2)

Results

More than 90% of the tomato fruits were between 100 and 300 g in mass (Fig. 3.6). Respectively, the weight of the mesocarp and gel accounted for 75.1–80.0% and 20.0–24.9% of the total weight. The densities of mesocarp and gel for three-locular and four-locular tomatoes from Fenguan 906 tomato fruits were 1.05~0.96 g/cm3 and 1.09~1.04 g/cm3 , while the densities of those for five-locular and six-locular tomatoes from Jinguang28 tomato fruits were 1.01~0.99 g/cm3 and 0.95~1.07 g/cm3 . The porosity for three-locular and four-locular tomatoes from Fenguan906 tomato fruits was 6.49–11.5%, and this for five-locular and six-locular tomatoes from Jinguang28 tomato fruits was 9.39%–4.78% (Table 3.3).

134

3 The Physical and Mechanical Properties of Tomato Fruit and Stem

Fig. 3.6 Normal distribution of tomato fruit mass

Table 3.3 Physical properties of two varieties of tomato fruit Parameter

Fenguan906

Jinguang28

Significance test

Three-locular

Four-locular

Five-locular

Six-locular

Porosity e (%)

6.49 ± 5.47

11.50 ± 8.77

9.39 ± 7.40

4.78 ± 5.91

ns

Bulk density ρ b

0.95 ± 0.07

1.04 ± 0.06

0.92 ± 0.09

0.93 ± 0.07

ns

Mesocarp density ρ p

1.05 ± 0.07

1.09 ± 0.12

1.01 ± 0.14

0.95 ± 0.07

Gel density 0.96 ± 0.16 ρg

1.04 ± 0.13

0.99 ± 0.06

1.07 ± 0.12

ns means not significant

3.4 Mechanical Properties of Tomato Fruit Components [3] 3.4.1 Material, Equipment, and Method 1.

Material

The selected materials were tomato fruit of two varieties in “Light red” stage, Fenguan906 and Jinguang28. When the fruits were shipped to the laboratory, their surface was washed and air dried. As shown in Fig. 3.7, each tomato fruit was cut into quarters with a sharp knife along the longitudinal axis of the tomato; the exocarp, mesocarp, and gel tissue were separated from each quarter, and finally a standard sample was prepared. The determination of mechanical properties of tomato exocarp, mesocarp, and locular gel tissue was performed within 24 h at room temperature (25.9 °C:56.6% RH). 2.

Equipment

3.4 Mechanical Properties of Tomato Fruit Components [3]

135

Fig. 3.7 The exocarp, mesocarp, and locular gel sample of tomato fruit

The equipment used in the test are TA-XT2i Texture Analyzer (SMS company, UK); 500-196-20 Electronic digital vernier caliper (Mitutoyo company, Japan, accuracy 0.01 mm); SF-400 Electronic scale (Botai Weiye Electronic Technology Ltd., Suzhou, range: 0–1000 g, accuracy: 0.1 g); Kestrel 4000 Handheld meteorological station (temperature range: −29–70 °C, accuracy: 0.1 °C, humidity range: 0–100%, accuracy: 0.1%). In this test, P50 probe, A/TG probe, HDP/KBS probe, and HDP/3 PB probe were used for the corresponding compression test, tensile test, shear test, and bend test. 3.

Method

The mechanical properties of tomato exocarp tissue were determined by tensile testing and shear testing, respectively; the mechanical properties of mesocarp tissue were determined by compression testing, tensile testing, shear testing, and bend testing, respectively; the mechanical properties of locular gel tissue were determined by compression testing and shear testing, respectively. The analyzer was calibrated with a 5 kg weight before the first test, and set the speed before and after loading to 1 mm/s while the loading speed was set to 0.1 mm/s. The samples were labeled before test and its thickness d was measured with an electronic digital caliper to an accuracy of 0.01 mm. 10 replications were performed for each group of trials. 1)

Compressive properties

The dimensions (length × width) of the prepared tomato mesocarp tissue and locular gel tissue specimens were 10 mm × 5 mm and 16 mm × 8 mm for compression testing respectively. After prepared, put it at the center of base plate, and the specimen was compressed along its length direction by P50 parallel plate probe, the compressibility level was set to 60%, as shown in Fig. 3.8. The force–deformation data were recorded by computer in real time. Subsequently, the mechanical parameters of tomato mesocarp and locular gel tissue such as compression elastic modulus, failure stress, failure strain, and failure energy were extracted from the obtained force–displacement curve.

136

3 The Physical and Mechanical Properties of Tomato Fruit and Stem

(a) mesocarp specimen

(b) locular gel specimen

Fig. 3.8 Compression testing

Equations for mechanical parameters were shown as follow [9–11]: σc =

Fcmax Fcmax = Ac wd

(3.3)

where σ c —the failure stress during compression testing (MPa ); F cmax —the elastic peak force (N); Ac– Ac is the specimen’s cross-sectional area (mm2 ); w—the specimen’s cross-sectional width (mm); d—the specimen’s cross-sectional thickness (mm). εc =

L Ls

(3.4)

where εc —the failure strain during compression testing (%); L—the length different before and after test (mm); L s —the initial length of specimens (mm). Ec =

σc 2ε0.5σc

(3.5)

where E c —the compression elastic modulus (MPa); ε0.5σ c —the strain at 50% of the yielding point stress on the stress–strain curves (%).

3.4 Mechanical Properties of Tomato Fruit Components [3]

137

εc Er ec =

σ dε

(3.6)

0

where E rec —the failure energy during compression testing(kJ/m3 ); σ —the normal stress during compression(MPa ); ε—the normal strain during compression (%). 2)

Tensile properties

The dimensions (length × width) of the prepared tomato exocarp tissue and mesocarp tissue specimens were 40 mm × 3 mm for tensile testing, respectively. After prepared, the specimen was tensiled along its length direction by A/TG probe, and the tensile displacement was set to 10 mm. The distance between the upper and lower probes was measured before testing, as shown in Fig. 3.9. As the test was going on, the force–deformation data were recorded by computer in real time. Subsequently, the mechanical parameters of tomato exocarp and mesocarp tissue such as tensile elastic modulus, failure stress, failure strain, and failure energy were extracted from the obtained force–displacement curve. Equations for mechanical parameters were shown as follow [11–16]: σl =

Flmax Flmax = Al wd

where σ l —the failure stress during tensile testing (MPa );

(a) exocarp specimen Fig. 3.9 Tensile testing

(b) mesocarp specimen

(3.7)

138

3 The Physical and Mechanical Properties of Tomato Fruit and Stem

F lmax —the elastic peak force (N); Al —the specimen’s cross-sectional area (mm2 ); w—the specimen’s cross-sectional width (mm); d—the specimen’s cross-sectional thickness (mm). εl =

L Ld

(3.8)

where εl —the failure strain during tensile testing (%); L—the length different before and after test (mm); L d —the initial distance between the upper and lower probes (mm). El =

σl 2ε0.5σl

(3.9)

where E l —the tensile elastic modulus (MPa); ε0.5σ l —the strain at 50% of the yielding point stress on the stress–strain curves (%). εc Er el =

σ dε

(3.10)

0

where E rel —the failure energy during tensile testing (kJ/m3 ); σ —the normal stress during tensile (MPa ); ε—the normal strain during tensile (%). 3)

Shear properties

The dimensions (length × width) of the prepared tomato exocarp tissue and mesocarp tissue specimens were 40 mm × 5 mm for shear testing, respectively, and the dimensions of locular gel tissue specimens were 40 mm × 10 mm. After prepared, the specimen was sheared along its thickness direction by HDP/BS probe until it fractured, as shown in Fig. 3.10. As the test was going on, the force–displacement data were recorded in real time. Subsequently, the shear strength of tomato exocarp, mesocarp, and locular gel tissue was extracted from the obtained force–displacement curve. Equations for mechanical parameters were shown as follow [12, 14, 17–19]: τs =

Fsmax Fsmax = As wd

(3.11)

3.4 Mechanical Properties of Tomato Fruit Components [3]

(a) exocarp specimen

(b) mesocarp specimen

139

(c) locular gel specimen

Fig. 3.10 Shear testing

where τ s —the shear strength during shear testing (MPa ); F smax —the rupture force in shear (N); As —the specimen’s cross-sectional area (mm2 ); w—the specimen’s cross-sectional width (mm); d—the specimen’s cross-sectional thickness (mm). 4)

Bending properties

The dimensions (length × width) of the prepared tomato mesocarp tissue specimens were 40 mm × 5 mm for bend testing. After prepared, the specimen was loaded along its thickness direction based on three-point bending test by HDP/3 PB probe until it fractured. The span of specimen was set to 16 mm, as shown in Fig. 3.11. As the test was going on, the force-deflection data were recorded by computer in real time. Subsequently, the bend mechanical parameters of tomato mesocarp tissue such as bend strength and maximum deflection were extracted from the obtained

Fig. 3.11 Bending testing

140

3 The Physical and Mechanical Properties of Tomato Fruit and Stem

Fig. 3.12 Bending moment diagram

force–displacement curve. The theoretical analysis of three-point bending test was shown as follow [20–22]: (1)

The maximum bending moment of the beam Mmax is shown as follow:

The three-point bending test can be simplified as Fig. 3.12a. The bending moment diagram of the beam was shown as Fig. 3.12b. Bending moment function: M=

c1  F  x 0 “Green” stage > “Pink” stage > “Light red” stage while the ratio of rupture force to peak force N c0 /N cp is exactly the opposite: “Breakers” stage < “Green” stage < “Pink” stage < “Light red” stage.

3.5.2 Creep Properties [33] 1.

Materials

The material selected for the test was Jinpeng5 tomato, which was hand-picked from the Guantang Bridge Tomato Planting Base in Zhenjiang City. According to the definition of tomato ripeness in Chinese National Standard SB/T10331-2000 and United States Standard 51.1860 Color Classification (There are some differences), 10 tomatoes were randomly picked for each of the “Breakers” stage (the surface of the tomato is completely green in color and turns white-green), “Turning” stage (less than 10% of the surface, in the aggregate, shows a definite change in color from green to tannish-yellow, pink, red, or a combination thereof), the “Pink” stage (more than 10% but not more than 40% of the surface, in the aggregate, shows pink or red color), and the “Light red” stage (more than 40% but not more than 60% of the surface, in the aggregate, shows pink or red color). The tomatoes had a radial diameter of 67.25–82.77 mm (perpendicular to the direction of the stem), an axial diameter of 55.84–75.12 mm (parallel to the direction of the stem), and a mass of 143.9–250.8 g. The test was completed within 36 h after harvest. 2.

Methods and equipment

The test was carried out on a TA-XT2i Texture Analyzer (Stable Micro-Systems, UK) selecting a 5 mm diameter cylindrical probe for puncture test of tomato fruit. Set the mode to creep test, first loaded the fruit at the equatorial position to a constant load value of 5 N according to the puncture loading pre-test quickly, kept the constant load time for 50 s, and then quickly unloaded, and the return speed before and after the test was 1.0 mm/s. A single sample test was completed after 50 s of unloading. The data about force, deformation, and time during the test were automatically recorded and saved by the device. The data sampling frequency of the device was 200 points/s. 3.

Results

20,000 data points were obtained for each sample. The force–deformation and deformation–time curves of creep and unloading tests of tomato fruits with different

154

3 The Physical and Mechanical Properties of Tomato Fruit and Stem

ripeness were shown in Fig. 3.23a and b, respectively. During the test, the load was first quickly loaded to a constant load (loading stage), and the fruit produced an instantaneous initial deformation, the probe remained a constant load for 50 s (creep stage), the tomato fruit was creep-deformed, the deformation gradually increased, and then unloaded quickly (unloading stage), the fruit recovered for a certain instantaneous deformation, and then recovered slowly until a certain permanent deformation was maintained. According to Fig. 3.23, as the ripeness increased, the instantaneous initial deformation, creep deformation, and permanent deformation increased too. The typical creep properties of tomatoes of four different ripeness were shown in Fig. 3.23. It can be seen that the ripeness has a significant influence on the creep curve. The critical creep strain of the two ripeness stages of tomato increased significantly with the increase of ripeness. The relationship between the strain and the ripeness is “Light red” stage > “Pink” stage > “Turning” stage > “Breakers” stage, which reflects that the firmness of tomato fruit decreases with the increase of ripeness. In the process of tomato creep, the deformation is divided into two parts, the compression deformation and creep deformation. The creep deformation of different ripeness stages was calculated by the software of the texture analyzer, as shown in Table 3.10. Obviously, the creep deformation increases as the ripeness level increases for same constant loading force. This is because the content of insoluble pectin in the tomato with lower ripeness is higher, so the tissue is hard and not deformed. As the ripeness of the tomato increases, the insoluble pectin in the fruit gradually degrades

(a) Force-deformation curve

(b) Deformation-time curve

Fig. 3.23 Creep-recovery test curve of tomato fruit

Table 3.10 Creep deformation of different ripeness stages Ripeness stage

Maximum (mm)

Minimum (mm)

Mean (mm)

Standard deviation

Breakers

0.277

0.225

0.248

0.027

Turning

0.478

0.333

0.405

0.0642

Pink

0.593

0.44

0.504

0.0559

Light red

0.602

0.485

0.553

0.0471

3.5 Compressive Mechanical Properties of the Whole Tomato

155

soluble pectin, and the fruit gradually changes soft and elastic [34, 35]. Therefore the fruit is more susceptible to deformation and mechanical damage under the same force, which is consistent with the research by Liu [36] and others. Therefore, in the process of mechanized harvesting, packaging, and transportation, the higher the tomato ripeness, the more likely it is to cause mechanical collision and loss.

3.5.3 Stress Relaxation Properties [33] The stress relaxation refers to the phenomenon that the viscoelastic object is loaded in a moment to produce an instantaneous deformation, and then the stress is reduced with time as the deformation is maintained. 1. 1)

Materials and methods Materials and equipment

The material selected for the test was Jinpeng5 tomato, which was hand-picked from the Guantang Bridge Tomato Planting Base in Zhenjiang City. According to the definition of tomato ripeness in Chinese National Standard SB/T10331-2000 and United States Standard 51.1860 Color Classification, 10 tomatoes were randomly picked for each of the green ripe stage, discoloration stage, the pre-term of red ripe stage, and the mid-term of red ripe stage. The tomatoes had a radial diameter of 62.9–79.75 mm (perpendicular to the direction of the stem), an axial diameter of 56.13–69.47 mm (parallel to the direction of the stem), and a mass of 136.0–239.8 g. The test was completed within 36 h after harvest. The equipment was same as the creep test. 2)

Methods

The preparation before the test was same as the creep test. The test was carried out on TA-XT2i Texture Analyzer (Measure Force in Compression, test mode: Hold until Time, pre-test, and post-test speed: 1.0 mm/s, hold time: 100 s, probe type: P100, stop plot at: Target). Other options were default. Then opened the Run a Test dialog and select the saved path. The data was analyzed by SPSS 17.0 mathematical statistics software. During the stress relaxation test, the probe compressed the sample to a set time after the set deformation and tried to realize the force of the probe over time to output the F-t curve. The compression-stress relaxation test can be divided into two parts: the compression stage and the relaxation stage. From the compression stage, parameters such as compression slope, peak force, and compression work (compression curve area), which reflect the rheological properties of the material, can be analyzed. The data obtained during the experiment needed to be exported from the dedicated software Texture Expert to Excel, and then the rheological model was fitted.

156

3 The Physical and Mechanical Properties of Tomato Fruit and Stem

Fig. 3.24 The stress relaxation curves of different ripeness stages

2. 1)

Results and analysis Stress relaxation test curve of tomato fruit with different ripeness

In order to remove the influence of the size of the tomato on the test, the compression amount was set to 5% of the diameter in the stress relaxation test, and the instantaneous compression deformation of the test was discarded, and the ratio of the measured force to the compression deformation of the corresponding tomato was used as the vertical axis. The stress relaxation curves at four different ripeness stages were shown in Fig. 3.24. It can be seen that the stress relaxation of tomatoes in different ripeness stages has a similar relationship. After loading to a set compression deformation, the amount was kept constant, then the stress relaxation of the tomato appeared, and the ratio of force to deformation was continuously reduced, that is, for the same deformation amount, the internal stress was continuously reduced. Ripeness has a significant effect on the stress relaxation properties of the whole tomato fruit. With the increase of ripeness level, the ratio of force to deformation of tomato stress relaxation process is decreasing, and the rate of decrease is also decreasing, indicating that the stress reduction of the stress relaxation process is getting slower. The stress relaxation curves of “Pink” stage were similar to that of “Light red” stage. The reason may be that the tomato is basically formed in the “Pink” stage, and the composition and pectin content are not significantly different. 2)

Maximum stress analysis of tomato stress relaxation at different ripeness

When the viscoelastic body is deformed by force, there is elastic stress that recovers deformation. However, since the internal particles also have a flowing property, when the internal particles are under the action of internal stress, the flow of each part reaches an equilibrium position. And when permanent deformation occurs, the internal stress disappears. The stress relaxation test process can be divided into the compression stage and the stress relaxation stage. When turning from the compression stage to stress relaxation stage, the maximum force appears, that is, the initial stress of relaxation. Figure 3.25 reflects the maximum force distribution of tomatoes

3.5 Compressive Mechanical Properties of the Whole Tomato

157

Fig. 3.25 The peak force of stress relaxation process of tomatoes in different ripeness stages

in four different ripeness stages in the test and the mean force of each ripeness. It can be seen that during the stress relaxation period, the maximum force reduces sharply during the green ripe stage to the discoloration stage. After that, the maximum force is slowly decreasing, reflecting the tendency of the force for a same compression rate decreases as the ripeness level increases.

3.5.4 Load–Unload Properties [33] The load–unload test is to determine the elastoplastic parameters of the material. From the load–unload cycle curve, the compression force, stiffness, elasticity, and hysteresis loss of the loading stage can be analyzed. 1. 1)

Materials and methods Materials and equipment

The material selected for the test was Jinpeng5 tomato, which was hand-picked from the Guantang Bridge Tomato Planting Base in Zhenjiang City. According to the definition of tomato ripeness in Chinese National Standard SB/T10331-2000 and United States Standard 51.1860 Color Classification, 10 tomatoes were randomly picked for each of the green ripe stage, discoloration stage, the pre-term of red ripe stage, and the mid-term of red ripe stage. The tomatoes had an radial diameter of 61.22–80.82 mm (perpendicular to the direction of the stem), an axial diameter of 52.97–69.36 mm (parallel to the direction of the stem), and a mass of 132.2–241.5 g. The test was completed within 36 h after harvest. The equipment was same as the creep test.

158

2)

3 The Physical and Mechanical Properties of Tomato Fruit and Stem

Methods

The preparation before the test was same as the creep test. The test was carried out on TA-XT2i Texture Analyzer (Measure Force in Compression, pre-test and post-test speed: 1.0 mm/s, compression amount: 12%, hold time: 100 s, probe type: P100, stop plot at: Final). Other options were default. Then opened the Run a Test dialog and select the saved path. The data was analyzed by SPSS 17.0 mathematical statistics software. 2. 1)

Results and analysis Typical analysis curve of tomato load–unload process

Typical load–unload curves (F-D) for tomatoes of different ripeness are shown in Figs. 3.26 and 3.27. Tomatoes of different ripeness stages have similar curve linear relationships in the load–unload test. The curve can be divided into two parts: the loading period (AB) and the unloading period (BC). The area of the closed loop is the plastic strain energy of the tomato. AD is the total deformation D generated during the loading process, AC is the plastic deformation Dp generated during the loading process, and CD is the elastic deformation De generated during the loading Fig. 3.26 Typical load–unload curve

Fig. 3.27 Typical load–unload curves for tomatoes of different ripeness

3.5 Compressive Mechanical Properties of the Whole Tomato

159

process, and there is an equation D = Dp + De for them. The abscissa of point B is the deformation of tomato under the corresponding compression rate, and the ordinate is the maximum force F max of tomato at the same compression rate. 2)

Comparison of maximum force of tomatoes in different ripeness stages

The peak force changes during the loading process of tomatoes in different ripeness stages are shown in Fig. 3.28. It can be seen that the ripeness has a significant effect on the peak force in the period from “Breakers” stage to “Turing” stage. But during the transition of “Turning” stage, “Pink” stage, and “Light red” stage, there was no significant change in peak force. 3)

Comparison of hysteresis loss of tomatoes in different ripeness stages

Hysteresis loss is the energy absorbed by the fruit during the load–unload cycle and can be used as a measure of damping capacity. The closed area enclosed by load–unload curves is the value of the hysteresis loss. It can be seen that during the “Breakers” stage to the “Turning” stage, the hysteresis loss is significantly reduced, while in the “Turning” stage, the “Pink” stage, and the “Light red” stage, the hysteresis loss decreases slowly (Fig. 3.29). Fig. 3.28 Comparison of maximum loading force of tomatoes in different ripeness stages

Fig. 3.29 Comparison of hysteresis loss of tomatoes in different ripeness stages

160

3 The Physical and Mechanical Properties of Tomato Fruit and Stem

Fig. 3.30 Comparison of elasticity of tomatoes in different ripeness stages

4)

Comparison of elasticity of tomatoes at different ripeness stages

The elasticity indicates the degree of recovery after load–unload. It can be seen from Fig. 3.30 that the elasticity of tomato has a significant increase with the increase of ripeness. The ability of tomato fruit to recover after loading is “Light red” stage > “Pink” stage > “Turning” stage > “Breakers” stage.

3.6 Frictional Mechanical Properties of Tomato Fruits [3] 3.6.1 Static and Sliding Friction Coefficients 1.

Materials and methods

Three materials, namely, 307 stainless steel, Lacquered stainless steel and Rubber were tested in this trial to determine the friction and rolling resistance coefficients of tomato fruits on these surfaces. Friction coefficients were measured through an electric horizontal single-column test stand (model AEL, Ali Instrument Co., Ltd., Zhejiang, China), as shown in Fig. 3.31. First, a material board (90 × 200 mm) on the test stand with bolts was fixed, the fruit sample was put on the material surface and the adhesive tape that pasted on the equatorial region of fruit was connected to the hook of HF-50 digital push–pull gauge (measuring range: 0–50 N, sensitivity: 0.01 N) with a cord. Subsequently, the horizontal movement velocity of push–pull gauge was set at 1.5 mm/s by control panel. The static and sliding friction coefficients were calculated by the following equations [37, 38]: μs = μd =

Fmax FN

(3.23)

Fd FN

(3.24)

3.6 Frictional Mechanical Properties of Tomato Fruits [3]

161

Fig. 3.31 The static and sliding friction coefficients

where μs —the static coefficient of friction; μd —the sliding coefficient of friction; F max —the maximum value of static friction force, which is equal to the horizontal pulling force as the tomato started to move(N); F d —the sliding friction force, which is equal to the average value of horizontal pulling force during the movement of tomato(N); F N —the normal force between the surfaces, which is equal to the weight of the tomato fruit (N). 2.

Results

Table 3.11 shows the calculated static and sliding friction coefficients between the tomato fruit and any of the stainless steel, lacquered stainless steel, and rubber surface. For Fenguan906 tomato fruit, the static friction coefficients ranged from 0.375 to 0.488 for stainless steel, 0.408–0.641 for lacquered stainless steel, and 0.396–0.503 for rubber. While for Jinguan28 tomato fruit, it was 0.437–0.483 for stainless steel, 0.612–0.622 for lacquered stainless steel, and 0.511–0.534 for rubber. For Fenguan906 tomato fruit, the sliding friction coefficients ranged from 0.352 to 0.47 for stainless steel, 0.387–0.618 for lacquered stainless steel, and 0.383–0.474 for rubber. While for Jinguan28 tomato fruit, it was 0.409–0.453 for stainless steel, 0.587–0.593 for lacquered stainless steel, and 0.491–0.507 for rubber. Two-factor analysis of variance showed the variety, locule number and material had a significant effect on the static and sliding friction coefficients of tomato fruits according to the F test. In the mean comparison test of Fisher’s LSD, the Jinguan28 variety was statistically higher than the Fenguan906 variety in respect to static and sliding friction coefficients of tomato fruits, their means of static and sliding friction coefficients were 0.5394–0.4434, and 0.5120–0.4227, respectively. This can be attributed to its round shape, water content, and firm texture of tomato fruits according to Alayunt’s observations [39], because the sphericity and loading slope of Jinguan28 tomato fruits were higher than Fenguan906 in the obtained physical parameters. The friction coefficients of four-, five-, and six-locular tomatoes had no

0.612 ± 0.170

0.622 ± 0.144

0.437 ± 0.091

0.483 ± 0.122

Five

Six

0.534 ± 0.063

0.511 ± 0.060

0.503 ± 0.068 0.453 ± 0.137

0.409 ± 0.074

0.470 ± 0.088

0.352 ± 0.090

0.641 ± 0.027

0.488 ± 0.089

Four

0.396 ± 0.067

0.408 ± 0.053

0.375 ± 0.077

Three

0.593 ± 0.143

0.587 ± 0.173

0.618 ± 0.002

0.387 ± 0.056

Lacquered stainless steel

Sliding friction coefficient Stainless steel

Lacquered stainless steel

Stainless steel

Rubber

Static friction coefficient

Locule number

Table 3.11 Static and sliding friction coefficients of tomato fruits

0.507 ± 0.062

0.491 ± 0.054

0.474 ± 0.036

0.383 ± 0.060

Rubber

162 3 The Physical and Mechanical Properties of Tomato Fruit and Stem

3.6 Frictional Mechanical Properties of Tomato Fruits [3]

163

significant differences from each other, but the friction coefficient of three-locular tomato fruits was significantly different from four-locular tomato fruits. The mean friction coefficient of three-locular tomato fruits was less than four-locular tomato fruits. The highest static and sliding friction coefficients were on lacquered stainless steel, and their means were 0.595 and 0.5681, respectively. The rubber and stainless steel showed no significant differences in static friction coefficients of tomato fruits, whereas, the higher sliding friction coefficient was on rubber than stainless steel. This is due to the properties of friction surfaces. Similar results were found by Kabas and Ozmerzi for cherry tomato fruits, Jannatizadeh et al. for Iranian apricot, Naderiboldaji et al. for sweet cherry, Jahromi et al. for date, and Caliir et al. for wild plum [29, 40–43].

3.6.2 Measurement of Rolling Resistance Coefficient 1.

Materials and methods

As can be seen from Fig. 3.32, the slope of the surfaces can be changed continuously by the vertical movement of digital push–pull gauge. The tomato sample was placed as shown in the figure; the slope was tangent to the fruit shoulder and the longitudinal axis of fruit was perpendicular to the slope. The slope increased following the rise of push–pull gauge. The angle θ at which the initial movement of the tomato was recorded by a protractor and the rolling resistance coefficient μR was calculated as the tangent of the angle θ. These methods of determining the friction and rolling resistance coefficients have been used by several researchers [29, 39].

Fig. 3.32 The rolling resistance coefficients

164

3 The Physical and Mechanical Properties of Tomato Fruit and Stem

Table 3.12 Rolling resistance coefficient of tomato fruits Variety Fenguan906 Jinguang28

2.

Locule number

Rolling resistance coefficient Stainless steel

Lacquered stainless steel

Rubber

Three

0.516 ± 0.061

0.524 ± 0.104

0.435 ± 0.027

Four

0.531 ± 0.021

0.577 ± 0.033

0.488 ± 0.061

Five

0.530 ± 0.080

0.577 ± 0.089

0.512 ± 0.089

Six

0.505 ± 0.059

0.533 ± 0.077

0.519 ± 0.049

Results

In respect to rolling resistance coefficients of tomato fruits (Table 3.12), three-factor analysis of variance showed that the variety, locule number and material had no significant effect on the rolling resistance coefficients of tomato fruits according to the F test. The mean of rolling resistance coefficient was 0.5208. During the process of grasping tomato fruit using robot’s finger, the surface materials that contact directly with tomato fruit usually include stainless steel, lacquered stainless steel, and rubber. Therefore, in order to prevent that the grasped tomato slip through the robot’s finger, the grasping force should not be less than 1.51 N for stainless steel, 1.15 N for lacquered stainless steel, and 1.46 N for rubber as the mass of Fenguan906 four-locular tomato fruits is 150 g according to viewpoints of Chen P. and Glossas N. of sliding detection [44, 45].

3.7 Mechanical Structure Model of the Whole Tomato Fruit 3.7.1 The Wheel-like Simplification Mechanical Structure of Fruit [4, 46] 1.

Anisotropy of tomato fruit resistance to compression

The anti-compression ability of tomato fruit is mainly determined by the internal structure of the fruit, except for the strength of the exocarp. The pericarp is composed of exocarp, mesocarp, and endocarp. The mesocarp is succulent and usually composed of several layers. The interior of the fruit is divided into 5–8 locules by the dividing wall connected with the mesocarp and the core. The locule will develop as a placenta with seeds surrounded by a layer of gel inside [47] (Fig. 3.33a). The cross section of the fruit can be simplified as a wheel-like force structure (Fig. 3.33b), where the mesocarp, the radial wall, and the core correspond to the rim, the spoke, and the hub, respectively. From the “Green” stage to the “Breakers” stage, the internal structure of the fruit gradually develops and the load capacity of the wheel structure reaches the greatest. With the increase of ripeness, the core and

3.7 Mechanical Structure Model of the Whole Tomato Fruit

165

Exocarp Mesocarp Rim

Endocarp

Spoke

Radial wall Placenta Seeds Locule tissue Hub Core

Vascular bundle

(a) Structure of cross section

(b) Wheel-like structure of cross section

Fig. 3.33 The cross section and its wheel-like structure of tomato fruit

the radial wall continued to fluidize, and the spoke and hub functions in the wheellike structure disappear rapidly. The mesocarp is also softened and the fruit carrying capacity is declining. 2.

Relationship between compression strength and ripeness of tomato fruit

Whether the compression test carried out in axial or radial, the concaves appear in the axial section, so the axial section of the fruit is further simplified to the ringshaped structure shown in Fig. 3.34a, b, or the symmetrical structure of two arched beams. The fruit consists of several ring-shaped micro-elements g (Fig. 3.34c). When the fruit is axially loaded, its anti-compression ability Nz is a superposition of the anti-compression ability of all axial ring-shaped micro-elements as follow: Nz =

n 

N (gi )

i=1 π

=

N0 dϕ 0

= π N0

(3.25)

where ϕ—the angle of projection of the ring-shaped structure g in the radial plane xOy and x-axis; N(g) = N 0 ϕ—the anti-compression ability of the ring-shaped structure g(N). When loaded radially, the bearing capacity N h is only a superposition of the radial loading direction component of the bearing capacity of several axial ring-shaped structures:

166

3 The Physical and Mechanical Properties of Tomato Fruit and Stem

Fig. 3.34 The simplified ring-shaped mechanical structure of tomato fruit

Mesocarp

Fh

Fz Mesocarp

Core Core

(a) The axial loading mechanical structure

(b) The radial loading mechanical structure

(c) The ring-shaped micro-element Δg

Nh =

n 

N (gi ) sin ϕi

i=1 π

=

N0 sin ϕdϕ 0

= 2N0

(3.26)

When loading on the whole fruit axially and radially, the fruit core is subjected to compression and tensile force respectively. As the viscoelastic body, the compression strength of the fruit core is much higher than the tensile strength. Based on the above two factors, the axial anti-compression ability of tomato is significantly stronger than that of radial.

3.7.2 Mechanical Properties of Tomatoes with Different Numbers of Locules [3] The effects of different locule numbers on the mechanical properties of tomato fruit were obtained by load–unload tests.

3.7 Mechanical Structure Model of the Whole Tomato Fruit

1. 1)

167

Materials and methods of test 1 Materials and equipment

The test was carried out in the laboratory of the College of Food and Bioengineering, Jiangsu University. Fresh tomato fruits from Fenguan906 tomato of two structure types were used in this study: three-locule (T ) and four-locule (F). Tomatoes were uniformly grown at the Ruijing Vegetable Research Institute. 50 tomatoes of three-locule and four-locule were hand harvested respectively at the semi-ripe stage according to United States standards for grades of fresh tomatoes [48, 49]. The test was carried out on a TA-XT2i texture tester (SMS company, UK), with the 500-196-20 electronic digital display vernier caliper (Japan Sanfeng Measuring Tool Factory, precision 0.01 mm) and SF-400 electronic scale (Suzhou Botai Weiye Electronic Technology Co., Ltd. (range: 0–1000 g, accuracy: 0.1 g)) used meanwhile. In the load–unload test, the parameters of the TA-XT2i texture tester were set as follows (test mode: Measure Force in Compression, the running program: Return to Start, the test parameters were set as follows: The loading speed was 0.5 mm/s, the speed before and after the test was 2 mm/s, the initial distance of the probe from the tomato was 10 mm, the curve recording method was Final, the probe type of the instrument was P100, a flat-panel probe with a diameter of 100 mm, the compression rate and loading position were shown in the test design). The texture is calibrated with a weight of 5 kg before testing. 2) (1)

Test design Test factors and levels

Locule number and loading position are two points which can reflect the internal structure of tomatoes. To obtain the mechanical properties of tomatoes with different structures, a full factorial design was used. The factors were including: ➀ ➁

➂

(2)

Two structure types: three-locular and four-locular tomatoes. As shown in Fig. 3.35. Two loading positions: locular: L and radial wall: CW tissue. Locular tissue is the pericarp over the locules (position2), whereas radial wall tissue is the pericarp located over the septum (position1). Position 1 (Fig. 3.35b and c) corresponded to the valley between two adjacent fruit shoulders (Fig. 3.35a), and position 2 (Fig. 3.35b and c) corresponded to the middle of one fruit shoulder (Fig. 3.35a). Five compression rate: 4, 8, 12, 16, and 20%. All tomatoes were individually labeled and grouped randomly before the test. All loadings were located at equatorial region. Physical parameters of tomatoes

At first, the physical properties of tomato fruits were measured at room temperature. The principal dimensions of tomatoes in each group, namely, the longitudinal axis through the stem containing the height, H, the transverse axis containing the

168

3 The Physical and Mechanical Properties of Tomato Fruit and Stem

(a) Three-locular tomato (left) and four-locular tomato (right)

(b) Equatorial cross section (left) and its simplified structure (right) of three-locular tomato

(c) Equatorial cross section (left) and its simplified structure (right) of four-locular tomato 1-Loading position of CW tissue 2-Loading position of L tissue

Fig. 3.35 Three-locular and four-locular tomato

3.7 Mechanical Structure Model of the Whole Tomato Fruit

169

maximum diameter, L max , the minimum diameter, L min , and the compression diameter, L c were measured by using a micrometer to an accuracy of 0.01 mm. The fresh mass m (g) of fruit was measured by an electronic balance to an accuracy of 0.01 g. From the principal dimensions, the geometric mean diameter, Dg (mm), the arithmetic mean diameter, Da (mm), and sphericity, ϕ [50], were calculated by using the Eqs. 2.3–2.5. The sphericity is an indicator of the shape of a spherical fruit, which indicates the degree of difference between the actual shape of the fruit and the sphere [51]. The geometric mean diameter and the arithmetic mean diameter are indicators of the particle diameter of the fruit, which combines the size in all directions of the fruit [52]. (3)

Mechanical parameters of tomatoes

After the test, the mechanical parameters of the tomato were extracted from the force–displacement curve, such as strain energy Es, elastic strain energy E e , plastic strain energy E p , peak force F max , elasticity rc, and loading slope rk . The typical load–unload force–displacement curve (F–D) was shown in Fig. 3.36 [53–56]. AB is the loading line of the loading period, and BC is the unloading line of the unloading period. The area of the closed-loop ABC is the plastic strain energy of the tomato, and its value E p can be calculated by the equation E p = E s − E e . The strain energy E s is the deformation potential energy that the tomato stores in its interior during the loading period, and its value is the area enclosed between the loading line and the displacement line in the force–displacement curve. The elastic strain energy E e is the energy released by the tomato during the unloading period and is the area enclosed by the unloading line and the displacement line in the force–displacement curve. AD is the total deformation amount D of the tomato during the loading period. AC is the plastic deformation amount Dp during the loading period. CD is the elastic deformation amount De during the loading period. And D equals to Dp plus De . The Fig. 3.36 Load–unload curve

170

3 The Physical and Mechanical Properties of Tomato Fruit and Stem

elasticity rc indicates the degree of recovery of the tomato after unloading, and its value is De /(De + Dp ). The slope of the loading line AB is the loading slope r k , which is the ratio of the loading force of the tomato to the corresponding displacement during the elastic deformation period. The abscissa of point B is the deformation of tomato produced under the corresponding compression rate, and its value is D = De + Dp . The ordinate of point B is the peak force F max of tomato applied at the corresponding compression rate. 3)

Statistical analysis

After the test, statistical analysis of variance was performed on the obtained data using SAS 9.1.3 software, assume that the significance level was α = 0.05. 2. 1)

Materials and methods of test 2 Materials and equipment

The test was carried out in the laboratory of the College of Food and Bioengineering, Jiangsu University. Fresh tomato fruits from Fenguan906 and Jinuang28 tomato of two structure types were used in this study: three-locule (T) and four-locule (F). Tomatoes were uniformly grown at the Ruijing Vegetable Research Institute. 50 tomatoes of threelocule and four-locule were hand harvested respectively at the “Light red” stage according to Chinese standard and United States standards for grades of fresh tomatoes. The equipment and its parameters were the same to test 1. 2)

Test design

In order to obtain the mechanical properties of tomatoes with different varieties of tomatoes and their internal structure, a full factorial design was used. The factors were including: (1) (2) (3)

Two varieties of tomatoes: Fenguan906 and Jinguang28. Four structure types: three-locular, four-locular, five-locular, and six-locular tomatoes. Two loading positions: locular: L and radial wall: CW tissue, as shown in Fig. 3.37. All tomatoes were individually labeled and grouped randomly before the test. All loadings were located at equatorial region. The tomato sample was placed on the base plate and pressed by the moving parallel plate probe until the fruit ruptured, and the force–deformation curve was recorded in real time. Subsequently, the mechanical properties such as rupture energy E r (equals to the E s in Fig. 3.36), rupture force F F (equals to the F max in Fig. 3.36), compressibility ε, and loading slope r k were extracted from each recorded curve.

3.7 Mechanical Structure Model of the Whole Tomato Fruit

a. Three-locular

b. Four-locular

171

c. Five-locular

d. Six-locular

1-Loading position of CW tissue 2-Loading position of L tissue

Fig. 3.37 Tomatoes with different locule number and their simplified structures of equatorial plane

3. 1) (1)

Results and analysis Result and analysis of test 1 Load–unload test

The mechanical and physical parameters extracted from the load–unload test were shown in Table 3.13. Each cross-term data represents the mean ± standard deviation of the corresponding property parameters of all tomatoes at the corresponding compression rate. Multivariate analysis of variance showed that the compressibility Table 3.13 Mechanical and physical properties of tomatoes Mechanical Compressibility ε and 0 4% physical properties

8%

12%

16%

20%

E p (mJ)◆

0

7.21 ± 1.97 42.16 ± 15.41 101.17 ± 35.99 209.09 ± 59.38 368.73 ± 128.9

F max (N)◆

0

9.44 ± 2.55 25.97 ± 8.16

38.54 ± 10.16

54.88 ± 13.47

63.13 ± 13.5 0.41 ± 0.05

rc ◆

0.63 ± 0.09

0.59 ± 0.07

0.55 ± 0.05

0.5 ± 0.05

rk ◆

3.62 ± 0.89

4.85 ± 1.29

4.59 ± 1.02

4.53 ± 1.03

4.5 ± 1.13

64.40 ± 4.65 65.56 ± 6.47

67.27 ± 5.98

65.83 ± 4.32

67.13 ± 5.91

L c (mm)◇ ϕ◇

0.92 ± 0.02

0.91 ± 0.02

0.93 ± 0.03

0.92 ± 0.02

Dg (mm)◇

61.30 ± 3.77 62.88 ± 4.96

0.92 ± 0.04

63.04 ± 5.52

62.47 ± 3.88

63.62 ± 4.79

Da (mm)◇

61.56 ± 3.82 63.16 ± 5.06

63.43 ± 5.57

62.75 ± 3.90

63.92 ± 4.89

◆Mechanical parameters ◇Physical parameters

172

3 The Physical and Mechanical Properties of Tomato Fruit and Stem

level had a significant effect on the mechanical property parameters (E p , F max and r c ) of tomato. The plastic strain energy and peak force increase with the increasing compressibility level, which is the same as Linden’s point of view [57]. The elasticity decreases as the compressibility increases. When the compressibility level is 8%, the loading slope is the largest, and when the compression rate level is 4%, the loading slope is the smallest. Compared with the mechanical parameters of tomato, the physical parameters of tomato (L c , ϕ, Dg and Da ) did not change significantly with the increase of compressibility. This shows that the grouping of fruits is symmetrical, and there is no abnormal data in the load–unload test results. Otherwise, the abnormality of the test results may be caused due to the difference in the physical properties of the fruit. (2)

Mechanical properties of three-locular tomatoes

Mechanical property parameters of three-locular and four-locular tomatoes obtained from load–unloading test were shown in Fig. 3.38, such as E p , F max , r c , and r k . T cw , T l , F cw , and F l indicated the loading positions of CW and L tissue from three-locular tomatoes and the loading positions of CW and L tissue from four-locular tomatoes, respectively. The relationship between mechanical property parameters of three-locular tomatoes and compressibility was shown in Fig. 3.38a–d. ➀

Plastic strain energy E p (Fig. 3.38a)

➁

The loading position had no significant effect on the plastic strain energy of the three-locular tomato. The plastic strain energy of tomato increased with the increase of compression rate during loading. Peak force F max (Fig. 3.38b) and elasticity r c (Fig. 3.38c) The loading position had no significant effect on the peak force and elasticity of the three-locular tomato. When the compressibility was less than 12%, the F max loading on the L tissue was slightly larger than the F max of the CW tissue, and when the compressibility is greater than 12%, the opposite was true. The elasticity rc of the tomato when loading on the L tissue was slightly larger than the r c loading on the CW tissue.

Fig. 3.38 Load–unload test results of three-locular tomato

3.7 Mechanical Structure Model of the Whole Tomato Fruit

173

Fig. 3.39 Load–unload test results of four-locular tomatoes

➂

Loading slope r k (Fig. 3.38d)

➃

When the compressibility was less than 12%, the r k loading on the L tissue was larger than the r k loading on the CW tissue. When the compressibility was larger than 12%, the condition was the opposite. Discussion The loading position has no significant effect on mechanical properties of threelocular tomatoes such as plastic strain energy, peak force, elasticity, and loading slope from the above results. This is because that the angle of adjacent radial walls is about 120°, so when loading on L tissue and CW tissue, the main effect of the two forces on the tomato is the same. For the difference of the tomato shapes, the angle may slightly larger or smaller than 120°, so the loading position has small local influences on the mechanical properties. For example, the elasticity loading on the L tissue is slightly larger than it loading on the CW tissue.

(3)

Mechanical properties of four-locular tomatoes

The relationship between mechanical property parameters of four-locular tomatoes and compressibility was shown in Fig. 3.39a–d. ➀

Plastic strain energy E a (Fig. 3.39a)

➁

The loading position had no significant effect on the plastic strain energy of the four-locular tomato when the compressibility was smaller than 12%. The effect was gradually significant when the compressibility was larger than 12%. The energy difference produced by the tomato being loaded at two positions increased with the increasing compressibility. When the compressibility was 16% and 20%, the energy of loading on the CW tissue was 1.22 and 1.47 times that loading on the L tissue, respectively. Therefore, the compressibility has significant effect on the plastic strain energy, and the energy of four-locular tomatoes increases with the increasing compressibility. Peak force F max (Fig. 3.39b) The loading position has no significant effect on the peak force of four-locular tomatoes. When the compressibility was smaller than 16%, the Fmax loading on the CW tissue was slightly larger than that loading on the L tissue. When

174

3 The Physical and Mechanical Properties of Tomato Fruit and Stem

➂

the compressibility was larger than 16%, there was nearly no change of the F max loading on the CW tissue, this may because of the expired CW tissue, but the F max loading on the L tissue was increasing still. The compressibility has significant effect on the peak force, and that of four-locular tomatoes increases with the increasing compressibility. Elasticity r c (Fig. 3.39c)

➃

Loading position and compressibility have a significant effect on the elasticity of four-locular tomatoes. For each compressibility, the elasticity r c loading on the L tissue was larger than r c loading on the CW tissue. When the compressibility was 4%, the ratio of the elasticity rc loading on the L tissue to the that loading on the CW tissue reached a maximum: 1.16:1. This also showed that the tomato’s elastic recovery ability of the four-locular tomato loading on the L tissue was greater than that loading on the CW tissue. The elasticity decreases as the compressibility increases. Loading slope r k (Fig. 3.39d)

➄

The loading position has a significant effect on the loading slope of the fourlocular tomato. When the four-locular tomato was loaded on the L tissue, r k was smaller than that loading on the CW tissue. When the compressibility was 20%, the ratio of r k when the tomato was loaded on the CW tissue to r k when loaded on the L tissue reached a maximum of 1.34:1. To produce the same deformation at the two loading positions, the grasping force exerted by the robot finger when grasping the CW tissue is larger than that when grasping the L tissue. Discussion From the above results, the compressibility of 12% is the turning point of the mechanical parameters of four-locular tomatoes. This is because when the compressibility is larger than 12%, the CW tissue of the four-locular may gradually being destroyed. As shown in Fig. 3.39c, when the compressibility is larger than 12%, the elasticity decreases sharply but the plastic deformation increases greatly, and the plastic stain energy gradually increases with it.

2)

Results and analysis of test 2

The results for Test 2 were shown in Table 3.14. The cross-term data indicates the mean ± standard deviation of the corresponding mechanical property parameters of each group of tomato under the corresponding variety, locule number, and loading position. (1)

Rupture energy E r

In all cases the highest rupture energy for tomato fruits (3.23 J) was obtained at loading position CW for Fenguan906 four-locular tomato fruits while the lowest for fruits (1.98 J) was at loading position CW for three-locular tomato fruits. The tomato fruit with symmetric internal structure required more rupture energy, especially Jinguan28 six-locular tomato fruits. The average rupture energy of the tomato

3.7 Mechanical Structure Model of the Whole Tomato Fruit

175

Table 3.14 Results of test 2 Variety

Locule Loading Mechanical parameters number position Rupture Rupture force energy (J) (N) CW

1.98 ± 0.43 42.84 ± 8.75

15.67 ± 1.38

3.74 ± 0.21

L

2.44 ± 0.37 43.56 ± 7.38

16.23 ± 1.25

3.3 ± 0.57

CW

3.23 ± 0.74 49.61 ± 4.69

14.31 ± 2.11

5.57 ± 0.69

L

2.11 ± 0.47 51.68 ± 9.88

17.85 ± 1.56

4.06 ± 1.02

Five

CW

2.29 ± 0.88

9.13 ± 3.66

11.73 ± 5.66

L

2.47 ± 0.27 72.27 ± 2.40

9.25 ± 1.17

10.76 ± 0.35

Six

CW

2.87 ± 1.60 76.74 ± 23.56

8.75 ± 2.48

13.38 ± 3.53

L

2.84 ± 1.60 85.78 ± 29.7

9.59 ± 1.49

11.02 ± 3.97

Fenguan906 Three Four Jinguang28

Compressibility Loading (%) slope

69.3 ± 16.66

fruits from cultivar Fenguan906 (2.44 J) was more than that of the fruits of Jinguan28 (2.62 J). Statistical analysis showed the locule number and loading position had no significant effect (P < 0.05) on the rupture energy of tomato fruits. In this research, the sizes such as height and diameter of tomato fruits with different locule numbers showed a significant difference; this indirectly indicated that the sizes had no significant effect on the rupture energy of tomato fruits. Kilickan and Guner reported that the sizes had no significant effect on the rupture energy of olive fruits [50]. (2)

Rupture force F F

The rupture force increased with increasing locule number of tomato fruits. The rupture force loaded at the position L was slightly greater than that loaded at the position CW. The average rupture force of the tomato fruits from cultivar Fenguan906 (46.92 N) was less than that of the fruits of Jinguan28 (76.02 N). Statistical analysis showed the locule number had a significant effect (P < 0.05) on the rupture force of tomato fruits but the loading position had not. (3)

Compressibility ε

The compressibility is relative deformation when the tomato ruptured [50]. Statistical analysis showed that the locule number and loading position had a significant effect (P < 0.05) on the compressibility of tomato fruits. The three- and five-locular tomato fruits have asymmetric internal structure and the four- and six-locular tomato fruits have symmetric internal structure (Fig. 3.37). The tomato fruit with asymmetric internal structure had similar compressibility values when it was loaded from CW and L. The reason that the compressibility values had little difference is that the real structure of tomato fruit is slightly different from the simplified structure (Fig. 3.35b and c). The tomato fruit with symmetric internal structure had higher compressibility value when it was loaded from L than when the tomato was loaded from CW. The reason is that the CW tissue has higher resistance than L tissue. The compressibility gradually decreased with increasing locule number of tomato fruits. The reason is that

176

3 The Physical and Mechanical Properties of Tomato Fruit and Stem

the radial wall tissue increased and its resistance was improved as the tomato has more locule numbers. The average compressibility for the initial rupture of tomato fruits from cultivar Fenguan906 (16.02%) was higher than that of the fruits of Jinguan28 (9.18%). The average compressibility loading from CW position for tomato fruits (11.97%) was always lower than those loading from L (13.23%). In all cases the highest compressibility for tomato fruits (17.85%) was obtained at loading position L for Fenguan906 four-locular tomato fruits. The loading slope is a measure of the fruit hardness. As can be seen from Table 3.14, the hardness of tomato fruits from cultivar Fenguan906 was lower than that of the fruits of Jinguan28 at the light red ripe stage.

3.8 Mechanical Damage in Tomato Fruits [3] 3.8.1 Mechanical Damage Mechanism of Tomato Fruit According to the existing literature, whether tomato is damaged depends mainly on two factors: (1) cell membrane rupture caused by external load; (2) the appearance and action of cell wall-modifying enzymes [58]. Tomato damage is usually divided into two steps: the first step is the mechanical damage of the cell wall and cell membrane of the fruit, and the second step is the enzymatic degradation (including the cell wall) of the damaged tissue. When the tomato is mechanically damaged, the cell structure is destroyed, and the enzyme released from the ruptured cells is more likely to meet the substrate, causing browning of the damaged tissue. At the same time the pectin and cellulose components of the cell wall are rapidly depleted with the effect of polysaccharide-digesting enzymes and acids, causing rapid enzymatic deterioration of cell wall polysaccharides, i.e., softening of damaged tissues and forming soft spots.

3.8.2 Physiological Change of Tomatoes After Being Compress 1. 1)

Materials and methods Materials and equipment

The test was carried out in the laboratory of the College of Food and Bioengineering, Jiangsu University. Fresh tomato fruits of two structure types from the variety of Fenguan906 tomato at the ripeness of “Light red” stage were used in this study: three-locule (T) and four-locule (F). 140 tomatoes were picked, 50 for three-locular tomato and 90 for four-locular tomato.

3.8 Mechanical Damage in Tomato Fruits [3]

177

The test was carried out on a TA-XT2i texture tester (SMS company, UK), with the 500-196-20 electronic digital display vernier caliper (Japan Sanfeng Measuring Tool Factory, precision 0.01 mm), SF-400 electronic scale (Suzhou Botai Weiye Electronic Technology Co., Ltd. (range: 0–1000 g, accuracy: 0.1 g)), DZF-6050 vacuum drying oven (Shanghai Kanglu Instrument Equipment Co., Ltd.) and Kestrel 4000 handheld weather station (Temperature range: −29–70 °C, accuracy: 0.1 °C, humidity range: 0–100%, accuracy: 0.1%) used meanwhile. 2) (1)

Methods Test design

The seeds and colloid are located inside of each locular cavity in the tomato fruits. A radial wall separates the locules. Different degrees of mechanical damage were caused as each four-locular tomato was loaded at the locular tissue and radial wall tissue. Thus two positions at the fruit surface (locular: L and radial wall: CW tissue) were loaded. The two positions on the cross section of tomato, position 1 at the radial wall tissue and position 2 at the locular tissue, are shown in Fig. 3.35. Locular tissue is the pericarp over the locules, whereas radial wall tissue is the pericarp located over the septum. Position 1 (Fig. 3.35b and c) corresponded to the valley between two adjacent fruit shoulders (Fig. 3.35a), and position 2 (Fig. 3.35b and c) corresponded to the middle of one fruit shoulder (Fig. 3.35a). Additionally, the degree of mechanical damage had little difference at the two positions for three-locular tomatoes because the angle nears 120° between radial wall tissues. To study the effect of structure on the water content and mass loss of tomato fruits, three loading positions were compared, which can be described as radial wall tissue of three-locular tomato (FCW ), radial wall tissue of four-locular tomato (TCW ), and locular tissue of four-locular tomato (TL ). Compressibility is the most important explanatory variable in the model of the degree of mechanical damage of tomato fruit [59]. It shows a significant positive effect on the degree of mechanical damage for the same loading position and structure type. Four compressibility levels: 4, 8, 12, and 16% were used in this test, which indicated four increasing degrees of mechanical damage of tomato. A full factorial design was performed, as shown in Table 3.15. In total, 120 tomatoes (10 tomatoes × 3 loading positions × 4 compressibility levels) were loaded in the test. (2)

Test steps

The test was finished mainly by the following four steps: Table 3.15 Test factors and their levels Factors

Levels 1

2

3

Loading position

TCW

FCW

FL

Compressibility

4%

8%

12%

4 16%

178

➀

➁

➂

3 The Physical and Mechanical Properties of Tomato Fruit and Stem

Measuring the physical properties of tomato fruits. The axial diameter(H), radial diameter 1 D1 , and radial diameter 2 D2 were measured by using a vernier caliper and the fresh mass M 0 of the three-locular and four-locular tomato was measured by an electronic balance [50]. Then three-locular tomatoes and four-locular tomatoes were sorted into five groups and 9groups, respectively, and labeled. The first group had 5 mediumsize tomatoes and five large tomatoes, respectively, other groups were medium tomatoes. The first group was defined as group 1 and group 2, respectively. The 2–5 groups of three-locular were redefined as group 3–group 6. The 2–9 groups were redefined as group 7–group 14. After being grouped, the load–unload tests of tomatoes from group 3 to group 14 were conducted at room temperature with a TA-TX2 Texture Analyser (Texture Technologies Corp., NY, USA). The analyser was calibrated with a 5 kg weight prior to the first test. It was equipped with a 100 mm diameter plate for the load–unload test. Equipment settings were as follows:

Test speed—0.5 mm/s; Initial distance—10 mm from the tomato; Recording method for curve1—Final, for curve 2–Target. The compressibility and loading positions on a tomato sample are shown in Table 3.15. ➃ At last, all grouped tomatoes were put in a phytotron for 5 days of storage at 24 and 26.2% RH, and the mass of the fruits M 1 , M 2 , M 3 , M 4, and M 5 was measured and recorded once a day. Subsequently, the tomatoes were placed in a vacuum drying oven (DZF-6050) set to 85 °C to determine the dry mass M d of tomato fruit. The water content, WC, and mass loss, ML, were determined using the following Equations [60]: M0 − Md M0

(3.27)

M0 − Mn , n = 1, 2 . . . 5 M0

(3.28)

WC = ML =

where W C —the water content of tomato fruit (%); M L —the loss weight of tomato fruit (%); M 0 —the fresh weight of tomato fruit (g); M d —the dry weight of tomato fruit (g); M n —the weight of tomato fruit on the nth day (g).

3.8 Mechanical Damage in Tomato Fruits [3]

179

Table 3.16 Fresh mass and water content of group1 and group2 Structure type

Parameters 1

2

3

4

5

6

7

8

9

10

Three-locular M 0 (g) tomato W C (%)

104.2 118.0 145.3 154.4 146.2 162.0 162.3 169.1 165.1 147.4

Four-locular tomato

100.5 151.3 121.4 134.5 157.8 139.2 169.1 181.6 170.8 137.8

2. 1) (1)

M 0 (g)

95.0

W C (%)

95.3

94.8 95.4

94.9 94.3

95.4 94.9

94.8 95.5

95.7 95.6

95.1 95.1

95.3 95.5

95.2 95.3

95.1 95.2

Results and discussion The water content of damage-free tomato The water content of three-locular and four-locular tomatoes

The fresh mass and water content of 10 three-locular tomatoes (group 1) and 10 four-locular tomatoes (group 2) are shown in Table 3.16. These tomatoes were not damaged. Values of the coefficient of variation of the fresh mass and water content of group 1 and group 2 are presented in Fig. 3.40. Obviously, the coefficient of variation of the fruit fresh mass was bigger than that of fruit water content, whether for threelocular tomato or four-locular tomato, and the difference was slight in the coefficient of variation of fruit water content. The fresh mass of tomato had no significant effect on the water content. The average water content of three-locular and four-locular tomato was 95.13 ± 0.28% and 95.21 ± 0.38%, respectively. The fruit structure type had no significant effect (P > 0.05) on the water content of tomato at a = 0.05 according to one-way ANOVA. This also illustrates that the fruit structure had no significant effect on the dry matter content of tomato. (2)

The water content of medium and large tomato fruit

Group 1 and group 2 were sorted into two sets based on the masses according to the CNS SB/T 10331-2000; medium fruit: 100 ≤ M 0 ≤ 149 and large fruit: 150 ≤ M 0 Fig. 3.40 The coefficient of variation of the fresh mass and water content of group 1 and group 2

180

3 The Physical and Mechanical Properties of Tomato Fruit and Stem

Fig. 3.41 Mean water content of medium and large fruit of three-locular and four-locular tomato

≤ 199, respectively. The water content of medium and large fruit for three-locular and four-locular tomatoes is presented in Fig. 3.41. The size of the tomatoes had no significant effect on water content of three-locular tomato and four-locular tomato. This was in accordance with the obtained conclusion by coefficient of variation in last section. Therefore, this further illustrated that the fresh mass of tomato had no significant effect on water content. (3)

Discussion

The water content of fruits and vegetables is mainly affected by their growing environment [61, 62]. In this experiment, the study tomatoes were from the same growing station, so the water content of the fruit had small variance and was not significantly affected by the structure types of the fruits. This also showed that the volume of three-locular tomato would be bigger than four-locular tomato when the tomatoes had the same fresh mass. The relationship between water content and fresh mass had been widely studied in fruits. Wu F. reported the water content of “Classical 1” cucumber fruit had no relation with the fresh mass [62] and Liu M. et al. reported the water content decreased with increasing fresh mass of “Zuohe 2” strawberry [63]. Akar R. and Aydin C., Aviara N. et al., and Sessiz A. et al. showed the water content of gumbo, cactus pear, capper, guna, pistachio nut, and barberry increased with fresh mass, respectively, and these followed linear regression equations [60, 64–68]. The tomato fruit had the same property of water content as the cucumber but was different from the above-mentioned other fruits in this paragraph. Therefore, the tomato fresh mass increased with dry mass but not water content, and this followed linear regression equations.

3.8 Mechanical Damage in Tomato Fruits [3]

181

Table 3.17 Water content of loaded tomatoes after 5 days of storage Loading position

4%

8%

12%

16%

TCW

0.954 ± 0.002

0.955 ± 0.003

0.949 ± 0.002

0.948 ± 0.011

FCW

0.959 ± 0.001

0.950 ± 0.003

0.956 ± 0.001

0.952 ± 0.002

FL

0.955 ± 0.001

0.955 ± 0.013

0.952 ± 0.001

0.956 ± 0.006

2)

Water content of loaded tomatoes

The water content of loaded tomatoes after 5 days of storage is presented in Table 3.17. Tomatoes were loaded at all the combinations of four compressibility levels: 4, 8, 12, and 16% and three positions: TCW , FCW, and FL . The data in Table 3.17 represent average values ± standard deviations of the water content for 10 tomatoes at different compressibility level and loading position. According to a MANOVA, the compressibility and the loading position had no significant effect (P > 0.05) on the water content of tomato at a = 0.05. The most important physiological processes in postharvest fruits are respiration and transpiration. Respiration is the process by which fruits take in oxygen and give out carbon dioxide. The oxygen from the air breaks down carbohydrates in the fruit into carbon dioxide and water [69]. The carbohydrate content in fruit decreases with increase in storage period, so the overall fruit quality is reduced. The fruit respiration rate increases after mechanical damage, and the consumption rate of carbohydrate dry matter such as cellulose and pectin substance in the fruit is raised. For the water content of damaged tomato after 5 days of storage, the M d was residual dry matter mass in tomato. Transpiration is a process of water evaporation in fruit [70]. Transpiration accounts for most of the mass loss in the majority of horticultural produce. In tomatoes, transpiration accounts for 92–97% of mass loss. The mass loss due to respiration is considered negligible compared to that due to transpiration [71]. So, the change of tomato dry matter mass due to respiration was slight in 5 days. Therefore, compressibility and loading position showed no significant effect on the tomato dry matter content, which also had no significant effect on the tomato water content accordingly. 3) (1)

Mass loss of loaded tomatoes The effect of compression on the mass loss

The mass loss on the fifth day after the tomatoes were loaded under the conditions of four compressibility levels and three loading positions is presented in Table 3.18. The data in Table 3.18 represent average values ± standard deviations of the mass loss for ten tomatoes at different compressibility levels and loading positions. The highest mass loss in tomatoes was observed at FCW for 12 and 16%, followed at TCW , and lowest mass loss at F*L. However, the mass losses were no significantly different among three loading positions for 4 and 8%. For instance, the ratio of mass losses was 1.48 at 16% for FCW and FL , and about 1 at 4% for FCW and FL . The reason might be that the deformations were in the elastic region of tomato at 4 and 8% but

182

3 The Physical and Mechanical Properties of Tomato Fruit and Stem

Table 3.18 The mass loss of loaded tomatoes after 5 days of storage Loading position

4%

8%

12%

16%

FCW

0.034 ± 0.046

0.046 ± 0.002

0.078 ± 0.001

0.121 ± 0.001

TCW

0.035 ± 0.006

0.046 ± 0.001

0.055 ± 0.014

0.107 ± 0.052

FL

0.035 ± 0.027

0.049 ± 0.007

0.044 ± 0.014

0.082 ± 0.030

beyond the elastic region at 12 and 16%. Certainly, there would be an elastic limit value between 8 and 12% for tomato fruit. At three loading positions, the degree of mechanical damage was the greatest at FCW and the lowest at FL when one tomato had the same deformation which was beyond the elastic region. Mechanical damage showed a significant positive effect on mass loss. Therefore, loading position had no significant effect on the mass loss at 4 and 8% but had a significant effect on it at 12 and 16%. The mass loss rate was the largest at FCW and the lowest at FL for 12 and 16%. The result was same to the idea from Assi et al. [72]. Furthermore, compressibility showed a significant effect (P > 0.05) on the mass loss in tomato fruit at a = 0.05 according to one-way ANOVA. Obviously, the mass loss increased with higher compressibility at the same loading position. The change of mass loss rate was found to be higher when the compressibility ranged from 12 to 16% compared to the compressibility ranging from 4 to 8%. For example, the ratio of mass loss rate was 1.55 at F*RW for 16 and 12%, but it was only 1.35 for 8 and 4%. Four different compressibility levels at the same loading position corresponded with four degrees of mechanical damage in tomato fruit. Consequently, this further illustrated that the relationship between the degree of mechanical damage and the mass loss did not follow the simple linear regression equations. However, not much is known about the detailed relationship at various positions so far. Finally, compared with several nonlinear regression equations, the best relationships between these mass losses on the fifth day against various positions and compressibility levels of tomato were shown in below equations: M L 5FC W = 0.021e10.84C

R 2 = 0.99

(3.29)

M L 5T C W = 0.023e8.83C

R 2 = 0.92

(3.30)

M L 5F L = 0.027e6.12C

R 2 = 0.77

(3.31)

where ML 5FCW —the mass loss of tomato fruit at the fifth day at F CW ; ML 5TCW —the mass loss of tomato fruit at the fifth day at T CW ; ML 5FL —the mass loss of tomato fruit at the fifth day at F L ; C—the compressibility (%). The regression equation will be necessary in the quality evaluation of mechanical properties of postharvest tomato during storage.

3.8 Mechanical Damage in Tomato Fruits [3]

183

The mass loss after storage increased with increase in compressibility. The reason for the increased mass loss at higher compressibility might be that mechanical damage of tomato fruits broke down the surface organization of the tissues, thereby leading to greater flux of water vapor through the damaged area [73]. As the degree of mechanical damage increased, the damaged area increased, leading to greater evaporation during transpiration. Thus mechanical damage greatly accelerates the rate of mass loss from fruit. (2)

The effect of storage time on the mass loss

The cumulative mass loss rate of tomatoes at 4, 8, 12, and 16% for 5 days of storage is presented in Fig. 3.42. The mass loss increased with the storage period at all the combinations of compressibility and loading position. It has been shown that storage period has a significant effect on mass loss [74, 75]. Similar trend in mass loss rate of tomato fruits with storage period is in agreement with previous studies on eggplants, apples, oranges, and mango [14, 76–78]. Transpiration rate is influenced by factors such as temperature, humidity, surface area, respiration rate, and air movement [79]. The mass loss due to respiration is considered negligible compared to that due to

(a) Compressibility—4%

(b) Compressibility—8%

(c) Compressibility —12%

Fig. 3.42 The relationship of mass loss with the storage period

(d) Compressibility —16%

184

3 The Physical and Mechanical Properties of Tomato Fruit and Stem

transpiration. So, the transpiration rate was nearly constant when the tomato was stored in a phytotron at 24 and 26.2% RH. In addition, tomatoes from group 3 to group 14 had no significant difference (P > 0.05) in the geometric mean diameter Dg, sphericity F, and surface areas S at a = 0.05 according to a MANOVA. Therefore the trend in mass loss with storage period followed linear regression equations. (3)

The effect of loading position on the mass loss

The loading position had no significant effect on mass loss during storage at 4 and 8% compressibility; the rate of mass loss at 4% was 0.63% per day and for 8% was 0.84%. But the loading position had a gradual significant effect on mass loss during storage at 12 and 16%. FCW showed a significant greatest mass loss during 5 days at a rate of 1.04% per day at 12 and 2.14% per day at 16%. FL showed a significant lowest mass loss during 5 days in a rate of 0.83% per day at 12 and 1.54% per day at 16%. At the end of 5 days storage, the cumulative mass losses were 12.1, 10.7, and 8.2% at FCW , TCW, and FL for 16% compressibility, respectively. Obviously, the loading position showed a significant effect on the split probability of tomato. The surface wax is broken down after tomatoes have cracked, thereby leading to the transpiration rate being increased. As a consequence, a crack greatly accelerates the rate of mass loss from tomato fruit [80]. Therefore, the loading position can have a considerable effect on mass loss in tomato fruit during storage.

3.9 The Properties of Tomato Stem 3.9.1 Stem System [5, 81] 1.

Geometric model of tomato stem system

A tomato infructescence refers to a group of tomato fruits arranged on a stem that is composed of a main branch or a complicated arrangement of branches. Then, a tomato infructescence may be composed of one or more clusters, which are termed a single cluster and multi-cluster, respectively. In a single cluster, each fruit connects to the main stalk through a pedicel and a peduncle (in sequence); however, in a multi-cluster, each fruit connects to the main stalk through a pedicel, a rachis, and a peduncle (in sequence) (Fig. 3.43). 2.

Mechanical model of stem system

To describe the bending stiffness of the stem, each part of the stem may be assumed as a rigid body, and each end of this body is connected to adjacent objects via a spring hinge (Fig. 3.44). Then, the equivalent bending stiffness coefficients for the single cluster and multi-cluster are respectively given as

3.9 The Properties of Tomato Stem Fig. 3.43 Cluster structure of tomato

Fig. 3.44 Simplified mechanical model

185

186

3 The Physical and Mechanical Properties of Tomato Fruit and Stem

Single cluster: k0 =

1 ka

1 +

(3.32)

1 kc

Multi-cluster: k0 =

1 1 ka

+

1 kb

+

1 kc

(3.33)

3.9.2 Mechanical Properties of Tomato Fruit System [4, 5] 1.

Construction of analytical method for abscission layer strength

When the fruit is ripe, the cells on the stem begin to age, forming a layer of so-called “abscission layer” where the stem is connected to the branches. As shown in Figs. 3.1 and 3.43, there is an abscission layer between the tomato pedicel and the upper stem. The presence of the delamination is important for mechanized harvesting because the strength of the abscission layer is weak and easy to separate, while retaining the pedicel at the calyx [1]. Therefore, measuring various deformation strengths of the abscission layer is of great significance for carrying out the picking operation. The strength of the fruit-stem abscission layer is the key to the fruit harvesting operation, but the physical and mechanical properties of the fruit currently have a relatively uniform measurement method, while the mechanical properties and strength of the different layers under different loads are often difficult to use in the ready-made standard methods. Therefore, this chapter first proposes a method for determining and analyzing the mechanical properties of fruit-stalk (abscission layer) based on standard mechanical testing equipment. 1) (1)

Bending strength of the abscission layer Materials and equipment

10 tomatoes with stem were taken from each of the green fruit stage, the green ripe stage, the first ripeness stage, and the semi-ripe stage for the pedicel-breaking test. The fruit stems (pedicels) were folded from the flower buds. (2)

Methods

In this test, the length of the pedicel was measured by a vernier caliper, and the angle at the abscission layer was measured by a protractor. The sample was loaded on the WDW30005 type micro-control electronic universal testing machine as shown in Fig. 3.45. The horizontal distance from the end of the pedicel to the outer edge of the

3.9 The Properties of Tomato Stem

187

Fig. 3.45 Loading for the bend-off test of the abscission layer

lower plate was measured with the vernier caliper, and a downward force was applied to the abscission layer until it was broken. This force was automatically recorded by a computer. The 100 N range force sensor was selected for the test with an accuracy of ±0.5% and a loading rate of 0.25 mm/s. (3)

Principle

The method can be simplified to the angled simplified beam structure shown in Fig. 3.46, A is the end of the pedicel, and the base is connected by a fixed hinge, and the C end is supported by the movable hinge and is broken by the force FM at the point B from the abscission layer. By measuring the length of the pedicel, the distance between the support and the angle αl , the required bending moment of the fruit stem (pedicel) from the abscission layer can be obtained by the following Equation: [M] = FM · AD = FM · AB · cos ∠B AC

(3.34)

In the triangle ABC, according to the cosine theorem and the sine theorem, respectively, 2

2

AB + BC − 2 AB · BC · cos αl = AC Fig. 3.46 Simplified force diagram for the bend-off test

2

(3.35)

188

3 The Physical and Mechanical Properties of Tomato Fruit and Stem

Fig. 3.47 Load figure for tensile test

AC BC = sin αl sin ∠B AC

(3.36)

∠BAC can be calculated by the above two equations, and then the breaking bending moment can be obtained by Eq. (3.34). 2) (1)

Tensile strength of abscission layer Materials and equipment

10 tomatoes with stem were taken from each of the green fruit stage, the green ripe stage, the first ripeness stage, and the semi-ripe stage for the stem (pedicel) tensile test. (2)

Methods

The test was carried out on a WDW30005 type micro-control electronic universal testing machine, the tensile test was performed by changing the fixture (Fig. 3.47). The 500 N range force sensor was selected for the test with an accuracy of ±0.5% and a loading rate of 0.25 mm/s. 3) (1)

Twist strength for abscission layer Materials and equipment

10 tomatoes with stem were taken from each of the green fruit stage, the green ripe stage, the first ripeness stage, and the semi-ripe stage for the stem (pedicel) twist test. (2)

Methods

The test was carried out on a WDW30005 type micro-control electronic universal testing machine (range: 2Nm, resolution: 0.001Nm, precision: ±0.5%). Hold the tomato on the torque tester, ensure that the fruit stem (pedicel) was at the center of the twist, applied the torque slowly on the fruit stem until the fruit stem (pedicel) was disconnected, and the peak torque was automatically recorded by the torque tester.

3.9 The Properties of Tomato Stem

189

Fig. 3.48 Principle of measurement

2.

Method for determination and analysis of joint bending resistance

The mechanical structure of the stem system and the mechanical properties of the joints at each level are the basis for describing the deformation response of the stem system under different loads. Similarly, the flexural properties of joints at various levels of the stem system are difficult to achieve with current standard clamping and loading methods. 1)

Principle

As shown in Fig. 3.48, when a force is applied perpendicularly to the loading part, the moment applied to the join point is a product of the force and the moment arm, as Mt = F pt · L 0

(3.37)

The angle of rotation at join point B is αt = arctan

xt L0

(3.38)

The bending stiffness coefficient is kt =

Mt αt

(3.39)

Therefore, the bending stiffness coefficient k of the tested join point can be deduced by solving Eqs. (3.37)–(3.39):

190

3 The Physical and Mechanical Properties of Tomato Fruit and Stem

Fig. 3.49 Test loading

kt = 2)

F pt · L 0 arctan Lxt0

(3.40)

Material and method

To measure the bending stiffness coefficient of each join point of stem, a digital force meter (model HF-50, Ali Instrument Co., Ltd., Wenzhou, China) was mounted on the mobile station of an electric single-column test stand (model AEL, Ali Instrument Co., Ltd., Wenzhou, China) (Fig. 3.49) and held the fixed part of the join point with a vertical clamp to ensure that the loading part was aligned horizontally and in a direction perpendicular to the center line of the force meter. Then, we applied a tensile force to the loading part with a hooked test head and measured the length between the join point and the initial loading point with a Vernier caliper (model 91512, SATA Tools Co., Ltd., Shanghai, China). Finally, we moved the force meter back at a speed of 0.2 mm/s and recorded the movement distance and the corresponding force automatically. 20 groups of tests were carried out for k a , k b , and k c , respectively.

3.9.3 Results [4, 5] 1.

Breaking strength of abscission layer

The average bending angle α l of the four sets of fruit stalks is 130.26°, both of which are broken from the abscission layer. The average breaking moment is 161.29mNm, and the maximum and minimum bending moments are 65.81 and 308.72mNm, respectively. The average breaking displacement is only 2.61 mm, indicating that fruit picking by breaking the fruit stem (pedicel) is a convenient way. The breaking strength of different ripeness fruit stems (handles) has the following regularity (Fig. 3.50): “Breakers” stage > “Pink” stage > “Light red” stage > “Green” stage

3.9 The Properties of Tomato Stem

191

Fig. 3.50 Average bending moment of different ripeness tomatoes

The results of the “Green” stage, “Pink” stage, and “Light red” stage are close, which indicates that the force and motion parameters between the robot and the fruit are similar when harvesting at different ripeness stages, which provides great convenience for the design and control of the picking robot. 2.

Tensile strength of abscission layer

The measured fruit stems (pedicels) were fractured from the abscission layer, the average breaking force was 22.53 N, and the maximum breaking force reached 34.29 N. The pull-off force of the fruit stems (pedicels) was similar under different ripeness (Table 3.19). There was no significant difference in the pull-off force of different ripeness fruit stems (pedicels). 3.

Twist strength of abscission layer

The average twisting torque is 51.9 mNm, the maximum is 88 mNm, the minimum is 22 mNm, and the twist angle is between 360 and 1200°. Most of the fruit stems (pedicels) are broken by the abscission layer, and very few occur at the flowerbed. There is no significant relationship between the torque and the fruit ripeness, but it is linearly related to the diameter of the stem (pedicel). 4.

Bending mechanical properties of stem

Through measurement and calculation, we find that the moment is positively related to the included angle before the join point was broken, but it was not a linear relation Table 3.19 Tensile strength of tomato abscission layer with different ripeness

Minimum (N)

Average (N)

Maximum (N)

Green

11.84

21.89

31.05

Breakers

12.97

23.40

29.82

Pink

15.18

22.99

30.07

Light red

13.63

23.77

34.29

192

3 The Physical and Mechanical Properties of Tomato Fruit and Stem

Fig. 3.51 Typical rotation angle–moment curve of joint of stem

Table 3.20 Statistical results of measurement of bending stiffness coefficient Maximum (mNm/rad)

Average (mNm/rad)

Minimum (mNm/rad)

Standard deviation

ka

2802.50

673.84

89.9

644.5

kb

1205.20

406.69

71.69

401.2

kc

321.37

185.34

38.00

k 0 (Single cluster)

288.31

107.08

26.71

k 0 (Multi-cluster)

232.65

145.36

19.46

87.02

(Fig. 3.51). Thus, we concluded that the joint between adjacent components of the tomato stem possesses not only elasticity but also viscosity properties. As shown in Fig. 3.51, to describe the elasticity of this joint, we linearly fitted the relationship curve between the moment before it reached its peak value and the corresponding included angle to the original point. The statistical results of measurement are listed in Table 3.20.

References 1. Szymkowiak E, Irish E (1999) Interactions between jointless and wild-type tomato tissues during development of the pedicel abscission zone and the inflorescence meristem. Plant Cell 11(2):159–175 2. Li Z, Zhang X (1984) Plant anatomy. Higher Education Press 3. Li Z (2011) Based on the biomechanical characteristics of tomato harvesting robot grasping injury research. Jiangsu University 4. Liu J, Li P, Li Z et al (2008) Experimental study on mechanical properties of tomatoes for robotic harvesting. Trans CSAE 24(12):66–70 5. Liu J (2010) Analysis and optimal control of vacuum suction system for tomato harvesting robot. Jiangsu University

References

193

6. Mpotokwane S, Gaditlhatlhelwe E, Sebaka A et al (2008) Physical properties of bambara groundnuts from Botswana. J Food Eng 89(1):93–98 7. Owolarafe O, Olabige M, Faborode M (2007) Physical and mechanical properties of two varieties of fresh oil palm fruit. J Food Eng 78(4):1228–1232 8. MINISTRY OF HEALTH, PRC (2003) Determination of moisture in food. China National Standardization Administration Committee 9. Kheiralipour K, Tabatabaeefar A, Mobli H et al (2008) Some mechanical and nutritional properties of two varieties of apple in Iran. American-Eurasian J Agric Environ Sci 3(3):343–346 10. Masoudi H, Tabatabaeefar A, Borghaee A (2007) Determination of storage effect on mechanical properties of apples using the uniaxial compression test. Can Biosyst Eng 49(3):329–333 11. Alamar M, Vanstreels E, Oey M et al (2008) Micromechanical behaviour of apple tissue in tensile and compression tests: Storage conditions and cultivar effect. J Food Eng 86(3):324–333 12. Fidelibus M, Teixeira A, Davies F (2002) Mechanical properties of orange peel and fruit treated pre-harvest with gibberellic acid. Trans ASAE 45(4):1057–1062 13. Vanstreels E, Alamar M, Verlinden B et al (2005) Micromechanical behaviour of onion epidermal tissue. Postharvest Biol Technol 37(2):163–173 14. Singh K, Reddy B (2006) Post-harvest physico-mechanical properties of orange peel and fruit. J Food Eng 73(2):112–120 15. Mali S, Grossmann M, Garcia M et al (2005) Mechanical and thermal properties of yam starch films. Food Hydrocolloids 19(1):157–164 16. Srinivasa P, Ravi R, Tharanathan R (2007) Effect of storage conditions on the tensile properties of eco-friendly chitosan films by response surface methodology. J Food Eng 80(1):184–189 17. Montero-Calder N, Rojas-Grau M, Martin-Belloso O (2010) Mechanical and chemical properties of gold cultivar pineapple flesh. Eur Food Res Technol 230(4):675–686 18. Harker F, Stec M, Hallett I et al (1997) Texture of parenchymatous plant tissue: a comparison between tensile and other instrumental and sensory measurements of tissue strength and juiciness. Postharvest Biol Technol 11(2):63–72 19. Wei Y, Wang C, Wu J (2001) Flow properties of fruit fillings. Food Res Int 34(5):377–381 20. Pitts M, Davis D, Cavalieri R (2008) Three-point bending: an alternative method to measure tensile properties in fruit and vegetables. Postharvest Biol Technol 48(1):63–69 21. Prasad A, Rao K, Nagasrinivasulu G (2009) Mechanical properties of banana empty fruit bunch fibre reinforced polyester composites. Indian J Fibre Text Res 34(2):162–167 22. Gong L, Xiong Y (2006) Engineering mechanics. Tsinghua University Press 23. Sadrnia H, Rajabipour A, Jafari A et al (2008) Internal bruising prediction in watermelon compression using nonlinear models. J Food Eng 86(2):272–280 24. Bargel H, Neinhuis C (2005) Tomato fruit growth and ripening as related to the biomechanical properties of fruit skin and isolated cuticle. J Exp Bot 56(413):1049–1060 25. Batal K, Weigele J, Foley D (1970) Relation of stress-strain properties of tomato skin to cracking of tomato fruit. HortScience 5(1):223–224 26. Matas A, Cobb E, Bartsch J et al (2004) Biomechanics and anatomy of Lycopersicon esculentum fruit peels and enzyme-treated samples. Am J Bot 91(3):352–360 27. Koch J, Nevins D (1989) Tomato fruit cell wall: I. Use of purified tomato polygalacturonase and pectinmethylesterase to identify developmental changes in pectins. Plant Physiol 91(3):816– 822 28. Bargel H, Neinhuis C (2004) Altered tomato fruit cuticle biomechanics of a pleiotropic non ripening mutant. J Plant Growth Regul 23(2):61–75 29. Kabas O, Ozmerzi A (2008) Determining the mechanical properties of cherry tomato varieties for handling. J Texture Stud 39(3):199–209 30. Gładyszewska B, Ciupak A (2009) Changes in the mechanical properties of the greenhouse tomato fruit skins during storage. Tech Sci 12(1):1–8 31. Thompson D (2001) Extensiometric determination of the rheological properties of the epidermis of growing tomato fruit. J Exp Botany 52(359):1291–1301

194

3 The Physical and Mechanical Properties of Tomato Fruit and Stem

32. Wang F, Wang C, Yang X (2008) Mechanical characteristics and finite element analysis of watermelon. Trans CSAE 24(11):118–121 33. Bai X (2012) Dynamic collision simulation and parameter optimization of rapid gripping process of tomato harvesting robot. Jiangsu University 34. Wang H (2001) Chinese tomato. Heilongjiang Science & Technology Press, Harbin 35. Wang W (1993) Effects of pectin and pectinase on tissue softening of fruits and vegetables. J Shandong Polythechnic Univ (3): 37–38 36. Liu J, Li P, Li Z et al (2008) Experimental study on mechanical properties of tomatoes for robotic harvesting. Trans CSAE 24(12):67–70 37. Altuntas E, Sekeroglu A (2008) Effect of egg shape index on mechanical properties of chicken eggs. J Food Eng 85(4):606–612 38. Coskuner Y, Karababa E (2007) Physical properties of coriander seeds (Coriandrum sativum L.). J Food Eng 80(2):408–416 39. Alayunt F, Çakmak B, Can H (1998) Friction and rolling resistance coefficients of fig. In: Proceedings of the I international symposium on international society for horticultural science, pp 301–304 40. Caliir S, Haciseferoullari H, Zcan M, et al (2005) Some nutritional and technological properties of wild plum (Prunus Spp.) fruits in Turkey. J Food Eng 66(2):233–237 41. Jahromi M, Jafari A, Rafiee S, et al (2008) Some physical properties of date fruit (Cv. Lasht). Int Agrophysics 22(3):221–224 42. Jannatizadeh A, Naderi BM, Fatahi R et al. (2008) Some postharvest physical properties of Iranian apricot (Prunus Armeniaca L.) fruit. Int Agrophysics, 22(2):125–131 43. Naderiboldaji M, Khub A, Tabatabaeefar A et al (2008) Some physical properties of sweet cherry fruit. American-Eurasian J Agric Environ Sci 3(4):513–520 44. Chen P, Hasegawa Y, Yamashita M (2006) Grasping control of robot hand using fuzzy neural network. Lecture notes in computer science, vol 3972, no 2, pp 1178–1187 45. Glossas N, Aspragathos N (2001) Fuzzy logic grasp control using tactile sensors. Mechatronics 11(7):899–920 46. Liu J, Li P, Li Z et al (2009) Design of the vacuum suction device for tomato harvesting robot. In: Proceedings of the CSAE annual meeting 47. Takashi T, Kataok J (1981) Physiological basis of tomato. Shanghai Scientific & Technical Publishers 48. China Domestic Bureau (2000) SBT 10331-2000 Tomato 49. US Department Of Agriculture (1997) United States standards for grades of fresh tomatoes 50. Kilickan A, Guner M (2008) Physical properties and mechanical behavior of olive fruits (Olea Europaea L.) under compression loading. J Food Eng 87(2):222–228 51. Zhou Z (1994) Agricultural materials science. Agricultural Press 52. Tu K, Zhu W, Jiang S (2006) Food physics. Southeast University Press 53. Kubilay V, Faruk O (2003) determining the strength properties of the Dixired peach variety. Turk J Agric For 27(3):155–160 54. Baltazar A, Espina-Lucero J, Ramos-Torres I et al (2007) Effect of methyl jasmonate on properties of intact tomato fruit monitored with destructive and nondestructive tests. J Food Eng 80(4):1086–1095 55. Wang H (2004) Study on rheological properties and quality of kiwi apple. Jiangsu University 56. Williams S, Wright B, Truong V et al (2005) Mechanical properties of foods used in experimental studies of primate masticatory function. Am J Primatol 67(3):329–346 57. Linden V, Ketelaere B, Desmet M et al (2006) Determination of Bruise susceptibility of tomato fruit by means of an instrumented pendulum. Postharvest Biol Technol 40(1):7–14 58. Linden V, Baerdemaeker J (2005) The phenomenon of tomato bruising: where biomechanics and biochemistry meet. In: Proceedings of the V international postharvest symposium, international society for horticultural science, pp 925–934 59. Li Z, Li P, Liu J (2010) Effect of tomato internal structure on its mechanical properties and degree of mechanical damage. Afr J Biotech 9(12):1816–1826

References

195

60. Sessiz A, Esgici R, Kizil S (2007) Moisture-dependent physical properties of caper (Capparis ssp.) fruit. J Food Eng 79(4):1426–1431 61. Rashid A, Frederick S, Bruce S et al (2005) Effect of soilwater depletion on growth, yield and fruit quality of carambola in gravelly loam soil. In: Proceedings of the Florida state horticultural society annual meeting, pp 237–241 62. Wu F (2009) A study on biomass accumulation of cucumber plants in controlled environment. Jiangsu University 63. Liu M, Takayuki K, Munehiro T et al (2002) Effect of fruit water content on fruit property of strawberry. Acta Agriculturae Boreali-Sinica 17(3):114–117 64. Akar R, Aydin C (2005) Some physical properties of gumbo fruit varieties. J Food Eng 66(3):387–393 65. Aviara N, Shittu S, Haque M (2007) Physical properties of guna fruits relevant in bulk handling and mechanical processing. Int Agrophysics 21(1):7–16 66. Fathollahzadeh H, Mobli H, Jafari A et al (2008) Effect of moisture content on some physical properties of barberry. J Agric Environ 3(5):789–794 67. Kabas O, Ozmerzi A, Akinci I (2006) Physical properties of cactus pear (Opuntia ficus india L.) grown wild in Turkey. J Food Eng 73(2):198–202 68. Razavi S, Emadzadeh B, Rafe A et al (2007) The physical properties of pistachio nut and its kernel as a function of moisture content and variety: part I. Geometrical properties. J Food Eng 81(1):209–217 69. Paliyath G, Murr D, Handa A et al (2008) Postharvest biology and technology of fruits, vegetables, and flowers. Wiley-Blackwell Publishing 70. Bartz J, Brecht J (2003) Postharvest physiology and pathology of vegetables. Marcel Dekker, Inc. 71. Shirazi A, Cameron A (1993) Measuring transpiration rates of tomato and other detached fruit. HortScience 28(10):1035–1038 72. Assi N, Jabarin A, Al-Debei H (2009) Technical and economical evaluation of traditional vs. advanced handling of tomatoes in Jordan. J Agron (Pak) 8(1):39–44 73. Elshiekh F, Abu-Bakr A (2008) Effect of harvesting method on quality and storability of grape fruits. Univ Khartoum J Agric Sci (Sudan) 16(1):1–14 74. Kumar A, Ghuman B, Gupta A (1999) Non-refrigerated storage of tomatoes-effect of HDPE film wrapping. J Food Sci Technol 36(5):438–440 75. Javanmardi J, Kubota C (2006) Variation of lycopene, antioxidant activity, total soluble solids and weight loss of tomato during postharvest storage. Postharvest Biol Technol 41(2):151–155 76. Khan M, Ahmad I (2005) Morphological studies on physical changes in apple fruit after storage at room temperature. J Agric Soc Sci 1(2):102–104 77. Mccornack A (1975) Postharvest weight loss of Florida citrus fruits. In: Proceedings of the Florida state horticultural society annual meeting, pp 333–334 78. Abbasi N, Iqbal Z, Maqbool M et al (2009) Postharvest quality of mango (Mangifera indica L.) fruit as affected by Chitosan coating. Pak J Bot 41(1):343–57 79. Mahajan P, Oliveira F, Macedo I (2008) Effect of temperature and humidity on the transpiration rate of the whole mushrooms. J Food Eng 84(2):281–288 80. Bauer S, Schulte E, Thier H (2004) Composition of the surface wax from tomatoes. Eur Food Res Technol 219(3):223–228 81. Liu J, Li P, Mao H (2015) Mechanical and kinematic modeling of assistant vacuum sucking and pulling operation of tomato fruits in robotic harvesting. Trans ASABE 58(3):539–550

Chapter 4

Development of Damage-Free Hand–Arm System for Tomato Harvesting

4.1 Summary 4.1.1 Research Significance The harvesting robot is usually composed of mobile platform, manipulator, endeffector, vision system, and control system. The mobile harvesting operation of the fruits and vegetables needs complex coordination among different modules. As a hand–arm system which handles the fruit and vegetable targets directly, the principle of the harvesting action, the ability of perception and judgment, and the hand–arm coordination are the key to ensure the speedy damage-free harvesting. The development of new-principle and new-structure damage-free harvesting hand–arm system has important academic and practical significance.

4.1.2 Content and Innovation (1)

(2)

(3)

A damage-free harvesting end-effector with the multi-sensing ability and integrated laser-vacuum-mechanical action was developed, which provides key equipment support for not only speedy damage-free harvesting operation but also the innovative research of laser cutting of stem and complex modeling of vacuum sucking and pulling; The equipment and method of damage-free fruit harvesting with the passive– active combined compliance control was put forward, and the force feedback control was combined with and the passive flexibility adaptation to position– posture deviation by the floating rotary support structure were combined to effectively solve the flexible and damage-free harvesting in the complex actual environment; Based on the structure of the commercial manipulator and the independently developed damage-free end-effector, the hand–arm system of the tomato fruit

© Science Press, Beijing and Springer Nature Singapore Pte Ltd. 2021 J. Liu et al., Rapid Damage-Free Robotic Harvesting of Tomatoes, Springer Tracts in Mechanical Engineering, https://doi.org/10.1007/978-981-16-1284-8_4

197

198

4 Development of Damage-Free Hand–Arm …

harvesting robot was constructed, which provides a technical idea to realize the short-cycle development and high performance operation of the robotic harvesting equipment.

4.2 Development of Damage-Free Harvesting End-Effector 4.2.1 System Scheme Design of Damage-Free Harvesting End-Effector The performance of the end-effector plays a decisive role to guarantee the success rate and efficiency of robotic fruit harvesting, which is installed on the manipulator’s wrist and directly contacts the fruit. The industrial end-effector is too simple to meet the needs of fruit harvesting. At present, end-effectors of harvesting robot developed in Japan, Holland, the United States, and other countries still have a low success rate and efficiency, and easily cause fruit damage. Because of the particularity of the object, the end-effector of the harvesting robot must be designed according to the properties of the fruit, so as to achieve the target of reliable and efficient harvesting. For this reason, the design and development of the end-effector were carried out based on the property analysis of tomato fruit and stem. 1.

General Principles for Designing of End-Effector for Robotic Harvesting (1)

(2)

(3)

Due to the great difference between the size and mechanical properties of tomato fruit and stem, the power and stroke of each working part of the end-effector should meet the needs of the harvesting of most fruits. For example, the diameter of the most fruits is between 50 ~ 90 mm, but very few too large or too small “abnormal-shape fruit” cannot be excluded. If the end-effector is to meet the needs of any fruit harvesting, the size and power of the end-effector will be too large, which is not conducive to the optimization of the structure, the power consumption, and the harvesting performance; The performance of the end-effector is mainly characterized by the success rate of damage-free rate, operation efficiency, and the level of energy consumption, and the above performance is highly dependent on the operation principle, overall design scheme, hardware structure, and optimal control of the end-effector; Under the premise of satisfying the performance requirements of the harvesting operation, the end-effector should be “light and small”, that is, small volume and lightweight. Due to the limitation of the operating space in the plant canopy and the load-carrying capacity of the manipulator, the lighter and compact end-effector has a greater advantage in the operational flexibility and is more conducive to the energy saving of the operation process.

4.2 Development of Damage-Free Harvesting End-Effector

2.

199

Comparison of Existing Structural Schemes

The various structural schemes of current end-effectors of fruit harvesting robot are shown in Table 4.1.

4.3 Motion Configuration Scheme In Table 4.1, it is found there are so many types of end-effectors for different fruit. Even for the same fruit, different configuration and structures are available. Generally speaking, different end-effectors may mainly be classified into four operational principles: single detaching, single gripping, and gripping then detaching, stabling than gripping then detaching. 1)

Single detaching The fruit is detached directly, then it falls to the ground or into a tube below till into a container. It can avoid the problems of the complex mechanism, gripping failure, and damage caused by the gripping of fruit or stem. However, there are obvious defects in this type of end-effector, which is difficult to effectively harvest tomato: (1)

(2)

(3)

2)

First, because of the flexibility of fruit branches and stems, the fruit on the plant is in a semi-positioning state, which will swing with the wind blowing or touching, which brings difficulties to the separation. Second, the way of detaching is limited, and the action of twisting and bending cannot be applied, and the cutting action is also restricted. Failure of the cutting might happen with either scissors or disk cutting tools, due to the lateral force applied to the stem in cutting (Fig. 4.1). Third, the swinging and position change of the fruit bring difficulties to the collection with the tube or tray, and the success rate is greatly reduced. The direct fall to the ground not only causes a lot of damage to the fruit but also cannot be collected, it is unacceptable for the robot harvesting.

Single gripping The fruit is gripped and then detached by different wrist motions of the manipulator. The detachment is realized by pulling, bending, or twisting instead of cutting, which may be named as non-tool detaching methods. This type of endeffector is just a simple gripper, which has lower costs and energy consumption and is more versatile. However, there are some limitations for the tomato harvesting by single gripping: (1)

(2)

Different from the manual harvesting, it is impossible to find a pivot on the abscission layer in robotic bending, twisting, or pulling, so it is not easy to load on the abscission layer effectively; For any single gripping, the load applied on the abscission layer is transmitted from the gripping force on the fruit. As a result, the gripping force

Sweet pepper

Sweet pepper

[3, 4]

[5]

Cut

Motor

Cut

Cut

–

–

–

–

Detaching Stable method method

Pneumatic Cut

Motor

Peduncle Motor

–

General-purpose –

–

Gripping Power position source

[2]

Target

Tomato

Picture

[1]

Source of literature

Table 4.1 Scheme analysis of current end-effectors for robotic harvesting

1

1

1

1

2

–

–

–

–

–

–

–

DOF Finger Finger type number

–

–

–

Limit switch

Sensing ability

(continued)

80*45

–

–

–

Size (mm)

200 4 Development of Damage-Free Hand–Arm …

Spherical fruit

Cherry tomato

Cherry tomato

Tomato

[7, 8]

[9]

[10]

[11]

Target

Cherry tomato

Picture

[6]

Source of literature

Table 4.1 (continued)

Peduncle Motor

Peduncle Step motor

Cut

Cut

Bend

Pneumatic Cut

Suction pad

–

Pipe

–

–

Detaching Stable method method

Pneumatic Cut

Peduncle Motor

-

-

Gripping Power position source

2

1

2

2

2

2

2

2

–

–

Rigid

Arc rubber pad

Arc rubber plate

–

–

DOF Finger Finger type number

Negative pressure sensor

–

–

Near infrared proximity sensor pressure sensor

Photoelectric sensor

Sensing ability

(continued)

240*130*110

–

–

–

210*75*125

Size (mm)

4.3 Motion Configuration Scheme 201

Tomato

Tomato

Tomato

[15–19]

[20, 21]

[22–25]

Target

Tomato cluster

Picture

[12–14]

Source of literature

Table 4.1 (continued)

Fruit

Fruit

Fruit

Stem

Motor

Motor

Motor

Motor

Gripping Power position source

Bend, Twist

Twist

Twist

Cut

Suction pad

–

Suction pad

Push plate

Detaching Stable method method

2

2

2

3

4

2

2

2

Under actuated

–

Rigid

Rigid

DOF Finger Finger type number –

Size (mm)

(continued)

L300

Potentiometer –

Limit switch

Touch sensor 250*120*130 + negative pressure sensor + potentiometer

Limit switch + Photoelectric sensor

Sensing ability

202 4 Development of Damage-Free Hand–Arm …

Tomato

Cherry tomato

Spherical fruit

[27]

[28, 29]

[30]

Target

Tomato

Picture

[26]

Source of literature

Table 4.1 (continued)

Fruit

Stem

Fruit

Fruit

Cut

Bend, Twist

–

–

Suction pad

Suction pad

Detaching Stable method method

Pneumatic –

Motor

Solenoid pneumatic

Gripping Power position source

1

2

4

2

–

4

4

Artificial muscle

–

Under actuated

Under actuated

DOF Finger Finger type number

CCD camera

–

Limit switch

CCD camera vacuum pressure sensor

Sensing ability

(continued)

270*100*100

–

254*139*140

–

Size (mm)

4.3 Motion Configuration Scheme 203

Tomato

Strawberry

Strawberry

Strawberry

[32]

[33–35]

[36]

[37, 38]

Target

Tomato

Picture

[31]

Source of literature

Table 4.1 (continued)

–

–

Motor

Bend

Pneumatic Push

Motor

Motor

Suction Device

–

Suction tube

–

–

Detaching Stable method method

Peduncle Pneumatic Cut

Fruit

Fruit

Fruit

Fruit

Gripping Power position source

3

1

2

1

4?

2

3

2

2

4

Urethane Layer

Latex overlay

–

V-shape rubber suction pad

–

DOF Finger Finger type number

Photoelectric Sensor

–

–

–

Touch switch

Sensing ability

–

–

(continued)

180*170*170

Size (mm)

204 4 Development of Damage-Free Hand–Arm …

Strawberry

Strawberry

Strawberry

Strawberry

[40, 41]

[42]

[43]

[40, 44, 45]

Target

Strawberry

Picture

[39]

Source of literature

Table 4.1 (continued)

Motor

Peduncle Motor

Peduncle –

Peduncle Motor

-

Peduncle Motor

Gripping Power position source

Cut

Cut

Cut

Cut

Cut

2

2

钩子

–

1

1

1

2

2

2

–

2

–

–

–

–

–

DOF Finger Finger type number

–

–

–

Detaching Stable method method

2 Photoelectric Sensors

–

CCD camera

3 Photoelectric sensors Limit switch

Limit switch

Sensing ability

–

–

(continued)

130*60*50

–

–

Size (mm)

4.3 Motion Configuration Scheme 205

Asparagus

Citrus

Citrus

Citrus

[48]

[49]

[50–53]

[54]

Target

Kiwifruit

Picture

[46, 47]

Source of literature

Table 4.1 (continued)

Motor

Step motor

–

–

–

Hydraulic cylinder

Peduncle Pneumatic motor

Stem

Fruit

Gripping Power position source

Cut

Cut

Cut

Cut

Vacuum suction pad

–

–

–

–

Detaching Stable method method

3

1

6

2

2

–

–

2

2

2

–

–

–

Arc surface

DOF Finger Finger type number

Proximity switch

CCD camera ultrasonic sensor

6 CCD cameras

–

Infrared sensor Force sensor

Sensing ability

–

–

(continued)

2060*310*610

–

–

Size (mm)

206 4 Development of Damage-Free Hand–Arm …

Citrus

Citrus

Citrus

Sweet pepper

[55]

[55]

[57, 58]

[5, 59, 60]

Target

Citrus

Picture

[55, 56]

Source of literature

Table 4.1 (continued)

Stem

Fruit

-

–

Fruit

Cut

Motor

Electric arc thermal cutting

Pneumatic Twist motor

Pneumatic Cut

–

–

–

Pneumatic jaw

–

-

Detaching Stable method method

Pneumatic Cut

Gripping Power position source

2

1

3

–

2

2

3

–

–

3

Thermocol layer

Elastomer padding

–

–

Inflated

DOF Finger Finger type number

–

CCD camera

–

–

6-axis force sensor

Sensing ability

–

–

–

(continued)

Size (mm)

4.3 Motion Configuration Scheme 207

Cucumber

Strawberry

Eggplant

Eggplant

[63, 64]

[65]

[66, 67]

[68, 69]

Target

Citrus

Picture

[61, 62]

Source of literature

Table 4.1 (continued)

Motor

Peduncle Motor

Fruit

Peduncle Motor

Cut

Cut

Electric hot wire cutting

Electrode thermal cutting

–

Suction pad

–

Suction pad

–

Detaching Stable method method

Pneumatic Cut

Peduncle Motor

–

Gripping Power position source

1

2

2

3

1

2

4

2

2

–

–

–

Sensing ability

Rigid

Rubber

CCD camera + ultrasonic sensor

In hand camera Photoelectric Sensor

Phenolic Limit contact plastic sheet

Rigid

–

DOF Finger Finger type number

–

–

–

–

–

(continued)

Size (mm)

208 4 Development of Damage-Free Hand–Arm …

Apple

Apple

Apple

Apple

[73]

[75]

[76]

[77]

Target

Apple

Picture

[70–74]

Source of literature

Table 4.1 (continued)

Fruit

Fruit

–

Step motor

Step motor

–

Peduncle Motor

Peduncle Motor

Gripping Power position source

Cut

Twist

2

–

–

5

1

1

4

2

–

2

2

–

CCD camera

–

–

Sensing ability

Tendon – transmission

Silicone padding

–

Rigid

Rigid

DOF Finger Finger type number

Supporting 2 bar

–

Suck then – pull

Twist

Twist

Detaching Stable method method

–

–

–

–

–

(continued)

Size (mm)

4.3 Motion Configuration Scheme 209

Sweet pepper

Cucumber

[78]

[79, 80]

Target

Sweet pepper

Picture

[78]

Source of literature

Table 4.1 (continued)

Fruit

Fruit

Fruit

Cut

Cut

–

Suction pad

–

Detaching Stable method method

Pneumatic Cut

–

–

Gripping Power position source

2

2

2

2

–

2

Pneumatic flexible

Lip-type

Fin-ray

DOF Finger Finger type number

Micro switch

TOF Camera + CCD camera

TOF Camera + CCD camera

Sensing ability

–

–

–

Size (mm)

210 4 Development of Damage-Free Hand–Arm …

4.3 Motion Configuration Scheme

211

(a) Scissor

(b) Disc cutter

Fig. 4.1 Lateral force applied on the stem in cutting

(3)

3)

for detachment is much larger than that of stable gripping, and the risk of fruit damage increases; As a result of lack of the pivot, the force applied not only to the target fruit but also to the whole plant. Manage to the plant might happen if it was not well supported.

Gripping then detaching Gripping then detaching is a more general way of harvesting for various endeffectors. Compared with fruit gripping, the stem gripping has the advantage of avoiding damage to fruit, but there are also some limitations: (1)

(2)

First, the stem gripping and cutting is very convenient and suitable for the fruit with stronger stem, such as cucumber and eggplant (Figs. 4.2 and 4.3). However, the stem gripping is often not easy to be realized for tomato, whose stem is very short and the growth posture is very different. Second, as an abscission layer exists in the tomato stem, the accidental fall of tomato fruit might happen during harvesting and transportation.

(a) eggplant

(b) cucumber

Fig. 4.2 Comparison of stem of different fruits

(c) tomato

212

4 Development of Damage-Free Hand–Arm …

(a) End-effector for robotic cucumber harvesting [63,64]

(b) End-effector for robotic eggplant harvesting[68,69]

Fig. 4.3 Stem gripping end-effector

(3)

4)

Third, the method of stem gripping then cutting is all used for the existing gripping then detaching end-effectors, while other detaching methods are limited.

Sucking then gripping then detaching

For the sucking then gripping then detaching end-effector are to the auxiliary sucking and pulling is added with the vacuum suction pad. In the end-effectors for the robotic harvesting of spherical fruit developed in different countries, the vacuum suction system has been widely used because of its high adaptability to the shape error of the fruit, the low damage of suction action, the convenience, and simplicity of applying. If replacing the mechanical gripping with the vacuum sucking, the reliability and stability of fruit holding are limited. In order to obtain a high enough suction force, larger suction pad, higher vacuum pressure, and vacuum flow will be necessary, but it is difficult to realize the coordination of them. Furthermore, the problem of the suction damage will occur. However, the vacuum suction system, as the auxiliary device of the end-effector to grip then detach the fruit, can pull the target fruit away from the cluster to avoid the damage to the adjacent fruit, expand the movement space of the finger and increase the success rate of the gripping. At the same time, the auxiliary positioning is enhanced and the probability of grip failure cause by fruit swinging is reduced. Based on the above analysis, fruit sucking then gripping then detaching is adopted as the configuration of the end-effector. However, in view of the shortcomings of the existing detaching methods, the laser cutting technology is applied to stem cutting for the first time.

4.4 System Components of the End-Effector

213

4.4 System Components of the End-Effector All the different modules compose this end-effector system for tomato harvesting robot, which are an actuator unit, a laser unit, a driving/control unit, a power supply unit, and an industrial laptop, as shown in Fig. 4.4. The actuator unit connects with the wrist of an industrial manipulator, the YASKAWA MOTOMAN SV3X, performing the task of fruit singulating, gripping, and cutting. Distributed sensors on the body carry out data acquisition. The laser unit, including a protection/control circuit for optical laser, performs the task of laser beam generation and control. The industrial laptop and driving/control unit compose open architecture of control system to data processing and control of external devices. The power supply unit utilizes highenergy lithium battery to supply power for the generation of the different voltages needed to drive the system by DC/DC convert. In the process of harvesting, there are several steps: (1)

(2) (3)

When the manipulator sends the end-effector to the predetermined position near the fruit, the position, distance, shape information of the target fruit is obtained through the sensing system; The feed mechanism drives the vacuum suction pad forward; When the sensing system perceives the suction pad to suck the fruit successfully, the suction pad pulls the fruit to move backward, so that the fruit is isolated from the cluster;

Fig. 4.4 Schematic representation of the system architecture

214

4 Development of Damage-Free Hand–Arm …

(4)

The fingers close, the sensing system detects and feeds back the gripping force and sliding information; The fingers are feedback controlled to realize the smooth and reliable gripping of the fruit; The motor drives the laser focusing lens to focus on the stem and cut off the stalk by the laser beam; The manipulator drives the end-effector to put the fruit into the fruit box so that the fruit harvesting is finished.

(5) (6) (7)

4.4.1 Mechanism Design of End-Effector [81] Actuators of the end-effector for spherical fruit harvesting robot include a gripper, a laser cutting device, and vacuum suction pad device, performing fruit gripping, peduncle cutting, and fruit singulation, respectively. 1.

Design of Gripper

1)

Configuration of gripper

The end-effector of tomato harvesting robot developed by Monta M. of Japan and Ling P. of USA adopted simple multi-joint flexible fingers [22, 26], the curve of the finger was smooth and had a certain compensation ability to be well adapted to the difference of the size of the fruit. But this mechanism, which is highly under actuated, drives all the joints of the 4 fingers by one driving part. When there are any obstacles, such as branch, between the fruit and the end-effector, the flexible finger will bend, resulting in the failure of the fruit gripping. In other fields, the humanoid multi-joint dexterous hand and different kinds of flexible fingers driven by pneumatic artificial muscle, shape memory alloy and electroactive polymers, etc. are one of the hotspots of current research [82–86], but its high complexity and high cost are still far from the practical application, especially in the agricultural application. By contrast, the pneumatic or motor-driven two-finger or three-finger gripper has been widely used in various fields, such as industry, architecture, logistics, and so on. It has also become the main scheme in the field of robotic harvesting because of its simple structure, flexible application, and convenient manipulation. The spherical fruit can be gripped by 2 fingers or multiple fingers. The less the number of fingers, the worse the stability of gripping. The grip with multi-finger end-effector is more stable and reliable, but the complexity of the mechanism and control is increased, and the interference with the stem, branches, and leaves in the harvesting space will also increase. Therefore, the regular-shape and regular-size fruit may be more suitable for grasping with fewer fingers. 2)

Type of finger

According to the motion characteristics of fingers, the 2-finger gripper can be classified into three types: parallel-linear, translational, and rotational (Table 4.2):

4.4 System Components of the End-Effector

215

Table 4.2 Type and characteristic of gripper Type

Parallel-linear

Translational

Rotational

Characteristics

Two fingers are moving parallel along same straight line

Two fingers move symmetrically in plane and keep parallel

Two fingers rotate symmetrically around fixed axis

Gripping error

Small

Large

Large

Lateral force

No

Yes

Yes

Action range

Small

Small

Large

Mechanism type

Less

A few

More

Graph

(1) (2) (3)

Parallel-linear type: Two fingers are moving parallel along the same straight line; Translational type: Two fingers move symmetrically in a plane and keep parallel; Rotational type: Two fingers rotate symmetrically around a fixed axis.

The opening degree of the rotational type is larger, and the kinds of its transmission mechanism are more than 60 [87]. The translational type also has a variety of transmission forms, while the transmission mechanisms of parallel-linear type are relatively small. In the industrial field, the above three kinds of gripping mechanisms are all widely used. But when it is used for harvesting of fruit with the size difference, as the center position of the translational and rotary clamping mechanism is constantly changing during the movement process, it is difficult to keep coincidence of each fruit center and gripping center and realize precise gripping. At the same time, the lateral force exists during the gripping process with the translational or rotational type gripper, which affects the reliability of the gripping and may cause the sliding friction damage of fruit surface. As a result, parallel-linear type gripper is adopted in this end-effector. 3)

Mechanism scheme of parallel-linear gripper

The commonly used mechanism scheme of parallel-linear type gripper is shown in Table 4.3. Compared with the gear-double rack and double dialing fork mechanism, although the speed and efficiency of the bi-directional screw mechanism are low, the motion precision is very high. Also, the opening degree is obviously larger index same space limit, and its gripping force is far greater as a result of the force amplification of the spiral transmission. Finally, the self-locking performance of spiral transmission is more valuable. Once it grips the fruit, it will not get loose due to the reaction force

216

4 Development of Damage-Free Hand–Arm …

Table 4.3 Frequently used scheme of parallel-linear gripper Type

Bi-directional screw

Gear-double rack

Double dialing fork

Motion accuracy

High

Medium

Low

Speed

Medium

High

Medium

Open degree

Large

Medium

Small

Efficiency

Low

High

Medium

Self-lock

Able

Unable

Unable

Gripping force

Large

Medium

Small

Graph

of the fruit, and no additional reversing brake devices and measures are needed. Therefore, the bi-directional screw scheme of parallel-linear gripper is chosen. 4)

Shape of finger surface

The shape of the finger surface has a decisive influence on the needed gripping force and contact stress to the fruit. The reduction of the critical contact stress to the fruit will reduce the probability of fruit damage. Compared with the line contact between plane finger surface and spherical fruit surface, the contact area between the arc finger surface and spherical fruit surface is greatly increased, thus the critical contact stress is greatly reduced. As shown in Table 4.4, when using the circular arc surface, the critical gripping force is far less than the plane type. Table 4.4 Shape type of finger surface Type

Plane

Arc Surface Horizontal

Vertical

Graph

Equation

N=

Critical gripping force N(N)

2.90

G 2f

N= 0.27

G·ctg(θ +arctan f ) 2

N=

G 2f

· cos θ

1.79

Note θ is the center angle of circular arc surface. The fruit weight N is 3.14 N according to the maximum weight of 98% samples of tomato fruit, the coefficient of friction f between the finger surface and fruit surface with dew is 0.54 according to the test results, the center angle θ is 50°

4.4 System Components of the End-Effector

2.

Design of the Vacuum Suction System

1)

Components of the compressor-tank-ejector vacuum suction system

217

The new vacuum suction system is composed of a vacuum generation sub-system, sucking and pulling sub-system, and monitor and control sub-system. The vacuum generation sub-system generates a vacuum, and the suction pad is driven by the feeding mechanism to move forward. The suction pad seals and is attached to the fruit surface, then pulls the fruit back by the suction feed mechanism, thus realizing the isolation of the target fruit from the adjacent fruit. 2)

Vacuum generation system

A vacuum ejector relies on the flow of compressed air as the “motive” fluid to create the vacuum at a desired port due to the Venturi effect. Compared with the vacuum pump, the vacuum generator has faster vacuum production and relieving speed. It is more suitable for frequent switching occasions to avoid the pulsation of vacuum pressure. Furthermore, it can make use of the compressed air through the built-in electromagnetic valve to switch the suction/blow of the suction pad, thus control the rapid relief of the vacuum between the sucker and the fruit at any time. Therefore, it can better satisfy the needs of sucking and holding in tomato fruit harvesting. Contrary to regular applications in workshops, it is impossible to rely on a constant compressed air supply system for mobile harvesting robots. Thus, a compact vacuum ejector with a 1 mm nozzle diameter (model SCP10, Schmalz Co., Ltd., Glatten, Germany) is applied, and a mini 0.1875 kW air compressor compacted with an air tank (model AC-001, Deli Pneumatic Tools Co., Ltd, Taiwan, and China) is used to supply compressed air to the vacuum ejector. The speed of the air compressor is 2,800 rpm, and the capacity of the air tank is 6 L. 3)

Sucking and pulling sub-system

A 2.5-fold round bellows suction pad is attached to the front end of a rack, which is driven back and forth by a mini DC motor (model RE-max 24, Maxon Motor Ag, Sachseln, Switzerland) through a pinion (Fig. 4.5). Compared with the flat suction pad, the multi-fold bellows structure can produce larger contraction, bending, and even torsional deformation. It has good cushioning in contact with the target object and in followed sucking and pulling. It has a strong adaptability to different sizes and shapes and has a good compensation for position error. Therefore, it is especially suitable for sucking and pulling when the fruit is harvested. It is found in related research that, larger cups covered more surface area of the fruit and could transmit much more pulling force to the fruit once attached, they proved very difficult to attach to a free-hanging fruit under field conditions [26]. Any small variation in the fruit surface curvature would not allow the cup to seal and attach [26]. For the size of tomato fruit, the diameter less than 19 cm is more suitable [1]. Therefore, three suction pads with different diameter of 20, 14, and 9 cm are

218

4 Development of Damage-Free Hand–Arm …

Fig. 4.5 The working principle of Venturi effect in vacuum generator

Fig. 4.6 Vacuum circuit diagram of the vacuum suction device

4.4 System Components of the End-Effector

219

Table 4.5 Main parameters of the bellows vacuum suction pads Folds

Outer diameter (mm)

Effective diameter (mm)

Inner diameter (mm)

Suction force (N)

Pull-out force (N)

Internal volume (cm3 )

Maximum deformation (mm)

2.5

9

7

3.8

0.68

2.3

0.15

3

2.5

14

11

5

1.17

5.7

0.975

8

2.5

20

16

9

3.8

12.1

2

8

used, and the appropriate size of the suction pad is selected by comparison. The main parameters of the three suction pads are shown in Table 4.5. 4)

Monitor and control system

As shown in Fig. 4.6, an integrated pressure switch stops the air compressor when the tank pressure (pT ) rises to 0.8 MPa (ABS) and starts it up again when the tank pressure drops to 0.6 MPa (ABS) to avoid electric energy waste due to idling. A pressure control valve is used to adjust the differential between tank pressure (pT ) and air supply pressure (pS ), which are monitored with a pressure gauge 1 and pressure gauge 2, respectively. A solenoid valve to start and stop air supply to the vacuum ejector and a pressure sensor to monitor the vacuum degree in the suction pad provide feedback control. The check valve integrated into the compact ejector maintains a certain negative pressure for some time, which helps avoid unexpected fruit detachment from the suction cup and save air consumption during harvesting since a certain suction force is maintained even when the air supply is stopped. 5)

Hose size

The size of the compressed air supply hose and vacuum hose has an important influence on the response speed of the vacuum system. The initial hose size of this vacuum suction system is shown in Table 4.6. 3.

Design of the Peduncle Laser Cutting System

1)

Comparison and evaluation of present peduncle detachment methods

To perform tasks of fruit robotic harvesting, a fruit must be detached from the plant during robotic harvesting with appropriate methods. Undoubtedly, fruit detachment Table 4.6 Initial hose size of this vacuum suction system

Compressed air supply hose

Vacuum hose

Inner diameter (mm)

Length (m)

Inner diameter (mm)

Length (m)

4

9

4

3.1

220

4 Development of Damage-Free Hand–Arm …

from plants is a basic and essential step either for manual or robotic harvesting. Labors can use scissors to cut fruits off, or by various wrist motion to detach fruits, which may be classified into tool detaching methods. Similarly, robotic detaching of fruits may be performed by mechanical or thermal cutting, or by applying certain wrist motion, which may be classified into three main types of detaching methods: (1)

Wrist motion

Fruit detaching by wrist motion is much simpler which only need to grip the fruit and then to bend or twist it off relying on the friction force [81, 88]. However, its success rate is limited related to the posture of the fruit, joint structure of the peduncle. Meanwhile, how to avoid bruise of fruits by the fingers is still a large challenge. So for more harvesting robots, a variety of tools are adopted to cut peduncles. (2)

Mechanical cutting

Mechanical cutting devices are widely used in present harvesting robots. Any mechanical cutting device may be consists of the cutter or scissors, the transmission, and the driving system, so complexity, size, weight, and energy consumption of the end-effector will increase obviously. Meanwhile, the complicated space distribution of fruits, branches, and leaves in canopies of some fruits may limit the motion space of robot and lead to failure of the cutting operation. So mechanical cutting is suitable for plants whose xylems of peduncles are not developed and the direction of peduncles is nearly uniform, e.g., cucumbers (Fig. 4.7). In addition, it is inevitable for the transportation of viruses from one plant to the other and the fruit’s water loss after the peduncle is cut [63].

(a) Cucumber[92] Fig. 4.7 Mechanical cutting for robotic harvesting

(b) Eggplant[68,69]

4.4 System Components of the End-Effector

(3)

221

Thermal cutting

So, new thermal cutting techniques with electrodes, heating wires, or laser were adopted in peduncle cutting of cucumbers, strawberries, respectively. The endeffector of cucumber harvesting robot developed in Wageningen University in Holland [63] and the end-effector of sweet pepper harvesting robot developed by the University of Kochi University of Technology in Japan [59, 60] have applied thermal cutting technology with two electrodes instead of the traditional method of stem detaching. It employs two electrodes carrying a high-frequency electrical potential [63]. Once a stalk contacts the electrodes, it is cut by the high-frequency current between the electrodes [63]. This method avoids the problem of virus transportation from one plant to the other and water loss from the fruit [63]. However, it requires that the two electrodes must be reliably contacted with the stem to form a conductive path so that the demand for accurate visual and mechanical positioning is too high. Furthermore, this contact thermal cutting technology is limited by the stem length, the canopy space. The strawberry harvesting end-effector developed by China Agricultural University uses a thermal cutting method with electric hot wire to burn off the stem. The paper shows that the average detaching speed is 2.86 s, and the success rate is above 90% [89] (Fig. 4.8). However, as the stem needs larger bending deformation in the process of cutting, it is only suitable for long thin stems. 2)

System components of the peduncle laser cutting

Compared with the above methods, laser cutting has several advantages. A large amount of energy is concentrated in a high-power laser beam which is focused by a lens on a small enough diameter, so the material is melt or vaporized quickly in the laser beam path [90]. Firstly, the tool-free laser cutting means there are no physical contacts between tools and cutting materials [91], so the transportation of viruses with the cutting tool is avoided. Secondly, there is no interaction force between the tool and the material [92], which means the great gripping force that loaded to fruit and manipulator is avoided, so it is very flexible and versatile to different fruits and vegetables without tool changes [93] Another most important advantage is that this method is safer than the mechanical cutting since the high thermal effect only appears in focusing spot and then danger outside of the focus range is avoided [90]. (1)

Laser selection

In material processing and medical operation, CO2 laser, Nd:YAG laser, and KTP laser are usually used, while their power conversion efficiency is only about 10, 1, and 0.25%, respectively. Therefore, they have to be served by alternating current supply power to get higher optical power and huge water-cooling systems are absolutely necessary to deal with their high heat exhaust. Thus they are impossible to be applied in mobile robots.

222

4 Development of Damage-Free Hand–Arm …

So a 30 W high-power fiber-coupled laser diode (nLight) is adopted in this laser stem-cutting device. Its power conversion efficiency is 49% and can be supplied with DC batteries. It is smaller enough and the fiber tip can reach anywhere in 3D space. Meanwhile, it is easy-controlled. So it is ideal for this device. 3)

System components

The device includes the following two parts (Figs. 4.9 and 4.10): (1)

Laser generation and control unit

This unit mainly consists of the laser, battery, and a protection/control circuit. Meanwhile, a heat sink is necessary. This unit will generate laser beam reliably, and perform on/off and intensity control of the laser beam. Its fiber bundle is connected with the focusing lens by standard SMA-905.

Fig. 4.8 Thermal cutting method with electric hot wire [89]

Fig. 4.9 Block diagram of the laser stem-cutting device diagram

4.4 System Components of the End-Effector

223

Fig. 4.10 Structure diagram of laser peduncle-cutting device. 1. Focusing lens 2. Fasten ring 3. Thrust bearing 4. DC servo motor 5. Fiber 6. Protection/control circuit 7. Lithium battery pack

(a) Sharp GP2Y0D02YK0F

(b) Sharp GP2Y0D340K

(c) Omron E32-DC200E

Fig. 4.11 Sensors for distance and position detect

(2)

Bearing and rotation structure

A focusing lens is installed on the end-effector of harvesting robot, which is connected to the laser by a fiber. A mini DC motor will drive the focusing lens to rotate to cut stems. A reliable bearing structure is vital here.

4.4.2 Design of the Sensing System [81] It is an important feature to be able to percept interactive information in an unconstructive environment to an intelligent harvesting robot, especially external information such as distance, proximity, and force. At present, few robotic end-effectors for fruit harvesting have the ability to percept internal information and most of

224

4 Development of Damage-Free Hand–Arm …

(a) 3-axis force sensor

(b) 6-axis force sensor

Fig. 4.12 Sensors for force detect

them rely on visual system to percept external information completely. In order to percept sufficient information needed in intelligent control, a multi-sensor method is applied. Except for encoder of the servomotor and pressure sensor in the vacuum system, many types of sensors are used to information acquisition of the object fruit, plant, and environment. 1.

Sensors for Distance and Position Detect

Sensors for distance and position detect are applied for sufficient information acquisition working together with visual system. The visual system acquires characteristic information and guide sensors for distance and position detect to aim at the object plant and fruit, sensors for distance and position detect help the visual system to judge position, distance, and shape of the object plant and fruit, and guide the autonomous vehicle with the manipulator and the end-effector to move towards the object and fruit, meanwhile to harvest the fruit with proper posture. So a distance sensor (Sharp GP2Y0D02YK0F) is mounted in the middle part between the two fingers to localize the object plant and fruit in far distance; a distance sensor (Sharp GP2Y0D340K) is mounted on the front of each finger, and these two distance sensors are used for accurate localization of the object fruit in near distance; a proximity sensor (Omron E32-DC200E) is mounted on the front of each finger too, and these two proximity sensors are used to feel the proximity of the finger to the object fruit in 10 mm distance and compensate the positioning error of the visual system, which would help the end-effector to adjust its position and posture to avoid collision with the object fruit. Selection and match of these sensors are based on the following principles: (1) Match of detecting distance and range; (2) higher detecting accuracy; (3) lighter and smaller; (4) detecting accuracy would not be influenced by shape and material of the fruit; (5) better ability to adapt to the hostile environment.

4.4 System Components of the End-Effector

2.

225

Sensors for Force Detection

Accurate force control is most important for successful damage-free harvesting, so a 3-axis force sensor (Sunsor Smart300-50) is mounted in each inner finger surface which can detect all forces in three-dimensional space (F x , F y , F z ), and a 6-axis force sensor (ATI Nano25) is amounted in the wrist which can detect all force and moment information in three-dimensional space (F x , F y , F z , M x , M y , M z ). Information fusion of these three-finger forces can be helpful in accurate gripping force and end-effector posture detection. Meanwhile can weigh fruits to preliminary fruit grading.

4.4.3 Design of Control System [81] 1.

Structure of the Control System

In view of the tasks of multi-axis motion and multi-valve control, multi-sensor information input, and communication with the arm controller, an open-architecture master–slave control system is constructed based on a PMAC2A-PC104 multi-axis motion controller. Figure 4.13 shows the system architecture and main components of the hand controller. Different modules of the PMAC multi-axis motion controller are used to perform the above tasks, respectively. The primary machine interface connector JMACH1 is used to accept incremental encoder inputs of DC motors, and the analog input of the vacuum negative pressure sensor is also accepted via analog-to-digital converters installed in JMACH1. The machine interface connector JMACH2 supplies pulse-and-direction output signals to control the motion of DC motors. The EPOS small-sized full digital smart motion controllers are adopted as slave nodes in motors control, which have a variety of operating modes and can be flexibly assembled using positioning, speed, and current regulation. The multisensor information input and multi-valve control are mainly through JOPTO generalpurpose digital input and output connector. A control program of the end-effector is written in PComm32PRO communication software installed on PC. 2.

EPOS Driving Unit

The EPOS position controller of MAXON motor 5 is a small-sized full digital smart motion controller. It not only serves as a motor driver but also provides a powerful motion control function. A number of operating modes, such as point to point, position control, speed control, torque control, step control, and so on, provide a flexible application in a wide range of drive systems. In the developed damage-free harvesting end-effector, two-position controllers of EPOS 24/5 and EPOS 24/1 are separately configured according to the different current requirements of the motors in the gripper, sucking and pulling system, and laser stem-cutting device (Fig. 4.14). Their core component is TMS320LF2406 DSP (digital signal processing) chip, the instruction execution cycle is only 33 ns, the flash memory area of 32 K, 544B

226

4 Development of Damage-Free Hand–Arm …

PC

DPRAM

PMAC2A-PC/104 JMACH1 Encoder Analog input input

JMACH2 Pulse Direction output output

Negative pressure sensor

Interface board

Epos controller 1/2

DC motor 1/2

Proximity sensor 1/2

Distance Sensor 1/2/3

JOPTO Digital Digital input output

Interface board

Photoelectric isolation plate Encoder 1/2

Force sensor 1/2

Controller board of the vacuum ejector

Air supply solenoid valve

Suction solenoid valve

Blowing solenoid valve

Vacuum switch

Fig. 4.13 Control system diagram

double access RAM, 2 KB single access RAM, and 8 Channel 10-bit A/D converter are included in the chip. The two controllers both provide different communication modes of CAN bus and RS-232 serial port. The hardware connection of EPOS position controller using pulse/direction mode is shown in Fig. 4.15. 3.

PMAC2A-PC104 Motion Controller

The PMAC2A-PC104 motion controller is one of the open motion controllers developed by the Delta Tau Data System Company in the United States (Fig. 4.16). With the aid of Motorola’s DSP56311 digital signal processor, the CPU master frequency is 40MHZ. It has great flexibility and can manipulate 1–4 axis simultaneously, allowing each shaft to move independently. It has the main following features: 128 k × 24 SRAM user memory. 512 k × 8 flash memory for user backup and firmware. 4 Channels Axis-Interface Circuitry, Each Including:

4.4 System Components of the End-Effector

Fig. 4.14 EPOS 24/5 digital motion controller

Fig. 4.15 The hardware connection of the EPOS position controller Fig. 4.16 PMAC2A-PC104 motion controller

227

228

4 Development of Damage-Free Hand–Arm …

Fig. 4.17 The hardware connection of PMAC motion controller

• • • •

12-bit ± 10 V differential analog output (filtered PWM) 3-channel differential single-ended encoder input Pulse-and-direction output pair 5 input flags, 2 output flags.

50-pin IDC header for amplifier and encoder interfaces. 34-pin IDC header for the flag, pulse-and-direction interfaces. RS-232 serial port with 10-pin IDC header. The hardware connection of PMAC motion controller using pulse/direction mode is shown in Fig. 4.17.

4.4.4 Design of Power Supply System [81] To satisfy the demand of mobile harvesting robot, the power supply system uses high-energy lithium battery pack as a power source. The volume and weight of the lithium battery of the same capacity are only about 1/3–1/4 of the ordinary lead-acid battery. Therefore, the power supply of the 24V10Ah lithium battery is used to reduce the load and space occupancy of the harvesting robot and meet the demand for long time power supply. The system uses a series of DC/DC power modules to convert the gradually falling voltage of the lithium battery into a stable 24, 12, and 5 V voltage to supply stable power for all motor systems, control, and connecting plates, sensors, control valves, and laser driving circuits of the end-effector (Figs. 4.18 and 4.19).

4.4 System Components of the End-Effector

229

15 14

Voltage (V)

13 12

Stable working stage

11 10 9 8 7

0

20

40

60

80

100

Capacity Percentage (%) Fig. 4.18 Discharge characteristic curve of lithium battery Vacuum Switch

24V

Vacuum Suction/blowing Valve DC Motor Driver 1/2 PMAC Motion Controller Proximity Switch 1/2

Lithium

Emerg-

Battery ency Pack

DC/DC Power Switch

24V

5V

Distance Sensor 1/2/3

Power Module 34 Channel Terminal Block

Switch 50 Channel Terminal Block DC Motor Driver 3

12V

Force Sensor 1/2/3 Laser Driver A/D Converter

DC/AC Inverter

220V

Fig. 4.19 Structure diagram of the power supply system

Compressor

230

4 Development of Damage-Free Hand–Arm …

4.4.5 Structure Design of the End-Effector [81] 1.

Mechanical Structure of the Gripper

The body of the end-effector includes front shell and rear shell, and light highstrength aluminum–magnesium alloy material is used. The removable structure of the fingers brings convenience to the expansion and maintenance of the end-effector. The fingers are 45 mm wide with an arc surface and 5 mm-thick rubber (Fig. 4.20). The gripper is driven by Maxon DC servomotor. 2.

The Suction Pad Feeding Mechanism

The gear and rack mechanism is selected as the suction pad feeding mechanism. The suction pad is mounted on the front end of a guide rod. The vacuum hose is connected to the rear end of the guide rod. The guide rod is used as a channel for the flow of the vacuum gas. The nylon rack is fixed on the guide rod with the nylon rack to avoid the damage to the motor by accident. 3.

Bearing and Rotation Structure for the Laser Focusing Lens [81, 94]

When an accurate position of a stem is decided with the help of vision and sensor system of a harvesting robot, the focusing lens should rotate to aim at and cut it off. So the actuating mechanism composed of the focusing lens, DC motor, and the

Fig. 4.20 Mechanical structure of the end-effector for tomato harvesting

4.4 System Components of the End-Effector

231

Fig. 4.21 Bearing and rotation structure for the laser focusing lens

bearing system performs actions of rotating the focusing lens precisely. The laser stem-cutting device with its actuating mechanism is shown in Fig. 4.21. 1)

DC Motor

A DC motor integrated with a gearbox is used to drive the focusing lens. For there is no working load on the focusing lens, a small power motor may be able to be selected. In this device, a Maxon mini DC motor integrated with a mini gearbox is adopted whose total weight and length are only 9.8 g and 43.5 mm, respectively. 2)

Bearing Structure

As the mini motor with mini gearbox used in this device is small power, output shaft diameter of the gearbox is only 2 mm, both permissible radial and an axial load of which are Delay performs complicated collimating and focusing of the laser beam is about 250 g, a reliable bearing structure is essential. The bearing structure mainly bears radial and axial load to the output shaft with a thrust bearing composed of the bearing saddle, bearing ring, balls, and cover (Fig. 4.21). A fastening ring is fastened on the focusing lens with two screws. To satisfy the needs of harvesting fruits of different varieties, the tilt angle of the focusing lens can also be adjusted by adjusting screws.

4.4.6 Prototype and Its Performance Indicators [81] The prototype of the end-effector for robotic tomato harvesting developed is shown in Fig. 4.22. In order to effectively reduce the size and weight of the end-effector

232

4 Development of Damage-Free Hand–Arm …

Fig. 4.22 The prototype of end-effector for robotic tomato harvesting

and improve the dynamic performance, the Maxon micro DC motor systems with integrated incremental encoder and reducer are selected as the power source. The motor system with the power of 60, 11, and 0.75 W are used in the gripper, the vacuum sucking and pulling system, and laser stem-cutting device, respectively. The main parameters of Maxon motors and gearboxes are listed in Table 4.9. Design mass of the body of end-effector is only 1.3 kg. The end-effector can be mounted on different manipulators by changing a connecting plate only. The main performance indices of the end-effector are shown in Table 4.8. The test shows that the main indices, motion precision, and stability of the end-effector meet the design expectations. It provides a good hardware platform for further research. Table 4.7 The main parameters of Maxon motors and gearboxes Motor

Gearbox

Power (W)

Nominal torque (mNm)

Nominal speed (rpm)

Weight (g)

Reduction

Weight (g)

Number of gear teeth

pitch of screw (mm)

Gripper

60

85.0

8050

238

4.8:1

118

20

1

Vacuum sucking and pulling system

11

12.1

7670

71

24:1

55

30

–

Laser cutting system of peduncles

0.75

2790

7

64:1

2.8

–

–

0.741

4.4 System Components of the End-Effector

233

Table 4.8 Performance indices of the end-effector for robotic tomato harvesting Gripper

Vacuum suction system

Laser cutting system of peduncles Total

Open degree (mm)

Maximum gripping force (N)

Maximum gripping speed (mm/s)

Rated gripping speed (mm/s)

20–100

123.4

30.6

28.0

Maximum stroke of suction pad (mm)

Maximum pulling force (N)

Maximum pulling speed (mm/s)

Rated pulling speed (mm/s)

110

65.6

1000

401.4

Adjustable angle of Maximum speed focusing lens (°) of focusing lens (rad/s)

Rated speed of focusing lens (rad/s)

−10 ~ + 10

18.8

4.6

Weigh (kg)

Size (mm)

1.3

309 × 160 × 134

4.4.7 Upper Lower Type End-Effector Especially, a special mode of fruit gripping is applied in view of body size and weight of the end-effector and how to locate the peduncle, which means the gripper grips the fruit with an upper finger and a lower finger. The lower finger bears a gravity load of the fruit and the end-effector picks the fruit by non-contact laser cutting instead of twist or scissors cutting, which means small gripping forces are needed. So the strength of the mechanical part can be decreased, together with its weight. Furthermore, the torque and power of the motor needed by the end-effector can also be decreased so that it is possible to select a smaller motor. For locating laser beam on the peduncle accurately, a groove is opened on the upper finger with V-shape front part to compensate a certain degree of the position error of the peduncle (Fig. 4.23). When the gripper grips the fruit, the peduncle enters the groove of the upper finger and then reaches the aim position.

4.4.8 Passive–active Coupled Compliant End-Effector for Robot Tomato Harvesting [95] When the end-effector carries out the harvesting operation on the fruit, there are the complicated posture and swinging of the fruit, the branch and leaf obstacle, the positioning error of the robotic eye–hand system. As a result, the posture deviation and constant change between the end-effector and the fruit are often caused in the sucking and gripping of the fruit, which will seriously affect both the success of the operation and the realization of damage-free gripping based on force feedback.

234

4 Development of Damage-Free Hand–Arm …

Fig. 4.23 Upper lower type end-effector for robotic tomato harvesting

On the basis of the development of the damage-free end-effector, the passive– active coupled compliance scheme of the end-effector is proposed. By the effective coupling of multi-position, multi-dimensional force sensing, and passive compliant structure, both the adaptability of the complex canopy environment, fruit posture and fruit tenderness, and the performance of the speedy damage-free harvesting operation will be greatly improved. The passive compliant structure includes a vacuum compliant structure composed of vacuum bellows suction pad and a spherical joint, a floating-rotation bearing structure of 3-axis force sensor on a front part of the finger (Fig. 4.24). An arc finger surface is installed on the loaded part of the finger 3-axis force sensor, and the base of finger 3-axis force sensor is connected to the finger through a floatingrotation bearing structure. Through the floating-rotation bearing structure, the posture deviation between the finger and the fruit can be automatically adjusted to achieve the passive compliance of the gripping. The vacuum bellow suction pad is connected to the rack through a spherical joint. By replacing the fixed connection with the spherical joint, the position, angle deviation and change between the fruit and the vacuum bellows suction pad can be automatically adapted. At the same time, through the installation structure of the finger 3-axis force sensor with the arc finger surface, all force and friction information in gripping between the fruit and the finger surface can be obtained. The active compliance control is realized through the sensing and feedback of the 3-axis force sensors installed on each of the two fingers and the 6-axis force sensor installed at the wrist. By combining the passive compliant structure of the end-effector with the active compliance control, the position and angle deviation between the suction pad and the fruit, the position and angle deviation between the finger and the fruit are both automatically adjusted to prevent the fruit from being broken or bruised in the harvesting

4.4 System Components of the End-Effector

235

(a) Schematic diagram of main body structure

(b) Schematic diagram of the installation and floating-rotation bearing structure of 3-axis force sensor on front part of finger (Main view)

(c) Schematic diagram of the installation and floating-rotation bearing structure of 3-axis force sensor on Front part of finger (Top view)

(d) Schematic diagram of compliance of the floating-rotation bearing structure to posture deviation 1. Finger, 2.Bi-directional, 3.Guide rod, 4.Gear, 5.Rack, 6.Wrist 6-axis force sensor, 7.Motor, 8.Bevel gear, 9.Bevel gear, 10.Spherical joint, 11.Vacuum bellows suction pad, 12.Finger 3-axis force sensor, 13.Loaded part of finger 3-axis force sensor, 14.Base of finger 3-axis force sensor, 15.End cover, 16.Extension spring, 18.Ball, 19.Front part of finger, 20.Finger base, 21.Arc finger surface, 22.Elastic layer, 23. Screws, 24.Lug, 25.Bolt/nut, 26.Lug.

Fig. 4.24 Passive–active coupled compliant end-effector

236

4 Development of Damage-Free Hand–Arm …

process. And at the same time, either the collision in gripping and potential bruise of the fruit or the collision between the end-effector and the plant is avoided.

4.5 Damage-Free Harvesting Hand–arm System Based on Commercial Manipulator [96] 4.5.1 Background and Needs One typical harvesting robot usually consists of an autonomous vehicle, one or more manipulator with certain end-effectors, and corresponding vision/sensing system and control system. One manipulator (arm) with its end-effector (hand) may be named as a harvesting hand–arm system, which is the vital component that executes the task of harvesting by series motions, usually including reaching object fruit or vegetable, gripping, detaching, and sending the fruit into some container. Several hand–arm systems of harvesting robots have been developed in different countries. Although there are large differences between any two different prototypes, they may be classified into two types: integrated-type and respected-type. The first one refers to a hand–arm system whose manipulator and end-effector were designed and produced interactively, and the latter refers to a hand–arm system that was constructed with industrial manipulator and self-designed or modified endeffector. Either type has its advantages and disadvantages. The integrated-type could be designed for a particular production system, which might gain better coordination of both mechanical and control systems between manipulator and end-effector. However, specific mechanism and structure had to be designed to satisfy the needs of robotic harvesting of different fruits and vegetables, and different cultivation system, which might led to longer development cycle, higher development difficulty and risk, more complicated control algorithm, and worse performance and reliability. Compared with the integrated-type, the respected-type adopted commercial multiaxis articulated manipulator, which could supply enough freedom and dexterity with ideal performance and reliability. As a result, the development cycle, difficulty, and risk could be shortened or reduced greatly. However, as the harvesting task is very complicated which usually includes compliant gripping, detaching, and some assistant motion, the end-effector that usually contains several actuators and sensors is more complicated than that of an industrial robot. So a control unit for the end-effector is necessary to perform tasks of information input and control command output, and then integration between hand controller and an enclosed arm controller would pose a challenge. Finally, the hand–arm coordinative control is vital in harvesting operation in view of the feedback multi-axis motion of both the end-effector and the manipulator.

4.5 Damage-Free Harvesting Hand–arm System …

237

4.5.2 The Control System Structure of Commercial Manipulator [31] A Motoman-SV3J10 6-axis articulated industrial manipulator is adopted to construct the hand–arm system with the self-designed end-effector (Fig. 4.25). The manipulator has vertical and horizontal workspace range of 1019–677 mm respectively (Fig. 4.26), the allowable wrist load reaches 3Kg, and the repeatability of motion is up to 0.03 mm. The JRC controller of the manipulator is free-standing, enclosed type which uses “PAC” language to control the manipulator (Fig. 4.27). Control program of the manipulator is written with “WINCAPS” software installed on PC.

U axis R axis L axis

B axis T axis

S axis

Fig. 4.25 Motoman SV3J10 manipulator

(a) Vertical direction Fig. 4.26 Workspace range of the manipulator

(b) Horizontal direction

238

4 Development of Damage-Free Hand–Arm …

Fig. 4.27 The JRC controller of the manipulator

The functions of each interface are as follows (according to JRC Installation & Maintenance Guide): CN1—RS232C serial interface connector; CN2—CRT display connector; CN3—Keyboard connector; CN4—Mouse connector; CN5—Connector for teaching pendant; CN6—Printer interface; CN7—Power connector for I/O, 9 pins; CN7/1, CN7/2, Internal power source output + 24 V. CN7/3, CN7/4, Internal power source output 0 V. CN7/5, FG. CN7/6, CN7/7 Power input E0V. CN7/8, CN7/9 Power input E24V. CN8—Connector for user input or system input, 50 pins; CN8/1, Power for robot stop (internal + 24 V). CN8/2, Manipulator stop. CN8/3, Power for Enable Auto (internal + 24 V). CN8/4, Enable Auto. CN8/1 –N8/4, for switching between manual mode and automatic mode of the manipulator. CN9–HAND I/O, Connector for end-effector I/O, 20 pins; CN9/1–CN9/8, Port number 64 –71, end-effector output.

4.5 Damage-Free Harvesting Hand–arm System …

239

CN9/9–CN9/16, Port number 48 –55, end-effector input. CN9/17, Power E0V for end-effector. CN9/18, Power E24V for end-effector. CN9/19, CN9/20, not connected. CN10—Connector for user output or system output; CN11—Power connector; CN12—Motor connector; CN13—Encoder connector.

4.5.3 Control System Integration Between the Manipulator and the End-Effector [31, 34] To fulfill communication between the hand controller and the arm controller, effective interface connection has to be constructed firstly, and then the correct relation between programs of PComm32PRO and WINCAPS is also essential. As both the JRC controller and the PMAC motion controller have input and output interface to send and receive signals, while the intermediate relay can realize the on–off of the control circuit, a convenient communication structure is constructed by two relays to complete the information transfer between the two to compose an integrated harvesting hand–arm system. The hardware structure and communication principle are shown in Figs. 4.28 and 4.29, respectively. Two relays are used to judge and transmit each other’s state between PComm32PRO and WINCAPS, respectively: (1)

(2)

A Certain command in arms control program in WINCAPS is used to set digital output from the hand I/O connector of JRC as “ON”, then contacts of relay1 are closed to make digital input from JPOT general-purpose digital I/O of PMAC as low level “0”. Once the digital input from JPOT is detected as “0” in hand control program in PComm32PRO, following hand control program in PComm32PRO will be executed; Also, certain command in hand control program in PComm32PRO is used to set M-variable as high level “1” to change the output of multiplexer port connector JTHW of PMAC from + 5v to 0 V, then contacts of relay2 are opened to make digital input from the hand I/O of JRC as “ON”. As soon as the digital input from JPOT is detected as “ON” in arms control program in WINCAPS, following arm control program in WINCAPS will be executed.

The dual relay communication scheme effectively realizes the coordinated control between the closed controller of the commercial manipulator and the open controller of the developed end-effector (Fig. 4.30).

240

4 Development of Damage-Free Hand–Arm …

PC RS232

USB Relay1

Digital output

Digital input GND

24V

JRC controller

PMAC controller Relay2 GND

GND Digital input

Digital output

Fig. 4.28 Schematic diagram of hand–arm communications

Run arm control program in WINCAPS

No

Run hand control program in PComm32PRO

Set digital output interface of JRC=ON Digital input interface of JRC=ON? Yes

Digital input interface of PMAC=0? Yes

Set digital output interface of PMAC=1

Run following arm control program in WINCAPS

Run following hand control program in PComm32PRO

Reset digital output interface of JRC=OFF

Digital input interface of PMAC=1

Fig. 4.29 Flow chart of hand–arm communication

No

4.5 Damage-Free Harvesting Hand–arm System …

(a) The hand-arm system

241

(b) Test verification

Fig. 4.30 The hand–arm system and test verification

References 1. Takahashi Y, Ogawa J, Saeki K (2001) Automatic tomato picking robot system with human interface using image processing. In: Proceedings of the 27th annual conference of the IEEE of industrial electronics society 2. Kahya E, Arim S (2014) Research on robotics application in fruit harvesting. J Agric Sci Technol b(5):386–392 3. Kitamura S, Oka K (2005) Recognition and cutting system of sweet pepper for picking robot in greenhouse horticulture. In: Proceedings of the IEEE international conference on mechatronics and automation 4. Oka K, Kitamura S. Sweet pepper picking robot in greenhouse horticulture. https://www.lab. kochi-tech.ac.jp/oka-lab/society/ron_pdf/okalab20051010a.pdf 5. Bachche S (2013) Automatic harvesting for sweet peppers in greenhouse horticulture. Kochi University of Technology 6. Kondo N, Nishitsuji Y, Ling P et al (1996) Visual feedback guided robotic cherry tomato harvesting. Trans ASAE 39(6):2331–2338 7. Ceres R, Pons JL, Jiménez AR et al (1998) Design and implementation of an aided fruitharvesting robot (Agribot). Ind Robot 25(5):337–346 8. Ceres R, Pons J, Jiménez A et al (1998) Agribot: a robot for aided fruit harvesting. Ind Robot 25(5):337–346 9. Tanigaki K, Fujiura T, Akase A et al (2008) Cherry-Harvesting robot. Comput Electron Agric 63(1):65–72 10. Ji C, Zhang Z, Yu J et al (2013) Research on key technology of truss tomato picking robot in greenhouse. Sciencepaper Online 11. Hayashi S, Sakaue O (1997) Basic operation of tomato harvesting system using robot: manufacture of two-finger harvesting hand with auxiliary cutting device and basic experiment for harvest. Bull Nat Res Inst Veg Ornam Plant Tea.(Jpn) 133–142 12. Kondo N, Yata K, Iida M et al (2010) Development of an end-effector for a tomato cluster harvesting robot. Eng Agric Environ Food 3(1):20–24 13. Kondo N, Yamamoto K, Shimizu H et al (2009) A machine vision system for tomato cluster harvesting robot. Eng Agric Environ Food 2(2):60–65 14. Kondo N, Taniwaki S, Tanihara K et al (2007) An end-effector and manipulator control for tomato cluster harvesting robot. In: Proceedings of the ASABE Annual International Meeting

242

4 Development of Damage-Free Hand–Arm …

15. Kondo N, Monta M, Fujiura T et al (1993) Study on control method for redundant manipulator— control of tomato harvesting manipulator with 7 degrees of freedom. Shita J 5(1):44–53 16. Kondo N (1994) Basic components of robot for bioproduction. J Robot Soc Jpn 12(7): 952–955 17. Monta M, Kondo N, Shibano Y et al (1992) Studies on tomato harvesting hand (2). J Jpn Soc Agric Mach 56(Supplement):391–392 18. Kondo N, Shibano Y, Mohri K et al (1992) Studies on tomato harvesting hand. J Jpn Soc Agric Mach 54(Supplement):235–236 19. Kondo N (1996) Problems and views on robotic hand for agricultural use—fruit harvesting robotic hands as examples. J Jpn Soc Agric Mach 58(1):139–144 20. Kawamura N, Fujiura T, Ura M et al (1985) Fruit harvesting robot. J Jpn Soc Agric Mach 47(2):237–241 21. Kawamura N, Namikawa K, Fujiura T et al (1984) Study on agricultural robot (Part 1)— microcomputer-controlled manipulator system for fruit harvesting. J Jpn Soc Agric Mach 46(3):353–358 22. Kondo N, Ting K (1998) Robotics for bioproduction systems. Am Soc Agric Eng (ASAE) 23. Krikke J (2005) Robotics research exploits opportunities for growth. Pervasive computing, vol 4(3). IEEE, pp 7–10 24. Monta M, Kondo N, Ting K et al (1998) Harvesting end-effector for inverted single truss tomato production systems. J Jpn Soc Agric Mach 60(6):97–104 25. Monta M, Kondo N, Ting K et al (1997) Single truss upside down tomato production system (2). J Jpn Soc Agric Mach 59(Supplement):323–324 26. Ling P, Ehsani R, Ting K et al (2004) Sensing and end-effector for a robotic tomato harvester. In: Proceedings of the ASAE annual meeting 27. Chiu Y, Yang P, Chen S (2013) Development of the end-effector of a picking robot for greenhouse-grown tomatoes. Appl Eng Agric 29(6):1001–1009 28. Fujiura T, Wada T, Nishiura Y et al (2010) Robot for harvesting cherry tomatoes in mobile cultivation facilities, 1: outline of the robot and 3-D image recognition. J Soc Agric Mach 72(2):143–151 29. Fujiura T, Wada T, Nishiura Y et al (2010) Robot for harvesting cherry tomatoes in mobile cultivation facilities, 2: harvesting experiment. J Soc Agric Mach 72(2):152–159 30. Kobayashi T, Kijima R, Ojika T et al (1996) Development of fruits harvesting robot system aided by virtual reality techniques (Part 2). Vis Feedback Control 8(4):264–270 31. Ota T, Yamashita T, Hayashi S et al (2010) Development of a tomato harvesting robot with a vision system using specular reflection. (Part 1)—selection of hand-approach direction by image processing. J Stat Mech Theory Exp (5):1155–1172 32. Zhao J (2006) Research on tomato harvesting robot technology based on 3D vision. China Agricultural University 33. Yamamoto S, Hayashi S, Yoshida H et al (2009) Development of a stationary robotic strawberry harvester with picking mechanism that approaches target fruit from below (Part 1)—development of the end-effector. J Jpn Soc Agric Mach 71(6):71–78 34. Yamamoto S, Hayashi S, Yoshida H et al (2010) Development of a stationary harvester with picking mechanism that approachestarget fruit form below(Partv2)—construction of the machine’s optical system. J Jpn Soc Agric Mach 72(2):133–142 35. Yamamoto S, Hayashi S, Yoshida H et al (2010) Development of a stationary harvester with picking mechanism that approachestarget fruit form below(Partv3)—performance test with a movable bench system. J Jpn Soc Agric Mach 72(5):479–486 36. Dimeas F, Sako DV, Moulianitis V et al (2013) Towards designing a robot gripper for efficient strawberry harvesting. In: Proceedings of 22nd international workshop on robotics in alpeadria-danube region 220–226 37. Hayashi S, Shigematsu K, Yamamoto S et al (2010) Evaluation of a strawberry-harvesting robot in a field test. Biosys Eng 105(2):160–171 38. Shiigi T, Kurita M, Kondo N et al (2008) Strawberry harvesting robot for fruits grown on table top culture. In: Proceedings of the ASABE annual international meeting

References

243

39. Wang L, Zhang T, Chu J et al (2014) An efficient strawberry harvesting end-effector with large misalignment tolerance. Trans Chinese Soc Agric Mach (s1):252–258 40. Kondo N, Ninomiya K, Hayashi S et al (2005) A new challenge of robot for harvesting strawberry grown on table top culture. In: Proceedings of the ASABE annual international meeting 41. Yamamoto S, Hayashi S, Yoshida H et al (2014) Development of a stationary robotic strawberry harvester with a picking mechanism that approaches the target fruit from below. Jpn Agric Res Q 48(3):261–269 42. Han K (2012) Strawberry harvesting robot for bench-type cultivation. J Biosyst Eng 37(1):65– 74 43. Arima S, Shibusawa S, Kondo N et al (2003) Traceability based on multi-operation robot; information from spraying, harvesting and grading operation robot. In: Proceedings of IEEE/ASME international conference on the advanced intelligent mechatronics 44. Kondo N, Monta M, Hisaeda K (2001) Harvesting robot for strawberry grown on annual hill top, 2: manufacture of the second prototype robot and fundamental harvesting experiment. J Soc High Technol Agric (Japan) 45. Cui Y, Nagata M, Gejima Y et al (2007) Study on strawberry harvesting robot using machine vision for strawberry grown on annual hill top (Part 2). J Jpn Soc Agric Mach 69(2):60–68 46. Zhang F, Li Z, Wang B et al (2014) Study on recognition and non-destructive picking endeffector of kiwifruit. In: Proceedings of the 11th world congress on the intelligent control and automation (WCICA) 47. Zhang F (2014) Research and design on the nondestructive end-effector of kiwifruit harvesting robot. Northwest A & F University 48. Irie N, Taguchi N, Horie T et al (2009) Asparagus harvesting robot coordinated with 3-D vision sensor. In: Proceedings of the IEEE international conference on industrial technology 49. Aloisio C, Mishra RK, Chang C et al (2012) Next generation image guided citrus fruit picker. In: Proceedings of the IEEE international conference on technologies for practical robot applications (TePRA) 50. Pool T, Harrell R (1991) An end-effector for robotic removal of citrus from the tree. Trans ASAE 34(2):373–0378 51. Harrell R, Slaughter D, Adsit P (1989) A fruit-tracking system for robotic harvesting. Mach Vis Appl 2(2):69–80 52. Harrell R (1987) Economic analysis of robotic citrus harvesting in Florida. Trans ASAE 30(2):298–304 53. Burks T, Villegas F, Hannan M et al (2005) Engineering and horticultural aspects of robotic fruit harvesting: opportunities and constraints. HortTechnology 15(1):79–87 54. Kondo N, Monta M, Noguchi N (2012) Agricultural robot: mechanisms and practice. Agricultural information research, vol 21(2). Kyoto University Press, p 50 55. Armada MA, Muscato G, Prestifilippo M et al (2005) A prototype of an orange picking robot: past history, the new robot and experimental results. Ind Robot: Int J 32(2):128–138 56. Allotta B, Buttazzo G, Dario P et al (1990) A force/torque sensor-based technique for robot harvesting of fruits and vegetables. In: Proceedings of the IEEE international workshop on intelligent robots and systems’ 90. ‘Towards a New Frontier of Applications’ 57. Hannan MW, Burks T (2004) Current developments in automated citrus harvesting. In: Proceedings of ASAE/CSAE annual international meeting 58. Flood S (2006) Design of a robotic citrus harvesting end effector and force control model using physical properties and harvesting motion tests. University of Florida 59. Bachche S, Oka K (2013) Performance testing of thermal cutting systems for sweet pepper harvesting robot in greenhouse horticulture. J Syst Des Dyn 7(1):36–51 60. Bachche S, Oka K, Sakamoto H (2012) Development of current based temperature arc thermal cutting system for green pepper harvesting robot[C]. In: Proceedings of the Shikoku-section joint convention of the institute of electrical and related engineers, vol 29 61. Lee B, Rosa U (2006) Development of a canopy volume reduction technique for easy assessment and harvesting of valencia citrus fruits. Trans ASABE 49(6):1695–1703

244

4 Development of Damage-Free Hand–Arm …

62. Lee BS, Rosa U, Cheet-Ancheri K (2006) End-Effector for automated citrus harvesting. In: Proceedings of the ASABE annual international meeting, No. 061143 63. van Henten E, Hemming J, van Tuijl B et al (2002) An autonomous robot for harvesting cucumbers in greenhouses. Auton Robot 13(3):241–258 64. van Henten E, van Tuijl B, Hemming J et al (2003) Field test of an autonomous cucumber picking robot. Biosys Eng 86(3):305–313 65. Zhang K, Yang L, Zhang T (2009) Design of transmission mechanism for vegetable dehydrator test-bench. J Agric Mech Res 31(4):54–56 66. Hayashi S, Ganno K, Ishii Y et al (2002) Robotic harvesting system for eggplants. Jpn Agric Res Q: JARQ 36(3):163–168 67. Hayashi S, Ganno K, Ishii Y et al (2001) Development of a harvesting end-effector for eggplants. J Soc High Technol Agric 13(2):97–103 68. Hayashi S, Ganno K, Kurosaki H et al (2003) Robotic harvesting system for eggplants trained in V-Shape (Part 2). J Soc High Technol Agric 15(4):211–216 69. Hayashi S., Ota T, Kubota K et al (2005) Robotic harvesting technology for fruit vegetables in protected horticultural production. Inf Technol Sustain Fruit Veg Product 227–236 70. Bulanon D, Kataoka T (2010) Fruit detection system and an end effector for robotic harvesting of Fuji apples. Agric Eng Int CIGR J 12(1) 71. Bulanon DM, Kataoka T, Okamoto H et al (2005) Feedback control of manipulator using machine vision for robotic apple harvesting. In: Proceedings of ASAE No.053114 72. Bulanonet D, Kataoka T, Okamoto H (2004) Determing the 3-D location of the apple fruit during harvest. In: Proceedings of the ASAE conference 73. Kataoka T, Ishikawa Y, Hiroma T et al (1999) Hand mechanism for apple harvesting robot. J Jpn Soc Agric Mach 61(1):131–139 74. Zhang S, Kataoka T (2000) Robotic challenges in apple harvesting. J Jpn Soc Agric Mach 62(3):18–21 75. Baeten J, Donné K, Boedrij S et al (2008) Autonomous fruit picking machine: a robotic apple harvester. In: Proceedings of the field and service robotics 76. Zhang Q (2011) Design and research of end-effector for apple harvesting robot. Nanjing Agricultural University 77. Cui P (2010) Research and design of the end-effector of an apple-picking robot. Chinese Acad Agric Mech Sci 78. Hemming J, Bac C, Van Tuijl B et al (2014) A robot for harvesting sweet-pepper in greenhouses. In: Proceedings of the international conference of agricultural engineering 79. Qian S, Yang Q, Wang Z et al (2010) Research on holding characteristics of cucumber and end-effector of cucumber picking. Trans CSAE 26(7):107–112 80. Wang Y (2010) Research on motion planning and control system of the cucumber picking robot. Zhejiang University of Technology 81. Liu J (2010) Analysis and optimal control of vacuum suction system for tomato harvesting robot. Jiangsu University 82. Liu J, Zhang Y (2003) Design and implement a system of grasp identification for dexterous robot hand. Robot 25(3):259–263 83. Yamano I, Maeno T (2005) Five-Fingered robot hand using ultrasonic motors and elastic elements. In: Proceedings of the IEEE international conference on robotics and automation 84. Peng G, Yu L, Liu H (2006) Structural design of a dexterous hand actuated by pneumatic artificial muscle. Trans Beijing Inst Technol 26(7):593–597 85. Barcohen Y, Shahinpoor M (1998) Flexible low-mass robotic arm actuated by electroactive polymers. Proc SPIE—Int Soc Opt Eng 3329:1–6 86. Yang K, Wang Y (2008) Design, drive and control of a novel sma-actuated humanoid flexible gripper. J Mech Sci Technol 22(5):895–904 87. Belfiore NP, Pennestrì E (1997) An atlas of linkage-type robotic grippers. Mech Mach Theory 32(7):811–833 88. Liu J, Peng Y, Faheem M (2020) Experimental and theoretical analysis of fruit plucking patterns for robotic tomato harvesting. Comput Electron Agric 173:105330

References

245

89. Zhang K, Yang L, Zhang T (2009) Design of an end-effector for strawberry harvesting robot. J Agri Mech Res 54–56:60 90. Heisel T, Schou J, Christensen S et al (2001) Cutting weeds with a Co2 laser. Weed Res 41(1):19–29 91. Halim A (2008) Carbon Dioxide (Co2 ) laser cut quality of acrylic using different feed rate and canstant power. (Bachelor)]. University Malaysia Pahang 92. Bruzzone L (2002) Modelling and design of a Parallel Robot for Laser-Cutting applications. In: Proceedings of the IASTED international conference modelling, identification and control 93. Choi H, Li X (2006) Experimental study of laser-cheese interaction for making of thin cheese slices with complex shapes. J Food Eng 75(1):90–95 94. Liu J, Li Z, Li P et al (2008) Design of a laser stem-cutting device for harvesting robot. In: Proceedings of the IEEE international conference on automation and logistics 95. Liu J, LI P, Mao H (2008) Compliant end-effector of fruit and vegetable harvesting robot. ZL200810019826.7. Accessed13 Aug 2008 96. Liu J, Li Z, Wang F et al (2013) Hand-Arm coordination for a tomato harvesting robot based on commercial manipulatort. In: Proceedings of the IEEE international conference on robotics and biomimetics

Chapter 5

Mathematical Modeling of Speedy Damage-Free Gripping of Fruit

5.1 Summary 5.1.1 Research Significance The collision in speedy gripping has two characteristics: the constant active energy input and the restricted contact collision. The law of gripping force changing in this dynamic process is quite different from either the static gripping or passive falling collision. Experiment–simulation combined research is essential to discover this basic law. Also to realize the optimized control of speedy damage-free gripping, it is necessary to find out the relationship between either of the gripping force, the peak collision force and either of the input current, the gripping speed, and the gripping position.

5.1.2 Content and Innovation (1)

(2)

(3) (4)

The changing curve of the gripping force at low, middle, and high speed was obtained, which provides a basis for establishing the dynamic model of gripping collision and analyzing the mechanism of speedy gripping collision. The three-phase change law of collision, rebound, and stress relaxation was found, which provides an important basis for revealing the dynamic mechanism of the specially restricted collision between the active electro-mechanical gripper system and viscoelastic fruit. The energy principle for the speedy gripping collision phenomena between the active electro-mechanical gripper system and viscoelastic fruit was proposed. The relationship between either of the gripping force, the peak collision force and either of the input current, the gripping speed, and the gripping position was obtained, which lays the foundation for the establishment of the optimal control mode of speedy damage-free gripping.

© Science Press, Beijing and Springer Nature Singapore Pte Ltd. 2021 J. Liu et al., Rapid Damage-Free Robotic Harvesting of Tomatoes, Springer Tracts in Mechanical Engineering, https://doi.org/10.1007/978-981-16-1284-8_5

247

248

5 Mathematical Modeling of Speedy Damage-Free Gripping of Fruit

Fig. 5.1 Experiment of speedy fruit gripping. 1. End-effector 2. Computer and EPS interface 3. Manipulator 4. Control box 5. Amplifier 6. Data acquisition board 7. Force sensor

5.2 Experiment of Speedy Fruit Gripping and Special Collision Characteristics 5.2.1 Experiment of Speedy Fruit Gripping [1, 2] The Jinpeng 5 green-ripening tomato fruits, which were manually picked from a tomato planting base in Dantu District, Zhenjiang City, were used as experimental materials. The experiment of speedy fruit gripping was carried out with the selfdeveloped end-effector system of robotic fruit harvesting (Fig. 5.1). The equatorial diameter, height, and mass of the tomato fruits are 69.84–82.09 mm, 50.35– 65.22 mm, and 29.8–217.8 g, respectively. The finger of the end-effector is driven by a 60 W Maxon DC motor, and the speed setting of the motor is realized through the interface program of EPOS controller. In each finger, a micro-force sensor is installed (MDS, measuring range: 100 N, precision: 0.01 N). Five speed levels (0.9, 2.1, 5.2, 12.2, and 17.4 mms−1 ), converted based on transmission ratio, was set, and 45 tomato fruits (9 tomato fruits for each speed level) were randomly selected and loaded on the equatorial plane of the fruit. The collision experiment was completed in 36 h after picking and the peak collision force was recorded.

5.2.2 Collision Characteristics of Speedy Fruit Gripping The low-speed gripping process can be regarded as a (quasi)static loading process, gripping force can be analyzed and judged by the static force balance relationship and be given to the control of gripping operation. However, it is found in the experiment of speedy fruit gripping, instantaneous peak force is caused by the collision

5.2 Experiment of Speedy Fruit Gripping and Special Collision Characteristics

249

Fig. 5.2 Collision force curve of speedy gripping

between fingers and target fruit, which may cause fruit damage (Fig. 5.2). Due to the viscoelastic characteristics of the fruit and the continuous energy input from the fingers, the collision between the fruit and the finger is the active complete-inelastic collision, which is completely different from the free complete-elastic collision. The collision time is from contact between fingers and fruit to stop of finger movement, and its peak force also appears after a certain deformation of the fruit. The results show that the peak collision force has a positive correlation with the collision speed, and the collision time is negatively correlated with the collision speed. Furthermore, the peak collision force is positively correlated with the elastic stiffness of the fruit, while the collision time is negatively correlated with the elastic stiffness of the fruit. The relationship between peak collision forces, collision time, and collision speed under maximum stiffness is shown in Fig. 5.3.

Fig. 5.3 Relationship curve between peak collision force and speed and that between collision time and speed

250

5 Mathematical Modeling of Speedy Damage-Free Gripping of Fruit

5.3 The Special Collision Issue of Speedy Fruit Gripping In the speedy gripping operation, the collision will cause transient overloading and fruit damage. The collision mechanism and pattern of speedy fruit gripping is different from the collision either between fruit and environment of among fruits in harvesting, transporting, grading, and falling: (1)

(2)

The contact collision, between fruit and ground (box bottom) occurred during fruit harvesting, transporting, grading, and falling, is manifested as the process of potential-kinetic energy conversion and loss. By contrast, the continuous energy input from the electro-mechanical gripper system enables the speedy fruit gripping as a continuous process of contact, collision, and grip tightening. Different from the one-side loading and bouncing off in free contact collision, the speedy fruit gripping is characterized by the double-side symmetrical loading and constraint deformation.

Therefore, the mechanism of the collision in speedy fruit gripping with robotic end-effector must be explained by establishing the interaction law between electromechanical gripper system and the viscoelastic fruit.

5.4 Dynamic Characteristics in Different Phases of Speedy Fruit Gripping [1] In the process of gripping, the motor drives the fingers to close, contact the fruit, and continue to input energy to make the fruit deform, until the finger stops moving under the counterforce. The process consists of four phases and each of them has the following kinetic characteristics: (1)

Phase of no-load feeding: Under the drive of the motor, the gripper closes at a certain speed v0 until the finger contacts with the target fruit (Fig. 5.4a).

Fig. 5.4 Dynamic characteristics of different phases of speedy fruit gripping

5.4 Dynamic Characteristics in Different Phases of Speedy Fruit Gripping [1]

(2)

Phase of constant-speed loading: When the fingers contact the fruit and continue to close at speed v0 , the fruit has viscoelastic deformation and produces deformation resistance (N) (N ≤ F 0 ) (Fig. 5.4b). The conditions for the deformation of the fruit at the uniform loading stage are 

(3)

D˙ = v0 D¨ = 0

(5.1)

where D is the fruit deformation, which is equal to the finger displacement from the fingers contacting the fruit, mm; v0 is the constant speed of the fingers, mms−1 ; and F 0 is the maximum static gripping force that is decided by the motor torque and the transmission, N. Phase of collision decelerating: The fruit deformation continues to increase until the deformation resistance (N) exceeds the maximum static gripping force (F0 ), and the fingers slow down until they stop (Fig. 5.4c). According to Newton’s second law, the dynamic characteristics of this phase of collision decelerating is m e D¨ = F0 − N

(4)

251

(5.2)

where me is the equivalent mass of the electro-mechanical gripper system to the finger, kg. Phase of stress relaxing: The gripper is in a self-locking state, and the fruit is subjected to stress relaxation under constant compression deformation (Fig. 5.4d). The condition of fruit deformation at this phase is

D = 0

(5.3)

In the two phases of constant-speed loading and collision decelerating, the motor carries on the continuous energy input, while the fruit is deformed with the closing of the two fingers. But in free contact collisions, there is only the initial kinetic energy when fruit contacts, while one-sided inelastic collision deformation of the fruit happens. Therefore, there is a great difference between the collision mechanism and pattern. It is of great value to describe the special law of the collision of speedy fruit gripping.

252

5 Mathematical Modeling of Speedy Damage-Free Gripping of Fruit

5.5 Fruit Compression Model [1, 3] 5.5.1 The Viscoelastic Properties of Fruit and the Characterization of Constitutive Model 1. (1)

Rheological model of tomato whole-fruit creep characteristics Burger’s model

Burger’s model has been widely used as a classical model for characterizing the creep properties of viscoelastic materials (Fig. 5.5). Burger’s four-element model is composed of a Maxwell body, which is a series structure of an elastic element and a viscous element, and a Kelvin body, which is a parallel structure of an elastic element and a viscous element. The basic rheological differential equation of Burger’s model is     E1 E2 ˙ 1 ¨ E1 E1 E2 E2 ˙ ¨ + + N D= N+ N+ D+ η2 E1 η2 η1 η2 η1 η2

(5.4)

where E 1 is the instantaneous elastic coefficient, N/mm; η1 is the viscosity coefficient of the series viscous element, N·s/mm; E 2 is the delayed elastic coefficient, N/mm; and η2 is the viscosity coefficient of the parallel viscous element, N·s/mm. The creep deformation and unloading deformation recovery formulas of Burger’s model are  F E F0  F0 0 − 2t 1 − e η2 + t (0 ≤ t ≤ ta ) + E1 E2 η1  E2 E F0  F0 − 2t − (t−t ) D= 1 − e η2 e η2 a + ta (t > ta ) E2 η1

D=

(5.5) (5.6)

where t a is time of peak deformation of creep, s. Also in Eqs. (5.5) and (5.6), the ratio of E2 to η2 can be defined as the creep delay Fig. 5.5 The four-element Burger’s model

5.5 Fruit Compression Model [1, 3]

253

τK =

η2 E2

(5.7)

where τ k is the creep delay time, s. As can be seen from Eqs. (5.5) and (5.6), when the load (F0 ) is applied on the fourelement Burger’s model, the instantaneous elastic deformation and unrecoverable permanent viscosity deformation are produced by the elastic element and viscous element in the Maxwell body, respectively, and the delayed elastic deformation is produced by the Kelvin body [4–7]. (2)

Creep parameters of tomato whole fruit of different ripeness

The parameters of Burger’s model of tomato fruit at different ripeness stage are obtained by fitting the experimental creep curves in Fig. 3.23 with Eqs. (5.5) and (5.6). As shown in Table 5.1, the ripeness level has a significant effect on the creep parameters of tomato (E 1 , E 2 , η1 , and η2 ). It can be seen that with the increase in ripeness level, all parameters (E1 , E2 , η1 , and η2 ) decrease significantly. It is indicated that the instantaneous elastic deformation, the viscoelastic deformation, and the final plastic deformation of the tomato fruit should be increased with the increase of ripeness. However, there is no a significant difference in the creep delay time (τk ) for different ripeness. These phenomena may be due to the changes in their internal tissues. With the increase of ripeness, the insoluble pectin in tomato fruit gradually hydrolyzed into soluble pectin under the action of pectin hydrolase. It was accompanied by the dissolution of the colloid and the destruction of the primary wall in the cell wall. The appearance showed the change of fruit substance, which resulted in a significant change of mechanical indices of viscoelasticity. In the mechanical harvesting, transporting, or sorting of tomato fruit, different parameters for different ripeness levels should be used to reduce possible bruise in fruit. 2. (1)

Rheology model of stress relaxation of whole tomato fruit Maxwell model

According to the change in the trend of an experimental curve of stress relaxation, the stress relaxation process of tomato can be seen qualitatively, but it is difficult to explain the stress relaxation quantitatively. But the rheology model can solve this problem. The Maxwell model can better describe the relaxation characteristics of agricultural products, so it is often used to fit the relaxation curve. The elastic modulus, relaxation time, and viscosity coefficient of the model can reflect the rheological properties of the material. Generalized Maxwell five-element model is shown in Eq. 5.8 [7]. F(t) = D0 E 0 + D0

2 i=1

E Mi e

−τ t

Mi

(5.8)

3.837 ± 0.174a

η2 (N·s/mm)

τ K (s)

115.599 ± 23.234a

η1 (N·s/mm) 3.924 ± 0.152a

76.391 ± 6.731b

1760.850 ± 119.566b

19.505 ± 1.640b

29.646 ± 5.150a

2751.091 ± 323.909a

E 2 (N/mm)

5.501 ± 0.230b

Turning

6.068 ± 0.514a

E 1 (N/mm)

Breakers

Table 5.1 Parameters in Burger’s model for tomato fruit of different ripeness

3.944 ± 0.211a

57.479 ± 7.471c

1379.300 ± 64.453c

14.4010 ± 1.223c

4.784 ± 0.306c

Pink

3.725 ± 0.147a

42.783 ± 1.251d

1276.600 ± 50.040d

11.541 ± 0.470d

4.192 ± 0.217d

Light red

254 5 Mathematical Modeling of Speedy Damage-Free Gripping of Fruit

5.5 Fruit Compression Model [1, 3]

255

where F(t) is the force applied on the model, N; D0 is the constant deformation of the model, mm; E 0 is equilibrium modulus of elasticity, MPa; E Mi is the modulus of elasticity of the ith Maxwell body; and MPa; τ Mi is the relaxation time of the ith Maxwell body, τ Mi = ηMi /E Mi , s. The five-element Maxwell model is made up of two Maxwell bodies and a Hooke body in parallel. Its schematic diagram is shown in Fig. 5.6. Using the least square method of quasi-Newton method (BFGS) and the general global optimization method, the stress relaxation model of the tomato fruits of four ripeness levels was obtained by fitting the selected data. The comparison of the typical simulated curve with the experimental curve is shown in Fig. 5.7. The correlation of the model reached 0.98, which proves that the five-element Maxwell model can well describe the stress relaxation characteristics of tomato fruits. (2)

Stress relaxation model parameters of whole tomato fruit at different ripeness levels

The Maxwell model stress relaxation equation as shown in Eq. (5.8) is used to fit the experimental creep curves of tomato at different ripeness levels as shown in Fig. 3.24, Fig. 5.6 The five-element Maxwell model

Fig. 5.7 Comparison of the typical simulated curves with the experimental curves

256

5 Mathematical Modeling of Speedy Damage-Free Gripping of Fruit

Table 5.2 Stress relaxation model parameters of whole tomato fruit at different ripeness levels Parameter

E 0 (MPa)

E 1 (MPa)

τ 1 (s)

η1 (N·s/mm)

E 2 (MPa)

τ 2 (s)

η2 (N·s/mm)

Breakers

8.478

7.226

2.165

15.613

1.841

40.069

74.220

Turning

3.133

5.901

1.928

8.747

0.950

35.700

33.956

Pink

2.629

5.447

1.836

9.950

0.824

35.021

28.853

Light red

2.445

5.532

1.725

9.534

0.778

34.359

26.728

and the fitting results are listed in Table 5.2. The results show that the ripeness level has a significant influence on the model parameters (E 0 , E 1, and E 2 ). The viscosity coefficient (η1 and η2 ) and the relaxation time (τ 1 and τ 1 ) of “Breakers” stage is significantly higher than that of the other three ripeness level whose parameters are similar.

5.5.2 Burger’s Modified Model for Characterization of Creep Properties of Whole Fruit 1.

The deficiency of Burger’s model

As shown in Eqs. (5.5) and (5.6), the key defect of Burger’s model is that when the creep exceeds a certain time, the viscoelastic deformation at the end time of creep, −

E2

t

F 0 (1 − e η2 )/E 2 , is assumed to be maximum (the exponential term approaching 0). All the increments of the deformation are derived from the viscosity deformation (F0t /η1 ), and thus the later creep deformation is near linear rise, which deviates from the objective facts of “strain saturation when the time is infinite” [8–11]. As a result, the value of the highest point is always overestimated [11, 12], and the increasing trend of creep deviated from the actual situation. Therefore, although Burger’s model can fit the data well, it cannot meet the need of “prediction” [11, 13]. 2.

The structure and defects of the existing modified model

Based on the deficiency that the viscosity deformation of Burger’s model is overestimated in the later period of creep, different researchers have modified the series viscosity element (η1 ) (Fig. 5.8), and, respectively, put forward two different modified methods of “exponential” and “power function”, respectively. (1)

The exponential modified model

Many researchers have discussed the modified model whose viscosity coefficient changes exponentially [8, 10, 11, 14, 15] η1 (t) = η0 ekt (0 < k < 1)

(5.9)

5.5 Fruit Compression Model [1, 3]

257

Fig. 5.8 Modified Burger’s model with a time-dependent viscous element

where η1 (t) is the modified viscosity coefficient of the series viscous element in Burger’s model, N·s/mm, both η0 and k are constants. The expression of its creep phase is changed to D=

E F0 F0 F0 − 2t + (1 − e η2 ) + (1 − e−kt ) E1 E2 kη0

(5.10)

The expression in the unloading phase correspondingly becomes D=

E E F0 F0 − 2t − 2 (t−t ) (1 − e η2 1 )e η2 1 + (1 − e−kt1 ) E2 kη0

(5.11)

It can be found from the above equation, when the viscosity coefficient η1 increases exponentially with time, the rising rate becomes larger and larger. And the longer the creep time, the faster is the increase in the viscosity coefficient (η1 ). The accelerated infinitely increasing trend of the viscosity coefficient (η1 ) deviated from the reality. This causes a much lower ratio of viscous deformation when applying to long-term creep prediction, and the deviation of the prediction trend. (2)

The power-function modified model

Some researchers have proposed the modified method of viscosity coefficient changing according to power-function law [12] η1 (t) = η0 t p (0 < p < 1)

(5.12)

where p is a constant. The expression of its creep phase is changed to D=

E F0 F0 F0 − 2t t 1− p + (1 − e η2 ) + E1 E2 η0 (1 − p)

(5.13)

The expression in the unloading phase correspondingly becomes D=

E E F0 F0 − 2t − 2 (t−t ) 1− p t (1 − e η2 1 )e η2 1 + E2 η0 (1 − p) 1

(5.14)

258

5 Mathematical Modeling of Speedy Damage-Free Gripping of Fruit

Compared with the exponential law, with the creep, the power function of the viscosity coefficient (η1 ) will gradually increase and reach the peak value, which can effectively reflect the “strain saturation” law of creep. But because of its initial viscosity coefficient of 0, its creep deformation rate is D =

F0 − Eη 2 t F0 e 2 + t−p η2 η0

(5.14)

It is found that the initial creep rate of the modified model is infinite, which also deviates from reality and causes the very steep creep curve at the initial stage. 3.

The proposal of four-element and six-parameter model

Based on the above analysis, the power-function modified method can express the actual law of creep development more ideally, but it is necessary to be reconstructed according to the defect that the viscosity coefficient (η1 ) is 0 at an initial time. The first consideration is to give the initial time a certain value that is η1 (t) = m + η0 t p (0 < p < 1)

(5.15)

where m is a constant, N·s/mm. However, there is a problem of integral failure in its viscous deformation term

t F0 ( m+η0 t p dt). Therefore, the viscous deformation item in Burger’s model is modi0

fied by an inverted treatment that is to transform the viscosity coefficient η1 in the denominator into a power-function structure with constant term D=

E F0 F0 F0 − 2t + (1 − e η2 ) + t E1 E2 m + η0 t p

(5.16)

The expression in the unloading phase correspondingly becomes D=

E E F0 F0 − 2t − 2 (t−t ) (1 − e η2 1 )e η2 1 + p t1 E2 m + η0 t1

(5.17)

Thus, the “four-element and six-parameter” model is formed, whose viscosity coefficient η1 changes according to the law of Eq. (5.15). And the model is evaluated through the comparison of the fitting and prediction results. Next, this new model is evaluated through comparison of fitting and prediction results. 4. (1)

Comparison of fitting and prediction performance between Burger’s model and its different Modified models of the curve fitting precision of different models

Using the Curve Fitting Toolbox of MATLAB, the above models were used to fit the experimental data of creep phase of tomato fruit with different ripeness, and the

5.5 Fruit Compression Model [1, 3]

259

creep trend and unloading recovery were predicted with the fitting parameters. The typical curve fitting and prediction effect of each model were shown in Fig. 5.9. It can be seen from the mean coefficient of determination (R2 ) and sum of squared error (SSE) for different ripeness (Fig. 5.10a, b), and the goodness of fit of Burger’s model is obviously inferior to that of other modified models, indicating that all the above modifications to Burger’s model have played a significant effect. The mean coefficient of determination (R2 ) and the sum of squared error (SSE) of the fourelement and six-parameter model were 0.9975–0.9994 and 0.04047–0.07633 for different ripeness levels, respectively. Therefore, the best curve fitting accuracy is achieved in this four-element and six-parameter model.

Fig. 5.9 Comparison between typical model curves and experimental measurement curves

Fig. 5.10 Curve fitting effect of different models

260

(2)

5 Mathematical Modeling of Speedy Damage-Free Gripping of Fruit

Creep deformation rate

The creep deformation rate represents the slope of creep deformation curve and the speed of creep change. According to the definition of creep rate, creep deformation rate calculation of experimental data is carried out as D ≈

D t

(5.18)

Due to the fluctuation of the experimental data, the interval t = 0.1 s is taken to calculate the creep deformation rate, and the smooth processing of 100-point moving average is carried out. Because of the time interval and moving average processing, the starting point of the curve is t = 0.6 s. The approximate creep deformation rate curve obtained from the experimental data is shown in Fig. 5.9. In each fitting model, the creep deformation rate of Burger’s model, the exponential modified model, and the four-element and six-parameter model are as follows: Burger’s model: F0 − Eη 2 t F0 e 2 + η2 η1

(5.19)

F0 − Eη 2 t F0 e 2 + e−kt η2 η0

(5.20)

D = Exponential modified model: D =

Four-element and six-parameter model: D =

F0 − Eη 2 t F0 mk F0 t k e 2 + − k η2 η0 + mt (η0 + mt k )2

(5.21)

It is known from Eq. (5.14) that there is a fatal defect for the power-function correction model that the initial creep deformation rate is infinite. It can be seen from the comparison of creep deformation rate of different models (Fig. 5.11) that the creep deformation rate of Burger’s model at the beginning of the creep is too low and that of the exponential modified model is also lower than the reality. Compared with the above models, the four-element six-parameter model is the most accurate expression of the change rule of curve slope. (3)

Expression precision of key indices

The creep deformation rate is the proportion of the delayed elastic deformation and permanent viscous deformation to the total deformation, and the elastic degree is the proportion of the recoverable part (instantaneous elastic deformation and delayed elastic deformation) to the total deformation [16]. The above two key indices of

5.5 Fruit Compression Model [1, 3]

261

Fig. 5.11 Creep deformation rate of different models

creep are combined to reflect the characteristics and laws of creep. According to the fitting results of different models for experimental data, the mean relative fitting error of different ripeness levels is used to evaluate the accuracy of the key indicators. Its expression is  10   1  D − D0 δ= × 100%  10 i=1 D0

(5.22)

where δ is the mean relative fitting error, % and D0 is the experimental value of deformation The mean relative fitting error of the creep deformation rate of Burger’s model at different ripeness levels is 7.46–15.41%, and that of the power-function model is up to 17.18–25.21%. The key reason is that the viscosity coefficient of the initial time (η1 ) is 0 and the creep rate is infinite, which causes the curve steepness of the initial time and the delay projectile. The exaggeration of sexual deformation makes the instantaneous elastic deformation seriously underestimated, resulting in a serious departure from the expression of creep deformation rate. The average relative errors of the exponential correction model and the four-element six-parameter model are 1.91% to 5.27% and 2.54% to 3.04%, respectively, only 1/2–1/3 of Burger’s model, and the expression precision is greatly improved. As shown in Fig. 5.12b, the average relative fitting error of the elastic degree of Burger’s model in different ripeness levels is up to 18.52%–30.37%, and that of the exponential modified model is also 14.29%–24.33%. Compared with the abovementioned models, the average fitting errors of the power-function modified model and the four-element and six-parameter model are only 4.35%–8.27% and 2.89%– 7.12%, respectively. The main reason is that Burger’s model and the exponential modified model have serious error for the estimation of permanent viscous deformation in unloading (Fig. 5.9). The power-function modified model and the fourelement and six-parameter model can reflect the real permanent viscosity, and thus more accurately express the elasticity of the tomato fruit.

262

5 Mathematical Modeling of Speedy Damage-Free Gripping of Fruit

Fig. 5.12 Average relative fitting error of different models

(4)

Prediction precision of creep

The prediction ability is the key to characterizing the accuracy and trend reliability of the model. The trend of long-term creep prediction shown in Fig. 5.9 shows that Burger’s model has a linear upward trend, and the exponential modified model descends too fast because of the infinite increasing trend of the viscosity coefficient η1 , which are obviously deviating from the actual state of “strain saturation”. To compare the prediction precision of creep of different models, the model fitting is carried out with experimental data within the initial 30 s of the creep section, and then the creep deformation of the 50th s is predicted with the fitting parameters. It can be found that the average relative prediction errors of Burger’s model, the exponential modified model, the power-function modified model, and the fourelement and six-parameter model are 3.39%–3.95%, 1.85%–2.13%, 0.39%–0.84%, and 0.29%–0.46%, respectively (Fig. 5.13). Compared with Burger’s model, the prediction accuracy of the exponential modified model is increased about one time, Fig. 5.13 Average relative error of creep deformation prediction

5.5 Fruit Compression Model [1, 3]

263

and that of the power-function modified model is increased 2–5 than that of the exponential modified model. The four-element and six-parameter model further reduces the average prediction error to less than 0.5%, which realizes a breakthrough in order of magnitude.

5.6 Complex Collision Model in Speedy Gripping of Fruit [1] 5.6.1 Phase of Constant-Speed Loading and Phase of Stress Relaxing Substituting the boundary condition of fruit deformation for phase of constant-speed loading, Eq. (5.1) and the boundary condition of fruit deformation for phase of stress relaxing, Eq. (5.3), into the basic rheological differential equation of Burger’s model, Eq. (5.3), the following equations can be obtained:  E1 E2 ˙ E1 E2 E1 E1 E2 + + N= v0 N+ η2 η1 η2 η1 η2 η2   E1 E1 E2 ˙ E1 E2 N¨ + + + N =0 N+ η2 η1 η2 η1 η2

N¨ +



(5.23) (5.24)

Equation (5.23) is the second-order constant-coefficient inhomogeneous equation for phase of constant-speed loading, while Eq. (5.24) is the homogeneous linear differential equation for phase of stress relaxing. The model solution for the two phases can be obtained based on their initial conditions, respectively [14, 17]. For phase of constant-speed loading, that is, N = v0 (η1 +

−E 1 − η1r2 r1 t E 1 + η1 r 1 r 2 t e + √ e ) √ 1 1

(5.25)

For phase of stress relaxing, that is, N=

D0 ( Eη1 2E2 + E 1r1 ) r (t−t ) D0 (− Eη1 2E2 − E 1r2 ) r (t−t ) e1 1 + e2 1 √ √ 1 1

And in Eqs. (5.25) and (5.26), there is  1 =

E1 E1 E2 + − η2 η1 η2

2 +4

E1 E2 η22

(5.26)

264

5 Mathematical Modeling of Speedy Damage-Free Gripping of Fruit

where t 1 is the time spent from fingers touching fruit surface until they stop moving, s; D0 is the peak collision deformation of fruit, mm; and Δ1 is the discriminant of the characteristic equations of Eqs. (5.23) and (5.24).

5.6.2 Phase of Collision Decelerating It can be obtained from the basic rheological differential equation of Burger’s model, Eq. (5.4), that     ... E 2 E1 E1 E2 ¨ 1 ... E1 E2 ˙ ¨ + + + D = N + N + N D η2 E1 η2 η1 η2 η1 η2

(5.27)

If substituting the dynamic characteristics for phase of collision decelerating, Eq. (5.3), into Eq. (5.27), it can be further obtained that ... N+



   E1 E1 E2 ¨ E1 ˙ E1 E2 E1 E2 + + + F0 = 0 N+ N− η2 η1 η2 η1 η2 me m e η2

(5.28)

Equation (5.28) is a third-order constant-coefficient inhomogeneous linear differential equation. For the convenience of analysis, to order ⎧ E1 E1 E2 ⎪ ⎨ p 1 = η2 + η1 + η2 q1 = Eη11 ηE22 + mE1e ⎪ ⎩ q2 = Em1eEη22

(5.29)

So Eq. (5.28) can be expressed as ... N + p1 N¨ + q1 N˙ + q2 N − q2 F0 = 0

(5.30)

According to literature [14], the discriminant of the characteristic equations of Eq. (5.29) is 2 = (

p1 q1 p13 q2 2 q1 p 2 − − ) + ( − 1 )3 > 0 6 27 2 3 9

The characteristic equation of Eq. (5.29) has one real root λ1 and two conjugate complex root λ2,3 = α ± iβ(α, β∈R), and the solution of Eq. (5.29) is N = F0 + [C1 cos(β(t − t0 )) + C2 sin(β(t − t0 ))] eα(tt0 ) −C1 eλ1 (tt0 ) The initial condition of the equation for phase of collision decelerating is

(5.31)

5.6 Complex Collision Model in Speedy Gripping of Fruit [1]

265

⎧  ⎨ N t=t0 = F0 = v0 D˙  ⎩  t=t0 D t=t0 = v0 t0

(5.32)

where t 0 is the time spent to complete phase of collision decelerating, s. According to the initial condition, Eq. (5.32), the solution of Eq. (5.31) can be obtained as follows: ⎧ αβλ2 (αβt0 −2) ⎨ C1 = mv0 2βλ2 +α2 λ2 +α1 2 β 2 −β 2 λ2 −2αβλ 1 1 1 1 (5.33) 3 2 2 2 2 3 αβ λ t −2β λ −α β λ1 t0 +2αβ 3 λ1 0 1 1 ⎩ C2 = mv0 (β + ) 2βλ2 +α 2 λ2 +α 2 β 2 −β 2 λ2 −2αβλ 1

1

1

1

It can be found that the constants C1 and C2 are proportional to the initial gripping speed v0 .

5.7 The Basic Law of Collision in Robotic Gripping of Fruit [1] 5.7.1 The Law of Collision Force in Robotic Gripping of Fruit By substituting Burger’s model parameters of tomato fruit in Table 5.1 at different ripeness levels into the complex three-phase collision model, including Eqs. (5.25), (5.31), and (5.26), the simulation curve of force change in robotic gripping of fruit is obtained. Taking the tomato fruit of “Breaker” stage as example (Fig. 5.14), the low-speed gripping of the fruit is a quasi-static loading process, and the proportion of phase of collision decelerating, t 1 − t 0 , and its contribution to the peak force are very little. At the same time, it needs a longer time of t1 to complete the gripping of the fruit. With the increase of initial gripping speed v0 , the time required to complete the contact and collision, t 1 , rapidly decreases, and the peak collision force Fig. 5.14 Simulated collision force curve in robotic gripping of fruit (green)

266

5 Mathematical Modeling of Speedy Damage-Free Gripping of Fruit

is significantly increased after the phase of collision decelerating. After reaching the peak collision force, the fingers stop moving and the fruit presents stress relaxation.

5.7.2 The Influence of Initial Gripping Speed and Fruit Ripeness on Gripping Collision Time The time required to complete contact and collision, t1 , consists of time spent in both phase of constant-speed loading and phase of collision decelerating. From Eq. (5.25) and the boundary condition, Eq. (5.32), the time spent in phase of constant-speed loading, t 0 , is nonlinear negative correlation with the initial loading speed v0 . The time spent in phase of collision decelerating, t 1 − t 0 , is determined by Eq. (5.30) according to both the following boundary conditions and Eq. (5.2):  D˙ t=t1 = 0

(5.34)

Formula (5.34) is difficult to solve out the explicit solution of t 1 − t 0 , but it can be found that t 1 − t 0 is the function with the real root, λ1 ; the real part of complex root, α; and the imaginary part of complex root, β, of the characteristic equation of Eq. (5.29) as variables t1 − t0 = f (α, β, λ1 )

(5.35)

Since λ1 , α, and β are determined by the four-element parameters of Burger’s model, the time spent in phase of collision decelerating, t 1 − t 0 , is a constant which is independent of the initial gripping speed, v0 . The following conclusions can be drawn from three-phase complex collision model: (1)

(2)

With the increase of initial gripping speed v0 , the time spent in phase of constant-speed loading, t 0 , is reduced rapidly, while the spent in phase of collision decelerating, t 1 − t 0 , remains constant. With the increase of initial gripping speed v0 , the time required to complete the contact and collision, t 1 , has an extremum, which is determined by phase of collision decelerating.

Figure 5.15 clearly reflects the above laws, and also indicates that the riper the fruit, the softer the fruit, and the longer the time to complete the contact and collision. At the initial gripping speed of 0.1 mm·s−1 , the time required for the fingers to complete contact and collision of tomato fruit at ripeness levels of breakers, turning, pink, and light red reach 6.17, 16.01, 19.29, and 20.81 s, respectively. And at the initial gripping speed of 6 mm·s−1 , the time required for the fingers to complete contact and collision of tomato fruit at the above ripeness levels in sequence are 0.62, 1.06, 1.19, and 1.24 mm·s−1 , respectively, which is only 10%, 6.6%, 6.2%, and 6.2% of the time required at the initial speed of 0.1 mm·s−1 , respectively. When the initial

5.7 The Basic Law of Collision in Robotic Gripping of Fruit [1]

267

Fig. 5.15 Relation between time required to complete contact and collision, t1 , and initial gripping speed, v0

gripping speed exceeds 10 mm·s−1 , the collision time tends to be determined by the extremum value determined by phase of collision decelerating, so it is very limited to continue to increase the initial gripping speed to improve the operation efficiency.

5.7.3 The Influence of Initial Gripping Speed and Fruit Ripeness on Gripping Collision Deformation The fruit deformation of phase of constant-speed loading is as  D t=t0 = v0 t0

(5.36)

Because of the nonlinear negative correlation between t0 and the initial gripping speed, v0 , the fruit deformation in phase of constant-speed loading is nonlinearly related with v0 as shown in Fig. 5.16. At lower speed, the fruit deformation in phase Fig. 5.16 Relation between gripping collision deformation in phase of constant-speed loading and initial gripping speed

268

5 Mathematical Modeling of Speedy Damage-Free Gripping of Fruit

Fig. 5.17 Relation between the collision deformation of tomato fruit and initial gripping speed

of constant-speed loading is larger at the uniform loading stage, but it tends to be stable with the increment of the initial gripping speed, v0 . From Eqs. (5.2), (5.31), and (5.33), the fruit deformation in phase of constantspeed loading can be obtained as  C2 C2 2C1 C1 sin(β(t1 − t0 )) + ( 2 D t=t1 = [− 2 + 2 − α β αβ α C1 2C2 C1 α(t1 −t0 ) ) cos(β(t1 − t0 ))]e + 2 + − 2 eλ1 (t1 −t0 ) β αβ λ1

(5.37)

As t 1 − t 0 is a constant, while C 1 and C 2 are proportional to the initial gripping speed, v0 , the fruit deformation in phase of constant-speed loading is also proportional to initial gripping speed, v0 . The gripping collision deformation of the fruit, D0 , is composed of the fruit deformation in both phase of constant-speed loading and phase of collision decelerating, so it has an approximate linear relationship with initial gripping speed, v0 , except the extremely low speed (Fig. 5.17). At the initial gripping speed of 0.1 mm·s−1 , the fruit deformations at ripeness levels of breakers, turning, pink, and light red are only 0.60, 1.57, 1.90, and 2.05 mm, respectively. And at the initial gripping speed of 30 mm·s−1 , the fruit deformations at the above ripeness levels in sequence are up to 10.77, 17.32, 19.20, and 19.91 mm, respectively.

5.7.4 The Influence of Initial Gripping Speed and Fruit Ripeness on Peak Collision Force By Eq. (5.31), the peak gripping collision force is  N0 = N t=t1 = F0 + [C1 cos(β(t1 − t0 )) + C2 sin(β(t1 − t0 ))] ea(t−t0 ) − C1 eλ1 (t−t0 )

(5.38)

5.7 The Basic Law of Collision in Robotic Gripping of Fruit [1]

269

Fig. 5.18 Relation between the peak gripping collision force and initial gripping speed

As t 1 − t 0 is constant, while C 1 and C 2 are proportional to the initial gripping speed, v0 , the peak gripping collision force is linear with the initial gripping speed v0 . At lower speed, the gripping process is approximately static loading, and the peak impact force is close to the maximum static gripping force F 0 . However, the high-speed collision will produce a great peak force. The relationship between the peak gripping collision force and initial gripping speed obtained from the complex three-phase model is shown in Fig. 5.18. At the initial gripping speed of 10 mm·s−1 , the average gripping collision forces at ripeness levels of breakers, pink, and light red are 43.96, 27.92, and 27.18 N, respectively. And at the initial gripping speed of 40 mm·s−1 , the average gripping collision forces at the above ripeness levels in sequence are up to 157.84, 93.69, and 90.72 N, respectively. By combining the above relation between the peak gripping collision force and initial gripping speed and the experimental results of tomato fruit rupture force [1, 2], the probability curve of the fruit rupture damage is obtained (Fig. 5.19). It can be found that when the initial gripping speed is 10 mm·s−1 , the rupture damage probabilities of tomato fruits at each ripeness level are 2%, 0.4%, and 0.4%, respectively.

Fig. 5.19 Probability of fruit rupture at different initial gripping speeds

270

5 Mathematical Modeling of Speedy Damage-Free Gripping of Fruit

However, too larger initial gripping speed will not be of benefit to the improvement of operational efficiency but only lead to a rapid increase of fruit damage probability. At the same time, when the initial gripping speed is lower than 20 mm·s−1 , it is surprising to find that the rupture damage probabilities of tomato fruits of the “Pink” stage and the “Breakers” stage are the lowest and the highest, respectively. It may be due to the double influence of the peak collision force and the rupture force threshold. However, when the initial gripping speed is greater, it is obvious that the riper the fruit, the higher is its rupture damage probability.

5.8 The Theoretical Calculation of the Time Consumption of Gripping [2] 5.8.1 The Stroke Composition of the Finger Gripping Process As shown in Fig. 5.20, the fingers are closed from a certain opening state to achieve reliable gripping. The stroke can be divided into two segments: approaching and gripping. In the approaching segment, the fingers are closed from a certain degree of opening to contact the fruit (w4 ), and the fingers are in no-load state. Then in the gripping segment, the fingers continue to close together and the fruit produces a certain deformation (δ), so that a certain force is produced between the fingers and fruit, so as to achieve the purpose of reliable gripping. Fig. 5.20 Parameter diagram of fruit gripping with robotic fingers

5.8 The Theoretical Calculation of the Time Consumption of Gripping [2]

271

5.8.2 Dimension Relation of Fruit Gripping with Robotic Fingers As shown in Fig. 5.20, to ensure that the fruit can safely enter the gripping position, the finger must ensure a certain degree of opening, so that the opening size, w1 , exceeds certain allowance of fruit diameter, which can be set as 10 mm: w1 = 2R + 10

(5.39)

The distance between finger and fruit, w4 , is related to the opening size, w1 w4 =

w1 + w3 − R 2

(5.40)

Substituting Eq. (5.39) into Eq. (5.40), it can be obtained as w4 = w3 + 5

(5.41)

According to the experiment results of mechanical properties of tomato fruit, as shown in Table 3.8, the relationship between the gripping force and the deformation of tomato fruit at ripeness level from breakers to light red is as follows: N = k g δ = (4.95−12.55)δ

(5.42)

The fruit stiffness k g varies significantly with ripeness levels and individual differences. When the inner surface of finger is a vertical arc (Fig. 5.21), the gripping force needed for the fingers to hold the fruit reliably is N = k gri ·

Fig. 5.21 Force balance relationship between the vertical arc curved surfaces and fruit

G · cos θ 2f

(5.43)

272

5 Mathematical Modeling of Speedy Damage-Free Gripping of Fruit

where k gri is a safety factor; G is the weight of tomato fruit, N; f is the sliding friction coefficient between fruit and finger surface; and θ is the center angle of the vertical arc, rad.

5.8.3 The Time Consumption Composition of the Finger Gripping Process According to the above analysis and Fig. 5.4, for the control of gripping, the finger gripping process can be divided into three stages: unload acceleration, unload constant-speed approaching, and collision. The time required to complete gripping is tgri = tacc + tg + t1

(5.44)

where t gri is the total time consumption of the finger gripping process, s; t acc is the time spent to accelerate the fingers unloaded from static state, s; and t g is the time spent in unload constant-speed stage until contacting the fruit, s.

5.8.4 Selection of Damage-Free Control Mode w4 is the unload feeding phase as shown in Fig. 5.4. Therefore, in theory, w4 can be completed as quick as possible, and then the finger speed can be reduced gradually after detection of the contact to fruit surface with force sensing. When the gripping force reaches a certain value, the fingers stop. But in fact, the above analysis is only applicable to the (quasi)static process at much lower speed, and the instantaneous peak force produced by the collision between the fingers and the fruit during rapid gripping may cause damage to the fruit (Fig. 5.2). Speedy damage-free gripping of fruit is a challenging task. It may be relatively simple and convenient to control the gripping speed to realize the reliable damagefree gripping. In speed control mode, when contacting the fruit, the fingers gradually slow down until stopped to grip the fruit firmly. According to the experimental curve of the peak collision force, collision time, and initial gripping speed (Fig. 5.3), in order to avoid the gripping collision damage, the peak collision force should be less than the minimum rupture force 46.8 N of the transverse (radial) compression shown in Table 3.9. So the theoretical speed threshold of the damage-free gripping is 14.1 mm/s which is known from Fig. 5.3.

5.8 The Theoretical Calculation of the Time Consumption of Gripping [2]

273

5.8.5 Time Calculation of Damage-Free Gripping 1.

Unload acceleration stage

If selecting linear acceleration in the starting of gripping, the displacement and time consumption of each finger in the unload acceleration stage are, respectively, as follows: tacc = v0 /ag

(5.45)

sacc = v02 /2ag

(5.46)

where v0 is the set speed of unload feeding, mm/s, v0 ≤ 14.1 mm/s; ag is the acceleration in the starting of gripping, mm/s2 ; and sacc is the displacement of each finger in the unload acceleration stage The gripping mechanism of the end-effector, driven by the motor, overcomes the load torque and friction torque to accelerate the finger. The maximum acceleration in theory is determined by the following equation: ag0 = i a

Ma − Ma f × 106 Jar

(5.47)

where ag0 is the maximum starting acceleration of each finger in theory, mm/s2 ; ia is the transmission ratio from the gripper motor to the finger, mm/rad; M a is the rated torque of the gripper motor, mNm; M af is the friction torque in the gripping mechanism, mNm; and J ar is the equivalent moment of inertia of the gripping mechanism, gmm2 . In Eq. (5.47), M af is determined by the following equation: Ma f = γa M · Ia f /1000

(5.48)

where γ aM is the torque constant of the gripper motor, mNm/A and I af is the current of the gripper motor required to overcome the friction torque in the gripping mechanism, A. In Eq. (5.48), the equivalent moment of inertia of the gripping mechanism can be obtained from the following equation: Jar = Ja1 + Ja2

1 ωa3 2 + Ja3 ( ) + i a2 m f i a2 ωa1

(5.49)

274

5 Mathematical Modeling of Speedy Damage-Free Gripping of Fruit

where J a1 is moment of inertia of the gripper motor, g·mm2 ; J a2 is moment of inertia of the gripper gearbox, g mm2 ; ia is reduction ratio of the gripper gearbox; J a3 is moment of inertia of the bi-directional screw, g·mm2 ; ωa3 /ωa1 is angular speed ratio between the bi-directional screw and the gripper motor; and mf is mass of each finger, g. 2.

Unload constant-speed approaching stage

The two fingers close to the fruit at a set high speed of vg , the time spent in this stage is tg =

sg vg

(5.50)

where sg is finger displacement in this unload constant-speed approaching stage, mm.

5.9 Collision Stage Based on the collision force curve of speedy gripping (Fig. 5.2) and the complex three-phase collision model (Fig. 5.14), and at the same time taking the difference of the elastic stiffness of different fruit into consideration, the time consumption in collision stage after contact the fruit is approximately tdis ≈ t1 × (0.56 ∼ 1.43)

(5.51)

The minimum time consumption for damage-free gripping of tomato fruit is influenced by the factors such as the stiffness of tomato fruit, the parameters of the motor system, the inertia of each component of the end-effector, the friction of the transmission, and so on. It can be concluded that both lightening of the key components, such as the finger of the end-effector and the increase of the transmission precision, have an important effect on shortening the time consumption of reliable damage-free fruit gripping and improving the efficiency of the robotic harvesting.

References 1. Liu J, Bai X, Li P et al (2014) Complex collision model in high-speed gripping of fruit. Trans Chin Soc Agric Mach 45(4):49–54 2. Liu J (2010) Analysis and optimal control of vacuum suction system for tomato harvesting robot. Jiangsu University 3. Liu J, Bai X, Li P (2013) Modified Burger’s model for describing creep behavior of tomato fruits. Trans CSAE 29(9):249–255 4. Ayman AE et al (2012) Mathematical evaluation changes in rheological and mechanical properties of pears during storage under variable conditions. J Food Sci Eng 2(10):564–575

References 5. 6. 7. 8. 9. 10. 11. 12. 13. 14. 15. 16. 17.

275

Steffe J (1996) Rheological methods in food process engineering. Freeman Press Zhou Z (1994) Agricultural materials science. China Agricultural Press Li L (2001) Food physics. China Agricultural Press Jiang S, Feng F, Zhao J (2006) Study on creep characteristics and rheological model of carrot. Jiangsu Agric Sci 5:133–135 Li X, Zhu J, Wang W et al (1997) A study on creep properties and static loading mechanisms of apple. Acta Univ Agric Boreali-occidentalis 6:64–68 Wu L, Peng X (1990) Modified burgers model and its application to hardened concrete. J Chongqing Inst Archi Eng 1:41–46 Tia M, Liu Y, Haranki B et al (2009) Modulus of elasticity, creep and shrinkage of concrete– phase Ii: Part 1–creep study. University of Florida Wang F (2005) Rheology of wood materials. Northeast Forestry University Press Li C, Aydin A, Shi X et al (2008) Comparison and modification of rock creep models. J Exp Mech 23(1):9–16 Zhang Y, Huang X (2008) Viscoelastic mechanical model of permanent deformation of asphalt mixtures under repeated load. J Highw Transp Res Dev 25(4):1–6 Zhang J, Xu L, Wang B et al (2010) Study on the steering chracture of six-wheel robot vehicle. J Wuhan Univ Technol Transp Sci Eng 4(4):699–702 Tu K, Zhu W, Jiang S (2006) Food physics. Southeast University Press Chen K (1989) Food rheology and Its measurement. China Light Industry Press

Chapter 6

Simulation of Damage-Free Robotic Gripping of Fruit

6.1 Summary 6.1.1 Research Significance Simulation studies can visualize complex physical models and inner interaction relationships and effectively expand the space-time scope of research. Different from apple and other fruits, the mechanical properties of various components of tomato fruit are extremely different. Therefore, it is impossible to explain the mechanism of the damage in the local gripping contact area under overall loading to the fruit, if we treat the fruit as homogeneous. Meanwhile, the simulation study of the constrained inelastic collision between the electromechanical system with energy input and the viscous-elastic multi-component object is of great significance to solve the damage-free fruit harvesting under the rapid gripping collision.

6.1.2 Content and Innovation (1)

(2)

(3)

The nonlinear multi-component finite element model of tomato fruit was constructed, which laid a foundation for solving the virtual simulation of precise gripping collision of finite element-virtual prototype. The stress distribution in the fruit under the gripping action has been proved by the static stable gripping simulation of the viscous-elastic multi-component fruit based on finite element, and the difference between loading on the whole fruit and the local stress under the gripping contact is confirmed. All these results providing a basis for revealing the mechanism of gripping damage. Through the static stable gripping simulation of viscous-elastic multicomponent fruit based on finite element, the stresses of different tissues of exocarp, mesocarp, and locular gel were revealed. And then the strength of each

© Science Press, Beijing and Springer Nature Singapore Pte Ltd. 2021 J. Liu et al., Rapid Damage-Free Robotic Harvesting of Tomatoes, Springer Tracts in Mechanical Engineering, https://doi.org/10.1007/978-981-16-1284-8_6

277

278

(4)

6 Simulation of Damage-Free Robotic Gripping of Fruit

component is combined to explain the bearing capacity of the fruit for different gripping positions and fruit ripeness level and the mechanism of damage. The dynamic simulation of speedy gripping collision by combining finite element with virtual prototype was realized, the change trend of gripping force during the rapid gripping collision was then found, and the mathematical relationship between the peak force of collision and the gripping speed was established to provide a support for realizing speedy damage-free gripping of fruit.

6.2 Finite Element Model of Fruit 6.2.1 Viscoelastic Finite Element Model of the Whole Tomato Fruit [1] The process of gripping a tomato with a finger is actually a collision process with active powered motion. Both the motion control strategy of the gripping system and the mechanical properties of the tomato have a significant impact on the collision process. Therefore, to study the gripping of the tomato, it must determine a reasonable gripping strategy for the end-effector based on the mechanical properties of tomato fruit. To study the loading-deformation of tomato during the gripping process, the viscoelastic finite element model can be established by ANSYS software to analyze the stress and strain changes of tomato fruit during loading. 1.

Representation of viscoelastic properties in ANSYS

ANSYS software is one of the world’s most popular finite element analysis software. It can interface with most CAD software and mechanical simulation software to realize data sharing and exchanging. It is one of the most valuable CAE tools in product design [2]. The software mainly includes three parts: pre-processing module, analysis calculation module, and post-processing module. The pre-processing module provides a powerful solid modeling and meshing tool that allows users to easily construct finite element models. The ANSYS program can also analyze large 3D soft body motions. When the accumulation effect of motion plays a major role, these functions can be used to analyze the motion characteristics of complex structures in space and to determine the resulting stresses, strains, and deformations in the structure. Under load, the response of the viscoelastic material includes an elastic portion that responds immediately and a sticky portion that takes a period of time to manifest. In the ANSYS, the shear relaxation kernel function G(t) and the volume relaxation kernel function K(t) can be used to describe viscoelasticity. There are two main ways of representation. One is the Maxwell used by the generalized Maxwell unit

6.2 Finite Element Model of Fruit

279

(VISCO88 and VISCO89), another is the Prony series form used by structural units (VISCO185 and VISCO186). These two representations are essentially the same, except that the specific mathematical expressions are slightly different. (1)

Prony series form Prony

The basic form of viscoelastic properties expressed in Prony series is  t G(t) = G ∞ + G i exp − G τi i=1   n K  t K (t) = K ∞ + K i exp − K τi i=1 nG 



(6.1)

where G∞ and Gi are shear moduli, K ∞ and K i are bulk moduli, τiG and τiK are the relative time of each Prony series component, then define the following relative modulus:  αiG = G i G 0  αiK = K i K 0

(6.2)

where G0 and K 0 are the transient moduli of the viscoelastic material and are defined as follows: G 0 = G(t = 0) = G ∞ + K 0 = K (t = 0) = K ∞ +

nG  i=1 nK 

Gi Ki

(6.3)

i=1

In ANSYS, the orders nG and nK of the Prony series may not necessarily be the same, and of course the relaxation time τiG and do not have to be the same. For the viscoelastic problem, Poisson’s ratio of the viscoelastic body is generally taken as a function of time μ = μ(t). However, sometimes the situation can be approximated as a constant, which is based on the elastic constant relationship: E(t) 2(1 + μ) E(t) K (t) = 3(1 − 2μ) G(t) =

(6.4)

where E(t) is the relaxation modulus which can be determined by experiments. The corresponding coefficient ratios of E(t), G(t), and K(t) are the same.

280

6 Simulation of Damage-Free Robotic Gripping of Fruit

This makes it possible to unify G(t) and K(t) in the E(t) form. If we express the relaxation modulus as the Prony series form, i.e., n 

E(t) = E ∞ +

i=1

  t E i exp − τi

(6.5)

So, for G(t) and K(t), there are n = nG = nK , τi = τiG = τiK for relative time and αi = αiG = αiK for relative modulus. Similar to G0 and K0 , we also define the transient relaxation modulus E0 : E 0 = E(t = 0) = E ∞ +

nG 

Ei

(6.6)

i=1

Therefore, equations as follows can be obtained from (6.3) to (6.6): E0 2(1 + μ) E0 K0 = 3(1 − 2μ) G0 =

(6.7)

There are three options for Shift function: (a) William–Landel, Ferry: timetemperature equivalent equation for polymers, (b) Tool-Narayanaswamy equation, (c) user-defined. When using PRONY simulation, Shift Function is not necessarily input. If the relaxation modulus E(t) is not related to temperature, you do not need to input shift function. (2)

Generalized Maxwell representation

The expression of viscoelastic properties in ANSYS can be input using generalized Maxwell units. The Maxwell model is a series of a spring and a damper. The generalized Maxwell model consists of k parallel springs and dampers. The generalized Maxwell model is used in ANSYS to represent viscoelastic behavior, which is a general-purpose model. Maxwell and Kelvin–Voigt are special cases. Maxwell, which characterizes viscoelastic properties in ANSYS, can only use VISCO88 (2D) and VISCO89 (3D) unit types. The Maxwell model characterizes the viscoelasticity of an object mainly by the following two kernel functions:   t G i exp − G τi i=1   n K  t K (t) = K ∞ + K i exp − K τi i=1 G(t) = G ∞ +

nG 

(6.8)

6.2 Finite Element Model of Fruit Table 6.1 Input parameters for Maxwell

281

Real constant

Bulk modulus

Shear modulus

N

C50

C71

G0

C46

C48

G∞

C47

C49

Ci

C51–C60

C76–C85

τi

C61–C70

C86–C95

where N is the number of Maxwell units, N ≤ 10, G0 is the initial shear/bulk modulus (i.e. solid phase), G0 or G∞ is the final shear/bulk modulus (i.e. liquid phase), C i is the relaxation coefficient of shear/bulk modulus, C i = 1.0, the constant τ i is the relaxation time

In the formula, each symbol has the same meaning as above, and τ i is the relaxation time of each component. The relative modulus of each component is defined in Eq. (6.9), which is slightly different from the Prony input definition.  CiG = G i (G 0 − G ∞ )  CiK = K i (K 0 − K ∞ )

(6.9)

The Maxwell input enters the viscoelastic property data by defining a constant 1–95. The usual parameters that need to be entered and their meanings are shown in Table 6.1. 2. (1)

Effect of ripeness on viscoelastic parameters of tomato fruit Burger’s model parameters and their relationship with ripeness

With the increase of ripeness, the instantaneous elastic coefficient E 1 decreases significantly, indicating that the instantaneous elastic deformation of tomato fruit increases significantly with the increase of ripeness. In the harvesting and packaging operations, the instant when tomato is loaded, the “Breakers” stage has a stronger ability of resistance to deformation than the tomato with higher ripeness level. The deformation can be restored after the load is removed. The delayed elastic coefficient E2 and the viscosity coefficient η2 associated with the high elastic deformation of the tomato are significantly reduced as the ripeness is improved. It indicates that the delayed deformation of tomato fruit produced by Kelvin model increases with the increase of ripeness level. That is, as the working time goes, the higher the ripeness level, the greater the deformation of the tomato. And this part of the deformation can be gradually recovered over time after the load is removed as well. As a typical viscoelastic body, the viscosity coefficient η1 of tomato after loading characterizes the ability of tomato to resist irreversible deformation after being

282

6 Simulation of Damage-Free Robotic Gripping of Fruit

loaded. It can be seen from Table 5.1 in Chap. 5 that the viscosity coefficient decreases significantly with the increase of ripeness level, indicating that the higher the ripeness level, the more susceptible the tomato is to plastic deformation and damage under the same load condition. The elastic lag time τ K is the time required for the strain to reach the final strain (1−1/e), and the speed of the elastic lag is characterized. There is no significant difference in elastic lag time τK , indicating that the ratio of viscosity coefficient η2 and delayed elastic coefficient E 2 of tomato fruit viscosity is not significantly changed, that is, with the increase of ripeness, the viscosity coefficient and the delay elastic coefficient are basically consistent. (2)

Maxwell model parameters and their relationship with ripeness

It can be seen from Table 5.2 in Chap. 5 that with the increase of ripeness, the equilibrium elastic coefficient E 0 decreases significantly, indicating that the stress relaxation deformation of tomato fruit increases significantly with the increase of ripeness. In the harvesting and packaging operations, the instant when tomato is loaded, and the “Breakers” stage has a stronger ability of resistance to deformation than the tomato with higher ripeness level. The elastic coefficients E 1 and E 2 of Maxwell’s unit decrease significantly with the increase of tomato ripeness level, indicating that with the increase of ripeness level, the tomato’s ability to resist elastic deformation gradually decreases, and the higher the ripeness level, the more susceptible the tomato to elastic deformation. The elastic modulus of each moment in the relaxation process is equal to the sum of the elastic moduli of a plurality of different frequency components, and thus the tomato with a higher ripeness level has a larger value of each instantaneous elastic modulus. The viscosity coefficients η1 and η2 decreases significantly during the “Breakers” stage to the “Turning” stage, but did not change significantly during the discoloration stage to the “Pink” stage and the “Light red” stage. It is indicated that the change of structure and composition of tomato from the “Breakers” stage to the “Turning” stage has a significant influence on the viscosity characteristics of stress relaxation characteristics. The change of viscosity coefficient with ripeness level indicates the degree of plastic deformation and damage of tomato under compressibility. The time required for the stress to completely disappear is very long. To express the stress relaxation rate, τ M is defined as the stress relaxation time. When t = τ M , σ = σ 0 e−1 , that is, the stress relaxation time is the time required the stress relaxation to 1/e of the initial value. The stress relaxation time τ M of the two Maxwell units decreases sharply during the transition from the “Breakers” stage to the “Turning” stage, but there was no significant change in the subsequent maturation process. The ratio of the viscosity coefficient ηM of tomato fruit to the delayed elastic modulus E M shows a significant change during the period of the transition from the “Breakers” stage to the “Turning” stage.

6.2 Finite Element Model of Fruit

3.

283

Establishment of a viscoelastic finite element model for tomato

The ultimate goal of finite element analysis is to restore the mathematical behavioral characteristics of an actual engineering system, that is, the analysis must be accurate for a physical prototype. The model in ANSYS generally refers to the process of generating the spatial domain and the actual system connection represented by nodes and units. Model generation includes geometric modeling of nodes and elements, material properties, real constants, and meshing. (1)

Geometric model of the whole tomato fruit

Tomatoes vary in size and shape, but tomatoes of the same variety and similar planting conditions are not significantly different. In order to more accurately simulate the mechanical gripping process of tomato, it is necessary to establish a finite element model of tomato close to the real thing. The real tomato geometry model was established by Li [3]. Firstly, the peripheral contour geometric curve of tomato whole fruit is obtained by depicting the drawing, then the coordinate system is used to extract the key points, the outer contour is fitted with Spline curve, and the tomato solid model is established by the surface operation and the body operation. The specific steps are (1)

(2)

(3)

The connecting line of tomato fruit umbilical to fruit stem is placed perpendicular to the test stand, cut with a knife along the longitudinal section of the tomato, and half of the section is placed down on the preset white paper 1-copy paper-white paper 2 with pencil depicting the profile curve of the section. Remove the tomato, white paper 1 and copy paper, and establish a coordinate system on the white paper 2. The origin is the position of the tomato fruit umbilical, the Y-direction is the line along the fruit umbilical to stem, the X is perpendicular to the Y-right, and the Z-direction is perpendicular to the direction of X and Y. Select the key points on the contour curve and extract the coordinates of the key points as shown in Table 6.2.

In ANSYS, the key points 1–21 are first generated, and then the Spline curve is generated to create a quarter cross section of the tomato, and then the geometric model of the tomato generated by rotation is shown in Fig. 6.1. (2)

Viscoelastic property setting

The whole tomato fruit is a typical viscoelastic body. The results of the corresponding test in the second chapter can be used to obtain the parameters of the tomato in the “Pink” stage (6.10). E(t) = E ∞ + E 1 e E(t) = 2.62929 + 5.44662 e

t − 1.83635

− τt

1

− τt

+ E2 e

+ 0.82280 e

2

t − 35.0209

(6.10)

284

6 Simulation of Damage-Free Robotic Gripping of Fruit

Table 6.2 The key points of tomato fruit outline No.

x (mm)

z(mm)

Numbers

x (mm)

y (mm)

z (mm)

1

0

y (mm) 0

0

12

37.5

44

0

2

7.5

1

0

13

36

50

0

3

12

2

0

14

33.5

55.5

0

4

18

5

0

15

28

60

0

5

22.5

8

0

16

23

62

0

6

26.5

12

0

17

17

62

0

7

30

17

0

18

10

61

0

8

33

22

0

19

6

58

0

9

35

26

0

20

3

56

0

10

37

32

0

21

0

55.5

0

11

38

39

0

Fig. 6.1 Geometric model of tomato fruit

By formula (6.4), Poisson’s ratio of tomato is 0.35, and the parameters of tomato shear modulus and bulk modulus are obtained as in Eq. (6.11). E0 t t = 0.97381 + 2.01727 e− 1.83635 + 0.30474 e− 35.0209 2(1 + μ) E0 t t = 2.92143 + 6.05180 e− 1.83635 + 0.91422 e− 35.0209 K0 = 3(1 − 2μ) G0 =

(6.11)

The unit properties of the material were defined for the established tomato geometric model and the viscoelastic VISCO89 generalized Maxwell unit (Fig. 6.2) was determined as the basic unit of tomato viscoelasticity. According to the definition of viscoelastic properties in ANSYS, the parameters of Maxwell are determined as shown in Table 6.3.

6.2 Finite Element Model of Fruit

285

Fig. 6.2 Viscoelastic unit type and attribute settings

(3)

Meshing

The finite element model is a mathematical representation of the actual structure and matter. ANSYS meshes the established solid model to produce a finite element model. Before the division, the meshing level needs to be set. The accuracy of the meshing determines the quality of the finite element calculation. In general, the finer the meshing, the more accurate the calculations, but the more computer resources

286

6 Simulation of Damage-Free Robotic Gripping of Fruit

Table 6.3 Parameters of Maxwell Parameter

Shear modulus

Bulk modulus

Number

Data

Number

Data

N

C50

2

C71

2

G0

C46

3295820

C48

9887460

G∞

C47

973810

C49

2921430

α1

C51

0.86876

C76

0.86876

α2

C52

0.13124

C77

0.13124

t1

C61

1.83635

C86

1.83635

t2

C62

35.0209

C87

35.0209

are spent. The maximum equatorial plane diameter of the tomato was 76 mm and the maximum height was 62 mm. In terms of the accuracy and calculation speed factors, determined the unit size to be 4 mm. The tomato solid model was freely meshed, and the total number of generated units was 27218 and the total number of nodes was 33891. The division result was shown in Fig. 6.3. (4)

Create a generic data file

In the simulation of tomato harvesting robot picking and gripping tomato, the established viscoelastic finite element tomato model needs to be imported into the dynamic simulation software ADAMS, so it is necessary to establish two software data exchange files. ANSYS and ADAMS can exchange data through modal neutral files, i.e., MNF (Model Neutral File) files. The main method of establishing the modal neutral file of the tomato finite element model is Main Menu-Solution-ADAMS Connection-Export

Fig. 6.3 Meshing of tomato viscoelastic finite element model

6.2 Finite Element Model of Fruit

287

to ADAMS, select the appropriate external action point, i.e., Marker point, and make the following settings (Fig. 6.4). Then the modal neutral files can be established. In the first 20 modes, the frequency values of the modes of the modal neutral files converted by the tomato viscoelastic finite element model are shown in Table 6.4.

Fig. 6.4 Establishment of mnf file

Table 6.4 Frequency values of the modes from the modal neutral files Mode

Frequency/Hz

Mode

Frequency/Hz

1

0.142

11

649.444

2

37.862

12

651.372

3

52.751

13

667.109

4

63.231

14

706.956

5

83.793

15

707.998

6

95.236

16

709.980

7

525.662

17

792.016

8

531.734

18

874.280

9

642.491

19

879.224

10

645.757

20

905.411

288

6 Simulation of Damage-Free Robotic Gripping of Fruit

6.2.2 Nonlinear Multi-component Finite Element Model of Tomato Fruit [3] 1.

Definition of material properties and unit types

The solid geometric model of tomato consists of three parts: exocarp, mesocarp, and locular gel. The properties such as elastic modulus, Poisson’s ratio, stress intensity, and density of the material of the three parts of the model are determined by the corresponding test results in Chap. 3 (Table 6.5). In Table 6.5, the FEM (E ave ) refers to finite element model used the mean value of elastic modulus of each tissue, the FEM (E max ) refers to finite element model used the maximum elastic modulus value of each tissue, and the FEM (E min ) refers to finite element model used the minimum elastic modulus value of each tissue. 2.

Meshing

The grasping of the tomato by the robot finger belongs to the contact problem between the rigid body and the soft body. It is not necessary to mesh the rigid body before establishing the rigid body-soft body contact pair. In order to mesh the soft body (tomato), the cell types of the three materials of the exocarp, the mesocarp, and the locular gel must be determined first. For different purposes, ANSYS cell library offers up to 200 different units to choose from. In this study, the structural unit Solid95 was selected as the unit type of the tomato exocarp tissue, and the structural unit Solid92 was selected as the unit type of the tomato mesocarp and the locular gel tissue. The Solid92 unit is defined by 10 nodes, and the Solid95 unit is defined by 20 nodes. Each node has three degrees of freedom: displacement in the X-, Y-, and Z-directions of the Table 6.5 Mechanical properties of Fenguan 906 tomato tissues Model

Tissues

Number

FEA (E avg )

Exocarp

1

FEA (E max )

FEA (E min )

Elastic modulus (MPa) 9.59

Poisson’s ratio

Stain strength (MPa)

Density (kg/mm3 )

Unit type

0.49

0.582

1000 × 10−9

Solid95

10−9

Solid92

Mesocarp

2

0.726

0.45

0.122

1070 ×

Locular gel

3

0.124

0.45

0.012

1010 × 10−9

Solid92

Exocarp

11

11.776

0.49

0.61

1000 × 10−9

Solid95

Mesocarp

12

0.868

0.45

0.152

1070 × 10−9

Solid92

10−9

Solid92

Locular gel

13

0.198

0.45

0.018

1010 ×

Exocarp

21

7.404

0.49

0.554

1000 × 10−9

Solid95

10−9

Solid92 Solid92

Mesocarp

22

0.584

0.45

0.092

1070 ×

Locular gel

23

0.05

0.45

0.006

1010 × 10−9

6.2 Finite Element Model of Fruit

289

node, and the unit has the capabilities of plasticity, creep, stress strengthening, large deformation, and large strain. The Solid92 unit has quadratic displacement and can better divide irregular grids. Solid95 units can have any spatial orientation, accept irregular shapes without reducing accuracy, and have a coordinated displacement function that is more suitable for simulating curve boundaries [5]. Therefore, Solid92 and Solid95 units are more suitable for simulating biomass materials with irregular shapes. The element edge length of tomato exocarp and mesocarp tissues were set to 2 mm, and the element edge length of locular gel tissue was set to 1 mm. The geometric models of tomato exocarp, mesocarp, and locular gel tissues were free meshed using tetrahedral elements. After the meshing operation, the exocarp part of Model 1 and Model 2 contains 2649 nodes and 4078 units, the mesocarp part contains 14118 nodes and 12849 units, the locular gel part contains 2489 nodes and 2475 units. The exocarp part of Model 3 and Model 4 contains 2914 Node and 4514 units, the mesocarp part contains 15809 nodes and 14630 units, the locular gel part contains 2878 nodes and 2679 units. The exocarp part of Model 5 and Model 6 contains 2971 nodes and 4600 units, the mesocarp part contains 16382 Node and 14573 units, and the locular gel part contains 3886 nodes and 3705 units.

6.3 Simulation of Static Gripping Process [3] 6.3.1 Geometry Model Finger-Fruit Contacting Process Integrating the design idea of the robotic grabbing mechanism in the above literature, we independently developed the finger picking mechanism of the tomato harvesting robot [6, 7]. There are two types of pointing surfaces, one is flat and the other is curved surface. The motor is gradually closed by dragging the two parallel fingers of the robot through the screw drive mechanism, so that the gripping force is applied to the tomato. The geometric dimensions of the flat fingers are 40 × 46 × 3 mm (length × width × thickness), and the geometrical dimensions of the curved fingers are: 54 × 45 × 3 mm (length × width × thickness). To simplify the model, the flat and curved fingers are reduced to the same size flat plate and curved thin plate (arc plate) rigid body. The initial contact position of the robot finger on the tomato is located on the equatorial plane of the tomato. Therefore, when establishing the geometric model of the finger, the maximum abscissa point and the minimum abscissa point of the tomato exocarp contour on the XOY plane (the opposite of the maximum abscissa) are taken as the initial point of contact between the finger and the tomato. In order to study the influence of finger type and grasping position on mechanical damage, six 3D solid grab models (Fig. 6.5) were established in this paper. The meanings of corresponding sub-icons are shown in Table 6.6. Considering the symmetry of the structure of the tomato solid model, in order to reduce the time and storage required

290

6 Simulation of Damage-Free Robotic Gripping of Fruit

Fig. 6.5 Grasp the model

Table 6.6 Meaning of each model icons Model

Corresponding Figure

Meaning

Model1

Figure 6.5a

The flat finger is grabbed from the position above the radial wall of the three-locular tomato (left side) or directly above the locule (right side)

Model 2

Figure 6.5b

The arc finger is grabbed from the radial wall (left side) or the locule (right side) of the three-locular tomato

Model 3

Figure 6.5c

The flat finger is grabbed from the radial wall of the four-locular tomato

Model 4

Figure 6.5d

The arc finger is grabbed from the radial wall of the four-locular tomato

Model 5

Figure 6.5e

The flat finger grabbing from the locule of the four-locular tomato

Model 6

Figure 6.5f

The arc finger is grabbed from the locule of the four-locular tomato

for computer analysis, the 1/4 body of the three-locular tomato geometric model and the 1/4 body of the four-locular tomato geometric model were selected.

6.3.2 Creating Contact Pair When the tomato fruit is compressed by the probe, the stiffness of the probe is much larger than that of exocarp tissue. Therefore, the probe was defined as a rigid body, while the exocarp tissue was defined as a flexible body. The rigid target surface was specified using the areas operation on the surface of the probe model, and the target type was defined using the “rigid w/pilot node” option. The flexible contact

6.3 Simulation of Static Gripping Process [3]

291

surface was specified using the areas operation on the external surface of the exocarp sub-model, and the contact type was defined as the “surface-to-surface” option. The TARGE170 element is used to represent various 3D target surfaces for the associated contact elements, and the CONTA174 element is a 3D, 8-node, higher order quadrilateral element that can be located on the surface of the 3D SOLID95 solid element with mid-side nodes. Thus, TARGE170 and CONTA174 were defined as the target surface element and contact surface element types, respectively. Some parameters were defined as follows: coefficient of friction, 0.375, contact algorithm, augmented Lagrange method, normal penalty stiffness, 1, contact detection, on gauss points, penetration tolerance, 0.1, automatic contact adjustment, close gap/reduce penetration. After the contact pair was created, the motion of the rigid target surface of the probe model was controlled using the defined pilot node.

6.3.3 Model Verification Method When the fingers clamp the tomato, as the compressibility increases, the cellular tissue of each component of the tomato begins to be destroyed, and then the internal structure of the tomato ruptures. The crack begins to cross the fruit between a fruit shoulder and another adjacent shoulder. The line expands and after the crack increases, the gel of the tomato gradually flows out of the crack. From the stable gripping test, it can be seen that when the grasping force exerted by the mechanical finger on the tomato reaches the grasping force required for stable grasping, the outside of the tomato does not crack, but internal mechanical damage may occur depending on the number of gripping force [8]. In order to accurately predict the mechanical damage of the inside of the tomato, this part verifies the correctness of the finite element model of the three-locular and four-locular tomato. The verification method is to use the rigid plate to carry out the whole fruit compression simulation test on the finite element model, and then compare it with the whole tomato fruit compression test data of the third chapter. The factors of the simulation test include: (1) (2) (3)

Two loading positions: the locule and the radial wall of the tomato. Two structural types: three-locular tomato and four-locular tomato. Four compressibility levels: 4%, 8%, 12%, and 16%. From the test results in Sect. 3.2, it can be seen that when the flat probes were compressed from the radial wall and the locule of the three-locular tomato, respectively, and the compressibility reached 15.67% and 16.23%, respectively, the three-locular tomato begins to crack. And when the compressibility reached 14.31% and 17.85% for radial wall and locule position of four-locular tomato, respectively, it began to crack.

Because this section mainly studies the internal mechanical damage of tomato before rupturing under external load, the simulation test only took the above four

292

6 Simulation of Damage-Free Robotic Gripping of Fruit

compressibility levels. In order to consider the fluctuations of the mechanical parameters of the various components of the tomato—the exocarp, the mesocarp, and the locule gel material, each model was simulated by the average elastic modulus, the maximum elastic modulus, and the minimum elastic modulus. Finally, the error between the predicted value and the actual value was calculated. The mechanical parameters of each model were set as shown in Tables 6.7, 6.8, 6.9 and 6.10. The maximum modulus of elasticity of the material was equal to the sum of the average modulus of elasticity of the material and its standard deviation, and the minimum modulus of elasticity was equal to the difference between the mean modulus of elasticity and its standard deviation. The analysis type of ANSYS used static large deformation analysis.

6.3.4 Prediction Method of Gripping Damage After the finite element model of the tomato was verified, the model was used to predict the internal mechanical damage of the tomato during the grabbing process. When a parallel robot finger grabs a tomato, a pair of opposite normal forces is applied to the left and right contact points of the finger and the tomato. If the normal force is applied to the contact point from the horizontal direction (Fig. 6.6), the finger must grasp the following formula when the tomato does not slide. 

f ≤ μN 2f = G

(6.12)

G 2μ

(6.13)

Simplified: N≥

where f —the friction between robot finger and tomato (N); μ—the static friction coefficient between robot finger and tomato; N—the grab force (normal force) applied by the robot finger to the tomato (N); G—weight of the fruit (N). According to the physical properties of tomato measured in the third chapter, the mass of the three-locular tomato is 130.69–188.59 g, and the mass of the four-locular tomato is 109.46–181.3 g. As can be seen from the literature reviewed, the main types of robot finger face materials are stainless steel, lacquered stainless steel, and rubber. According to the mechanical properties of the tomato measured in the third chapter, the average static friction coefficients of the three-locular tomato and the 307 stainless steel, lacquered stainless steel and rubber materials are 0.375, 0.408, and

0.124

7.28

5.28

9.59

FEA(E avg )

0.05

0.584

7.404

FEA(E min )

0.726

8.36 8.84

FEA(E max ) 11.776

0.198

0.868

Test data

12.9

36.8

-5.7

Error (%)

Compressibility-4% (2.44 mm)

E Exocarp E Mesocarp E Gel (MPa) Force (MPa) (MPa) (N)

Parameters

19.48

14.20

24.24

21.36

Force (N)

8.8

33.5

−13.4

Error (%)

Compressibility-8% (4.89 mm)

35.26

25.92

43.6

33.56

Force (N)

38.62 53.6

22.77

66.52

43.68

Force (N)

−22.71

11.58

−52.29

Error (%)

Compressibility-16% (9.77 mm)

−5.07

−29.92

Error (%)

Compressibility-12% (7.33 mm)

Table 6.7 Test and predicted value for locule position of three-locular tomato from Fenguan906 (flat)

12.37

26.18

25.36

Average error (%)

6.3 Simulation of Static Gripping Process [3] 293

0.124

7.28

5.28

9.59

FEA(E avg )

0.05

0.584

7.404

FEA(E min )

0.726

9.04 8.9

FEA(E max ) 11.776

0.198

0.868

Test data

19.47

41.59

1.55

Error (%)

Compressibility-4% (2.44 mm)

E Exocarp E Mesocarp E Gel (MPa) Force (N) (MPa) (MPa)

Parameters

19.44

14.22

24.24

22.88

Force (N)

15.04

37.85

−5.94

Error (%)

Compressibility-8% (4.89 mm)

35.06

25.74

43.64

35.54

Force (N)

1.35

27.57

−22.79

Error (%)

Compressibility-12% (7.33 mm)

Table 6.8 Test and predicted value for radial wall position of three-locular tomato from Fenguan906 (flat)

53.56

39.32

66.14

46.48

Force (N)

30.61 12.77

15.40

18.15

Average error (%)

−15.23

−42.29

Error (%)

Compressibility-16% (9.77 mm)

294 6 Simulation of Damage-Free Robotic Gripping of Fruit

0.124

6.34

4.58

9.59

FEA(E avg )

0.05

0.584

7.404

FEA(E min )

0.726

7.10 7.92

FEA(E max ) 11.776

0.198

0.868

Test data

10.7

35.49

−11.55

Error (%)

Compressibility-4% (2.44 mm)

E Exocarp E Mesocarp E Gel (MPa) Force (N) (MPa) (MPa)

Parameters

17.26

12.36

21.82

17.74

Force (N)

2.71

30.33

−22.99

Error (%)

Compressibility-8% (4.89 mm)

30.54

21.72

38.96

34.86

Force (N)

12.39

37.69

−11.76

Error (%)

Compressibility-12% (7.33 mm)

Table 6.9 Test and predicted value for locule position of four-locular tomato from Fenguan906 (flat)

46.94

33.96

59.6

43.48

Force (N)

31.35 8.44

21.9

20.85

Average error (%)

−7.96

−37.07

Error (%)

Compressibility-16% (9.77 mm)

6.3 Simulation of Static Gripping Process [3] 295

0.124

7.84

6.08

9.59

FEA(E avg )

0.05

0.584

7.404

FEA(E min )

0.726

7.18 9.60

FEA(E max ) 11.776

0.198

0.868

Test data 17.32 22.36

15.32

28.28

18.96

Force (N)

−17.93

8.65

−49.16

Error (%)

Compressibility-8% (4.89 mm)

−9.19

−33.70

Error (%)

Compressibility-4% (2.44 mm)

E Exocarp E Mesocarp E Gel (MPa) Force (MPa) (MPa) (N)

Parameters

42.18

32.84

53.40

44.28

Force (N)

4.74

25.84

−20.6

Error (%)

Compressibility-12% (7.33 mm)

Table 6.10 Test and predicted value for radial wall position of four-locular tomato from Fenguan906 (flat)

63.18

48.14

76.20

54.84

Force (N)

−15.21

12.22

−38.95

Error (%)

Compressibility-16% (9.77 mm)

11.77

15.51

35.60

Average error (%)

296 6 Simulation of Damage-Free Robotic Gripping of Fruit

6.3 Simulation of Static Gripping Process [3]

297

Fig. 6.6 Contact analysis

0.396, respectively, four-locular tomato and 307 stainless steel, lacquered stainless steel and rubber are 0.488, 0.641, and 0.503, respectively. Therefore, according to formula (6.13), when the robot fingers of 307 stainless steel, lacquered stainless steel, and rubber materials are used to grab the three-locular tomato respectively, the minimum stable gripping force required is 1.74–2.51 N, 1.60–2.31 N, and 1.65– 2.38 N, respectively. When the robot fingers of 307 stainless steel, lacquered stainless steel, and rubber materials were used to grab the four-locular tomato respectively, the minimum stable gripping force required was 1.12–1.86 N, 0.85–1.41 N, and 1.09–1.80 N, respectively. In order to predict the internal mechanical damage of the tomato during the grasping process, a simulation test of the robot finger grasping the tomato was designed according to the above test data. Test design factors include: (1) (2) (3) (4)

Two loading positions: the radial wall tissue and the locule tissue of tomato. Two structural types: three-locular tomato and four-locular tomato. Two types of finger faces: flat plates and arc plates. Five loading forces (grabbing force): 1 N, 10 N, 19 N, 28 N, 37 N.

All gripping force is applied to the contact surface of the tomato through the control node. After the simulation test, the stress and strain changes inside the tomato were observed through a specific section. There are six kinds of loading methods in the simulation test: (1)

(2) (3) (4)

(5)

Flat plate-locule (radial wall)-three-locular tomato, that is, using a rigid plate to load and compress from the locular or radial wall position of the three-locular tomato. Flat plate-locule-four-locular tomato, that is, using a rigid plate to load and compress from the locular position of the four-locular tomato. Flat plate-radial wall-four-locular tomato, that is, using a rigid plate to load and compress from the radial wall position of the four-locular tomato. Arc plate-locule (radial wall)-three-locular tomato, that is, using a rigid arc plate to load and compress from the locular or the arm position of the threelocular tomato. Arc plate-locule-four-locular tomato, that is, using a rigid arc plate to load and compress from the locular position of the four-locular tomato.

298

6 Simulation of Damage-Free Robotic Gripping of Fruit

Fig. 6.7 Initial state before the finite element model is loaded

(6)

Arc plate-radial wall-four-locular tomato, that is, using a rigid arc plate to load and compress from the radial wall position of the four-locular tomato. Figure 6.7 is the initial state of the finite element model in simulation test 1 (loading mode 1) after the constraint and force information are applied.

6.3.5 The Component Stress Simulation of Different Loading Methods 1.

Finite element model validation

The results of the real test and the simulation test are shown in Tables 6.7, 6.8, 6.9 and 6.10. EExocarp indicates the elastic modulus of tomato exocarp tissue, E Mesocarp indicates the elastic modulus of tomato mesocarp tissue, and EGel indicates the elastic modulus of tomato gel tissue. The parameter obtained by the real compression test is the experimental value, and the parameter obtained by the finite element model simulation test is the predicted value. The error term in the table indicates the relative error between the experimental value and the predicted value at the specified compressibility. The average error term indicates the average of the absolute values of the four error terms under the specified finite element model. Table 6.7 shows the experimental values and predicted values of the flat plate after compressing the three-locular tomato from the locule position. When the compressibility is 4%, the relative error between the experimental value and the predicted value is the smallest under the FEA (E max ) model, and the relative error between the experimental value and the predicted value is the largest under the FEA (E min ) model. When the compressibility is 8%, the relative error between the experimental value and the predicted value is the smallest under the FEA (E avg ) model, and the relative error between the experimental value and the predicted value is the largest under the FEA (E min ) model. When the compressibility is 12%, the relative error between the experimental value and the predicted value is the smallest under the FEA (E avg ) model, and the relative error between the experimental value and the predicted value is the largest under the FEA (E max ) model. When the compressibility is 16%, the

6.3 Simulation of Static Gripping Process [3]

299

relative error between the experimental value and the predicted value is the smallest under the FEA (E min ) model, and the relative error between the experimental value and the predicted value is the largest under the FEA (E max ) model. From the average error point of view, the average relative error between the experimental value and the predicted value in the FEA (E avg ) model is the smallest, 12.37%, in the FEA (E min ) model, the average relative error between the experimental value and the predicted value is the largest, it is 26.18%. Table 6.8 shows the experimental values and predicted values of the flat plate after compressing the three-locular tomato from the position of the radial wall. When the compressibility is 4% and 8%, respectively, the relative error between the experimental value and the predicted value is the smallest under the FEA (E max ) model, and it is the largest under the FEA (E min ) model. When the compressibility is 12%, the relative error between the experimental value and the predicted value is the smallest under the FEA (E avg ) model, and it is the largest under the FEA (E min ) model. When the compressibility is 16%, the relative error between the experimental value and the predicted value is the smallest under the FEA (E avg ) model, and it is the largest under the FEA (E max ) model. From the average error point of view, the average relative error between the experimental value and the predicted value in the FEA (E avg ) model is the smallest, which is 12.77%, in the FEA (E min ) model, the average relative error between the experimental value and the predicted value is the largest, which is 30.61%. Table 6.9 shows the experimental values and predicted values of the flat plate after compressing the four-locular tomato from the locule position. When the compressibility is 4% and 8%, respectively, the relative error between the experimental value and the predicted value is the smallest under the FEA (E avg ) model, and it is the largest under the FEA (E min ) model. When the compressibility is 12%, the relative error between the experimental value and the predicted value is the smallest under the FEA (E max ) model, and it is the largest under the FEA (E min ) model. When the compressibility is 16%, the relative error between the experimental value and the predicted value is the smallest under the FEA (E avg ) model, and it is the largest under the FEA (E max ) model. According to the average error, the average relative error between the experimental value and the predicted value in the FEA (E avg ) model is the smallest, which is 8.44%, in the FEA (E min ) model, the average relative error between the experimental value and the predicted value is the largest, which is 31.35%. Table 6.10 shows the experimental values and predicted values of the flat plate after compressing the four-locular tomato from the position of the radial wall. When the compressibility is 4%, the relative error between the experimental value and the predicted value is the smallest under the FEA (E avg ) model, and it is the largest under the FEA (E max ) model. When the compressibility is 12%, the relative error between the experimental value and the predicted value is the smallest under the FEA (E avg ) model, and it is the largest under the FEA (E max ) model. When the compressibility is 8% and 16%, respectively, the relative error between the experimental value and the predicted value is the smallest under the FEA (E min ) model, and it is the largest under the FEA (E max ) model. From the average error point of view, the average

300

6 Simulation of Damage-Free Robotic Gripping of Fruit

relative error between the experimental value and the predicted value in the FEA (E avg ) model is the smallest, which is 11.77%, in the FEA (E max ) model, the average relative error between the experimental value and the predicted value is the largest, which is 35.6%. In summary, the average relative error between the experimental and predicted values of the FEA (E avg ) model under the four loading conditions is the smallest, 12.37%, 12.77%, 8.44%, and 11.77%, respectively. Therefore, in the static large strain analysis of ANSYS, the finite element model corresponding to the elastic modulus of tomato exocarp, mesocarp, and locule gel, respectively, is suitable for predicting the mechanical damage of tomato. The finite element analysis method has been successfully applied to the internal mechanical damage prediction of watermelon, and the average relative error between the experimental value and the predicted value of the model is 7% and 11%, respectively. The reason for the error may be. (1)

(2) (3) (4) 2.

The exocarp, mesocarp, and locule gel of tomato are viscoelastic materials, which are simplified into linear elastic materials in the finite element model, ignoring the nonlinearity of the material. The difference between the simplified geometry and size of the tomato finite element model and the actual shape and size of the tomato. Small differences between the actual test loading position and the simulated test loading position. There are minor errors in the mechanical properties of tomato exocarp, mesocarp, and locule gel. Relationship between loading position and mechanical damage of tomato

The stress distribution results of tomato components under different loading positions after the simulation test are shown in Tables 6.11, 6.12 and 6.13. F1 , F2 , F3 , F4 , and F5 represent the first to fifth loading forces of the simulation test, respectively. σ max and σ min represent the maximum von Mises stress and the minimum von Mises stress of the material, respectively. Dmax represents the maximum integrated displacement of the material. εT represents the compressibility of the plate to tomato, which is the percentage of the ratio of the plate motion displacement to the equatorial plane diameter of the tomato. (1)

Loading mode 1: plate-locule (radial wall)-three-locular tomato

Table 6.11 shows the test results after loading from the locule (radial wall) position of the three-locular tomato using a flat plate, that is, the test result of the loading mode 1. It can be clearly seen from the table that with the gradual increase of loading force (F 1 → F 5 ), the maximum von Mises stress, minimum von Mises stress, and maximum comprehensive displacement of tomato exocarp, mesocarp, and locule gel tissue increase nonlinearly. For the same loading force, the maximum von Mises stress and the minimum von Mises stress of the tomato exocarp were larger than those of the tomato mesocarp, respectively. The maximum von Mises stress and the

6.3 Simulation of Static Gripping Process [3]

301

Table 6.11 Experimental value of loading mode 1 Component

Exocarp

Mesocarp

Locule gel

Mechanical properties

F 1 (1 N)

F 2 (10 N)

F 3 (19 N)

F 4 (28 N)

F 5 (37 N)

εT = 0.87%

εT = 5.11%

εT = 7.95%

εT = 10.38%

εT = 12.51%

σ max (MPa)

0.292

0.459

0.541

0.612

0.615

σ min (MPa)

2.32 × 10−3

2.32 × 10−2

4.52 × 10−2

6.73 × 10−2

8.53 × 10−2

Dmax (mm)

0.528

3.083

4.816

6.3

7.6

0.127

0.129

0.132

10−2

σ max (MPa)

3.1 × 10−2

8.7 ×

σ min (MPa)

1.9 × 10−4

1.88 × 10−3

3.67 × 10−3

5.48 × 10−3

7.23 × 10−3

Dmax (mm)

0.528

3.083

4.816

6.3

7.6

σ max (MPa)

2.13 × 10−3

1.22 × 10−2

1.24 × 10−2

1.25 × 10−2

1.26 × 10−2

σ min (MPa)

6.36 × 10−5

6.46 × 10−4

1.25 × 10−3

1.89 × 10−3

2.53 × 10−3

Dmax (mm)

0.354

2.524

4.155

5.598

6.868

Table 6.12 Experimental value of loading mode 2 Component

Exocarp

Mesocarp

Locule gel

Mechanical properties

F 1 (1 N)

F 2 (10 N)

F 3 (19 N)

F 4 (28 N)

F 5 (37 N)

εT = 1.03%

εT = 5.27%

εT = 8.07%

εT = 10.43%

εT = 12.49%

σ max (MPa)

0.457

0.427

0.509

0.538

0.616

σ min (MPa)

1.17 × 10−3

1.11 × 10−2

2.11 × 10−2

3.2 × 10−2 2.67 × 10−2

Dmax (mm)

0.64

3.216

4.928

6.364

7.628

σ max (MPa)

4.06 × 10−2

8.08 × 10−2

0.133

0.144

0.140

σ min (MPa)

5.18 × 10−5

5.74 × 10−4

1.04 × 10−3

1.37 × 10−3

1.70 × 10−3

Dmax (mm)

0.60

3.182

4.896

6.334

7.596

σ max (MPa)

2.22 × 10−3

1.24 × 10−2

1.23 × 10−2

1.23 × 10−2

1.24 × 10−2

σ min (MPa)

7 × 10−5

7.32 × 10−4

1.43 × 10−3

2.11 × 10−3

2.72 × 10−3

Dmax (mm)

0.22

2.02

3.56

4.9

6.08

302

6 Simulation of Damage-Free Robotic Gripping of Fruit

Table 6.13 Experimental value of loading mode 3 Component

Exocarp

Mesocarp

Locule gel

Mechanical properties

F 1 (1 N)

F 2 (10 N)

F 3 (19 N)

F 4 (28 N)

F 5 (37 N)

εT = 0.96%

εT = 4.68%

εT = 7.08%

εT = 9.06%

εT = 10.89%

σ max (MPa)

0.556

0.389

0.38

0.518

0.506

σ min (MPa)

1.31 × 10−3

1.4 × 10−2 2.77 × 10−2

4.21 × 10−2

4.36 × 10−2

Dmax (mm)

0.612

2.86

4.32

5.54

6.66

σ max (MPa)

3.86 × 10−2

8.19 × 10−2

0.101

0.122

0.134

σ min (MPa)

1.49 × 10−4

1.47 × 10−3

2.78 × 10−3

4.07 × 10−3

5.36 × 10−3

Dmax (mm)

0.58

2.82

4.28

5.5

6.62

σ max (MPa)

1.07 × 10−3

9.49 × 10−3

1.24 × 10−2

1.27 × 10−2

1.24 × 10−2

σ min (MPa)

7.51 × 10−5

7.51 × 10−4

1.42 × 10−3

2.09 × 10−3

2.71 × 10−3

Dmax (mm)

0.128

1.228

2.216

3.126

4.068

minimum von Mises stress of the tomato mesocarp were, respectively, greater than those of the locule gel of the tomato. In addition, when the loading force is the same, the maximum comprehensive displacement of the tomato exocarp and the mesocarp is greater than the maximum integrated displacement of the locule gel. These are caused by the different properties of the exocarp, mesocarp, and locule gel material of the tomato. When the loading force is 1 N, the maximum von Mises stress σmax of tomato exocarp, mesocarp, and locule gel tissue is 0.292 MPa, 3.1 × 10−2 MPa, 2.13 × 10−3 MPa, respectively, and the maximum von Mises stress of exocarp, mesocarp, and locule gel is less than the respective ultimate stress, and there is no mechanical damage inside the tomato. When the loading force is 10 N, the maximum von Mises stress σ max of tomato exocarp, mesocarp, and locule gel tissue is 0.459 MPa, 8.7 × 10−2 MPa, 1.22 × 10−2 MPa, respectively. The maximum von Mises stress of locule gel material is greater than its ultimate stress (0.012 MPa). So the tissue with stress value greater than the ultimate stress in the gel will appear mechanical damage, and the internal mechanical damage of the three-locular tomato is also generated. At this time the compressibility is 5.11%. When the loading force is 19 N, the maximum von Mises stress σ max of tomato exocarp, mesocarp, and locule gel tissue is 0.541 MPa, 0.127 MPa, 1.24 × 10−2 MPa, respectively, and the maximum von Mises stress of the mesocarp material is also greater than its ultimate stress (0.122 MPa). It indicates that in addition to mechanical damage caused by gel, mechanical damage occurred in the mesocarp tissue, and the compressibility was 7.95%. When the loading force is 28 N, the maximum von Mises stress σmax of tomato exocarp, mesocarp, and locule gel tissue is 0.612 MPa, 0.129 MPa, 1.25 × 10−2 MPa, respectively, and the

6.3 Simulation of Static Gripping Process [3]

303

maximum von Mises stress of the exocarp material is also greater than the ultimate stress of 0.582 MPa. It indicates that in addition to the mechanical damage of the gel and the mesocarp, the exocarp component is also suffered mechanical damage, and the compressibility is 10.38%. The nodal solutions of von Mises stress of exocarp, mesocarp, and locule gel tissue in three-locular tomatoes are shown in Fig. 6.8. Figure 6.8a–c visually express the von Mises stress change process of tomato in the exocarp, mesocarp, and locule gel tissue of the five loading force levels, respectively. The position of MX in the figure is the position of the maximum von Mises stress point of the material, and the position of MN is the position of the minimum von Mises stress point. Stress unit: KPa.

Fig. 6.8 Node stress cloud diagram of the exocarp, mesocarp, and gel tissue of the three-locular

304

6 Simulation of Damage-Free Robotic Gripping of Fruit

There are five values in icons in Fig. 6.8 (a1 –a3 ), (b1 –b2 ), and (c1 ), and the topdown stress increases from the minimum von Mises stress value to the maximum. There are six values in the icon in Fig. 6.8 (a4 –a5 ), (b3 –b5 ), (c2 –c5 ), the top-down stress increases from the minimum von Mises stress value to the maximum. The fifth stress value is the limit stress of the corresponding material. It can be clearly seen from the figure that when the loading force is F 2 , some gel tissue (blue region) begins to show mechanical damage, when the loading force is F 3 , part of the tomato mesocarp (blue region) also begins to appear mechanical damage, and when the loading force is F 4 , mechanical damage begins to occur in the inner part of the tomato exocarp (blue region). With the increase of loading force, the mechanical damage area of tomato exocarp, mesocarp, and locule gel tissue gradually increased. The mechanical damage of the tomato exocarp first appeared at the contact point of the left plate and the tomato (Fig. 6.8-a4 , MX), and then with the increase of the loading force, mechanical damage began to occur at the contact point between the right plate and the tomato. And gradually expand outward. The mechanical damage of the tomato mesocarp first appeared at the position of the inner surface of the locule adjacent to the loading point (Fig. 6.8-b3 , MX). As the loading force increased, the central tissue of the radial wall in the same longitudinal section also appeared to varying degrees of mechanical damage. The mechanical damage of the gel tissue first appeared in the MX of Fig. 6.8-c2 , and with the loading force increased, other parts of the gel also showed different degrees of mechanical damage. Figure 6.9 is a node displacement cloud diagram of the exocarp and mesocarp of the three-locular tomato equatorial plane. It is clear that when the plate compresses the three-locular tomato directly above the locule, because the mesocarp tissue is not supported by the radial wall, the displacement is the largest. Correspondingly, the portion of the peel directly below the radial wall is supported by the radial wall (the radial wall is perpendicular to the lower plate), and thus the displacement is minimized. The displacement of other parts is between the maximum displacement and the minimum. At the same time, with the increase of loading force (F1 → F5 ), the displacement of the same part from the exocarp and mesocarp tissue of the

Fig. 6.9 Node displacement cloud diagram of the exocarp and mesocarp of the equatorial plane of 1/2 the three-locular tomato (unit: mm)

6.3 Simulation of Static Gripping Process [3]

305

equatorial plane of the three-locular tomato gradually increased. The displacement of the mesocarp tissue from the loading position directly above the locule increased from 0.5425 to 7.637 mm. (2)

Loading mode 2: plate-locule-four-locular tomato

Table 6.12 shows the test results after loading from the locule position of the fourlocular tomato using a flat plate, that is, the test result of the loading mode 2. It can be clearly seen from the table that with the gradual increase of loading force (F 2 → F 5 ), the maximum von Mises stress, minimum von Mises stress, and maximum comprehensive displacement of tomato exocarp, mesocarp, and locule gel tissue increase nonlinearly. For the same loading force, the maximum von Mises stress and the minimum von Mises stress of the tomato exocarp were larger than those of the tomato mesocarp, respectively. The maximum von Mises stress and the minimum von Mises stress of the tomato mesocarp were, respectively, greater than those of the locule gel of the tomato. In addition, when the loading force is the same, the maximum comprehensive displacement of the tomato exocarp and the mesocarp is greater than the maximum integrated displacement of the locule gel. These are caused by the different properties of the exocarp, mesocarp, and locule gel material of the tomato. When the loading force is 1 N, the maximum von Mises stress σmax of tomato exocarp, mesocarp, and locule gel tissue is 0.457 MPa, 4.06 × 10−2 MPa, 2.22 × 10−3 MPa, respectively, and the maximum von Mises stress of exocarp, mesocarp, and locule gel is less than the respective ultimate stress, and there is no mechanical damage inside the tomato. When the loading force is 10 N, the maximum von Mises stress σmax of tomato exocarp, mesocarp, and locule gel tissue is 0.427 MPa, 8.08 × 10−2 MPa, 1.24 × 10−2 MPa, respectively. The maximum von Mises stress of locule gel material is greater than its ultimate stress (0.012 MPa). So the tissue with stress value greater than the ultimate stress in the gel will cause mechanical damage, and the internal mechanical damage of the three-locular tomato is also generated. At this time the compressibility was 5.27%. When the loading force is 19 N, the maximum von Mises stress σmax of tomato exocarp, mesocarp, and locule gel tissue is 0.509 MPa, 0.133 MPa, 1.23 × 10−2 MPa, respectively, and the maximum von Mises stress of the mesocarp material is also greater than its ultimate stress (0.122 MPa). It indicates that in addition to mechanical damage caused by gel, mechanical damage occurred in the mesocarp tissue, and the compressibility was 8.07%. When the loading force is 37 N, the maximum von Mises stress σmax of tomato exocarp, mesocarp, and locule gel tissue is 0.616 MPa, 0.14 MPa, 1.24 × 10−2 MPa, respectively, and the maximum von Mises stress of the exocarp material is also greater than the ultimate stress of 0.582 MPa. It indicates that in addition to the mechanical damage of the gel and the mesocarp, the exocarp component is also suffered mechanical damage, and the compressibility is 12.49%. The nodal solutions of von Mises stress of exocarp, mesocarp, and locule gel tissue in four-locular tomatoes are shown in Fig. 6.10. It can be clearly seen from the figure that when the loading force is F 2 , some gel tissue (blue region) begins to

306

6 Simulation of Damage-Free Robotic Gripping of Fruit

Fig. 6.10 Node stress cloud diagram of the exocarp, mesocarp, and gel tissue of the 1/4 four-locular tomato

show mechanical damage, when the loading force is F 3 , part of the tomato mesocarp (blue region) also begins to appear mechanical damage, and when the loading force is F 5 , mechanical damage begins to occur in the inner part of the tomato exocarp (blue region). With the increase of loading force, the mechanical damage area of tomato exocarp, mesocarp, and locule gel tissue gradually increased. The mechanical damage of the tomato exocarp first appeared at the contact point of the left plate and the tomato (Fig. 6.10-a5, MX). The position where the tomato mesocarp and locule gel tissue began to appear mechanical damage is the same as in the case of the loading mode 1.

6.3 Simulation of Static Gripping Process [3]

307

Fig. 6.11 Node displacement cloud diagram of the exocarp and mesocarp of the equatorial plane of 1/4 the four-locular tomato (unit: mm)

Figure 6.11 is a node displacement cloud diagram of the exocarp and mesocarp of the four-locular tomato equatorial plane. It is clear that when the plate compresses the four-locular tomato directly above the locule, because the mesocarp tissue is not supported by the radial wall, the displacement is the largest. For other parts, with the increase of the distance from the loading position, the displacement becomes smaller. At the same time, with the increase of loading force (F 1 → F 5 ), the displacement of the same part from the exocarp and mesocarp tissue of the equatorial plane of the four-locular tomato gradually increased. The displacement of the mesocarp tissue from the loading position directly above the locule increased from 0.6284 to 7.61 mm. (3)

Loading mode 3: plate-radial wall-four-locular tomato

Table 6.13 shows the test results after loading from the locule position of the fourlocular tomato using a flat plate, that is, the test result of the loading mode 3. It can be clearly seen from the table that with the gradual increase of loading force (F 1 → F 5 ), the maximum von Mises stress, minimum von Mises stress, and maximum comprehensive displacement of tomato exocarp, mesocarp, and locule gel tissue increase nonlinearly. For the same loading force, the maximum von Mises stress and the minimum von Mises stress of the tomato exocarp were larger than those of the tomato mesocarp, respectively. The maximum von Mises stress and the minimum von Mises stress of the tomato mesocarp were respectively greater than those of the locule gel of the tomato. In addition, when the loading force is the same, the maximum comprehensive displacement of the tomato exocarp and the mesocarp is greater than the maximum integrated displacement of the locule gel. These are caused by the different properties of the exocarp, mesocarp, and locule gel material of the tomato. When the loading force is 1 N and 10 N, the maximum von Mises stress of exocarp, mesocarp, and locule gel is less than the respective ultimate stress, and there is no mechanical damage inside the tomato. When the loading force is 19 N, the maximum von Mises stress σmax of tomato exocarp, mesocarp, and locule gel

308

6 Simulation of Damage-Free Robotic Gripping of Fruit

tissue is 0.38 MPa, 0.101 MPa, 1.24 × 10−2 MPa, respectively, and the maximum von Mises stress of the locule gel material is greater than its ultimate stress (0.122 MPa). So the tissue with stress value greater than the ultimate stress in the gel will cause mechanical damage, and the internal mechanical damage of the four-locular tomato is generated. At this time the compressibility was 7.08%. When the loading force is 28 N, the maximum von Mises stress σmax of tomato exocarp, mesocarp, and locule gel tissue is 0.518 MPa, 0.122 MPa, 1.27 × 10−2 MPa, respectively, and the maximum von Mises stress of the mesocarp material is also greater than its ultimate stress (0.122 MPa). It indicates that in addition to mechanical damage caused by gel, mechanical damage occurred in the mesocarp tissue, and the compressibility was 9.06%. When the loading force is 37 N, the maximum von Mises stress of the exocarp material is less than the ultimate stress of 0.582 MPa. It indicates that in addition to the mechanical damage of the gel and the mesocarp, the exocarp component is not suffered mechanical damage, and the compressibility is 10.89%. The nodal solutions of von Mises stress of exocarp, mesocarp, and locule gel tissue in four-locular tomatoes are shown in Fig. 6.12. It can be clearly seen from the figure that when the loading force is F 3 , some gel tissue (blue region) begins to show mechanical damage, when the loading force is F 4 , part of the tomato mesocarp (blue region) also begins to appear mechanical damage. With the increase of loading force, the mechanical damage area of tomato exocarp, mesocarp, and locule gel tissue gradually increased. But there was no mechanical damage to the exocarp tissue in this process. The mechanical damage of the tomato mesocarp first appeared in the central structure of the radial wall in the longitudinal section (Fig. 6.12-b4, MX), and then gradually expanded outward with the increase of the loading force. The mechanical damage of the gel tissue first appeared in the MX of Fig. 6.12-c3, and with the loading force increased, other parts of the gel also showed different degrees of mechanical damage. Figure 6.13 is a node displacement cloud diagram of the exocarp and mesocarp of the four-locular tomato equatorial plane. It is clear that when the plate compresses the four-locular tomato directly above the radial wall, the displacement is the largest on the loading position. For other parts, with the increase of the distance from the loading position, the displacement becomes smaller. At the same time, with the increase of loading force (F 1 → F 5 ), the displacement of the same part from the exocarp and mesocarp tissue of the equatorial plane of the four-locular tomato gradually increased. The displacement of the mesocarp tissue from the loading position directly above the radial wall increased from 0.5993 to 6.638 mm. (4)

Comparison and analysis of three loading modes

According to the above analysis, by comparing the results of the three loading modes, we can see that (1)

With the increase of loading force, the damage order of tomato internal tissues in the three loading modes was locule gel, mesocarp, and exocarp. The mechanical damage of gel and mesocarp tissue is not very serious, and the cell structure is destroyed, thereby destroying the separation between the substrate and the

6.3 Simulation of Static Gripping Process [3]

309

Fig. 6.12 Node stress cloud diagram of the mesocarp and gel tissue of the 1/4 four-locular tomato

enzyme, resulting in a series of enzymatic reactions that cause browning [9] at the loading position. After the exocarp tissue is damaged, the wound ruptures, exposing the inner layer of tissue, and the microorganisms directly propagate on the surface to form mildew. Therefore, compared with the damage of gel and mesocarp tissue, the damage of exocarp tissue will shorten the shelf life of tomato to a greater extent.

310

6 Simulation of Damage-Free Robotic Gripping of Fruit

Fig. 6.13 Node displacement cloud diagram of the exocarp and mesocarp of the equatorial plane of 1/4 the four-locular tomato (unit: mm)

(2)

(3)

(4)

3.

Under the same loading force, after the rigid flat plate is compressed from the radial wall position of the four-locular tomato, the compressibility of the plate to tomato is the smallest, and the compressibility is the largest when compressing from the locule position. For loading mode 1: When the compressibility is 5.11%, the mechanical damage firstly appears to the gel tissue, when the compressibility is 7.95%, the mechanical damage appears to the mesocarp tissue, and when the compressibility is 10.38% it appears to the exocarp tissue. For loading mode 2: when the compressibility is 5.27%, the mechanical damage firstly appears to the gel tissue, when the compressibility is 8.07%, the mechanical damage appears to the mesocarp tissue, and when the compressibility is 12.49%, it appears to the exocarp tissue. For loading mode 3: when the compressibility is 7.08%, the mechanical damage firstly appears to the gel tissue, when the compressibility is 9.06%, the mechanical damage appears to the mesocarp tissue, and when the compressibility is less than 10.89%, there was no mechanical damage to the exocarp tissue. Through this analysis, the critical value of the compressibility of mechanical damage under three loading modes can be predicted. When the loading force is the same, the flat plate is compressed from the radial wall position of the four-locular tomato (loading mode 3), the probability of mechanical damage of the tissue components is the smallest. The plate is compressed from the locule position of the four-locular tomato (loading method 2), the probability of mechanical damage to the tissue components is greatest. Relationship between finger type and mechanical damage of tomato

The previous section introduced the experimental results of the flat plate loading on the finite element model of tomato. This section describes the results of the loading of the arc plate on the finite element model of tomato. The stress distribution results of the components of the tomato after the test are shown in Tables 6.14, 6.15 and 6.16. Then the effect of finger type on mechanical damage of tomato was analyzed.

6.3 Simulation of Static Gripping Process [3]

311

Table 6.14 Experimental value of loading mode 4 Component

Exocarp

Mesocarp

Locule gel

Mechanical properties

F 1 (1 N)

F 2 (10 N)

F 3 (19 N)

F 4 (28 N)

F 5 (37 N)

εT = 0.7%

εT = 4.15%

εT = 6.39%

εT = 8.25%

εT = 9.86%

σ max (MPa)

0.177

0.413

0.539

0.563

0.578

σ min (MPa)

2.37 × 10−3

2.12 × 10−2

3.87 × 10−2

5.65 × 10−2

6.40 × 10−2

Dmax (mm)

0.411

2.514

3.878

5.011

5.995

σ max (MPa)

2.03 × 10−2

6.63 × 10−2

8.83 × 10−2

0.109

0.125

σ min (MPa)

1.93 × 10−4

1.73 × 10−3

3.16 × 10−3

4.63 × 10−3

6.03 × 10−3

Dmax (mm)

0.411

2.514

3.878

5.011

5.995

σ max (MPa)

1.93 × 10−3

1.22 × 10−2

1.22 × 10−2

1.24 × 10−2

1.25 × 10−2

σ min (MPa)

6.44 × 10−4

6.24 × 10−4

1.17 × 10−3

1.74 × 10−3

2.31 × 10−3

Dmax (mm)

0.304

2.12

3.41

4.505

5.473

F 4 (28 N)

Table 6.15 Experimental value of loading mode 5 Component Mechanical F 1 (1 N) properties ε = 0.81% T Exocarp

Mesocarp

Locule gel

(1)

F 2 (10 N)

F 3 (19 N)

εT = 4.16%

εT = 6.24% εT = 7.93%

εT = 9.39%

σ max (MPa) 0.287

0.341

0.379

0.433

0.467

σ min (MPa) 1.16 × 10−3

1.14 × 10−2

2.19 × 10−2

3.37 × 10−2

4.32 × 10−2

Dmax (mm)

1.266

1.901

2.417

2.861

0.108

0.125

0.247

10−2

F 5 (37 N)

σ max (MPa) 2.59 × 10−2

5.75 × 10−2

8.8 ×

σ min (MPa) 5.92 × 10−5

6.13 × 10−4

1.19 × 10−3

1.71 × 10−3

2.11 × 10−3

Dmax (mm)

1.252

1.887

2.404

2.849

1.26 × 10−2

1.24 × 10−2

1.23 × 10−2

0.234

10−2

σ max (MPa) 2.02 × 10−3

1.3 ×

σ min (MPa) 6.26 × 10−5

6.42 × 10−4

1.24 × 10−3

1.77 × 10−3

2.11 × 10−3

Dmax (mm)

0.848

1.408

1.889

2.315

0.109

Loading mode 4: arc plate-locule-three-locular tomato

Table 6.14 shows the test results after loading from the locule position of the threelocular tomato using an arc plate, that is, the test result of the loading mode 4.

312

6 Simulation of Damage-Free Robotic Gripping of Fruit

Table 6.16 Experimental value of loading mode 6 Component

Exocarp

Mesocarp

Locule gel

(1)

(2)

(3)

Mechanical properties

F 1 (1 N)

F 2 (10 N)

εT = 0.76%

εT = 3.80% εT = 5.78% εT = 7.38% εT = 8.77%

F 3 (19 N)

F 4 (28 N)

F 5 (37 N)

σ max (MPa)

0.386

0.409

0.405

0.425

0.475

σ min (MPa)

1.31 × 10−3

1.41 × 10−2

1.66 × 10−2

1.99 × 10−2

1.24 × 10−2

Dmax (mm)

0.468

2.32

3.524

4.5

5.352

σ max (MPa)

2.94 × 10−2

5.99 × 10−2

7.94 × 10−2

0.108

0.132

σ min (MPa)

1.47 × 10−4

1.42 × 10−3

2.62 × 10−3

3.78 × 10−3

4.9 × 10−3

Dmax (mm)

0.434

2.286

3.49

4.468

5.32

σ max (MPa)

1.03 × 10−3

8.31 × 10−3

1.23 × 10−2

1.23 × 10−2

1.25 × 10−2

σ min (MPa)

7.44 × 10−5

6.91 × 10−4

1.27 × 10−3

1.81 × 10−3

2.3 × 10−3

Dmax (mm)

0.126

1.132

1.984

2.736

3.418

When the loading force is 1 N, the maximum von Mises stress σmax of tomato exocarp, mesocarp, and locule gel tissue is 0.177 MPa, 2.03 × 10−2 MPa, 1.93 × 10−3 MPa, respectively, and the maximum von Mises stress of exocarp, mesocarp, and locule gel is less than the respective ultimate stress, and there is no mechanical damage inside the tomato. When the loading force is 10 N, the maximum von Mises stress σmax of tomato exocarp, mesocarp, and locule gel tissue is 0.413 MPa, 6.63 × 10−2 MPa, 1.22 × 10−2 MPa, respectively. The maximum von Mises stress of locule gel material is greater than its ultimate stress (0.012 MPa). So the tissue with stress value greater than the ultimate stress in the gel will appear mechanical damage, and the internal mechanical damage of the three-locular tomato is also generated. At this time the compressibility is 4.15%. When the loading force is 37 N, the maximum von Mises stress σmax of tomato exocarp, mesocarp, and locule gel tissue is 0.578 MPa, 0.125 MPa, 1.27 × 10−2 MPa, respectively, and the maximum von Mises stress of the mesocarp material is also greater than its ultimate stress (0.122 MPa). It indicates that in addition to mechanical damage caused by gel, mechanical damage occurred in the mesocarp tissue, but the exocarp component is not suffered mechanical damage. The compressibility was 9.06%.

6.3 Simulation of Static Gripping Process [3]

(2)

313

Loading mode 5: arc plate-locule-four-locular tomato

Table 6.15 shows the test results after loading from the locule position of the fourlocular tomato using an arc plate, that is, the test result of the loading mode 5. (1)

(2)

(3)

(3)

When the loading force is 10 N, the maximum von Mises stress σmax of tomato exocarp, mesocarp, and locule gel tissue is 0.341 MPa, 5.75 × 10−2 MPa, 1.3 × 10−3 MPa, respectively. The maximum von Mises stress of exocarp, mesocarp, and locule gel is less than the respective ultimate stress, and there is no mechanical damage inside the tomato. When the loading force is 19 N, the maximum von Mises stress σmax of tomato exocarp, mesocarp and locule gel tissue is 0.379 MPa, 8.8 × 10−2 MPa, 1.26 × 10−2 MPa, respectively. The maximum von Mises stress of locule gel material is greater than its ultimate stress (0.012 MPa). So the tissue with stress value greater than the ultimate stress in the gel will appear mechanical damage, and the internal mechanical damage of the four-locular tomato is also generated. At this time the compressibility is 6.24%. When the loading force is 37 N, the compressibility was 9.06%. The maximum von Mises stress σmax of tomato exocarp tissue is 0.467 MPa, which is less than the ultimate stress (0.582 MPa). It indicates that the exocarp component is not suffered mechanical damage under the 5 loading force. Loading mode 6: arc plate-radial wall-four-locular tomato

Table 6.16 shows the test results after loading from the radial wall position of the four-locular tomato using an arc plate, that is, the test result of the loading mode 6. It can be seen from the table that: (1)

(2)

(3)

When the loading force is 10 N, the maximum von Mises stress σmax of tomato exocarp, mesocarp, and locule gel tissue is 0.409 MPa, 5.99 × 10−2 MPa, 8.31 × 10−3 MPa, respectively. The maximum von Mises stress of exocarp, mesocarp, and locule gel is less than the respective ultimate stress, and there is no mechanical damage inside the tomato. When the loading force is 19 N, the maximum von Mises stress σmax of tomato exocarp, mesocarp, and locule gel tissue is 0.405 MPa, 7.94 × 10−2 MPa, 1.23 × 10−2 MPa, respectively. The maximum von Mises stress of locule gel material is greater than its ultimate stress (0.012 MPa). So the tissue with stress value greater than the ultimate stress in the gel will appear mechanical damage, and the internal mechanical damage of the four-locular tomato is also generated. At this time the compressibility is 5.78%. When the loading force is 37 N, the maximum von Mises stress σmax of tomato exocarp, mesocarp, and locule gel tissue is 0.475 MPa, 0.132 MPa, 1.25 × 10−2 MPa, respectively, and the maximum von Mises stress of the mesocarp material is also greater than its ultimate stress (0.122 MPa). It indicates that in addition to mechanical damage caused by gel, mechanical damage occurred in the mesocarp tissue, but the exocarp component is not suffered mechanical damage. The compressibility was 8.77%.

314

(4)

6 Simulation of Damage-Free Robotic Gripping of Fruit

Comparison of results loading by different fingers

By comparing the loading modes 1 and 4, the loading modes 2 and 5, and the loading modes 3 and 6, it can be seen that when the loading position, the structure type, and the loading force are the same, the compressibility of tomato is smaller when loading by the rigid arc plate (the arc finger) than the rigid flat plate (the flat finger). And the probability of mechanical damage in tomato internal tissue is also small.

6.4 Dynamic Simulation of Gripping Process [1] The simulation of the kinematics and dynamics of the gripping process is the basis for the motion planning and gripping control of the end-effector. The purpose of the simulation is mainly to detect the coordination of motion between the various mechanisms and the rationality of the modeling, to obtain the kinematics and dynamic state demonstration of the mechanism, and to measure the parameters of the motion and dynamics, then obtain the collision process information between the finger and the tomato. And optimize the gripping strategy based on the results. The gripping simulation analysis mainly uses ADAMS software to analyze the movement of the mechanism, measure the speed, acceleration, force, and other factors after simulating the picking and gripping process, and finally determine the collision force and strain energy of the tomato during the gripping process.

6.4.1 The Software Implementation of Dynamic Gripping Simulation 1.

Software interface

The kinematics simulation analysis software ADAMS is commercial software for simulating the kinematics and dynamics of mechanical systems. The analysis object is mainly multi-rigid body [10]. However, in the mechanical system, the flexible body will have an important impact on the motion of the whole system. Therefore, if the influence of the flexible body is ignored in the analysis, it will cause a large error. It also causes errors in the stress state and motion state of each component in the whole system, thus affecting the stress–strain distribution inside the component in the simulation. Therefore, to improve simulation accuracy, obtain accurate dynamic simulation results, and perform accurate stress–strain analysis on flexible bodies in the system, two softwares, ANSYS and ADAMS, are needed. After establishing a flexible finite element model, ANSYS can generate a flexible body modal neutral file (.mnf file) used by ADAMS, and then use the ADAMS/Flex module in ADAMS to load this file to generate the corresponding flexible body. The

6.4 Dynamic Simulation of Gripping Process [1]

315

modal superposition method is used to calculate the deformation during the dynamic simulation process and the force on the joint nodes. In this way, the flexibility properties of the components can be considered in the dynamic model of the mechanical system, and the accuracy of the system simulation is improved. Correspondingly, ADAMS can generate a load file (.lod file) used by ANSYS software for dynamic analysis, which can output the load spectrum and displacement spectrum information after dynamic simulation to ANSYS software. ANSYS can directly call this file to generate the boundary conditions of the force in the finite element analysis for the evaluation and analysis of stress, strain, and fatigue life, so that the results of stress and strain analysis based on accurate dynamic simulation results can be obtained, thereby improving the calculation accuracy. 2.

Modeling process for dynamic gripping simulation

The complete steps for stress–strain analysis of flexible body parts in a motion system using ADAMS in conjunction with the ANSYS program are as follows: (1)

(2)

Establish a flexible body model in ANSYS software, select the appropriate unit type to divide the unit, establish a finite element model of the flexible body, and use the ADAMS.mac macro file to generate the flexible body modal neutral file (.mnf file) required by the ADAMS software. Introduce the model of the rigid body in the ADAMS software, connect the assembly and add the modeling parameters. Read the modal neutral file, set the connection methods of the flexible body, and apply the load to perform the system dynamics simulation. In the post-processing, information such as stress and strain of the simulation process is analyzed.

6.4.2 The Establishment of System Virtual Prototype for Gripping The gripping process of the end-effector is actually a direct action of the fingers of the actuator and the tomato. In order to improve the calculation speed when studying the gripping collision process, it is necessary to simplify the simulation system. During the simulation process, the housing of the end-effector, the motor, the finger transfer mechanism, and the vacuum chuck system are discarded, and only the three-dimensional solid model of the finger is established and saved as an x_t file. 1.

Data exchange between Pro/E and ADAMS

Import the solid model of the finger in ADAMS. The import dialog box is shown in Fig. 6.14. Edit the color, position, material, and other attribute information of the finger model. The finger material selects the aluminum in the material library, the density

316

6 Simulation of Damage-Free Robotic Gripping of Fruit

Fig. 6.14 Import the geometric model of the finger

is 2.74 × 103 kg/mm3 , and the elastic modulus is 7.17 × 104 MPa, as shown in Fig. 6.15. Then the centroid position of each finger can be obtained automatically. 2.

Data exchange of ANSYS and ADAMS

Copy the modal neutral file generated in ANSYS to the working directory of ADAMS, import the tomato nonlinear viscoelastic modal neutral file, and create the tomato flexible body. The dialog box is shown in Fig. 6.16. Place the origin of the coordinate at the tomato fruit umbilical, and connection between the umbilical and the stem is coincident with the y-axis. The tomato flexible body is shown in Fig. 6.17. Edit the position of the tomato. During the actual picking process, the position of the tomato is mainly defined by the fruit stem and the vacuum chuck.

Fig. 6.15 The set of finger material

6.4 Dynamic Simulation of Gripping Process [1]

317

Fig. 6.16 Establishment of tomato flexible body

Fig. 6.17 Import the mnf file

The fruit stem limits three mobile freedom of the tomato. The vacuum corrugated suction cup limits three mobile freedoms and a rotational freedom. The combination of the fruit stem and the suction cup limits three mobile freedoms and the rotational freedom along the y-direction (as shown in Fig. 6.18). Therefore, when restraining the position of the tomato, it is necessary to set the ball sub-constrained motion pairs at the fruit stem and the suction cup. The fingers are moved in parallel during the gripping process, so the translational movement pairs are, respectively, set for the two fingers, and the DC motor drive is added, respectively. During the gripping process, the fingers contact with the surface of the tomato, thus creating a contact force at the point of contact. During the process of contact between the finger and the tomato, due to the relative motion, the material of the finger and the tomato will be compressed, and the kinetic energy of the finger will be converted into the compressive potential energy, accompanied by the loss of energy. The process and method of establishing contact between the finger and the tomato are shown in Fig. 6.19. The virtual prototype of the established end-effector finger is shown in Fig. 6.20.

318

6 Simulation of Damage-Free Robotic Gripping of Fruit

Fig. 6.18 Constraint conditions of the tomato fruit

Fig. 6.19 Contact between the finger and the tomato fruit

6.4.3 Simulation Analysis of Tomato Fruit Gripping with the End-Effector 1.

Simulation process

After completing the modeling of the end-effector gripping system and addition of the constraints, the motion pair, and the drive, according to the actual requirements of the tomato harvesting operation, the driving speed is driven by different fingers, and the simulation test is performed to set the simulation time and step size. Figure 6.21 shows the state when the tomato is clamped.

6.4 Dynamic Simulation of Gripping Process [1]

319

Fig. 6.20 Model of the gripping system

Fig. 6.21 Simulation of gripping process

Set the driving function of the end-effector finger movement as in Eq. 6.14, set the simulation movement time and the number of simulation steps, enter the ADAMS post-processing module, and run to obtain the contact force curve. (step(time, 0, 0, 1, 6) + step(time, 6, 0, 1.4, 1.2) + step(time, 1.4, 0, 3, 0)) (6.14) 2. (1)

Simulation operation and test verification Collision force variation law and verification

In terms of the simulation process, the contact force curve increased rapidly after the finger contacted the tomato, and it showed a slow downward trend after reaching a certain value, which effectively reflected the impact collision effect of the effector

320

6 Simulation of Damage-Free Robotic Gripping of Fruit

Fig. 6.22 Experimental verification of gripping force from the simulation results

finger on the tomato when gripping. While the finger remained clamped, the stress of the tomato presented a certain relaxation characteristic. During the gripping process, the contact force curve had certain volatility. The possible reason was that there was a certain collision between the finger and the tomato in the simulation, which led to certain flexible body vibration. Under the conditions of the initial opening of the finger and the same amount of tomato compression, the simulation and the collision test results of the collision force during the gripping process with the gripping speed were shown in Fig. 6.22. It can be seen that when the compression amount was constant, the collision force in the gripping process increased remarkably with the increase of the finger gripping speed, and the change basically conformed to the exponential function relationship. At the same time, the collision force-speed law of the clamp collision simulation was consistent with the test result, and the peak force obtained by the simulation was 16.1% higher than the test result. The reason for the error may be the existence of the transmission gap and the connection flexibility. The gripping system of the picking robot end-effector had a weakly flexible feature during the gripping collision, while the simulation was completely rigid. (2)

The change of strain energy

During the gripping process, the strain energy of the tomato flexible body reflects the deformation of the tomato and the energy transfer during the collision. After the ADAMS simulation, the strain energy curve obtained by the post-processing is shown in Fig. 6.23. The strain energy and its changing trend of the tomato flexible body during the gripping process are shown in Fig. 6.24 under the condition of finger movement speed of 1–6 mm/s. The strain energy reflects the absorption and transfer of energy during the gripping process. It can be seen that with the increase of the finger movement speed, the strain energy of the tomato increases.

6.4 Dynamic Simulation of Gripping Process [1]

321

Fig. 6.23 Strain energy of tomato flexible body at 3 mm/s

Fig. 6.24 Change in strain energy with speed

(3)

Application of collision simulation model (1)

(2) (3)

Combined with the three-stage composite collision mathematical model established in Chap. 4, compared with the test, it can effectively expand the factors such as gripping speed, object viscoelasticity, motor output range, and gripping finger surface materials. It also helps a depth analysis. Therefore provide a powerful tool for basic research and practical applications. It can intuitively analyze the stress distribution, stress intensity, and damage occurrence location of the fruit gripping collision. Based on the simulation model, the comparison of the structure and effect of the gripping control mode and strategy can be carried out to provide basic support for automatic fast harvesting operations.

322

6 Simulation of Damage-Free Robotic Gripping of Fruit

References 1. Bai X (2012) Collision simulation and parameter optimization of picking robot fast gripping[D]. Jiangsu University 2. Zhang H, Zhao Q (2008) Ansys finite element analysis complete self-study manual[M]. Mechanical Industry Press 3. Li Z (2011) Research on picking damage of picking robot based on biomechanical properties of tomato [D]. Jiangsu University 4. Sun Y, Yu D (1996) Agricultural biomechanics and agricultural bioelectromagnetism[M]. China Agriculture Press 5. Leng L, Zhao J, Zhang Y (2007) Fundamentals of finite element technology[M]. Chemical Industry Press 6. Li Z, Liu J, Li P et al (2008) Analysis of workspace and kinematics for a tomato harvesting robot[C]. Proceedings of the International Conference on Intelligent Computation Technology and Automation 7. Liu J, Li P, Li Z (2008) Hardware design of the end-effector for tomato-harvesting robot[J]. J Agricult Mach 39(3):109–112 8. Li Z, Liu J, Li P et al (2009) Study on the collision-mechanical properties of tomatoes gripped by harvesting robot fingers[J]. Afr J Biotech 8(24):7000–7007 9. Lin H, Xi Y, Chen S (2002) Enzymatic browning during fruit storage[J]. J Fuzhou Univ 30(Supplement):696–703 10. Li Z (2014) Detailed introduction and examples of Adams entry[M]. National Defense Industry Press

Chapter 7

Modeling of the Vacuum Sucked Pulling of Tomato Fruit

7.1 Summary 7.1.1 Function of Vacuum Sucked Pulling in Robotic Harvesting [1] 1.

Application of vacuum sucked pulling in robotic harvesting

Some challenges still persist in the robotic gripping and detachment of fruits individually. Firstly, because fruits such as strawberries and cherry tomatoes are extremely small, robotic gripping of individual fruit is a difficult task. Secondly, the space in a complicated dense canopy is usually insufficient to permit the robotic end-effector to perform the harvesting action. Thirdly, both positioning errors and the swing of free-hanging fruits are inevitable. These problems severely affect successful gripping and detachment and cause damage to adjacent fruits or stems [2]. As an auxiliary motion, the vacuum sucked pulling operation has proved to be effective in overcoming these problems, owing to its distinct advantages. Firstly, it enables the initial attachment of the end-effector to the target fruit, preventing movement during the cutting operation [2–6]. Secondly, it isolates the target fruit from other fruits of the same cluster so as to avoid damage to them during grasping [2–6]. This operation is being applied increasingly widely in the robotic harvesting of fruits [7–12]. 2.

Value of vacuum sucked pulling in robotic harvesting

According to the present application and researches of vacuum suction in robotic harvesting, we can draw the following conclusions: (1)

Vacuum suction can meet the needs of handling tender fruits and vegetables, so it has been widely applied in the development of harvesting robots. It is not reliable for picking fruit simply relying on suction, but as an auxiliary device, it can improve the performance of harvesting obviously.

© Science Press, Beijing and Springer Nature Singapore Pte Ltd. 2021 J. Liu et al., Rapid Damage-Free Robotic Harvesting of Tomatoes, Springer Tracts in Mechanical Engineering, https://doi.org/10.1007/978-981-16-1284-8_7

323

324

7 Modeling of the Vacuum Sucked Pulling of Tomato Fruit

(2)

The actual fruit harvesting is faced with complex dense canopy conditions, and the influence of the occlusion by branches and leaves and inter-touching of fruits is the main factor affecting the success rate of the harvesting. Through the cultivation mode modification, the interference of branches and leaves can be improved to a certain extent [5, 13–15]. However, inter-touching of fruits still becomes the key obstacle to the success rate of robotic harvesting of fruit, such as tomato, citrus and strawberry [5, 16, 17]. The vacuum sucked pulling is an effective way to solve this problem.

As the auxiliary action of the fruit harvesting operation of the end-effector, the purpose of sucked pulling is to isolate the target fruit from the multiple fruits squeezed from each other. This is helpful to avoid the block of the adjacent fruit in same the cluster to the finger movement and to avoid potential gripping bruise to the adjacent fruit in the same cluster. In one word, the vacuum sucked pulling may improve the success rate of fruit gripping in a complex canopy.

7.1.2 Research Significance [1, 18] 1.

Mechanical mechanism of on-plant fruit sucked pulling

Sucked pulling of on-plant fruit is a complex force kinematic process. Firstly, this process is quite different from gripped pulling, since it relies on the vacuum suction force between the suction pad and the fruit surface, which is decided by not only the degree of vacuum but also the sizes of both the suction pad and fruit. When the force required for the pulling motion exceeds the largest suction force, the fruit will break away from the suction pad. Second, corresponding displacements and forces will act on the whole fruit-stem system. Therefore, the success rate and efficiency of the process depend on not only the mechanical properties of the fruit-stem system but also the performance of the vacuum system and control optimization. To conclude, studies on the mechanics interaction between the suction pad and the fruit and on the corresponding response of the fruit-stem system are essential for optimizing the control strategies for realizing a precise, high-speed, and energy-saving sucked pulling operation. 2.

The actual needs of the sucked pulling stroke of on-plant fruit

As the auxiliary action of the tomato fruit picking operation of the end-effector, the key to judge the success of the sucked pulling operation is whether the sucked pulling distance can avoid the finger touching the adjacent fruit during the gripping process. At the same time, too large sucked pulling distance will lead to the sucking failure and the motor overloading. Therefore, according to the actual needs, it is critical to determine the appropriate sucked pulling distance to guarantee a success rate of fruit harvesting and reduce energy consumption.

7.1 Summary

325

The actual sucked pulling distance to avoid the gripping interference is related to the fruit number in the cluster. As the number is gradually declined in different harvesting rounds, the actual sucked pulling distance is also changing. At the same time, the permissible sucked pulling distance is also limited by properties of both the fruit-stem system and the end-effector. Therefore, the reliable analysis of the sucked pulling distance is the key to the problem of on-plant fruit sucked pulling, which has an important influence on the success of the fruit harvesting. 3.

Energy-saving sucked pulling operation of on-plant fruit

Present mobile vacuum systems usually rely on vacuum pumps to create a vacuum. Vacuum pumps are used widely in cleaning robots, climbing robots, and developed harvesting robot prototypes. However, vacuum pumps are more suitable for continuous running and lower demands for a stable vacuum degree. For fruit or food handling, frequent vacuum generation and release are necessary for frequent suction and release of different fruits, and the uncontrollability and pulsation of the vacuum degree may cause suction damage to fruits [3, 19, 20]. Thus, we developed a new vacuum suction device integrated into the end-effector that uses an air-powered vacuum ejector to create a vacuum, such that the vacuum degree is controllable and stable and the response speed is several times higher than that of a vacuum pump [11]. Since outdoor vegetable fields and orchards lack a constant air supply, the new device has one mini air compressor with an air tank to supply air to the ejector. For an outdoor robotic application, energy saving is a key issue for performing long-duration vacuum suction missions. Compared with a vacuum pump, the abovementioned compressor–tank–ejector vacuum system makes energy-saving possible by frequent intermittent work instead of continuous running. Furthermore, optimal control is undoubtedly necessary to save energy during fruit sucking and harvesting. The current research mainly focuses on the application of the vacuum suction system. Only a few literatures have studied the change of vacuum degree in the process of fruit sucked pulling and the relationship between vacuum degree and displacement of the suction pad [2], but the research on the basic law and control optimization of vacuum sucked pulling is extremely lack, which seriously affects its actual working performance and application.

7.1.3 Content and Innovation (1)

Mechanical and kinematic modeling of the vacuum sucked pulling of tomato fruits was performed on the basis of the physical and mechanical properties of both the fruit-stem system and the suction pad. And the relation among the vacuum suction, motion of the suction pad, and response of the fruit-stem system to the sucking effect was established. The modeling and findings of this study are valuable for both the optimal selection of the suction pad and

326

(2)

7 Modeling of the Vacuum Sucked Pulling of Tomato Fruit

the optimization of control strategies for realizing a precise, high-speed, and energy-saving operation. In view of the large difference of agricultural targets, the comprehensive indices of gripping success rate were established based on the probability theory for the first time. And then the optimization of control parameters and mode was completed, thus the high-success rate, high-efficiency, and energy-saving vacuum sucked pulling was realized in robotic tomato harvesting.

7.2 Modeling of Mechanical Behavior for Sucking with Suction Pad 7.2.1 Mechanical Relation Between Suction Pad and Spherical Surface [1] 1.

Theoretical model of the relation between suction force and negative pressure

The outer diameter of the vacuum sucker is called the nominal diameter, and the diameter of surface that is sucked effectively by the deformed suction pad is called the effective diameter. Theoretically, the pull-off force of suction pad is directly proportional to the vacuum negative pressure and its effective area. [Fs ] = 10−3 |pu |Ae = 10−3 |pu |π



Φe 2

2 (7.1)

where [F s ] is the pull-off force of suction pad, N; Pu is the negative pressure per area applied normally to object surface, kPa; Ae is the effective area of suction pad, mm2 ; e is the effective diameter of suction pad, which is defined as the diameter of the maximum suction area, mm. 2.

Normal force balance analysis of plane surface sucked pulling

As shown in Fig. 7.1, at any moment of constant-speed sucked pulling with the suction pad, the static force balance equation for the suction pad is Fc + F p = Fs

(7.2)

where F c is the normal contact force between the target and the suction pad that depends on the pressure and seal area (N); F p is the external force applied to the nipple of the suction pad in the positive axial direction, N; F s is the suction force between the target and the suction pad that depends on the negative pressure and the sucking area, N, it is as follows: pc (A2 − As ) + F p = 10−3 |pu |As

(7.3)

7.2 Modeling of Mechanical Behavior for Sucking with Suction Pad

327

Fig. 7.1 Force and deformation of suction pad in sucked pulling

where A2 is the outer sectional area of suction pad, mm2 ; As is the suction area, mm2 ; pc is the normal contact pressure per area, kPa. In Eq. (7.5), A2 = π Φ22 /4

(7.4)

As = π Φs2 /4

(7.5)

where 2 is the outer diameter of suction pad, mm2 , as known from Table 3.6, 2 = 20 mm; s is the diameter of actual sucking area, mm2 . Meanwhile, as shown in Fig. 7.1, the difference between the suction force F s and the axial tensile force F p leads to a compression deformation of the suction pad, expressed as 10−3 |pu | · As − F p = k p z

(7.6)

where k p is the compression stiffness coefficient of the suction pad (N/mm) and z is its axial compression deformation (mm). According to compression test results of the suction pad, k p = 0.406. According to Eq. (7.6), the axial compression deformation of the suction pad, z, will recover gradually with increasing axial tensile force F p . As a result, the contact area between the suction pad and the spherical surface will decrease and the suction area will conversely increase until it becomes of effective size. Then, the relation between z and Φ s is given as Φs = Φe −

z (Φe − Φ1 )(0 ≤ z ≤ z 0 ) z 0

(7.7)

328

7 Modeling of the Vacuum Sucked Pulling of Tomato Fruit

where z0 is the maximum axial compression deformation of the suction pad, mm, as known from Table 3.6, z0 = 8 mm; and 1 is the inner diameter of the suction pad, mm, as known from Table 3.6, 1 = 9 mm. Solving Eqs. (7.1)–(7.3), and (7.7) gives the tensile force, suction force, and normal contact pressure related to the degree of vacuum and the size and stiffness of the suction pad: Fp =

π z |pu |[Φe − (Φe − Φ1 )]2 − k p z (0 ≤ z ≤ z 0 ) 4 × 103 z 0 Fs =

π z |pu |[Φe − (Φe − Φ1 )]2 (0 ≤ z ≤ z 0 ) 4 × 103 z 0

pc = 3.

π {Φ22

4k p z (0 ≤ z ≤ z 0 ) z − [Φe − z (Φe − Φ1 )]2 0

(7.8) (7.9) (7.10)

Normal force balance analysis of spherical surface sucked pulling

As sown in Fig. 7.2, during sucking of a spherical fruit with a vacuum suction pad, the contact pressure is applied normally to the spherical surface in contact with the sealing lip, whereas negative pressure is applied normally to the spherical surface enveloped by the sealing lip. According to symmetry, the resultant of negative pressure is located in the axial direction of the suction pad. It is given as

Fig. 7.2 Normal force balance for spherical surface

7.2 Modeling of Mechanical Behavior for Sucking with Suction Pad

329

arcsin  Φs /2R

Fs = 10 |pu |

2π R 2 sin ξ cos ξ dξ dξ

3

(7.11)

0

Then according to symmetry, the resultant of contact pressure is located in the negative axial direction of the suction pad. It is given as arcsin  Φ2 /2R

Fc = pc

2π R 2 sin ξ cos ξ dξ

(7.12)

arcsin Φs /2R

Solving Eqs. (7.1), (7.2), (7.11), and (7.12) gives the pulling force, suction force, and normal contact pressure related to the degree of vacuum and the size and stiffness of the suction pad: Fp =

π |pu | z [Φe − (Φe − Φ1 )]2 − k p z (0 ≤ z ≤ z 0 ) 3 4 × 10 z 0

(7.13)

π z |pu |[Φe − (Φe − Φ1 )]2 (0 ≤ z ≤ z 0 ) 3 4 × 10 z 0

(7.14)

Fs =

pc =

4k p z  (0 ≤ z ≤ z 0 )  z 2 π Φ22 − [Φe − z (Φ − Φ )] e 1 0

(7.15)

The results show that the change of suction force, F s , pulling force, F p, and normal contact pressure pc is same for sucking of spherical surface and for sucking of plane sucking for spherical surface.

7.2.2 Experiment on Influence Factors of Suction Force 1.

Materials and methods

The experiment was carried out in the Key Laboratory of Modern Agricultural Equipment and Technology, Jiangsu University. The material was a pink tomato fruit (diameter: 62.62 mm) picked from Zhenjiang Vegetable Based. As shown in Fig. 7.3, the self-developed damage-free hand-arm system was applied to perform the experiment. The tomato fruit was suspended on the scaffold with a string. The HF-50 digital force meter was also mounted on the mobile station of the AEL electric single-column test stand. The fruit was gripped by a self-made transparent plastic clamp which is connected to the digital force meter. The manipulator was adjusted to make the center of the vacuum suction pad, the digital force meter, and the fruit on the same horizontal line.

330

7 Modeling of the Vacuum Sucked Pulling of Tomato Fruit

Fig. 7.3 Suction experiment of tomato fruit with suction pad

(1)

2.

Install the 2.5-fold bellows suction pad with a nominal diameter of 20 mm, start the air compressor and open the suction valve of the vacuum generator. Start the motor to drive the suction pad forward and suck and pull the fruit back at the speed of 2 mm/s until the suction pad detaches from the fruit surface. Record the whole process in real time with a Sony T10 digital camera, and find out readings of the vacuum pressure sensors and the digital force meter at the detaching moment by frame playing. Start the air compressor and open the suction valve of the vacuum generator until the vacuum pressure sensor shows a stable value. Replace the 2.5-fold bellows suction pads with a nominal diameter of 20, 14, and 9 mm, respectively, to start the motor to drive the suction pad to move forward at the speed of 2 mm/s. When there is a jump of the readings of the vacuum pressure sensor indicating that the suction pad has successfully sucked the fruit, stop the suction pad and record the stable vacuum pressure after the jump. Start the motor again to pull the fruit back with the suction pad at speed of 2 mm/s until the fruit was detached from the suction pad. The peak of the pull-off force was measured by the HF-50 digital force meter. For suction pad of each diameter, the experiment was repeated 30 times. Results and analysis

(1)

The influence of vacuum degree on pull-off force

(2)

It was found by experiment that when the vacuum negative pressure is higher than − 52 kPa (lower vacuum degree), there is a good linear relationship between vacuum negative pressure and pull-off force (Fig. 7.4a), and its linear fitting equation is as [Fs ] = 0.1984|pu | + 0.943

(7.16)

7.2 Modeling of Mechanical Behavior for Sucking with Suction Pad

(a) Linear fitting

331

(b) Guadratic curve fitting

Fig. 7.4 Relation curve between negative pressure and suction force

The goodness of fit, R2 , is up to 0.9933. At a higher vacuum degree, the data points are more discrete. In the total negative pressure range, there is a quadratic curve relation between vacuum negative pressure and pull-off force (Fig. 7.4b): [Fs ]= − 0.002|pu |2 + 0.3272|pu | − 0.9489

(7.17)

The goodness of fit, R2 , is up to 0.9854. (2)

The influence of diameter of the suction pad on pull-off force

It is found in the experiment that when the vacuum negative pressure is −13.6 kPa, the average pull-off force of suction pads of nominal 20, 14, 9 are 3.23, 1.58, and 0.58 N, respectively, and their ratio of the pull-off force is 5.58:2.73:1 (Fig. 7.5). The average relative errors of repeated measurements of suction pads of nominal 20, 14, 9 were 1.86%, 2.31%, and 1.67%, respectively, indicating that the repeated performance of the suction cup was stable.

Fig. 7.5 Relation between suction force and diameter of the suction pad

332

7 Modeling of the Vacuum Sucked Pulling of Tomato Fruit

7.2.3 The Effect of Fruit Surface Contour on Pull-off Force 1.

Difference between actual and theoretical pull-off force

The theoretical pull-off force of suction pads of different diameters can be obtained by Eq. (7.1). The theoretical pull-off force and the actual average pulling force measured in the tomato suction experiment, when the vacuum negative pressure is −13.6 kPa, are shown in Table 7.1. It is found from Table 7.1 that the actual pull-off force is slightly higher than the theoretical pull-off force. There is a certain difference between the pull-off force calculated from the theoretical model and the actual measured value. From Fig. 7.5, it is found that the actual pull-off force is higher than the theoretical value at a lower vacuum degree (Pu > −54 kPa). While the actual pull-off force is lower than the theoretical value and the data points are relatively discrete at the higher vacuum degree (Pu < −54 kPa). The possible reasons may be related to the adhesion property of peel, the impact on the digital force meter, and the leakage between tomato fruit and the suction pad. 2.

The effect of the shape of the fruit surface

The above pull-off force in Table 7.1 is the result of vacuum sucked pulling on the same fruit at the same location. In fact, the tomato fruit is not a regular ball, its surface shape changes. And the surface shape of a few fruits, especially malformed fruit, changes very much (Fig. 7.6). In the process of sucked pulling with suction pad, as there is an obvious difference in the radius of curvature of different parts of the fruit surface, even the inner concave, and the suction will be seriously affected. As the experiment shown in Fig. 7.6, the pull-off force at position 3 is relatively large and very stable at the same vacuum degree since the radius of curvature of the outward curve at position 3 is larger. While at position 1, because of the obvious change of the inner–outer curve and the radius of curvature, there is a larger change in the pull-off force. And as the presence of obvious inner concave at position 2, it is impossible to form a closed volume between the suction pad and the fruit epidermis, and the failure of suction happened (Fig. 7.7). It is found that the minimum radius of curvature of the workpiece is different for different sizes of vacuum suction pad (Table 7.2). The adaptability of the smaller Table 7.1 The theoretical and real pull-off force for different diameters of suction pad Nominal diameter, 2 (mm)

Effective diameter, e (mm)

Theoretical pull-off force (N)

Actual pulling force (N)

Rate of theoretical pull-off force to actual pulling force

20

16

2.73

3.23

1.18

14

11

1.29

1.58

1.22

9

7

0.52

0.58

1.11

7.2 Modeling of Mechanical Behavior for Sucking with Suction Pad

(a) Tomato fruit

333

(b) Contour of tomato fruit

Fig. 7.6 Contour extraction of tomato fruit

Fig. 7.7 The effect of contour shape of tomato fruit on the pull-off force

Table 7.2 Minimum radius of curvature of workpiece permitted by vacuum suction pad Nominal diameter of suction pad, 2 (mm)

50

30

20

14

9

Minimum radius of curvature of workpiece R (mm)

75

35

30

15

10

suction pad to the change of tomato fruit surface shape is stronger than that of the larger diameter sucker. Therefore, although the larger suction pad can greatly increase the effective suction area, the ability to adapt to the change of the shape of the fruit surface drops rapidly and the failure probability of suction can be greatly increased. Considering both the pull-off force and shape adaptability of suction pads, the diameter of 20 mm is an ideal choice to meet the needs of tomato fruit sucked pulling.

334

7 Modeling of the Vacuum Sucked Pulling of Tomato Fruit

7.3 Mechanical Model of Vacuum Sucked Pulling 7.3.1 Kinematic and Force Balance Analyses of Pulling of On-plant Fruit with Suction Pad 1.

Sucked pulling direction

Vacuum sucked pulling can isolate the target fruit from the cluster, which is helpful to obtain sufficient space and avoid damage to other fruits in the same cluster for subsequent grasping or cutting. When the on-plant tomato fruit is pulled away, the stem will simultaneously bear both tensile and torsional loads. As shown in Fig. 7.8, in up dip pulling, the weight of the fruit will have more influence on the suction and increase the risk of unexpected fruit detachment from the suction pad and fall to the ground. While in down dip pulling, tensile load instead of bending load will be mainly applied to the stem, which will lead to unexpected detachment in very limited movement distance. Therefore, a horizontal direction was selected in the assistant sucked pulling of on-plant fruit with the suction pad. 2.

Fruit displacement in vacuum sucked pulling

As shown in Fig. 7.8, the fruit-stem system may be assumed as being a combined rigid object that is connected to the rigid stalk through a spring hinge at point A. When the on-plant fruit is sucked and pulled away along the horizontal direction, the fruit-stem system will rotate around point A, and the movement distance of the center point O of the fruit in the x and y directions can be obtained by the following geometric relationships:

Fig. 7.8 Kinematic and static force analyses of fruit sucked pulling conditions in robotic harvesting

7.3 Mechanical Model of Vacuum Sucked Pulling

335

x = (L + R)(sin α − sin α0 )

(7.18)

y = (L + R)(cos α0 − cos α)

(7.19)

where x and y are the horizontal and vertical movement distances, respectively, of the fruit center during sucked pulling (mm); L is the length of the entire stem (mm); α is the angle between the stem and negative vertical direction (rad); and α 0 is the initial angle between the stem and negative vertical direction (rad).

7.3.2 Static Analysis of Pulling of On-plant Fruit with Suction Pad 1.

Force balance analysis

When the on-plant fruit is pulled away by the suction pad along the horizontal direction, the suction force F s and the contact force Fc are applied to the spherical fruit, the result of which is the axial tensile force F p , according to Eq. 7.5. As shown in Fig. 7.8, when the mass of the stem is ignored, it is necessary and sufficient that both the resultant force and the resultant moment of these forces (moments) are zero for a coplanar system of forces (moments) to be in equilibrium:   

Fx = 0 → Fr sin α + Ft cos α = F p

(7.20)

Fy = 0 → Fr cos α = G + Ft sin α

(7.21)

M A = 0 → F p [(L + R) cos α + y] = G(L + R) sin α + M

(7.22)

where F x is the horizontal force applied to the fruit-stem system (N), F t is the tangential force applied to the stem at point A (N), F r is the radial force applied to the stem at point A (N), F y is the vertical force applied to the fruit-stem system (N), G is the weight of the fruit (N), M A is the moment applied to the stem at point A (mNm), and M is the torque applied to the stem by the virtual spring hinge at point A (mNm). Solving Eqs. (7.19) and (7.22) gives Fp =

G(L + R) sin α + M (L + R) cos α0

(7.23)

336

7 Modeling of the Vacuum Sucked Pulling of Tomato Fruit

where the torque M is directly proportional to the rotation angle between the stem and the stalk in the elastic range: M = k0 (α − α0 )

(7.24)

Further, the angle between the stem and the stalk, α, can be obtained from Eq. (7.20) as: 

x + sin α0 α = arcsin L+R

 (7.25)

Finally, the tensile force in the process of sucked pulling of the on-plant fruit can be obtained by substituting Eqs. (7.20), (7.25) into Eq. (7.23) as Fp = 2.

x Gx + G(L + R) sin α0 + k0 [arcsin( L+R + sin α0 ) − α0 ]

(L + R) cos α0

(7.26)

Parameter identification

(1)

Physical properties of tomato fruit-stem system

Table 7.3 shows the physical properties of tomato fruit-stem system in the tests, where weight (G) is converted from mass (m). Figure 7.9 gives the frequency distribution curves for 100 samples of tomato for the initial angle between stem and negative vertical direction α0 , weight of fruit (G), length of entire stem (L), and radius of fruit (R) at the class intervals of 0.1 rad, 0.5 N, 20 mm, and 5 mm, respectively. Each frequency distribution curve of the dimensions shows a trend towards a normal distribution (Fig. 7.9). We may find from Table 7.3 and Fig. 7.9 that the frequency distribution of radius of fruit (R) is more concentrated, while that of length of the entire stem (L) is more decentralized due to the different standard deviation. Table 7.3 Measurement results of physical properties of tomato fruit-stem system Max.

Mean

Min.

Std. Dev.

α 0 (rad)

0.52

0.34

0

0.13

G (N)

3.97

1.91

0.75

0.60

L (mm)

152.74

77.68

25.40

29.73

R (mm)

52.33

35.66

22.95

4.12

7.3 Mechanical Model of Vacuum Sucked Pulling

337

Fig. 7.9 Probability distribution curve for different physical parameters

3. (1)

Compression experiment of suction pad Materials and methods

A 2.5-fold round bellows suction pad (FSG 20, Schmalz Co., Ltd., Glatten, Germany) was applied. The suction pad was made out of silicone. According to the product selection guide, the outer diameter (2 ), effective diameter (e ), and inner diameter (1 ) of the suction pad were 20, 16, and 9 mm, respectively. Meanwhile, axial length and rated maximum axial compression deformation z0 were 27.1 and 8 mm, respectively. To obtain the compression stiffness coefficient of the suction pad, kp , the suction pad was firstly mounted on the front end of a rack that could be driven back and forth by a DC motor (RE-max 24, Maxon Motor Ag, Sachseln, Switzerland) (Fig. 7.10), the HF-50 digital force meter was also mounted on the mobile station of the AEL electric single-column test stand, and a thin aluminum plate was pasted on the flat

Fig. 7.10 Compression experiment of suction pad

338

7 Modeling of the Vacuum Sucked Pulling of Tomato Fruit

Fig. 7.11 Relation between compression force and deformation of suction pad

top of the force meter. Then, the suction pad was adjusted such that it was aligned along the horizontal centerline of the force meter, and then the flat top was moved slowly until the surface of the thin aluminum plate touched the lip of the suction pad. Subsequently, the initial position from the scale plate on the test stand was recorded. Finally, the flat top was moved gradually until its movement distance was 8 mm, and the corresponding force in the force meter was recorded at displacement intervals of 0.1 mm. (2)

Experiment results

The experimental results indicate that there is a good linear relation between the axial compression force F A and the deformation of the suction pad provided the deformation was less than 7.2 mm. This relation is expressed as FA = 0.406z (0 ≤ z ≤ 7.2 mm)

(7.27)

As shown in Fig. 7.11, a very good agreement is obtained between test data and model predictions. However, in this experiment, the positioning of initial touch was decided by monitoring the displayed force data of the digital force meter and moving it carefully. So, the initial position error, which might be a delay, was inevitable mainly due to the resolution of the force meter. As a result, the pad was too compressed to surpass the elastic limit and the compression force increased sharply when the axial deformation of the suction pad exceeded 7.2 mm.

7.3.3 Discussion 1.

Mechanical relation and corresponding suction pad deformation and movement distance difference

7.3 Mechanical Model of Vacuum Sucked Pulling

339

Fig. 7.12 Relation between suction force and tensile force and deformation of suction pad (Pu = − 60 kPa)

According to Eq. (7.4), deformation of the suction pad, z, is caused by the difference between the suction force (F s ) and the tensile force (F p ). To observe the mechanical relation and corresponding suction pad deformation, substituting specification parameters of suction pad, including z0 , e , 1 , and k p into Eqs. (7.13) and (7.14), relation curves between F p and z, F s, and z, were derived, respectively (Fig. 7.12). As shown in Fig. 7.12, the largest deformation of the suction pad will occur at the beginning of the sucked pulling away of the fruit. This initial deformation will lead to an initial movement of the fruit, x a , and corresponding initial tensile force Fp0 applied to it according to Eq. (7.26). However, x a is just not enough especially to isolate the tomato fruit from the cluster. And then the tensile force (F p ) will increase continually with the pulling away of the fruit from the plant according to Eq. (7.26), and simultaneously the suction force (F s ) will increase more slowly according to Eq. (7.11). As a result, the deformation of the suction pad will recover gradually. The gradual restoration of deformation of the suction pad upon pulling away of the fruit will result in a difference in the movement distances of the fruit and the suction pad, expressed as s = x − z (z ≥ 0)

(7.28)

where s is the movement distance of the suction pad, which is a controlled variable in robotic harvesting (mm). As soon as the tensile force (F p ) becomes equal to the suction force (F s ), the axial deformation z of the suction pad will reach zero and the lip of the suction pad will lose contact with the surface of the fruit. The suction force F s at this detaching moment reaches the detachment force (F s ). According to Eqs. (7.11), (7.2), and (7.13): [Fs ] = F p |z=0 =

π |pu | 2 Φ 4 × 103 e

(7.29)

340

7 Modeling of the Vacuum Sucked Pulling of Tomato Fruit

In conclusion, to pull an on-plant fruit away in the horizontal direction, F p should be less than F s within a certain movement distance, otherwise, the fruit will detach from the suction pad unexpected. 2.

Permissible Movement Distance of Fruit in Sucked pulling Operation

The permissible maximum movement distance of fruit in the horizontal direction is limited by both geometrical and mechanical factors. It has been found in experiments that three phenomena will occur simultaneously with the position change of the fruit in horizontal sucked pulling. Firstly, sealing lip of the suction pad will slip along the fruit surface. Secondly, the sealing lip will incline to fit the fruit surface. Thirdly, the axis of the suction pad will bend to maintain a connection between the sealing lip and fruit surface. Ignoring axis bending of the suction pad, as soon as the center point of the fruit exceeds the effective diameter of the suction pad in the vertical direction, the suction pad cannot suck the fruit anymore (Fig. 7.13). This is expressed as ymax =

Φe 2

(7.30)

where ymax is the permissible maximum movement distance of the fruit in the vertical direction (mm). Therefore, the geometric permissible maximum movement distance of the fruit, which is decided by geometric constraints, can be acquired from the geometrical relation in Fig. 7.13, as Fig. 7.13 Theoretical maximum movement distance of fruit

7.3 Mechanical Model of Vacuum Sucked Pulling

 [x g ] = (L + R)( 1 − (cos α0 −

341

R − Φe /2 2 ) − sin α0 ) L+R

(7.31)

where [x g ] is the geometric permissible maximum movement distance of the fruit in the horizontal direction (mm). Secondly, according to Eq. (7.26) and Fig. 7.12, mechanical constraints that the tensile force (F p ) cannot exceed detachment force (F s ) will decide the mechanical permissible maximum movement distance of the fruit ([x m ]). Both the geometric and the mechanical permissible maximum movement distances must satisfy the need for fruit robotic harvesting. Through our calculations of Eqs. (7.26) and (7.31), we can conclude the following. (1)

(2)

(3)

With increasing stem length, both the geometric and the mechanical permissible maximum movement distances of the fruit ([x g ] and [x m ], respectively) will increase, which indicates that fruit with a longer stem can be sucked and pulled away more easily (Fig. 7.14). Meanwhile, [x m ] is affected much more easily by the length of the entire stem, so a higher degree of vacuum is necessary to suck and pull away a fruit with a shorter stem. With increasing fruit size, [x g ] increases, whereas [x m ] first increases and then decreases; the latter trend is attributed to the effect of the change in the weight of the fruit with a change in its size (Fig. 7.15). It then appears that the probability of failure of the sucked pulling operation is higher for oversized or undersized fruits. To suck ideally spherical fruits, the larger the size of the suction pad, the smaller will be [x g ] but the larger will be [x m ] (Fig. 7.16). Meanwhile, the minimum radius of curvature of the object is an important parameter to a

Fig. 7.14 Effect of length of entire stem on permissible movement distance (Pu = −50 kPa, R = 30 mm, G = 2 N, α 0 = 20°)

342

7 Modeling of the Vacuum Sucked Pulling of Tomato Fruit

Fig. 7.15 Effect of radius of fruit on permissible movement distance (Pu = −50 kPa, L = 50 mm, α0 = 20°)

Fig. 7.16 Effect of diameter of suction pad on permissible movement distance (Pu = −50 kPa, R = 30 mm, L = 50 mm, G = 2 N, α 0 = 20°)

certain suction pad, so the minimum size of the suction pad must be decided according to statistical data of fruit size of a certain variety. Therefore, the size of the suction pad should be decided carefully to ensure a successful fruit sucked pulling operation.

7.3 Mechanical Model of Vacuum Sucked Pulling

(a) Probability density

343

(b) Cumulative probability

Fig. 7.17 Distribution curve of static tensile force for various sucked pulling distances

3. (1)

Tensile force in low-speed and high-speed operations Normal distribution of static tensile force for various sucked pulling distances

Low-speed pulling motion with suction may be regarded as a static process. Substituting series of test data (100 samples) of each physical and mechanical parameter α 0 , G, L, R, and k 0 arranged randomly into Eq. (7.26), and different series of data of static tensile force F p were obtained when sucked pulling distance x was set as 5, 10, 15, 20, 25, 30 mm, respectively. The calculated result indicated that all probability density curves of static tensile force F p were not symmetrical (Fig. 7.17a). The cumulative probability represents the success rate of fruit sucked pulling for certain static tensile force F p (Fig. 7.17b). We find that a larger static tensile force (F p ) is necessary to achieve a higher success rate of pulling away an on-plant fruit to a certain distance and that a larger F p is necessary even for achieving a satisfactory success rate of pulling the on-plant fruit further away. A tensile force of 3.0 N can ensure a 98% probability of successful low-speed sucked pulling of on-plant tomato fruit to a distance of 30 mm. According to Eq. (7.15), when the compression deformation of the suction pad, z, is restored to 0, the pressure on the contact surface, pc , is changed to 0, and the suction pad is detached from the fruit. According to Eq. (7.26), the threshold of vacuum degree to satisfy the required sucked pulling distance, [pu ], is as [pu ] = −

x 1000{Gx + G(L + R) sin α0 + k 0 [arcsin( L+R + sin α0 ) − α0 ]}

64π(L + R) cos α0



pu0 > −89.4 kPa (7.32)

As shown in Fig. 7.18, the higher the sucked pulling distance, the higher the vacuum degree and success rate of sucked pulling. On the contrary, the larger the

344

7 Modeling of the Vacuum Sucked Pulling of Tomato Fruit

Fig. 7.18 Normal distribution of vacuum pressure thresholds for various sucked pulling distances

sucked pulling distance, the higher the vacuum degree required to meet a certain success rate of sucked pulling. However, if the size of tomato is not smaller enough, just a pulling force reaches 3.2 N and the vacuum reaches 16% can guarantee that almost 100% of the tomato fruit can be successfully pulled back to 30 mm. (2)

Negative pressure threshold in high-speed operation of fruit sucked pulling

In high-speed operation, a much larger positive acceleration in the initial phase is necessary to attain a higher speed in more limited time, and a much smaller negative acceleration in the terminal phase is similarly necessary. F p (x) = F p ± F p (a(t))

(7.33)

where F p  (x) is the tensile force in high-speed operation (N), a(t) is the acceleration of fruit in the horizontal movement direction (mm/s2 ), and F p (a) is the extra tensile force required to obtain this acceleration a(t), which was equal to the product of acceleration a and mass of the fruit m (N). According to Newton’s Second Law, the extra tensile force F p (a) was proportional to the acceleration. Take the constant acceleration and deceleration motion, for example, the tensile force in high-speed operation F p  (x) may be expressed as ⎧ ⎨ F p +ma0 F p (x)= F p ⎩ F p − ma0

(7.34)

where a0 is an absolute value of the constant acceleration (mm/s2 ). To avoid unexpected detachment in high-speed operation, the tensile force F p  (x) cannot exceed the detachment force [F s ] F p (x) ≤ [Fs ]

(7.35)

7.3 Mechanical Model of Vacuum Sucked Pulling

345

Fig. 7.19 Negative pressure threshold in fruit sucked pulling (R = 30 mm, L = 50 mm, G = 2 N, α0 = 20°)

So, it is necessary to supply the vacuum higher enough. According to Eq. (7.29), the threshold of negative pressure per area Pu may be expressed as |[Pu (x)]| =

4[Fs ] π Φe2

(7.36)

where [Pu (x)] is the threshold of the negative pressure per area relative to the horizontal movement distance x of the fruit (kPa). Solving Eqs. (7.26), (7.34), and (7.36), the theoretical curve of threshold of the negative pressure per area [Pu (x)] during constant acceleration and deceleration motion of horizontal fruit moving was obtained (Fig. 7.19). It indicates that in the initial acceleration phase, though the static tensile force (F p ) is much smaller, the tensile force F p  (x) and the corresponding threshold of the negative pressure per area [Pu (x)] in high-speed operation may be large enough. Undoubtedly, neglecting this special law of tensile force in high-speed operations will lead to an unexpected detachment of the fruit from the suction pad.

7.4 Probability Model of Sucked Pulling of On-plant Tomato Fruit [1] 7.4.1 Rate of Interference and Success of Fruit Gripping There are usually 3–5 tomato fruits in each cluster before harvesting (Fig. 7.20). The function of vacuum suction system is to avoid disturb to the finger grip motion from the adjacent fruit in the same cluster by pulling back the target fruit to a certain distance and to avoid the bruising to the adjacent fruit by the fingers. The key to the

346

7 Modeling of the Vacuum Sucked Pulling of Tomato Fruit

(a) Three fruits

(b) Four fruits

(c) Five fruits

Fig. 7.20 The tomato fruit cluster

success of the harvesting is whether the sucked pulling distance is large enough to avoid the collision between fingers and neighboring fruits in gripping. However, the larger the sucked pulling distance, the larger the suction force and the pulling force. It means not only higher vacuum degrees and higher energy consumption of the vacuum system, but also greater energy consumption for the motor. Meanwhile, the accident of the unwilling detachment between the fruit and the suction pad might happen. Therefore, it is critical to determine the appropriate sucked pulling distance according to the actual needs to ensure a higher success rate and lower energy consumption. For this reason, to define the rate of gripping interference as the interference rate of the adjacent fruit in the same cluster to the fingers of end-effector in gripping, and to define the success rate of gripping is the ratio of successful gripping of the fingers to avoid the interference. The success rate of gripping is a comprehensive characterization of successful sucked pulling and resulting in successful avoidance of gripping interference.

7.4.2 The Proportion of Fruit Number Per Cluster for Different Harvesting Rounds Because of the difference in maturity time of tomato fruit, several rounds of harvesting are usually required for each generation. The more the number of fruits per cluster, the more obstacles to the gripping, and the larger the difficulty of successful gripping. In different harvesting rounds, the proportion of fruit number per cluster is constantly changing, and the difficulty of successful gripping is also changing. The study begins with the following assumptions: (1) (2)

The probability of three fruits, four fruits, and five fruits grown in each tomato cluster is same. The probability of being harvested of each fruit is same.

7.4 Probability Model of Sucked Pulling of On-plant Tomato Fruit [1]

1.

347

The proportion of fruit number per cluster for the first harvesting round

According to the above assumption conditions, the probability of three fruits, four fruits, and five fruits in each cluster before the start of the first harvesting round are all 1/3. 2.

The proportion of fruit number per cluster for the second harvesting round

With the maturity and being harvested of tomato fruit, the proportion of one fruit and two fruits in each cluster appears and increases gradually, while the proportion of five, four, and three fruits changes continuously: (1)

The probability of five fruits in each cluster before the start of the second harvesting round is 1 1 · = 5.6% 3 6

(2)

The probability of four fruits in each cluster is the sum of the probability of maintaining four and five fruits reduced to four fruits after the first harvesting round 1 1 1 1 · + · = 12.2% 3 5 3 6

(3)

Similarly, the probability of three fruits in each cluster is the sum of three fruits, four fruits reduced to three,and five fruits reduced to three fruit after the first harvesting round 1 1 1 1 1 1 · + · + · = 20.6% 3 4 3 5 3 6

(4)

Similarly, the probability of two fruits in each cluster is the sum of three fruits reduced to two fruits, four fruits reduced to two fruits, and five fruits reduced to two fruits 1 1 1 1 1 1 · + · + · = 20.6% 3 4 3 5 3 6

(5)

Similarly, the probability of one fruit in each cluster is the sum of three fruits reduced to one fruit, four fruits reduced to one fruit, and five fruits reduced to one fruit 1 1 1 1 1 1 · + · + · = 20.6% 3 4 3 5 3 6

348

7 Modeling of the Vacuum Sucked Pulling of Tomato Fruit

(6)

Similarly, the probability of zero fruit in each cluster is the sum of three fruits reduced to zero fruit, four fruits reduced to xero fruit, and five fruits reduced to zero fruit 1 1 1 1 1 1 · + · + · = 20.6% 3 4 3 5 3 6

To conclude, in fruit clusters that have not been finished harvesting after the first harvesting round, the proportion of one–three fruits, four fruits, and five fruits are as follows, respectively: (1)

one fruit, two fruits, and three fruits (respectively): 20.6% = 25.9% 1 − 20.6%

(2)

four fruits: 12.2% = 15.4% 1 − 20.6%

(3)

five fruits: 5.6% = 7.0% 1 − 20.6%

3.

The proportion of fruit number per cluster for third harvesting round

Similarly, the proportion of three fruits, four fruits, and five fruits in each cluster before the start of the third harvesting round are as follows: (1)

five fruits: 7.0% ·

(2)

four fruits: 7.0% ·

(3)

1 1 + 15.4% · = 4.2% 6 5

three fruits: 7.0% ·

(4)

1 = 1.2% 6

two fruits:

1 1 1 + 15.4% · + 25.9% · = 10.7% 6 5 4

7.4 Probability Model of Sucked Pulling of On-plant Tomato Fruit [1]

349

Table 7.4 The proportion of fruit number per cluster for different harvesting rounds Harvesting round 1

Single fruit 0

Double fruits 0

Three fruits

Four fruits

Five fruits

33.3%

33.3%

33.3%

2

25.9%

25.9%

25.9%

15.4%

7.0%

3

47.7%

28.5%

15.8%

6.3%

1.7%

4

63.5%

24.5%

9.0%

2.5%

0.5%

5

74.7%

19.2%

4.9%

1.0%

0.1%

6

82.5%

14.4%

2.7%

0.4%

0.05

7.0% · (5)

1 1 1 1 + 15.4% · + 25.9% · + 25.9% · = 19.3% 6 5 4 3

one fruit: 7.0% ·

1 1 1 1 1 + 15.4% · + 25.9% · + 25.9% · + 25.9% · = 32.3%% 6 5 4 3 2

Furtherly, the probability of finishing harvesting in each cluster is 32.3%, too. Therefore, before the start of the third harvesting round, the proportion of 1–5 fruits are as follows, respectively: (1) (2) (3) (4) (5)

1.2% five fruits: 1−32.3% = 1.7% 4.2% four fruits: 1−32.3% = 6.3% 10.7% = 15.8% three fruits: 1−32.3% 19.3% two fruits: 1−32.3% = 28.5% 32.3% = 47.7% one fruit: 1−32.3%

The proportion of different fruit numbers per cluster can be obtained in the same way. The proportion of fruit number per cluster for different harvesting rounds is shown in Table 7.4.

7.4.3 The Required Sucked Pulling Distance and Its Probability for Different Fruit Number in Each Cluster When there is only one fruit in each cluster, the gripping process does not produce interference with adjacent fruit. The 2–5 fruits per cluster can be simplified as a planar problem that equators of each fruit in the same cluster are located on the same plane. In order to ensure that the fingertips do not touch the adjacent fruit when the target fruit is gripped, the required sucked pulling distance must be determined according to the known finger size and the size range of fruit.

350

7 Modeling of the Vacuum Sucked Pulling of Tomato Fruit

Fig. 7.21 Dimension parameter relation for two-fruit coplanar

(1)

Two-fruit coplanar

According to Fig. 7.21, xmin = a − c

(7.37)

where a is the distance from the centroid of the fruit to the nearest point of the adjacent fruit after sucked pulling, mm; c is the distance from the centroid of the fruit to the nearest point of the adjacent fruit before sucked pulling, mm; x min is the theoretical minimum sucked pulling distance, mm. When the fruit is sucked and pulled until its centroid reaches the gripping center of the arc finger surface, the fingers complete the gripping of the fruit. When the fingertips reach the adjacent fruit along the symmetric centerline, that is, when the distance between the fingertip and the gripping center along the symmetric centerline is just equal to a, the required sucked pulling distance of the target fruit is the minimum. According to the designed parameters, the distance between the fingertip and the gripping center along the symmetric centerline is known to be 26.3 mm or a = 26.3 mm.

7.4 Probability Model of Sucked Pulling of On-plant Tomato Fruit [1]

351

When two fruits in the same cluster is on the same plane, c is equal to the radius of the target fruit. It is known that the radius of 95% of tomato fruit is between 25 and 45 mm, and the minimum value of the theoretical minimum sucked pulling distance x min is −18.7 ~ 1.3 mm. Therefore, to two-fruit coplanar, sucked pulling distance smaller to only 1.3 mm can meet the needs of successful gripping. In the harvesting of double fruit per cluster, if the required is considered as a uniform distribution, the probability of the gripping interference, p2 , due to the smaller actual sucked pulling distance, x 0 , is as  p2 = (2)

1.3−x0 1.3+18.7

(0 ≤ x0 < 1.3) 0 (x0 ≥ 1.3)

(7.38)

Three-fruit coplanar

As shown in Fig. 7.22, according to geometric relations, for three-fruit coplanar, arcsin

R2 + c (R1 + R)2 + (R2 + R)2 − (R1 + R2 )2 R1 + c + arcsin + arccos =π R1 + R R2 + R 2(R1 + R)(R2 + R)

(7.39) The theoretical required sucked pulling distance for 95% of tomato fruit is calculated. In the case of three-fruit coplanar, the limiting size of the adjacent fruit and the target fruit has a 23 = 8 combination relationship. The possible value of parameter

Fig. 7.22 Dimension parameter relation for 3-fruit coplanar

352

7 Modeling of the Vacuum Sucked Pulling of Tomato Fruit

Table 7.5 Different combinations of the limiting size of fruit and corresponding required sucked pulling distance for three-fruit coplanar No.

R1 (mm)

R2 (mm)

R(mm)

c(mm)

x min (mm)

1

25

25

25

15.30

2

25

45

25

14.66

11.65

3

25

45

45

37.60

-11.30

4

25

25

45

40.39

-14.09

5

45

45

45

32.94

-6.64

6

45

25

45

37.60

-11.30

7

45

25

25

14.66

11.65

8

45

45

25

8.62

17.68

11.00

c for each combination can be obtained by Eq. (7.39), and the xmin can be obtained by Eq. (7.37) (Table 7.5). It can be seen from Table 7.5 that only when the target fruit is smaller and the adjacent fruit is larger, it is necessary to suck the target fruit and pull a certain distance. To three-fruit coplanar, sucked pulling distance of 17.68 mm can meet the needs of successful gripping. In the harvesting of three fruits per cluster, if the required sucked pulling distance is considered as a uniform distribution, the probability of the gripping interference, p3 , due to the smaller actual sucked pulling distance, x 0 , is as  p3 = (3)

17.68−x0 17.68+14.09

0

(0 ≤ x0 ≤ 17.68 mm) (x0 ≥ 17.68 mm)

(7.40)

Four-fruit coplanar

As shown in Fig. 7.23, for four-fruit coplanar, the limiting size of the adjacent fruit and the target fruit has a 24 = 16 combination relationship. The minimum required sucked pulling distance can be obtained by the geometric relationship, but the analytical equation is more complex. In the AutoCAD environment, the geometric figures are drawn according to the limiting dimensions (Fig. 7.24), and the value of parameter c is obtained with “size measure” command, and then, a variety of possible x min can be obtained by Eq. (7.36) (Table 7.6). Taking eighth combinations as an example, the value of parameter c is determined as shown in Fig. 7.24.

7.4 Probability Model of Sucked Pulling of On-plant Tomato Fruit [1]

353

Fig. 7.23 Dimension parameter relation for 4-fruit coplanar

(a) Limit state 1

(b) Limit state 2

Fig. 7.24 The possible value of c in case of 8th combinations for four-fruit coplanar

In the harvesting of four fruits per cluster, if the required sucked pulling distance is considered as a uniform distribution, the probability of the gripping interference, p4 , due to the smaller actual sucked pulling distance, x 0 , is as

45

45

45

45

45

14

15

16

25

8

13

25

7

12

25

6

45

25

5

11

25

4

45

25

3

45

25

2

10

25

1

9

R1 (mm)

No.

45

25

45

45

25

25

45

25

45

25

45

45

25

25

45

25

R2 (mm)

45

45

25

45

25

45

25

25

45

45

25

45

25

45

25

25

R3 (mm)

45

45

45

25

45

25

25

25

45

45

45

25

45

25

25

25

R(mm)

0 ~ 33.6

11.3 ~ 37.6

11.3 ~ 37.6

26.3 ~ −7.3

15 ~ −11.3

15 ~ −11.3

6.3 ~ −14.1 59.2 ~ 17.7

−32.9 ~ 8.6

44.2 ~ 11.6

−17.9 ~ 14.7 20.0 ~ 40.4

33.4 ~ 8 44.2 ~ 11.6

−7.1 ~ 18.3 −17.9 ~ 14.7

13.4 ~ −6.6

5.3 ~ −11.3

12.9 ~ 32.9

21.0 ~ 37.6

5.3 ~ −11.3

46.3 ~ 1.2

21.0 ~ 37.6

−0.8 ~ −14.1

−20.0 ~ 25.1

46.3 ~ 24.6

−20 ~ 1.7 27.1 ~ 40.4

26.3 ~ 8 46.3 ~ 24.6

−20 ~ 1.7

x min (mm)

0 ~ 18.3

cmin (mm)

Table 7.6 Different combinations of the limiting size of fruit and corresponding required sucked pulling distance for four-fruit coplanar

354 7 Modeling of the Vacuum Sucked Pulling of Tomato Fruit

7.4 Probability Model of Sucked Pulling of On-plant Tomato Fruit [1]

355

356

7 Modeling of the Vacuum Sucked Pulling of Tomato Fruit

Fig. 7.25 Dimension parameter relation for 5-fruit coplanar

(4)

Five-fruit coplanar

As shown in Fig. 7.25, for five-fruit coplanar, the limiting size of the adjacent fruit and the target fruit has a 25 = 32 combination relationship. The value of parameter c is also obtained in the AutoCAD environment, and then a variety of possible xmin can be obtained by Eq. (7.36) (Table 7.7). Taking 7–12th combinations as an example, the value of parameter c is determined as shown in Fig. 7.26. In the harvesting of five fruits per cluster, if the required sucked pulling distance is considered as a uniform distribution, the probability of the gripping interference, p5 , due to the smaller actual sucked pulling distance, x0 , is as

7.4 Probability Model of Sucked Pulling of On-plant Tomato Fruit [1]

357

Table 7.7 Different combinations of the limiting size of fruit and corresponding required sucked pulling distance for five-fruit coplanar No.

R1 (mm)

R2 (mm)

R3 (mm)

R4 (mm)

R(mm)

cmin (mm)

x min (mm)

1

25

25

25

25

52

25 ~ 0

1.3 ~ 26.3

2

25

45

25

25

25

18.3 ~ −32.9

8.0 ~ 59.2

3

25

25

45

25

25

4

25

25

25

45

25

5

45

25

25

25

25

6

25

25

25

25

45

40.4 ~ 7.0

−14.1 ~ 19.3

7

25

45

45

25

25

14.7 ~ −56.1

11.6 ~ 82.4

8

25

25

45

45

25

9

25

45

25

45

25

10

45

25

25

45

25

11

45

45

25

25

25

12

45

25

45

25

25

13

25

25

25

45

45

40.4 ~ −2.1

−14.1 ~ 28.4

14

25

45

25

25

45

15

25

25

45

25

45

16

45

25

25

25

45

17

25

25

45

45

45

37.6 ~ −15.6

−11.3 ~ 41.9

18

25

45

25

45

45

19

25

45

45

25

45

20

45

25

45

25

45

21

45

25

25

45

45

22

45

45

25

25

45

23

25

45

45

45

25

8.6 ~ −67.9

17.7 ~ 94.2

24

45

45

45

25

25

25

45

25

45

45

25

26

45

45

25

45

25

27

25

45

45

45

45

37.6 ~ −26.9

−11.3 ~ 53.2

28

45

25

45

45

45

29

45

45

25

45

45

30

45

45

45

25

45

31

45

45

45

45

25

8.6 ~ −80

17.7 ~ 106.3

32

45

45

45

45

45

45 ~ 0

−18.7 ~ 26.3

358

7 Modeling of the Vacuum Sucked Pulling of Tomato Fruit

(a) Limit state 1

(d) Limit state 4

(g) Limit state 7

(b) Limit state 2

(e) Limit state 5

(h) Limit state 8

(c) Limit state 3

(f) Limit state 6

(i) Limit state 9

Fig. 7.26 The possible value of c in case of 7–12th combinations for five-fruit coplanar

7.4 Probability Model of Sucked Pulling of On-plant Tomato Fruit [1]

359

360

7 Modeling of the Vacuum Sucked Pulling of Tomato Fruit

7.4.4 Theoretical Influence of Required Sucked Pulling Distance on the Rate of Gripping Interference Because the single fruit does not need vacuum sucked pulling and only 1.3 mm can achieve successful gripping for two-fruit coplanar, the rate of gripping interference for 3–5-fruit coplanar at the sucked pulling distance of H is as sg(i) = w3(i) p3 + w4(i) p4 + w5(i) p5 (x0 ≥ 15 mm)

(7.43)

where sg(i) is the total rate of gripping interference after the vacuum sucked pulling in ith harvesting round, %; w3(i) , w4(i) , w5(i) is the proportion of three, four, and five fruits per cluster in ith harvesting rounds, %. The relationship between the sucked pulling distance and the rate of gripping interference is shown in Fig. 7.26. As can be seen from Fig. 7.26a, the more the fruit per cluster, the higher the probability of gripping interference, and the larger the required sucked pulling distance to avoid gripping interference. As shown in Fig. 7.27b, the effect of vacuum sucked pulling off the fruit on avoiding gripping interference is very obvious. The interference probability of direct griping in the first harvesting round is up to 70.3%, and over 30% in the second harvesting round. In contrast, only sucked pulling distance of 10 mm can reduce the rate of gripping interference in the first and second rounds to 50.8% and 19.8%, respectively. Furthermore, the rate of gripping interference rate has been reduced to 32.2% and 10%, respectively, when the sucked pulling distance reaches 20 mm. At the same time, with the increase

7.4 Probability Model of Sucked Pulling of On-plant Tomato Fruit [1]

(a) Multi-fruit coplanar

361

(b) Different harvesting rounds

Fig. 7.27 Theoretical influence of required sucked pulling distance on the rate of gripping interference

of harvesting rounds, both the average fruit number per cluster and the proportion of multiple fruits may decrease, and the phenomenon of gripping interference will decrease significantly. In fact, because of the following factors, the actual rate of gripping interference will be lower than the theoretical value: (1)

(2)

The theoretical analysis is based on the assumption that equators of each fruit in the same cluster are located on the same plane. However, in practice, the probability of multi-fruit coplanar is very small. In most cases, the equatorial plane of each fruit in the same cluster has a certain angle and distance difference, which makes the space of gripping bigger and makes the success rate of gripping larger. According to the analysis of the morphological and structural characteristics of tomato, the diameter of the fruit basically conforms to the normal distribution, that is, most of the fruit diameter distribution is near the average. Therefore, in the above analysis, the range of the limiting size should also be a normal distribution, that is, the probability of limiting sucked pulling distance is very small. The probability of gripping interference pi (i = 3, 4, 5) obtained by the simplified uniform distribution should exceed the actual probability of gripping interference.

7.4.5 Determination of Sucked Pulling Distance The larger the sucked pulling distance, the higher the success rate of gripping after pulling, but the lower the success rate of sucked pulling. As a result, the actual success rate of gripping is as sr (i) = (1 − sm ) · (1 − sg(i) ) × 100%

(7.44)

362

7 Modeling of the Vacuum Sucked Pulling of Tomato Fruit

Fig. 7.28 Curve of actual success rate of gripping vs sucked pulling distance in different harvesting rounds

where sr(i) is the actual success rate of gripping in the ith harvesting round, %; sm is the failure rate of sucked pulling, %. Therefore, the determination of the sucked pulling distance must take into account the value of sg and sm to obtain a higher actual success rate of gripping. Comprehensive solving Eqs. (7.31), (7.43), and (7.44), the curve of actual success rate of gripping vs sucked pulling distance in different harvesting rounds is obtained as shown in Fig. 7.28. It can be found that with the increase of sucked pulling distance, the actual success rate of gripping first rises to a certain level, and then the failure rate of sucked pulling caused leads to the rapid decline of the actual success rate of gripping. When the sucked pulling distance is 32–34 mm in first harvesting round, the actual success rate of gripping rises from less than 30% to the highest 80%. And when the sucked pulling distance is 29–31 mm, 29–31 mm, and 28–30 mm in the second, third, and fourth rounds, respectively, the actual success rate of gripping is up to the highest. As the proportion of multiple fruits per cluster in the earlier harvesting round is larger, the effect of sucked pulling on the actual success rate of gripping is the most significant, and the most ideal sucked pulling distance is about 30 mm. With the increase of the harvesting round, the proportion of multiple fruits drops continuously, and the sucked pulling distance of 17–18 mm has a more obvious effect on the improvement of the success rate of gripping, and the effect of further increasing the sucked pulling distance is limited. For example, the actual success rate of gripping is 78.2% in first harvesting round when the sucked pulling distance is 30 mm, which is increased by 10.4% than that of 20 mm. When the sucked pulling distance is 20 mm, the actual success rate of gripping reaches 90.0%, 96.6%, and 98.8% in the first, second, and third harvesting round, respectively. If the sucked pulling distance is increased to 30 mm, the actual success rate of gripping reaches 93.5%, 97.8%, and 99.2% in the first, second, and third harvesting round, respectively. It increased only by 3.5%, 1.2%, and 0.4% than that of 20 mm, respectively.

References

363

References 1. Liu J (2010) Analysis and optimal control of vacuum suction system for tomato harvesting robot[D]. Jiangsu University 2. Monta M, Kondo N, Ting K (1998) End-effectors for tomato harvesting robot[J]. Artif Intell Rev 12(1–3):11–25 3. Ceres R, Pons J, Jiménez A et al (1998) Design and implementation of an aided fruit-harvesting robot (Agribot)[J]. Indus Rob 25(5):337–346 4. Ling P, Ehsani R, Ting K et al (2004) Sensing and end-effector for a robotic tomato harvester[C]. In: Proceedings of the ASAE annual meeting, No. 043088 5. Shiigi T, Kondo N, Kurita M et al (2008) Strawberry harvesting robot for fruits grown on table top culture[C]. In: Proceedings of the ASABE annual meeting, No. 084046 6. Chiu Y, Yang P, Chen S (2013) Development of the endeffector of a picking robot for greenhouse-grown tomatoes[J]. Appl Eng Agric 29(6):1001–1009 7. Hayashi S, Sakaue O (1997) Basic operation of tomato harvesting system using robot: manufacture of two-finger harvesting hand with auxiliary cutting device and basic experiment for harvest[J]. National Res Inst Veg Ornam Plants Tea 12:133–142 8. Reed J, Miles S, Butler J et al (2001) AE—Automation and emerging technologies: Automatic mushroom harvester development[J]. J Agric Eng Res 78(1):15–23 9. Hwang H, Kim S (2003) Development of multi-functional tele-operative modular robotic system for greenhouse watermelon[C]. In: Proceedings of the IEEE/ASME international conference on advanced intelligent mechatronicss 2:1344–1349 10. Baeten J, Donné K, Boedrij S et al (2007) Autonomous fruit picking machine: a robotic apple harvester[C]. In: Proceedings of the 6th international conference of field service robotics, pp 531–539 11. Liu J, Li P, Li Z (2008) Hardware design of the end-effector for tomato-harvesting robot[J]. Trans Chin Soc Agric Mach 39(3):109–112 12. Liu J, Li P, Li Z et al (2010) Design and test of the vacuum suction device for tomato harvesting robot[J]. Trans Chin Soc Agric Mach 41(10):170–173, 184 13. Kondo N, Ninomiya K, Hayashi S et al (2005) A new challenge of robot for harvesting strawberry grown on table top culture[C]. In: Proceedings of the ASABE Annual International Meeting 14. Arima S, Kondo N, Nakamura H (1996) Development of robotic system for cucumber harvesting[J]. Jpn Agric Res Quart 30:233–238 15. van Henten E, Hemming J, van Tuijl B et al (2002) An autonomous robot for harvesting cucumbers in greenhouses[J]. Autonom Rob 13(3):241–258 16. Lee F, Wang J (1999) Development of robotic arm for citrus fruit harvesting[J]. J Agric Mach 8(3):1–8 17. Arima S, Kondo N, Monta M (2004) Strawberry harvesting robot on table-top culture[C]. In: Proceedings of the 2004 ASAE annual meeting 18. Liu J, Li P, Mao H (2015) Mechanical and kinematic modeling of assistant vacuum sucking and pulling operation of tomato fruits in robotic harvesting[J]. Trans Asabe 58(3):539–550 19. Caldwell D, Davis S, Masey R et al (2009) Automation in food processing[M]. Springer, Berlin Heidelberg 20. Tomizawa T, Ohba K, Ohya A et al (2007) Remote food shopping robot system in a supermarket -realization of the shopping task from remote places[C]. In Proceedings of the international conference on mechatronics and automation

Chapter 8

Fruit Detaching Methods for Robotic Damage-Free Tomato Harvesting

8.1 Summary 8.1.1 Research Significance In robot operation, reliable gripping of fruit not only overcomes the effect of fruit gravity but also ensures the success of detaching fruit from the plant. The different detaching methods have different needs for reliable gripping force so that the choice of detaching methods has an important influence on damage-free harvesting.

8.1.2 Content and Innovation (1)

(2)

The detaching mechanism and compound mechanical analysis of various nontool detaching methods in the robot picking were carried out, and the comparative demonstration and optimal selection of all non-tool detaching methods were realized based on experimental results; The theoretical and experimental research on the feasibility of non-contact laser cutting of stem was carried out, and the relationship between the cutting efficiency and the multiple factors and the optimal control mode was established, so as to provide strong support for the realization of the speedy damage-free harvesting.

© Science Press, Beijing and Springer Nature Singapore Pte Ltd. 2021 J. Liu et al., Rapid Damage-Free Robotic Harvesting of Tomatoes, Springer Tracts in Mechanical Engineering, https://doi.org/10.1007/978-981-16-1284-8_8

365

366

8 Fruit Detaching Methods for Robotic …

8.2 Theoretical and Experimental Comparison of Non-tool Fruit Detaching Methods [1, 2] 8.2.1 Non-tool Fruit Detaching Methods Undoubtedly, fruit detachment from plants is a basic and essential step either for manual or robotic harvesting, which is essential to achieve a satisfactory success ratio of harvesting. Present detaching methods may be divided into tool and non-tool types. For tool methods, a peduncle is cut with a mechanical cutter or other cutting tools. Non-tool methods refer to detaching one fruit only by wrist motion of the manipulator after one fruit is gripped by the end-effector. When picking fruits from the plant manually, as the fingers grasp the target fruit, a person tends to adopt pulling, twisting, or bending method to detach the fruit (as shown in Fig. 8.1). Therefore, the robot can be designed to harvest fruits by similar motions. Selection between a tool and not-tool type methods of fruit detaching must be based on physical and mechanical properties of peduncles, which are significantly different for various types of fruits and vegetables. Some peduncles are much easier to be cut off, such as those of strawberries and cherry tomatoes, so tool-type methods were widely adopted by various harvesting robots of these varieties of fruits and vegetables [3–9]. End-effectors for robotic harvesting of tomatoes, cucumbers, eggplants, sweet peppers, etc., also used mechanical cutters to cut peduncles [10–21]. However, the force required to cut one tomato peduncle or one orange peduncle may reach 108.5–245 N and 60–290 N, respectively [22, 23], which need a much larger actuating mechanism and may cause a significant increment of size and weight of Fig. 8.1 Different actions in tomato fruit picking

8.2 Theoretical and Experimental Comparison …

367

the end-effector. Meanwhile, since working space in tomato canopies is so narrow that cutter motion is limited, it is often difficult to reach ideal position and posture to cut one peduncle. In addition, the transportation of viruses from one plant to the other and water loss from the fruit is inevitable [24]. Recently, different thermal cutting techniques with electrodes, heating wires, or laser were adopted in peduncle cutting of cucumbers, strawberries, and tomatoes, respectively [24–27]. However, the success rate and efficiency of these thermal cutting methods need further study, whose safety and energy consumption also deserve paying attention to. Some tomato harvesting robots that have been developed [28, 29] and apple harvesting robot developed by [30–32] detached fruits by twisting them off, while other tomato harvesting robot [3] and cherry-tomato harvesting robot [33] detached fruits by bending them off. Harvesting robots for citruses and apples [34, 35] also adopted non-tool detaching methods. Since there is an abscission layer in a peduncle of the tomato, it is easier to be detached by wrist motion theoretically. Compared with the above tool-type methods, the structure of end-effectors for non-tool methods is much simpler, and cost and energy consumption are much lower, meanwhile, they are more versatile. It is essential to achieve satisfactory robotic harvesting to select a proper fruit detaching method. Selection of different wrist motion of pulling, twisting, or bending to detach a tomato fruit may be decided by many aspects that are all different significantly, such as success rate, interference with adjacent fruits and stems, fruit bruise, and payload to the manipulator, etc. However, there is still not any literature focusing on this topic, so it is very valuable to perform the selection by a comparative study of different non-tool methods.

8.2.2 Experiments of Non-tool Detaching of Tomato Fruit 1.

2.

Experimental apparatus The pull-off force and displacement of fruits from plants were measured using an electric horizontal single-column test stand (Model AEL, Ali Instrument Co., Ltd., Wenzhou, China), as shown in (Fig. 8.2). The robotic harvesting system applied during the tomato fruit twist-off and bend-off tests included a two-finger end-effector developed by the authors and a six-axis articulated manipulator (Yaskawa Motoman Sv3x) (Fig. 8.3). The manipulator positioned the end-effector during harvesting operations and the end-effector gripped the fruit and detached it through a certain wrist or elbow motion. Materials and Methods Tests of pull-off, twist-off, and bend-off of tomato fruits from plants were carried out as follows:

368

8 Fruit Detaching Methods for Robotic …

Fig. 8.2 The diagram of pull-off test

Fig. 8.3 Six-axis articulated manipulator

1)

Pull-off test of tomato fruits from plants

2)

As shown in Fig. 8.2, each tomato plant was fixed on a metal rod and 20 randomly selected tomato fruits from the mature green to the lightred stage, and every fruit was gripped by an elastic gripper, which was hooked to an HF-50 digital push/pull force gauge (measuring range: 0– 50 N, resolution: 0.01 N). The force gauge was then moved backward horizontally at a speed of 2.0 mm/s until the fruit was detached. Data of pull-off force and displacement were recorded and transmitted to a computer via RS232. Twist-off and bend-off tests of tomato fruits from plants

8.2 Theoretical and Experimental Comparison …

(1)

(2)

369

Twist-off test of tomato fruits from plants In the twist-off test of tomato fruits from plants, a fruit was gripped firmly in the horizontal direction by the end-effector and twisted off by reciprocating swinging motion. Every ten samples of tomato fruits were selected randomly for a swing angle of 45°, 90°, 180°, and 350°, respectively, at the same speed of 300°/s. The whole testing process was recorded by Sony HDR-XR100E digital video camera and the swing cycles to detach one fruit were measured by video processing. Bend-off test of tomato fruits from plants Then in the bend-off test of tomato fruits from plants, the fruit was gripped firmly in the horizontal direction by the end-effector and bent off by rotation motion. Every ten samples of tomato fruit were selected randomly for an upward rotation angle of 20°, 30°, 40°, 50° and a downward rotation angle of −30°, respectively, at the same speed of 200°/s.

3.

Results (1)

(2)

Pull-off force and displacement of tomato fruits from plants The average pull-off force and displacement of tomato fruits from plants were found to be 10.47 N and 27.3 mm, with significant variation of 1.18–22.32 N and 3.0–68.1 mm, respectively (Fig. 8.4). Pull-off force and displacement of tomato fruits were positively correlated (R2 = 0.4034). Twist-off test result of tomato fruits from plants Test results indicated that it was not easy to detach a fruit by twisting motion of the wrist, though it has been adopted more widely than bending and pulling [28–32]. Average swing cycles to detach a fruit decreased more rapidly for larger swing angle (Fig. 8.5). But the difference was significant for any swing angle, and it might be a failure to detach a fruit

Fig. 8.4 Relationship between pull-off force and displacement of tomato fruits from plants

370

8 Fruit Detaching Methods for Robotic …

Fig. 8.5 Swing cycles for different swing angles when twisting tomato fruits off from plants

(3)

even after dozens of cycles of large-angle reciprocating swing. It was found that swing cycles to detach a fruit were mainly decided by the direct relation between the twist axis and the pedicel, instead of diameter and length of the peduncle. When the twist axis was perpendicular with the pedicel, the fruit might be detached much more easily merely after little cycles. In addition, interference of nearby peduncles and fruits with the fingers and unexpected bruise in twisting motion would increase obviously for larger swing angle. Bend-off test result of tomato fruits from plants It was found in the bend-off test that the success rate of break of abscission layers was much higher for larger rotation angle (Table 8.1), while the rate of unexpected events was also higher. For example, joints between the peduncle and the plant might be torn or broken which would lead to serious

Table 8.1 Bend-off test results of tomato fruits from plants Rotation angle °

Sample number

Number of successful break

Entire length of stem, Rate of mm unexpected event, % Success Failure

Overall success rate of detaching without unexpected event, %

−30

20

19

82.5

94.0

7

60.0

20

20

14

76.7

109.8

0

70.0

30

20

18

75.2

103.6

2

80.0

40

20

20

93.2

–

2

90.0

50

20

20

74.8

–

6

70.0

Mean

20

18.2

80.3

105.1

3.4

72.9

8.2 Theoretical and Experimental Comparison …

371

damage to the whole cluster and a loss of production (Fig. 8.6a). An interference of nearby peduncles and fruits with the fingers and unexpected bruise might also occur (Fig. 8.6b). Meanwhile, the rate of unexpected events of bending downward was much higher than that of upward bending motion (Fig. 8.7). It was also found that the success rate was related to the peduncle length, and it seemed that fruits with longer peduncles were more difficult to be detached by bending.

(a) Break of the whole cluster

(b) knocking down of the fruit

Fig. 8.6 Unexpected events

(a) Upward

(b) downward

Fig. 8.7 Bending motion to detach a tomato fruit and the force application

372

8 Fruit Detaching Methods for Robotic …

8.2.3 Theory of Strength and Detachment of Abscission Layers 1.

Load–deformation relation for stems and pedicels Relation between deformation and load to break the abscission layer under different loading types is decided by its stiffness. When the stem is loaded, deformation happens in the whole stem. Relation between the load, no matter what type and corresponding deformation is as follows: P =a

δ Lb

(8.1)

where P is load applied to the stem (N or mNm), δ is the corresponding deformation of the stem (mm or rad), a is the stiffness coefficient (N or Nmm2 rad−1 ). According to Eq. (8.1), the required deformation of the whole stem to break the abscission layer are as follows, respectively:

2.

[δ] =

L b [P] a

(8.2)

[δa ] =

La [δ] Lb

(8.3)

where [P] is the strength limit of the abscission layer (N or mNm), [◿δ] and [◿δ b ] is required deformation of the whole stem to break the abscission layer, respectively (mm or rad). Fatigue strength of abscission layers When the applied load reaches the strength limit of the abscission layer; meanwhile, the deformation reaches stiffness limit, the fruit is detached. If the abscission layer was wristed off by reciprocating swinging motion, the real break of the abscission layer is own to fatigue rupture of it. According to Eq. (8.2), the real torsion on the abscission layer is related to the swinging angle: T f = at

θ Lb

(8.4)

where θ is the actual torsion angle of the whole stem that is equal to the swinging angle of the wrist (rad), T f is the corresponding torsion applied on the abscission layer (mNm), at is the torsion stiffness coefficient (Nmm2 rad−1 ). According to fatigue rupture theory, the relation between fatigue limit and fatigue cycles follows the equation of fatigue curve: [T f ]m N = C

(8.5)

8.2 Theoretical and Experimental Comparison …

373

Abscission layer

Fig. 8.8 Combined load on the abscission layer

La α

Ra Fp

3.

where [T f ] is the fatigue limit for torsion of the abscission layer (mNm), m is the power exponent (m > 1), N is fatigue cycles, C is constant. Combined loading analysis During the practical operation of the non-tool detachment of fruit, the abscission layer is usually under combined loading due to the difference of direction relation between the wrist motion and the pedicel.

Take the pulling method as an example as shown in (Fig. 8.8), when the tomato fruit is pulled off from the plant, the direction of the pulling force applied by the robot was usually not similar to that of the pedicel because the postures of different pedicels were diverse and changed during detachment. Thus, the abscission layer was under the combined loading of tension -bending, and it was broken as soon as the combined load exceeds its strength limit. As the direction of pulling force is not the same as that of the pedicel, both tension and bending load will be applied on the abscission layer. The static mechanical relation is as follows: Ma = F p (L a + R) sin α

(8.6)

Fa ≈ F p cos α

(8.7)

where F p is the pulling force applied by the wrist to a tomato fruit (N), Ma is the bending moment applied on the abscission layer (mNm), F a a is the tension load applied on the abscission layer (N), α is the angle between the direction of the applied pulling force and the pedicel (0° ≤ α ≤ 90°), R is the radius of tomato, and L a is the pedicel length. So break of the abscission layer is because the combined load exceeds its strength limit: To pull the fruit off, the actual force needed to apply on the fruit by the wrist is as follows:

374

8 Fruit Detaching Methods for Robotic …

F p (L a + R) sin α + F p cos α ≤ [Fa ] da /2

(8.8)

where [F a ] is the tension limit of the abscission layer (N) and da is the diameter of the abscission layer. So, the actual pull-off force [F p ] can be deduced: [F p ] =

da [Fa ] 2(L a + R) sin α + da cos α

(8.9)

where [F p ] is a pull-off force applied by the end-effector (N). From Eq. 8.9, we can find that a pull-off force [F p ] is always smaller than the tension limit of the abscission layer [F a ] attributed to the combined loading. Furthermore, the larger is the angle α, the larger is the difference between pull-off force [F p ] and tension limit of the abscission layer [F a ]. Similarly, to detach the fruit by twisting, the abscission layer was also under combined loading of torsion-bending in actual cases. Thus, when the angle between the axis of the twisting motion and the pedicel is small, the break of the abscission layer is mainly decided by its torsion strength limit and stiffness, whereas when the angle is larger, the break of the abscission layer is mainly decided by its bending strength limit and stiffness. Besides, to detach a fruit by bending upward or downward, the abscission layer may be under the combined load of bending-compress or bending-tension, respectively, in actual cases.

8.2.4 Discussion 1.

Load and deformation (displacement) of the stem for different wrist motion (1)

Strength and stiffness of the stem under pure loading type As shown in Table 8.2, mean stem length is 6.2 times as long as that of pedicels, so the required deformation of the whole stem to break the abscission layer is also several times larger than that of the pedicel according

Table 8.2 Physical properties of tomato fruits and stems Physical properties

Number of observations

Pedicel length (L a ), mm

100

8.02

19.53

13.17

2.38

Stem length (L b ), mm

100

38.71

161.97

81.50

26.86

Pedicel diameter, 100 mm

1.99

5.05

3.26

0.63

45.94

91.48

71.45

6.89

Fruit diameter, mm

100

Minimum value

Maximum value

Mean value

Standard deviation

8.2 Theoretical and Experimental Comparison …

(2)

2.

375

to Eq. (8.3). Problems appear to twist motion that needs several or even dozens of continuous full rotation to detach the abscission layer according to Eq. (8.3), which far exceeds motion range of the wrist. So it is necessary to wrist off the abscission layer by reciprocating swinging motion. Since the swinging angle (θ ) is far less than the required torsion angle of the whole stem to break the abscission layer, corresponding torsion (T f ) applied on it is much smaller according to Eq. (8.4). According to Eq. (8.5), as Tf is much smaller, fatigue cycles will be much larger. Twistoff test results of tomato fruits from plants proved that it was difficult to ensure successful detaching of every fruit even after dozens of cycles of the maximum-angle reciprocating swing motion. Actual loading type applied to abscission layer for different wrist motion In practical, operation of non-tool detaching of fruit, the abscission layer is usually under combined loading, which is related to the relation between the direction of different wrist motion and that of the pedicel (Table 8.3). Previous test results indicated that the mean tensile strength of the abscission layer reached 23.01 N and the maximum value reached 34.29 N, which was much higher than the pull-off force of tomato fruits from plants. The significant difference might mainly be attributed to the actual combined loading on the abscission layer. In the pull-off test of tomato fruits from plants, the direction of pulling force applied by the robot was usually not the same with the pedicel as postures of different pedicels were diverse and changing in detaching operation, so the abscission layer was under combined loading of tensile-bending and it was broken as soon as combined load exceeded its strength limit, which leads to the result that pull-off force applied by the wrist was much lower than pure tensile limit. To detach the fruit by twist motion, the abscission layer was under combined loading of torsion-bending in most cases. So when an angle between the axis of twist motion and the pedicel was smaller, break of the abscission layer was mainly decided by its torsion strength and stiffness (Fig. 8.9a). While when the angle was larger, the break was mainly decided by its bending strength and stiffness (Fig. 8.9b). Detaching effect of different wrist motion (1) Load capacity of the wrist for different non-tool detaching method It is easily concluded that driving torque needed to twist off a fruit is far less than allowable torsion of the wrist, while the driving force needed to pull off a fruit may exceed the allowable force of the wrist in most cases, so it is not applicable. To detach a fruit by bending upward, the abscission layer is under combined loading of bending-compress in most cases, so break off the abscission layer is mainly decided by its bending strength and stiffness, required driving torque is also far less than an allowable bending moment of the wrist.

Load type on abscission layer

Load diagram

Wrist motion versus pedicel

Wrist motion type

Tensile-bending

Pulling

Pure tensile

%

Pure bending

∞

Torsion-bending

Twisting

Table 8.3 Load type on the abscission layer under different types of wrist motion

Pure torsion

%

Pure bending

∞

Bending-compress

Bending

Bending-tensile

376 8 Fruit Detaching Methods for Robotic …

8.2 Theoretical and Experimental Comparison …

(a) Smaller angle

377

(b) larger angle

Fig. 8.9 Relationship between the axis of twisting motion and the pedicel during the twisting off of fruit

(2)

(3)

Operating efficiency and operability of different non-tool detaching method Test result indicates that even when swinging angle reaches the rotation limit of the joint, mean swinging cycles needed to break the stem reaches 11.6 and the largest cycle in test is as higher as 36. Assumed a rotation speed of its limit, mean and maximum time needed to twist off a fruit is 27.1 s and 84.0 s, respectively. In fact, the needed time will be even larger in view of acceleration–deceleration time delay in reciprocating motion, which is unacceptable in practical harvesting operation. To bend off a fruit, displacement of the stem is directly proportional to moment arm (L) and rotation angle. As L is nearly 350 mm, displacement is enough to break the abscission layer if the rotation angle is beyond a certain value. Unexpected events of different non-tool detaching method To detach a fruit by pulling, a large force is necessary, which is applied not only on the target fruit but also on the whole plant. Lodging of the plant if it is not supported by some stake or trellis, might happen in practical harvesting, and a loss of production and economy is inevitable. To detach a fruit by twisting or bending, corresponding torsion or bending moment are also applied to the whole cluster, which may cause a detaching of or damage to it, as shown in Fig. 8.6. In conclusion, although twisting is the most widely applied detaching method in the present robotic harvesting of tomatoes, it was found not practically applicable in tests and was further proved in theory. Meanwhile, the pull-off method is also not applicable in practical. Pulling off is a preferred detaching method in view of its comprehensive advantages.

378

8 Fruit Detaching Methods for Robotic …

3.

Rotation direction and angle optimization of bend-off detaching method (1) Load capacity of the wrist for different bending direction As known from the above analysis, the required driving torque to detach a fruit by bending upward is much smaller. While to detach the fruit by bending downward, the abscission layer is under combined loading of bending-tensile in most cases, and break of the abscission layer is mainly decided by its tensile strength and stiffness as the angle between the direction of the friction force applied by the gripper and the pedicel is much larger, which may exceed the load capacity of the wrist. (2) Success rate and unexpected events of different bending direction It was found in tests that the success rate of detaching of bending upward was similar with bending downward for same rotation angle, while the rate of unexpected events of bending downward is much higher. This is mainly attributed to two aspects. Firstly, force applied on the whole cluster of bending downward (Fig. 8.7a) by the wrist is much larger than that of bending upward (Fig. 8.7b) to detach a fruit, which is combined with weight of the whole cluster to cause damage to it. Secondly, adjacent fruits are more easily to be bruised or even knocked off when bending downward, decided by postures of stems and distribution of fruits. In view of the above disadvantages of detaching a fruit by bending downward, it is preferred to select a bending upward as the ideal fruit detaching method. (3) Success rate and unexpected events for different rotation angles As shown in Table 8.1 and Fig. 8.10, the success rate of break of bending upward was higher for larger rotation angle and fruits with longer stems were more difficult to detach because the break of an abscission layer occurred only when bending motion range of the wrist exceeds required displacement of the stem. Meanwhile, 120

Fig. 8.10 Success rate of bending off upward for different rotation angles

Success rate of break

100

(%)

80

Success rate of detaching

60 40

Rate of unexpected events

20 0 -30

20

30

40

Rotation angle (°)

50

8.2 Theoretical and Experimental Comparison …

379

unexpected events also increase for larger rotation angle, due to the larger load and motion range. As shown in Fig. 8.10, the synthesis success rate of detaching operation, which refers to the success rate of bending off upward without any unexpected events, for rotation angle between 30° and 40° may reach more than 80%. So a 30°–40° bending upward motion is more satisfying to detach fruits in robotic harvesting.

8.3 Experimental Exploration of Laser Cutting of Stems [36, 37] 8.3.1 Put Forward Laser Cutting of Stems Because of the highly unstructured agricultural operation environment, the highly different and random of the fruit distribution, posture, and size, both the success rate and the dexterity of the mechanical detaching devices have a large distance from the actual application requirements. Researchers all over the world have never stopped the exploration of the new way of fruit detachment. Van Henten E.J., Bachche S., and Zhang K.L. applied electrodes or heating wires to detach cucumber and strawberry fruit [24, 38, 39], but its adaptability was still limited. Compared with the above methods, the greatest advantage of laser cutting technology is the non-contact cutting of the object through the focusing of high-energy laser beam. In recent years, it has been widely used in the processing of metal and non-metallic inorganic materials, and it has also shown a unique advantage in the cutting of wood and other organic materials [40–42]. The application of laser to the cutting of stems can effectively avoid the limit of spatial and unstructured environment. Additionally, it is hopeful to escape the current situation that all fruit detaching devices must be specially designed for specific fruit and vegetable varieties and cultivation patterns and to provide a common fruit detaching device and method to promote the application of robotic harvesting technology.

8.3.2 The Principle and Advantages of Laser Cutting of Biomaterials [36, 43] 1.

Principle of laser cutting

Laser cutting takes advantage of the high concentration of laser beam energy. By focusing on the surface of the object, the energy of laser beam is absorbed and high temperature is instantaneously produced, thus the cutting of the object is realized. Different from industrial materials, the laser cutting of biological tissues has a

380

8 Fruit Detaching Methods for Robotic …

succession process of photoluminescence, heat conduction, and tissue response [44]. Meanwhile, the different temperature response is produced due to the difference in laser wavelength, focal spot energy density, irradiation time, and tissue characteristics, which may lead to different thermal effects, such as carbonization, ablation, gasification, and non-thermal effects, such as photoetching and photo breakdown [45–49]. The temperature rise at the center of the laser focal spot can be approximately expressed as [50] 2 × 103 Aa ρ0 T = Λ



ad t π

(8.10)

where T is the temperature rise at the center of the laser focal spot, K; Aa is the absorptivity of materials to laser energy, %; ρ 0 is the thermal power density of focal spot, W/mm2 ; Λ is the thermal conductivity of materials, W/m·K; ad is the thermal diffusivity of materials, mm2 /s. 2.

The advantages of laser application in the cutting of stem The tissue structure of the stem is complex (Fig. 8.11), which is covered by the external phloem, the primary cortex, and the endothelium. Inside the stem, there are different tissues including xylem, internal phloem, and pulp. The material of stems determines the particularity of its light-thermal response parameters, which leads to the obvious difference between its laser cutting effect and that of industrial materials. Theoretically, the application of laser to stem cutting will have obvious advantages. (1)

The thermal conductivity of most dry wood and water is between 0.1– 0.2 W/m·K and 0.6 W/m·K, respectively, while the thermal conductivity of the metal is usually between 40 and 400 W/m·K. It can be inferred

Fig. 8.11 The internal structure of the tomato stem [50]

8.3 Experimental Exploration of Laser Cutting …

381

Fig. 8.12 The water content of cucumber stems

(2)

(3)

from Eq. (8.10) that the ability of the fresh stems to make full use of laser energy to increase the incisional temperature is much stronger than that of the metal material. But the water content of grown fresh stem is high, for example, the average water content of cucumber stems is above 90% (Fig. 8.12). The higher the water content, the worse the laser cutting effect will be. Too strong mirror reflection is always the key problem that affects the laser cutting efficiency of the metal sheet. The reflectivity of the metal surface to the laser wavelength is usually above 80%, which leads to the power of the laser processing equipment at the kW level. Even the laser processing of pure copper and pure aluminum is still a difficult problem [51]. Most of the agricultural materials are made up of innumerable small internal interfaces and are optically anisotropic. When a beam of light is irradiated to the fruit, only 4% of the incident can be mirrored reflected [52], and the more rough surface of the stem can more effectively absorb the laser irradiation. In addition to lead, aluminum, and other metal materials, most of the melting point is above 1000 °C. For example, the melting point of pure iron is more than 1500 °C, and its gasification point is up to 2740 °C. By contrast, the burning point of wood is usually 250–300 °C, and its gasification begins when the temperature exceeds about 500 °C. It can be inferred that the response temperature of stems is far lower than that of most metal materials. The limited power density of the laser focal spot can realize the stem cutting, which will help the laser type selection and the implementation of laser cutting operation at lower power.

382

8 Fruit Detaching Methods for Robotic … Peduncle

Fig. 8.13 Diagram of laser beam irradiation on columnar surface Laser beam

8.3.3 Particularity and Feasibility of Laser Cutting of Stem [36, 43] At the same time, some characteristics of the stems make its laser cutting present special rules: (1)

(2)

(3)

Unlike the large-sized flat sheet object in industrial processing, the surface of the stem is a small-diameter approximate cylinder (Fig. 8.13). At present, the analysis of heat transfer and temperature field of laser irradiation is mostly based on the hypothesis of semi-infinite surface [48, 51, 53, 54], but the heat transfer mode and the distribution of temperature field of the small effect surface of the stem are very different. At the same time, when the laser beam focuses on the approximate columnar surface of the stem, if the diameter of the stem is not larger enough than that of the focal spot then the difference of the incident angle of the laser beam will appear (Fig. 8.13). The diversity and inhomogeneity of the internal tissue of the stem form a stratification effect (Fig. 8.13), which result in the difference in the laser irradiation effect and the cutting mechanism at the different cutting depth of the stem. As a result, there is a difference in the efficiency, depth, quality of laser cutting. Due to the irregularity and individual differences of agricultural materials, any research on their properties and equipment development must be based on its statistical laws to complete. The adaptability of laser cutting is required by the individual differences in size and properties of stems.

8.3.4 Experiments on Laser Drilling and Cutting of Tomato Stems 1.

Plant materials One hundred “Jinpeng 5” tomato peduncles were collected randomly from a vegetable base in Zhenjiang City, Jiangsu Province, China, and transported to the Laboratory of Modern Agricultural Equipment and Technology in Jiangsu University on 16 May 2010. All tests were carried out within 24 h at room temperature 20–30 °C. The peduncles were labeled firstly, and their diameters were measured with a micrometer caliper with sensitivity of 0.01 mm.

8.3 Experimental Exploration of Laser Cutting …

383

1. Fiber 2. Focusing lens 3. DC serve motor 4. Bearing structure

Fig. 8.14 The stem laser cutting device mounted on the harvesting robot

2.

3.

Experimental apparatus A stem laser cutting device was applied in the experiments (Fig. 8.14). A fiber-coupled laser diode (Jitai GTDC0613T, power 30 W, wavelength 980 nm, threshold current 0.55 A) and a focusing lens (Daheng GCO-2901, rate1:1; focal length 50 mm) were adopted in this device. To satisfy the need of harvesting fruits of different varieties, the tilt angle of focusing lens can be adjusted vertically between ±10°. This device is supplied with a lithium battery and controlled by a motion controller integrated into control system of the harvesting robot. In these tests, the focusing lens is fixed on a testing bed by a fastening ring that could adjust the distance between the peduncle and the focusing lens (Fig. 8.15). Two PMMA plates are mounted on a cradle whose vertical position can be adjusted through a screw bolt driven by a DC motor. Lateral position and vertical angle of the two plates can be adjusted manually. Measurement of focal length and diameter of focal spot To measure the actual position of the focal spot, a 9 mm × 30 mm thin stainless plate was mounted on one mounting plate, which was then adjusted to be targeted by the laser beam. Subsequently, the distance between the stainless plate and the focusing lens was set 30 mm, the laser was turned on in continuous mode, the optical output power was set as 15 W, and then the focusing lens was moved back and forth slowly until the laser spot on the thin stainless plate reduced to the minimum, which was observed and recorded by Sony HDR-XR100E digital

384

8 Fruit Detaching Methods for Robotic …

Fig. 8.15 Test device of laser cutting of peduncles

1. Mounting plates 2. Graduated rule 3. Focusing lens 4. Base body 5. Screw bolt 6. Cradle

4.

video camera. The focal spot was observed, whose diameter was measured by image processing and the actual focal length was measured with the graduated rule. Effect of different factors on drilling through time Since any laser cutting operation begins with drilling, a series of laser drilling tests of tomato peduncles were performed. Firstly, the thin stainless plate was removed and two ends of one peduncle selected randomly from tomato peduncles that have been labeled were mounted on the two mounting plates, whose vertical position was adjusted slowly to be targeted by the laser beam by controlling the motor. (1)

(2)

To study the effect of peduncle diameter on drilling through time, end surface of the focusing lens was kept parallel to the mounting plate and the lateral distance from the focusing lens to the near surface of the peduncle was adjusted to the actual focal length that has just been measured, and then output optical power was set as 15 W. Eighteen peduncles were selected to be tested. To study the effect of output optical power on drilling through time, the end surface of the focusing lens was kept parallel to the mounting plate and the lateral distance from the focusing lens to the near surface of the peduncle was adjusted to the actual focal length. Nine different output optical powers of 1 W, 2.75 W, 3.75 W, 5 W, 6.25 W, 10 W, 15 W, 25 W,

8.3 Experimental Exploration of Laser Cutting …

(3)

(4)

385

and 30 W were tested. The test was performed three times at different positions along the longitudinal direction of peduncle for every output optical power. Defocusing distance refers to the distance from the actual focal spot to the near surface of peduncle, which is considered zero when it is set on the near surface, and above or below the surface is considered positive and negative, respectively [55]. To study the effect of defocusing distance on drilling through time, the end surface of the focusing lens was kept parallel to the mounting plate and output optical power was set as 15 W. Nine defocusing distances of −10 mm, −7 mm, −5 mm, −3 mm, 0 mm, +2 mm, +4 mm, +6 mm, and +8 mm were tested. The test was performed three times at different positions along the longitudinal direction of peduncle for every defocusing distance. To study the effect of the incident angle of laser beams on drill through time, the lateral distance from the focusing lens to the near surface of the peduncle was adjusted to the actual focal length and output optical power was set as 15 W. Nine incident angle of 0°, 10°, 20°, 25°, 35°, 45°, 50°, 60°, 65° were tested (Fig. 8.16). The test was performed three times at different positions along the longitudinal direction of peduncle for every incident angle.

The whole process of the above tests was recorded by Sony HDR-XR100E digital video camera and the drilling through time was measured by video processing. Fig. 8.16 Test of laser cutting of peduncles for different accident angle

386

8 Fruit Detaching Methods for Robotic …

8.3.5 Results and Discussion 1.

Feasibility of laser cutting to tomato stems (1)

Relation between operation performance and parameters of focal laser beam In most industrial applications of laser cutting, cutting surface quality is vital. However, in the laser cutting process of stems, only the feasibility and efficiency of stem laser cutting should be focused on. Both the feasibility and the efficiency are related to the parameters of focal laser beam which are decided by the quality of laser beam and the properties of the focusing system (Fig. 8.17). The diameter of focal spot is essential in any laser processing which is as follows [56]: d f = 4K f F

(8.11)

where d f is the diameter of focal spot (mm), K f is the beam parameter product (BPP), which is defined as the product of beam radius and the beam divergence half-angle and may quantify the quality of a laser beam and how well it can be focused to a small spot (mm·rad), F is the f-number, which is defined as the lens focal length divided by the beam diameter at the lens and may quantify the properties of a focusing system. df is the expected to be smaller enough to get a much higher power density. The relationship between power density of the focal spot and df is as follows: ρ0 = 4kd kt ka P/π d 2f

(8.12)

where ρ 0 is the power density of focal spot (W/mm2 ), P is the optical output power of laser (W), k d is the communication efficiency ratio in air of the output optical power of a laser, k t is the transmissivity of a focusing lens, k a is the laser absorptivity on the irradiated surface. The other important parameter in laser processing is the depth of focus which is as follows (Zuo 2008):

Quality of laser beam Properties of focusing system

Diameter of focal spot Depth of focus Actual focal length

Laser cutting feasibility Defocusing distance

Laser cutting productivity

Fig. 8.17 Relation between operation performance and parameters of focal laser beam

8.3 Experimental Exploration of Laser Cutting …

387

Fig. 8.18 Diameter distribution of tomato peduncles

z R f = 8K f F 2

(2)

(3)

(4)

(8.13)

where zRf is the depth of focus (DOF), which is the distance over which the focused beam can maintain satisfactory power density (mm). zRf is the expected to be large enough to cutting or drilling through the materials of a certain depth. While from Eqs. (8.11) and (8.12), size of d f is positively related to zRf , the feasibility of stem laser cutting should be judged with the test results and calculation. Diameter of tomato peduncles Results showed that the diameter of tomato peduncles varied within the ranges of 1.99–5.05 with a mean value of 3.26 mm (Fig. 8.18). The standard deviation may reach 0.63, so it is necessary for the focused laser beam to cutting off most of the peduncles in spite of the significant differences in diameter, which is entirely different compared with standard material processing. Focal length and diameter of focal spot A focal shift is inevitable in conventional lens systems which may cause a corresponding change in the laser focal position, so the actual focal length is necessary to be measured. Measurement results indicated that it was 50 mm which meant the focal shift of lens system was small enough to be ignored. Although the diameter of focal spot could not be observed with the naked eye since the wavelength of this laser was in the near-infrared range, it could be observed through CCD of cameras. Measuring results by image processing indicated that the diameter of focal spot was 2.10 mm (Fig. 8.19). Feasibility of laser cutting to tomato peduncles In Eq. (8.12), k d is given a value of 0.98 since the laser beam entered into the focusing lens directly and communication loss of the laser power in this device mainly came from communication distance from the focusing

388

8 Fruit Detaching Methods for Robotic …

Fig. 8.19 Near-infrared image of the focal spot

lens onto the surface of a peduncle. k t was given a value of 0.97 according to the instruction book of the focusing lens, and k a was given a value of 0.60 according to the absorption spectrum of green plants [57]. So, the maximum power density of focal spot was calculated out as 4.95 W/mm2 , which were lower than metal material processing and whose ability to drill or cut a tomato peduncle needs to be studied with the further test. Meanwhile, most tomato peduncles could not be cut off directly by just focusing the laser beam onto them compared with their diameter, so a rotation motion of the focusing lens was necessary.

2.

Known F was 2 according to the instruction book of focusing lens, DOF could be calculated out as 8.40 mm substituting Eq. (8.11) into (8.13), which ensured successful drilling or cutting through operation theoretically compared with the diameter of tomato peduncles. Effect of different factors on drilling through time (1)

Effect of peduncle diameter on drilling through time During the laser cutting operation, tomato peduncles will be drilled through at first. Drilling through time for different peduncles that were tested was between 6.8 s and 15.2 s when the optical output power was 15 W. Since the power density of focal spot was lower, the laser burned the peduncle to drill it, which led to carbonization in holes and lower productivity. To drill or cut flat materials, it has been widely proved that processing speed was negatively correlated with the depth of materials, while processing speed increased more slowly to larger diameter due to a decline of power density with an increase of depth [58–63]. However, the test result indicated that the drilling through a time of tomato peduncles was linearly correlated with the diameter, and a fitting equation is as follows (Fig. 8.20):

8.3 Experimental Exploration of Laser Cutting …

389

Fig. 8.20 Drilling through time as a function of the peduncle diameter

Tt = 2.9Ds

(2)

(8.14)

where T t is the drilling through time(s), Ds is the peduncle diameter (mm). The goodness of fit is 0.928. It may be attributed to the cylindrical surface of peduncle instead of a flat surface (Fig. 8.21). When the laser beam is focused on the cylindrical surface, a larger incident angle at the edge of the surface that is exposed to the laser beam reduces the average power density. To larger diameter, the incident angle at the edge may be smaller, which is helpful to make up for the nonlinear change of drilling through time with diameter. Effect of output optical power on drilling through time In the tests, in order to eliminate the effect of the peduncle diameter on the drilling through time, the test data of drilling through time were

Fig. 8.21 Schematic of the laser beam incidence to a peduncle in cross direction

390

8 Fruit Detaching Methods for Robotic …

Fig. 8.22 Drilling through time as a function of the laser power

corrected for different diameters firstly. Corrected test data of relation between drilling through time and the laser power performs a good fit to Eq. (8.15) (Fig. 8.22): Tt = 22.81P − 0.49

(3)

(8.15)

where P is the optical output power of laser (W). The goodness of fit is 0.967. When the laser power is larger, the effect of lessening drilling through time by increase power is limited. Test results showed that the effect of the laser power on the drilling or cutting efficiency for different materials in different studies was inconsistent. Similar non-linear law was got in laser drilling tests of 1Cr18Ni9Ti alloy, aerospace gas turbines, fiber-reinforced plastics, and laser cutting test of slates [64–67], while it was found in laser drilling test of stainless steel and laser cutting tests of die-boards, ST12 steel sheets that drilling through time or cutting speed was linearly correlated with the laser power [60, 68–70]. In another test of laser cutting Pseudotsuga sinensis wood, a power function with a power bigger than 1 between cutting speed and laser power was got [71]. So the relation between drilling through time and laser power in the laser cutting of peduncles and its basic principles need to be further studied. Effect of defocusing distance on drilling through time Test result of drilling through time versus defocusing distance performs a good fit to Eq. (8.16) (Fig. 8.23): Tt = 0.27Z d2 + 0.76Z d + 6.39

(8.16)

where Z d is the defocusing distance (mm). The goodness of fit is 0.893. It was widely accepted that locating the proper position of the focal spot on the workpiece surface and keeping it constant was essential in

8.3 Experimental Exploration of Laser Cutting …

391

Fig. 8.23 Drilling through time as a function of the defocusing distance

(4)

laser drilling and cutting either to improve operation efficiency, to extend processing depth, or to improve processing profile and quality. Usually, it was regarded as the best focal position near the under surface of an object since the power density was the highest on focal spot and locating it near the under surface improved the laser energy distribution along with material thickness [60, 72–77]. However, there were certain differences for different materials and needs, even contrary results might be got from some studies [78–82]. In the process of stems laser cutting, it is necessary to set a proper negative defocusing distance to improve the cutting efficiency and extend the cutting depth in view of the cylindrical surface instead of a flat surface, which makes use of the laser power more sufficiently than locating the focal spot on the near point of the stem surface. In practical robotic harvesting, it is impossible to achieve ideal negative defocusing distance at any time attributed to the mechanical and visual error of harvesting robot; therefore, the laser cutting system should have better error tolerance to cut through peduncles. Test result indicated that peduncles could be able to be cut through even under a positive defocusing distance of 8 mm by this laser cutting device, which was more favorable. Effect of incident angle of the laser beam on drilling through time Test result indicated that the drilling through time increased with the incident angle, and when the incident angle was larger than a certain value, the drilling through time increased more rapidly. This phenomenon was easily understood for focal spot becomes larger with an increase of the incident angle, the average power density of focal spot declines to prolong the drilling through time [56, 83, 84]. Meanwhile, the actual length of drilling through the path becomes longer to prolong drilling through time further.

392

8 Fruit Detaching Methods for Robotic …

Fig. 8.24 Schematic of the laser incidence to a peduncle in longitudinal direction

As shown in Fig. 8.24, when the laser beam is focused on a surface at an angle, the area of focal spot is stretched from a circle to an ellipse, which decreases laser intensity by a multiplier of the cosine of the incidence angle [85]. The distance through the object is also dependent on the cosine of incident angle, only inversely [85]. So, the actual area of focal spot for a certain incident angle and the length of drilling through a path in laser cutting of peduncles are as follows, respectively:  A = A0 cos λ

(8.17)

 L p = Ds cos λ

(8.18)

where A is the actual area of focal spot for a certain incident angle (mm2 ), A0 is the area of focal spot when the incident angle is 0° (mm2 ), L p is the actual length of drilling through path (mm), λ is the incident angle of the laser beam (°). The corresponding average power density of focal spot is as follows: ρ = ρ0 cos λ

(8.19)

where ρ is the average power density of focal spot for a certain incident angle (W/mm2 ). So, the simultaneous solution of Eqs. (8.15), (8.16), (8.18) and (8.19) is as follows: Tt = T0 cos−1.49 λ

(8.20)

where T 0 is the drilling through time when the incident angle is 0° (s). Test data of incident angle Vs drilling through time perform a good fit to Eq. (8.20) with a goodness 0.895 (Fig. 8.25). This result indicates that the incident angle 0° is best

8.3 Experimental Exploration of Laser Cutting …

393

Fig. 8.25 Drilling through time as a function of the laser incident angle

to improve the operation efficiency in laser cutting of peduncles, and the operation efficiency will fall down rapidly when the incident angle is above 35°. However, it is encouraged that there is not a threshold of incident angle, which means all peduncles can be able to be drilled through time for any incident angle smaller than 70°, thanks to the lower burning temperature of tomato peduncles and enough depth of focus. To actual laser cutting operation of peduncles in robotic harvesting, the ideal perpendicular posture and position are not always be able to reach due to the limited space in tomato canopy, so it is very important to minimize incident angle by collision-free optimizing posture and motion plan of the end-effector.

8.3.6 Realization of Laser Cutting of Peduncles [43] 1.

Speed of laser cutting of peduncles Since the diameter of the focal spot is smaller than the diameter of the peduncle, the cutting of the peduncles must be achieved through the movement of the focal spot. In the developed non-damage-free harvesting end-effector, the movement of focal spot is realized by the rotation of the focusing lens driven by the motor (Fig. 4.19, Fig. 4.20). The test device of laser cutting of peduncles is used to keep the focusing lens and mounting plates vertical. The laser driving current is 6A and the focal distance is set as standard. By setting the motor speed, the moving speed of cradle increased from 1.74 × 10−2 mm/s by every 1.74 × 10−2 mm/s gradually, until it is a failure to cut the peduncle off. The experimental results show that when the cutting speed of the laser beam exceeds 17.47 × 10−2 mms−1 , the peduncle cannot be cut off. Both the cutting

394

8 Fruit Detaching Methods for Robotic …

off time and the drilling through time are directly proportional to the diameter of the peduncle. Cutting off time at the same power is 3-4 times the drilling through time (Figs. 8.26 and 8.27). According to the law of laser drilling test, when the laser output power is increased and the negative defocus is used, the cutting speed can also be significantly accelerated. In the actual robot operation, because of the complex canopy environment, it is impossible to guarantee the vertical cutting between the laser beam and the peduncle. At the same time, the negative defocus measure also requires more precise visual and mechanical positioning of the peduncle.

Fig. 8.26 Relation between cutting off time and drilling through time

Fig. 8.27 Relation Between cutting off time and drilling through time

8.3 Experimental Exploration of Laser Cutting …

2.

395

Control strategy of laser cutting motion Based on the experimental results, the motion control strategy of laser cutting is determined as follows: (1) (2) (3)

(4) (5)

(6)

When the manipulator carries the end-effector to the harvesting position, the vision system measures the distance of the fruit. The end-effector sucks and pulls the fruit to the specified position, and the vision system performs the precise positioning of the peduncle. The manipulator adjusts the pose of the end-effector relative to the peduncle so that the focusing lens is vertical to the peduncle and the focal length is about 48 mm. Start the laser, set the current as 12A. At the same time, the motor drives the focusing lens to swing so that the laser focal spot sweeps through the peduncle. The motor speed is 50 rpm and the running time of the motor is 30 s. The vision system judges whether the peduncle is cut successfully or not. If the cutting is unsuccessful, the motor rotates backward, causing the laser beam to sweep through the peduncle again and complete the cutting.

8.4 Discussion (1)

(2)

This laser cutting device can be able to be applied in peduncle cutting based on two essential aspects. One is that the power density of focal spot of this device which decided by both laser beam quality and properties of the focusing system. To this laser cutting device, the diameter of focal spot was measured to be 2.10 mm, and the maximum power density of focal spot reached 4.95 W/mm2 , which was found enough to burn and further to cut off one peduncle. Although the absorption of 980 nm wavelength laser by green peduncles is lower, the rough surface and lower thermal conductivity of peduncles are helpful to improve absorption and utilization of laser power. The other is that tomato peduncles can be drilled or cut through even under a certain defocusing distance or incident angle due to enough depth of focus of laser cutting device that reached 8.4 mm and the lower burn temperature of tomato peduncles. All factors including peduncle diameter, laser power, defocusing distance, and incident angle have a significant effect on the efficiency of peduncle laser cutting. Test data indicated it is a good linear correlation between drilling through time and diameter of peduncle and the drilling through time was positively correlated with the incident angle and negatively correlated with the laser power. In addition, the defocusing distance of 1.4 mm is best to minimize the drilling through time. For practical operation in robotic harvesting, the peduncle diameter is not unique, only the defocusing distance and incident angle can be optimized to improve the laser cutting efficiency. Test results indicated that the drilling through speed was slow because the power density was very low. To shorten the drilling through time, increase the laser power

396

8 Fruit Detaching Methods for Robotic …

(3)

may be an option, but it is preferred to increase the power density by selecting better quality laser, such as fiber laser, to reduce the diameter of focal spot. Meanwhile, the pulse mode (PW) instead of the continuous wave mode (CW) might be applied to get higher peak power, or assist oxygen gas might be added to accelerate burn. It is usually impossible to cut off a peduncle directly just by focusing laser beam on it since the diameter of focal spot is smaller than that of most peduncles, so it is necessary to rotate the focusing lens driven by the mini motor. The rotation speed of motor should be decided based on further cutting speed test data and take into consideration of cylindrical surface of peduncles. Cutting off time at the same power is 3–4 times of the drilling through time.

From the test results, it is found that the speed of laser penetration and cutting is still too slow, and the efficiency is difficult to meet the requirements of actual operation. Although the cutting efficiency can be increased by more than 30% by using the optimum negative defocus according to Fig. 8.17, there is still a considerable distance from the actual operating efficiency. It is found that limited by the beam quality of a laser diode, the diameter of its focal spot is far greater than that of the widely used CO2 and Nd:YAG lasers for same focal depth. The thermal power density of the focal spot is too low, and the peduncle cutting is realized through the burning effect (Fig. 8.28). Under the burning effect, the peduncle is first dried and then burned, and the cutting efficiency is greatly reduced by the high moisture content of more than 90% of the fresh stem. The laser cutting experiments of biological tissue have proved that when the heat power density of the focal spot reaches a certain level, the gasification effect will be produced. That is, the boiling of the internal tissue, the steam breaking the cell wall, splitting the tissue, and taking the debris away to complete the cutting [47–49].

1

2

1. Burning flame 2. Peduncle

Fig. 8.28 The burning effect in laser cutting of peduncles

8.4 Discussion

397

Fig. 8.29 The laser cutting experimental platform of agricultural materials equipped with fiber laser in our laboratory

According to the operational flexibility of robotic harvesting, if a new fiber-coupled Nd:YAG laser with high beam quality or fiber laser with even higher beam quality (Fig. 8.29) is selected, it can improve the thermal power density of the focal spot by 102 –103 times when maintaining the ideal focal depth so that the fast gasification cutting can be achieved.

References 1. Liu J, Peng Y, Faheem M (2020) Experimental and theoretical analysis of fruit plucking patterns for robotic tomato harvesting. Comput Electron Agric 173: 2. Liu J (2010) Analysis and optimal control of vacuum suction system for tomato harvesting robot. Jiangsu University 3. Kondo N, Ting K (1998) Robotics for bioproduction system. ASAE 4. Kondo N, Ninomiya K, Hayashi S, et al (2005) A new challenge of robot for harvesting strawberry grown on table top culture. In: Proceedings of the ASAE annual international meeting

398

8 Fruit Detaching Methods for Robotic …

5. Kondo N, Nishitsuji Y, Ling P et al (1996) Visual feedback guided robotic cherry tomato harvesting. Trans ASAE 39(6):2331–2338 6. Cui Y, Kobayashi T, Nagata M (2005) Basic study on ultrasonic sensor for harvesting robot of strawberry. Bull Facul Agricul Miyazaki Univ 51(1/2):9–16 7. Hayashi S, Shigematsu K, Yamamoto S et al (2010) Evaluation of a strawberry-harvesting robot in a field test. Biosys Eng 105(2):160–171 8. Guo F, Cao Q, Masateru N (2008) Fruit detachment and classification method for strawberry harvesting robot. Int J Adv Rob Syst 5(1):41–48 9. Shiigi T, Kurita M, Kondo N, et al (2008) Strawberry harvesting robot for fruits grown on table top culture. In: Proceedings of the ASABE annual international meeting 10. Takahashi Y, Ogawa J, Saeki K (2001) Automatic tomato picking robot system with human interface using image processing. In: Proceedings of the 27th annual conference of the ieee industrial electronics society 11. Ceres R, Pons J, Jimenez A et al (1998) Design and implementation of an aided fruit-harvesting robot (Agribot). Indus Robot Int J 25(5):337–346 12. Hayashi S, Sakaue O (1997) Basic operation of tomato harvesting system using robot: manufacture of two-finger harvesting hand with auxiliary cutting device and basic experiment for harvest. Bull Natl Res Inst Veg Ornam Plants Tea (Japan) Ornam Plants Tea (12):133–142 13. Hayashi S, Ganno K, Ishii Y et al (2002) Robotic harvesting system for eggplants. Jpn Agricul Res Q 36(3):163–168 14. Arima S, Kondo N, Monta M (2004) Strawberry harvesting robot on table-top culture. In: Proceedings of the ASAE/CSAE annual international meeting 15. Arima S, Kondo N, Nakamura H (1996) Development of robotic system for cucumber harvesting. Jpn Agricul Res Q 30:233–238 16. Arima S, Shibusawa S, Kondo N, et al (2003) Traceability based on multi-operation robot; information from spraying, harvesting and grading operation robot. In: Proceedings of the IEEE/ASME international conference on advanced intelligent mechatronics 2:1204–1209 17. Kondo N, Ting K (1998) Robotics for plant production. Artif Intell Rev 12(1):227–243 18. Allotta B, Buttazzo G, Dario P, et al (1990) A force/torque sensor-based technique for robot harvesting of fruits and vegetables. In: Proceedings of the IEEE international workshop on intelligent, robots and systems 19. Lee B, Rosa U (2006) Development of a canopy volume reduction technique for easy assessment and harvesting of valencia citrus fruits. Trans ASABE 49(6):1695–1703 20. Lee B, Rosa U, Cheet-Ancheri K (2006) End-effector for automated citrus harvesting. In: Proceedings of the ASABE annual international meeting 21. Burks T, Villegas F, Hannan M et al (2005) Engineering and horticultural aspects of robotic fruit harvesting: opportunities and constraints. HortTechnology 15(1):79–87 22. Jia B (2009) Integrated gripper and cutter in a mobile manipulation robotic system for harvesting greenhouse products. In: Proceedings of the IEEE international conference on robotics and biomimetics 23. Raparelli T, Beomonte Z, Durante F (2005) Development of a picking device of an orange harvesting machine. In: Proceedings of the 6th international conference on fluid power, transmission and control 24. van Henten E, Hemming J, van Tuijl B et al (2002) An autonomous robot for harvesting cucumbers in Greenhouses. Autonom Robots 13(3):241–258 25. Zhang K, Yang L, Zhang T (2009) Design of an end-effector for strawberry harvesting robot. J Agricul Mech Res 54–56:60 26. Liu J, Li Z, Li P et al (2008) Design of a laser stem-cutting device for harvesting robot. In: Proceedings of the IEEE international conference on automation and logistics 27. Liu J, Li P, Li Z (2008) Hardware design of the end-effector for tomato-harvesting robot. Trans Chin Soc Agricul Mach 39(3):124–127 28. Monta M, Kondo N, Ting K (1998) End-effectors for tomato harvesting robot. Artif Intell Rev 12(1):11–25

References

399

29. Ling P, Ehsani R, Ting K et al (2004) Sensing and end-effector for a robotic tomato harvester. In Proceedings of the ASAE annual international meeting 30. Bulanon D, Kataoka T (2010) A fruit detection system and an end effector for robotic harvesting of Fuji Apples. Agricul Eng Int CIGR J 12(1):203–210 31. Bulanon D, Kataoka T, Okamoto H et al (1998) Determining the 3-D location of the apple fruit during harvest. In: Proceedings of the automation technology for off-road equipment, 91–97 32. Bulanon D, Kataoka T, Okamoto H et al (2005) Feedback control of manipulator using machine vision for robotic apple harvesting. In: Proceedings of the ASAE annual meeting, No. 053114 33. Tanigaki K, Fujiura T, Akase A et al (2008) Cherry-harvesting robot. Comput Electron Agric 63(1):65–72 34. Hannan M, Burks T (2004) Current developments in automated citrus harvesting. In: Proceedings of the ASAE annual meeting, No. 43087 35. Baeten J, Donné K, Boedrij S et al (2007) Autonomous fruit picking machine: a robotic apple harvester. In: Proceedings of the 6th international conference on field and service robotics 36. Liu J, Hu Y, Xu X et al (2011) Feasibility and influencing factors of laser cutting of tomato peduncles for robotic harvesting. Afr J Biotech 10(69):15552–15563 37. Liu J, Xu X, Li P (2014) Analysis and experiment on laser cutting of fruit peduncles. Trans Chin Soc Agricul Mach 45(1):59–64 38. Zhang K, Yang L, Zhang T (2009) Design of transmission mechanism for vegetable dehydrator test-bench. J Agricul Mech Res 31(4):54–56 39. Bachche S, Oka K (2013) Performance testing of thermal cutting systems for sweet pepper harvesting robot in Greenhouse horticulture. J Syst Design Dyn 7(7):36–51 40. Hernandez-Castaneda J, Sezer H, Li L (2010) Dual gas jet-assisted fibre laser blind cutting of dry pine wood by statistical modelling. Int J Adv Manuf Technol 50(1–4):195–206 41. Barcikowski S, Koch G, Odermatt J (2006) Characterisation and modification of the heat affected zone during laser material processing of wood and wood composites. Holz als Rohund Werkstoff 64(2):94–103 42. Nukman Y, Ismail S, Azuddin M et al (2008) Selected Malaysian Wood Co2-laser cutting parameters and cut quality. Am J Appl Sci 5(8) 43. Xu X (2012) Study on the laser cutting technology of peduncles for harvesting robot. Jiangsu University 44. Dua R, Chakraborty S (2005) A novel modeling and simulation technique of photo-thermal interactions between lasers and living biological tissues undergoing multiple changes in phase. Comput Biol Med 35(5):447 45. Ferraz A, Mittal G, Bilanski W et al (2007) Mathematical modeling of laser based potato cutting and peeling. Biosystems 90(3):602–613 46. Liu P, Liu G (1989) Mechanism of laser biological action. Science Press 47. Niemz M (2019) Laser-tissue interactions-fundamentals and applications. Springer 48. Lu J (1996) Physics of laser material interaction. China Machine Press 49. Wang K (2002) New technology of ophthalmic laser. People’s Military Medical Press 50. Saito T, Kataoka J (1981) Physiological basis of tomato. Shanghai Science and Technology Press 51. Zheng Q (1999) Advanced laser manufacturing technology. Inf Dev (4):6–9 52. Zhou Z (1994) Agricultural materials science. Agricultural Press 53. Chen Q (2001) Laser material interaction and thermal field simulation. Yunnan Science and Technology Press 54. Wang J (1992) Laser processing technology. China Metrology Press 55. Han W (2004) Computational and experimental investigations of laser drilling and welding for microelectronic packaging. Worcester Polytechnic Institute 56. Zuo T (2008) Beam quality, communication quality, focusing quality of laser for manufacture. Science Press 57. Kondo N, Monta M, Noguchi N (2004) Agri-robot-fundamentals and theory. Corona Publishing Co., Ltd, Tokyo

400

8 Fruit Detaching Methods for Robotic …

58. Luo L, Zhou J, Liu C (2005) Theoretic model of making vessels in myocardium by infrared laser and experiment validating. Acta Photonica Sinica 34(6):817–819 59. Hermanns C (2000) Laser cutting of glass. In: Proceedings of the Inorganic Optical Materials II. International Society for Optics and Photonics 4102:219–226 60. Xie X, Wei X, WH (2008) Theoretical model of Co2 laser cutting non-metal material. Tool Engineering 42(5):19–21 61. Olfert M (2000) Fundamental processes in laser drilling and welding. University of Waterloo 62. Jin Y (2008) High-efficiency laser drilling for metal and alloy materials. Ome Inf (2):26–29 63. Shao D, Hu B, Zheng Q (2009) Advanced laser manufacturing technology and equipment integration. Science Press 64. Liao J, Chen Q, Mei Y et al (1997) Laser Drilling of Instant Rice Noodle Sieve[J]. Applied Laser 17(3):125–127 65. Boutinguiza M, Pou J, Lusquinos F et al (2002) Co2 laser cutting of slate. Opt Lasers Eng 37(1):15–25 66. Naeem M (2008) Advancement in laser drilling for aerospace gas turbines. In: Proceedings of the 3rd Pacific international conference on application of lasers and optics (1):197–202 67. Ishide T, Shirata H, Matsumoto O et al (1996) Development of high-speed and high-quality laser cutting process for fiber reinforced plastics. Tech Rev 33(2):83–87 68. Walther K, Brajdic M, Kreutz E (2008) Enhanced processing speed in laser drilling of stainless steel by spatially and temporally superposed pulsed Nd: Yag laser radiation. Int J Adv Manuf Technol 35(9):895–899 69. Hata K, Shibata K, Okabe T et al (2000) Influence of laser beam irradiation conditions on the machinability of medium density fiberboard impregnated with phenolic resin. J Porous Mater 7(4):483–490 70. Liu H, Xu L (2003) Laser two-dimentional cutting technology optimization. Hongdu Sci Technol 1:16–20 71. Peters C (1977) Cutting wood and wood-based productions with a Multikilowatt Co2 laser. Forest Products J 27(1): 41–45 72. Hernandez J, Sezer H, Li L (2010) Dual gas jet-assisted fibre laser blind cutting of dry pine wood by statistical modelling. Int J Adv Manuf Technol 50:1–12 73. Lamikiz A, Lacalle L, Sanchez J et al (2005) CO2 laser cutting of advanced high strength steels (Ahss). Appl Surf Sci 242(3–4):362–368 74. Zhang X, Xie S, Zhan Z (2008) Influence of different defocus conditions on bone hard tissue ablation with pulsed Co2 laser. Chin J Lasers 35(7):1116–1120 75. Tong G, Xu H (2009) Compensation for optical distance and cutting quality in the process of laser cutting. J Nanjing Instit Indus Technol 9(4):8–9, 20 76. Xie X, Li L, Wei X et al (2008) Evaporative front of laser cutting Pmma. Chin J Lasers 35(6):925–930 77. Zhang Y, Fang M (2001) The research of the ceramic drilling by Nd: Yag laser. Laser Infrared 31(3):161–162 78. Qu C, Wang Y (2001) Study on the slit technology of laser cutting. J Heilongjiang Commer College (Natural Sciences Edition) 17(1):67–69, 87 79. Guan B, Liao J, Meng H (2005) Study of cutting precision gear wheel with fiber laser. Appl Laser 25(6):365–368 80. Ge Y, Wang W, Cui Z et al (2008) Corresponding experimentation research of aluminium alloy cutting using pulsed solid Nd: Yag laser. Weld Technol 37(5):20–24 81. Chen H, Yang R (2009) Test study on application of laser processing technology to cut chestnut shell. In: Proceedings of the cereals and oils 82. Zhu W (2009) Research of the cutting technology on silicon wafer by Yag laser. High Power Convert Technol (3):1–4, 8

References

401

83. Yang S, Zhang Z, Kou S (2009) Technological parameters in the laser cutting process of starting notch of connecting rod. J Harb Instit Technol 1:126–129 84. Chen J, Xu X, Xiao R (2007) Modern laser manufacturing technology. National Defense Industry Press, Beijing 85. Porter J, Louhisalmi Y, Karjalainen J et al (2007) Cutting thin sheet metal with a water jet guided laser using various cutting distances, feed speeds and angles of incidence. Int J Adv Manuf Technol 33(9):961–967

Chapter 9

Control Optimization and Test Study

9.1 Summary 9.1.1 Research Significance The positioning of the end-effector, the detachment, and the transportation of the fruit must be accomplished by the coordinated action of the hand–arm system. In view of the cluster growth characteristics of tomatoes, the auxiliary action of the adjacent fruit isolation also plays an important role. Therefore, the speedy fruit harvesting with a high success and low damage rate must be realized relying on the coordinated action of both the end-effector components and the hand–arm system.

9.1.2 Content and Innovation (1)

(2)

(3)

Combining the energy consumption of acceleration/deceleration with the optimal control of gripping collision force, the current value guaranteeing stable and reliable gripping, the control mode and parameters of highprobability damage-free gripping were obtained based on PMAC + EPOS control system; The optimal control mode of multi-action coordination of the end-effector was established by the parameter optimization of the sucked pulling model of the auxiliary vacuum suction process; Based on the damage-free harvesting system composed of the commercial manipulator and self-developed end-effector, a “hybrid” hand–arm coordinated control model was constructed and verified by experiments, which provides a direct basis for the speedy robotic damage-free harvesting of tomato fruits.

© Science Press, Beijing and Springer Nature Singapore Pte Ltd. 2021 J. Liu et al., Rapid Damage-Free Robotic Harvesting of Tomatoes, Springer Tracts in Mechanical Engineering, https://doi.org/10.1007/978-981-16-1284-8_9

403

404

9 Control Optimization and Test Study

9.2 Parameter Optimization of Speedy Flexible Gripping [1] 9.2.1 PID Parameter Adjustment of the Motion Control System In a mechatronics system, in order to obtain good steady-state performance and dynamic characteristics, it is necessary to adjust the control loop of the system. When the motor system and mechanical transmission structure of the system are determined, it is necessary to adjust the servo loop PID (proportional–integral–differential) filter according to the dynamic performance of the controlled physical system, so as to achieve good system stability, fast response, and small follow-up error. 1.

The Control structure of the motion control system

The motion control system of the gripping mechanism adopts compound control. The PMAC2A-PC104 motion controller has a built-in digital PID + feedforward servo control algorithm to close the position loop and speed loop of the system. The EPOS-24/5 driver closes the current loop of the system through an analog current regulator, as shown in Fig. 9.1. 2. (1)

PI adjustment of current loop Analysis of current loop

The main function of the current loop is to limit the maximum current of the motor and ensure the constant torque output of the motor: (1) (2)

To guarantee the maximum allowable dynamic current of the motor during the starting process; To make the current changes with the given value in the process of speed adjustment;

Fig. 9.1 The compound control of the motion control system of the gripping mechanism. Note ACR is a current regulator, UPE is a DC PWM converter, SM is a motor, BQ is an encoder

9.2 Parameter Optimization of Speedy Flexible Gripping [1]

405

Fig. 9.2 Step signal parameter setting

(3)

To supply the function of automatic overload protection and resume normal work after the overload fault disappears automatically;

The current loop belongs to the inner loop of the motion control system. Therefore, when adjusting the PID of the system, the current loop should be adjusted first, and then the speed loop and the position loop of the system should be adjusted in turn. (2)

PI adjustment of current loop (1)

(2)

Setting the parameters of the step signal As shown in Fig. 9.2, the step current is 1000 mA and the adjustment time is 20 ms. PI adjustment Let the initial P = 800, I = 200, the step response and the error curve are as shown in Fig. 9.3. Automatically adjust first, then manually adjust until the requirements are met. In order to minimize the evaluation error and adjustment time, the commonly used IAE (Integral Absolute Error) is used to measure.

By automatically adjusting PI (proportional term and integral term), the system automatically determines the PI value under the optimal performance according to the size of IAE. After automatic adjustment, I value is increased step by step, the step size is 50. The PI parameters and step response index of manual adjustment are shown in Table 9.1. After adjustment, when P = 2400, I = 900, the minimum value of IAE is 20017, the rise time is 0.8 ms, no overshoot, the adjustment time is 1.1 ms, a better regulation effect is achieved. 3.

Open-Loop Test

The purpose of an open-loop test of PMAC is to test the performance of a closed-loop PID controller. In Pulse + Direction control mode, an open-loop test can also test the operation of PFM (Pulse frequency modulation) signal generator in PMAC [2]. The test method is to write a value to the output register with the O instruction, which represents the proportion of the specified value, and then detect the operation of the

406

9 Control Optimization and Test Study

Fig. 9.3 Step response and error curve

Table 9.1 PI parameters and step response indices PI PI parameters

Response indices

P

I

IAE

Rise time/ms

Overshoot/%

Adjustment time/ms

2400

900

20017

0.8

0

1.1

2400

850

20706

0.9

0

1.3

2400

800

20427

1

0

1.4

2400

750

20350

1.1

0

1.5

2400

700

23368

1.2

0

1.6

2400

650

24663

1.5

0

2

2400

600

24543

1.8

0

2.4

2400

550

26506

2

0

2.8

motor in the report window. The default PFMCLK is 9.83 MHz and Ix69 is 20480 (equivalent to voltage 6.25 V). According to I x69 =

MaxFreq(kHZ) ∗ 65536 PFMCLK(kHZ)

(9.1)

The maximum allowable output frequency of the PMAC controller (MaxFreq) is about 3.07 MHz. The input instructions and test results are shown in Table 9.2. From the test results, the average error of rotational speed can be calculated to be 0.8%. It shows that the system is connected normally, and the PID parameters of the closed-loop drive system have good response characteristics after the adjustment.

9.2 Parameter Optimization of Speedy Flexible Gripping [1]

407

Table 9.2 Open-loop test data Instruction value

Output voltage/v

O1

6.25 × 1%

30.7

31.2

935

O2

6.25 × 2%

61.4

60.1

1803

O3

6.25 × 3%

92.1

91.4

2711

O4

6.25 × 4%

122.8

123.8

3714

O5

6.25 × 5%

153.5

154.7

4642

O6

6.25 × 6%

184.2

185.9

5577

O7

6.25 × 7%

214.9

215.1

6512

O8

6.25 × 8%

245.6

244.8

7345

4.

Output frequency/KHZ

Actual frequency/KHZ

Actual speed/rpm

PID adjustment of speed loop and position loop (1)

PID control algorithm for motion controller [2–5] (1) Working principle of PID filter

The standard PMAC controller provides a PID position loop servo filter to close system position loop and speed loop (Fig. 9.1). The filter is adjusted by setting the appropriate I variable of each motor: ➀ ➁ ➂ ➃ ➄ ➅

(2)

The function of proportional gain “K p ” -Ix30 is to provide the rigidity required by the system; The function of the differential gain function of “K d ” -Ix31 is to provide sufficient damping to ensure the stability of the system; The function of integral gain “K I ” -Ix33 is to eliminate steady-state error; The function of integration mode Ix34 is to determine whether the integral gain is effective throughout the whole process or only effective when the control speed is 0; The function of velocity feedforward gain “K vff ” -Ix32 is to reduce the following error caused by the introduction of differential gain and is proportional to the change speed of input signal; The function of acceleration feedforward gain “K aff ” -Ix35 is to reduce or eliminate the following error caused by the inertia of the system (proportional to the change of the acceleration of the input signal). The main characteristics of the PID control algorithm in PMAC ➀

Dead zone filtering

In order to avoid too frequent control action and eliminate the oscillation caused by the frequent action, the PID control algorithm with the dead zone is adopted, and the control formula is as follows:

408

9 Control Optimization and Test Study

 e(k) =

|e(k)| ≤ |e0 | 0 e(k) |e(k)| > |e0 |

(9.2)

where e(k) is the position tracking error; e0 is the adjustable parameter whose specific value can be determined by experiment according to the actual control object, too small value will lead to too frequent control action to reach the goal of stabilizing the controlled object, too large value will produce a larger lag to the system. The control system with dead-time filtering is a nonlinear system. When |e(k)| ≤ |e0 |, the output of the digital regulator is zero; when |e(k)| > |e0 |, the digital regulator has PID output. ➁

Integral separation

In ordinary PID control, the purpose of integrating part is to eliminate the static error. However, when the process starts, ends, increases or decreases by a large margin, the output of the system will deviate greatly in a short time. It will result in the accumulation of integrals in the PID operation, and the control amount will exceed the limit control amount corresponding to the maximum range of action that the actuator may allow, resulting in a large overshoot or even oscillation of the system. The basic idea of integral separation is to cancel the integral function when the controlled variable is deviated from the set value, so the stability of the system decreasing and the overshoot increasing is to avoid by the integral effect. ➂

Differentiation first

The characteristic of differentiation first is that it only differentials the output and does not differentiate the given variables. Therefore, when changing the given value the controller’s output u will not change significantly. The control strategy is suitable for the occasion of frequent rise and fall of the given value and can avoid the system oscillation caused by the rise and fall of the given value, thus significantly improving the dynamic characteristics of the system. The transfer function of the differential part is ud (z) = Kd (1 − z −1 ) y(z) ➃

(9.3)

Feedforward compensation

The introduction of velocity feedforward and acceleration feedforward increases the zero of the system, but the poles of the system remain unchanged, so the stability of the system remains unchanged. The steady and dynamic performances of the system are improved by adjusting the position of the zero in the z plane.

9.2 Parameter Optimization of Speedy Flexible Gripping [1]

(3)

409

The actual algorithm of PID

DACout (n) =219 ∗ Ix30∗     Ix08 ∗ FE(n) + (Ix32 ∗ CV (n) + Ix35 ∗ CA(n))/128 + Ix33 ∗ IE(n)/223 − Ix31 ∗ Ix09 ∗ AV (n)/128

(9.4) where DAC out (n) is The 16-bit servo cycle output command (−32768 to +32767) which is converted to an output of −10 to +10 V, and the value of DAC out (n) is defined by Ix69; Ix08 is an internal location magnification factor of motor X (default is 96); Ix09 is an internal amplification factor for the speed loop of the motor X (default is 96); FE(n) is the following error in the servo cycle n, which is the difference between the instruction position and the actual position in the cycle [CP(n) − AP(n)]; AV (n) is the actual speed in the servo cycle n, which is the difference between the last two actual positions of each servo cycle [AP(n) − AP(n − 1)]; CV (n) is the instruction speed in the servo cycle n, which is the difference between the last two instruction positions for each servo cycle [CP(n) − CP(n − 1)]; CA(n) is the instruction acceleration in the servo cycle n, which is the difference between the last two instruction speeds for each servo cycle [CV (n) − CV (n −  1)]; IE(n)  is the  integral of the following error in the servo cycle n, which equals to n−1 j=0 FE(j) . (4)

Transfer function

According to the above actual PID algorithm in the PMAC, the transfer function of the proportional term in the speed loop is obtained as Gb (z) = 2−19 Kp

(9.5)

The transfer function of differential terms is GD (z) = β × 2−7 Kd (1 − z −1 )

(9.6)

The transfer function of the PI term in the position loop is GP (z) = α ×

z(1 + 2−23 Ki ) − 1 z−1

(9.7)

In Eqs. (9.6) and (9.7), α = Ix08; β = Ix09. (2)

DAC calibration

Generally, the main error of DAC (Digital to analog conversion) device is a zero offset error, so it is necessary to calibrate DAC zero offset first. The internal DAC calibration of PMAC is completed by a pre-programmed internal PLC program. The calibration process is as follows:

410

9 Control Optimization and Test Study

Fig. 9.4 Calibration results

(1)

(2)

(3)

Set repeat test times and calibration step size Test for ten times repeatedly with calibration step 0.01 to detect zero offset error of the DAC device in PMAC, and its value is the average of ten test results. Execution of calibration The zero offset error of the DAC device in PMAC is stored in variable I129. The test result shows that the zero offset error is −1 (16 DAC bits) and the open-loop dead zone size is 69 DAC bits (Fig. 9.4). The dead zone is caused by the connection of the controller, driver, and motor. In order to achieve accurate control, compensation is needed. Offset compensation After compensation is selected, the value of variable I129 becomes the recommended value -1, which is stored in the register of PMAC. The output instruction value is added with the value of I129 before writing to the output instruction register to realize automatic compensation.

(3)

PID adjustment of speed loop and position loop

(1)

Close the motor

Use the #1j/online command to close the loop of motor 1 and use #1? online command inquires the status of motor 1. If the online window returns the status word 812…, it means that the motor has a closed loop; otherwise, use the #1j/online command repeatedly to close the loop. (2)

Adjusting servo ring PID with the step response ➀

➁

Enter the “PID Interactive Tuning Motor #1” dialog box as shown in Fig. 9.5. In “Left Axis Plot” select the type of response curve to draw, including position, velocity, acceleration, follow-up error, and DAC output. For the step response, the “position” type is chosen. Set Ixx31 = 0, Ixx34 = 1, Ixx33 = 0, Ixx32 = 0, Ixx35 = 0, Ixx30 = 500, enter Step Size and Step Time, press “Do a Step” to perform a step response.

9.2 Parameter Optimization of Speedy Flexible Gripping [1]

411

Fig. 9.5 PID interactive tuning

➂

➃

➄

➅

(3)

The collected data is automatically plotted as a step response curve, and the rise time, overshoot, and adjustment time are calculated. As shown in Fig. 9.6a, the actual position is far less than the command position, and the response is slow. The reason is that the rigidity of the system is too small, it is necessary to increase the K p . Increase the K p (Ix30) with a step size of 500, the adjustment process is as shown in Fig. 3.6b–d. Figure 3.6d in the case of a minimum rise time has a greater overshoot because the system damping is too small, it is necessary to increase the Kd . Increase K d (Ix31) gradually to reduce the overshoot. The adjustment process is as shown in Fig. 9.7a–c. Finally, the response curve without overshoot is obtained as shown in Fig. 9.7c. However, it can be seen from the figure that there are still steady-state errors. K I (Ix33) is gradually increased to reduce the steady-state error of the system. The adjustment result is shown in Fig. 9.7d, the rise time t r is 0.016 s, the adjustment time t s is 0.027 s, and the overshoot delta is 0. The corresponding response characteristics are shown in Table 9.3.

Parabolic Response feed forward control ➀

Enter “PID Interactive Tuning Motor #1” dialog box (Fig. 9.8), select “Parabolic Velocity” in “Trajectory Selection”, select speed in “Left Axis Plot”, and select follow error in “Right Axis Plot”.

412

9 Control Optimization and Test Study

(a) Kp =500

(c) Kp =1500

(b) Kp =1000

(d) Kp =2000

Fig. 9.6 The step response curve

(a) Kd =100, KI =0

(c) Kd =300, KI =0

Fig. 9.7 Impulse response curve

(b) Kd =200, KI =0

(d) Kd =300, KI =50

9.2 Parameter Optimization of Speedy Flexible Gripping [1]

413

Table 9.3 Step response adjustment parameters KI

KD

K vff

K aff

tr

ts

δ

A

500

0

0

0

0

0.154

0.505

0

B

1000

0

0

0

0

0.073

0.292

0

C

1500

0

0

0

0

0.045

0.167

0

D

2000

0

0

0

0

0.016

0.029

0

E

2000

0

100

0

0

0.016

0.029

6.3%

F

2000

0

200

0

0

0.016

0.029

5.9%

G

2000

0

300

0

0

0.013

0.027

0

H

2000

50

300

0

0

0.016

0.027

0

Fig. no.

KP

Fig. 9.8 Adjustment interface

➁ ➂

Input the “Move Size” and “Move Time” of the parabola and press “Do a Parabolic” to perform a parabolic response without changing the existing parameters. Waiting for the host to download data, collect data to draw into a curve, and compare it with the command curve.

Increase K vff (Ix32) to repeat the response process, the adjustment process is as shown in Fig. 9.9, and the parameter setting is as shown in Table 9.4, In Fig. 9.9a– f, the speed following error is too large because of the introduction of damping, so speed feedforward must be increased. In Fig. 9.9g, the speed following error is reversed because the speed feedforward is too large, so the speed feedforward must be reduced. In Fig. 9.9h, the shape of the following error curve is close to the square

414

9 Control Optimization and Test Study

(a) Kvff =0

(c) K vff =2000

(e) Kvff =4000

(g) Kvff =6000

Fig. 9.9 Sine wave response curve

(b) Kvff =1000

(d) Kvff =3000

(f) Kvff =5000

(h) Kvff =5600

9.2 Parameter Optimization of Speedy Flexible Gripping [1]

415

Table 9.5 Parameter setting Fig. no.

a

b

c

d

e

f

K vff

5600

5600

5600

5600

5600

5600

K aff

0

4000

8000

10000

15000

20000

E max

13

12

11.6

11.2

10.8

10.4

wave and close to the ideal situation. At this time, the following error is reduced to the minimum, and concentrated in the middle, along with the trajectory of the uniform distribution. Fig. No.

a

b

c

D

e

F

G

H

K vff

0

1000

2000

3000

4000

5000

6000

5600

K aff

0

0

0

0

0

0

0

0

E max

240

196

156

112

72

33

27

12.4

Select “Parabolic Velocity” in the “Trajectory Selection”, select acceleration in the “Left Axis Plot” and select follow error in the “Right Axis Plot”. Add K aff (Ix35) to observe the change of response curve and acceleration following error. Adjustment parameter setting and error were as shown in Table 9.5. When K aff = 20000, the dynamic following error reached a smaller value of 10cts, most of which was caused by noise or mechanical friction.

9.2.2 Energy Consumption Analysis of Acceleration and Deceleration Stage Acceleration/deceleration control is one of the key technologies in the field of motion control. PMAC motion controller has three kinds of acceleration/deceleration control algorithms for actuator: trapezoidal, semi-S curve, and S curve. These algorithms are realized by setting the time parameters of each acceleration stage of the motor, but the time parameter setting has a great influence on the energy consumption of the motor. Therefore, the relationship between the time parameter setting and energy consumption of different acceleration/deceleration control algorithms is studied. 1.

Acceleration/deceleration process of the gripping system

As shown in Fig. 9.10, the acceleration/deceleration process of the gripping system includes three stages: acceleration, constant speed, and deceleration.

416

9 Control Optimization and Test Study

Fig. 9.10 The acceleration/deceleration process of gripping system

(1) (2)

(3) 2.

In the first acceleration stage (oa stage), the speed of the motor increases from 0 to the set value v1 , and the finger is dragged to accelerate the movement; In the second constant-speed stage (ab stage), the speed of the motor v1 keeps constant and the fingers move at a uniform speed. When the fingers touch the tomato fruit, it enters the third deceleration stage (bd stage), and the speed of the motor decreases gradually from v1 to 0. In the third deceleration stage, the kinetic energy of the finger is converted to the elastic potential energy of tomato fruit, as shown in Fig. 9.11. The Implementation process of acceleration/deceleration control algorithm

The implementation of S curve acceleration/deceleration control algorithm in the PMAC motion is shown in Fig. 9.12.

F1 - The axial force exerted by the screw on the finger F2 – Tomato fruit's reaction force applied to the finger

Fig. 9.11 The stable gripping process

9.2 Parameter Optimization of Speedy Flexible Gripping [1]

417

Fig. 9.12 Flow chart of the S curve acceleration/deceleration algorithm of PMAC motion controller

In the initial stage of speed planning, the total displacement L and the constant v3 are set according to the specific conditions and are constant, while the jerk (the change of acceleration) time t 1 and the total acceleration time t 3 are related to the acceleration/deceleration algorithm adopted in speed planning (Fig. 9.13). For trapezoidal acceleration/deceleration algorithm, t 1 = 0 and t 2 = t 3 ; for semi-S curve acceleration/deceleration algorithm, t 1 = t 2 ; for S curve acceleration/deceleration algorithm, t 1 > 0 and t 3 = t 1 + t 2 . 3.

Energy consumption of different acceleration/deceleration algorithms

Because different acceleration/deceleration algorithms consume different energy during the acceleration and deceleration process when the time parameters are set differently, the relationship between the time parameters setting and the energy consumption of the motor is reduced in this section. (1)

Energy consumption of ideal acceleration/deceleration algorithm [6].

In the process of motor operation, the electrical energy Pel is converted into mechanical energy Pmech , and the thermal loss PJ is generated. The minimum energy consumption means average thermal loss PJ required in the process of motion control is minimized (Fig. 9.14).

418

9 Control Optimization and Test Study

Fig. 9.13 The time relation in acceleration/deceleration algorithm

(a) Acceleration

(b) Velocity

Fig. 9.14 The optimal speed profile

(a) Angular velocity

(b) Angular acceleration

9.2 Parameter Optimization of Speedy Flexible Gripping [1]

419

Theoretically, the best acceleration/deceleration speed profile with the smallest energy consumption is a parabola, but in practice, the generation of the parabola speed profile is more complex. As a result, the suboptimal speed profile (trapezoidal, quadratic spline curve, S curve) is often used instead. (2)

Energy consumption of trapezoidal acceleration/deceleration algorithm

Assuming that the motion time of the acceleration, constant speed, and deceleration stage of the best trapezoidal speed profile is shown in Fig. 9.15a, the speed curve equation in time [0,t 1 ] is as follows: v = At + B

(9.8)

Fig. 9.15 Trapezoidal speed profile

(a) Stage time division of trapezoidal speed profile

(b) Minimum energy trapezoidal speed profile

420

9 Control Optimization and Test Study

It meets the following boundary conditions: 

v|t=0 = v0 v|t=t1 = v1

The displacement equation of acceleration stage can be calculated according to the boundary conditions: t1 sa =

1 1 vdt = v0 t1 + (v1 − v0 )t1 = (v1 + v0 )t1 2 2

(9.9)

0

where sa is the displacement in the acceleration section. According to the symmetry of the acceleration and deceleration, the displacement in the deceleration stage is equal to the displacement sa in the acceleration stage. Suppose that in the total planned displacement L, the angular displacement of the required motor is θ, v0 = 0, it can be found θ =2sa + sc v1 =2 × t1 + v1 (T − 2t1 ) 2 =v1 (T − t1 )

(9.10)

where sc is the displacement in constant speed stage, mm. There are speed functions as follow:

ω(t) =

⎧ θ ⎪ ⎨ t1 (T −t1 ) · t ⎪ ⎩

θ T −t1 θ t1 (T −t1 )

0 < t ≤ t1 t 1 < t ≤ T − t1 · (T − t) T − t1 < t ≤ T

(9.11)

The heat loss E in a single cycle of a motor is known as [6]. J 2r E= 2 KM

T

dω dt

2 dt

(9.12)

0

where J is the total moment of inertia, kg·m2 ; K M is the torque constant of the motor. By substituting Eq. (9.11) into Eq. (9.12), the energy function of the trapezoidal curve is obtained E=

2 rJ 2 θ 2 · 2 t1 (T − t1 )2 KM

(9.13)

9.2 Parameter Optimization of Speedy Flexible Gripping [1]

421

The minimum value is obtained by Eq. (9.13) when t 1 = T /3 E=

13.5rJ 2 θ 2 KM2 T 3

(9.14)

That is, when the motion time of acceleration, constant speed, and deceleration is equal, the velocity speed obtained is the trapezoidal speed profile with the least energy consumption, as shown in Fig. 9.15b. (3)

Energy consumption of semi-S curve acceleration/deceleration algorithm

Assuming that the motion time of acceleration of acceleration, deceleration of acceleration, constant speed, acceleration of deceleration, and deceleration of deceleration stage of the best semi-S curve speed profile is shown in Fig. 9.16a, the acceleration curve equation in time [0, t 1 ] is a = At + B Fig. 9.16 The semi-S curve speed profile

(9.15)

422

9 Control Optimization and Test Study

According to the boundary conditions, the acceleration curve equation of the acceleration stage can be obtained, and then its speed and displacement curve equation can be obtained by integral: (1)

The acceleration curve equation  a=

(2)

0 < t ≤ t1 t+

4 v2 ( t1 2

− v0 ) t1 < t ≤ t2

(9.16)

The speed curve equation  v=

(3)

2 v2 ( − v0 ) · t t12 2 − t22 ( v22 − v0 ) · 1

1 v2 ( − v0 ) · t 2 + v0 t12 2 − t12 ( v22 − v0 ) · t 2 + t41 ( v22 1

0 < t ≤ t1 − v0 ) · t + 4v0 − v2 t1 < t ≤ t2

(9.17)

The displacement curve equation sa = (v2 + v0 )t1

(9.18)

Suppose that in the total planned displacement L, the angular displacement of the required motor is θ, v0 = 0, it can be found θ =2sa + sc =2 × v2 t1 + v2 (T − 4t1 ) =v2 (T − 2t1 )

(9.19)

By substituting Eq. (9.19) into Eq. (9.17), the speed function is obtained as ⎧ θ · t2 0 < t < t1 ⎪ 2t12 (T −2t1 ) ⎪ ⎪ ⎪ t2 θ ⎪ ⎨ (T −2t1 )t1 · (− 2t1 + 2t − t1 ) t1 < t < 2t1 θ ω(t) = 2t1 < t < T − 2t1 ⎪ T −2t1 ⎪ ⎪ ⎪ ω (t) T − 2t1 < t < T − t1 ⎪ ⎩ 4 ω5 (t) T − t1 < t < T

(9.20)

where ωn (t) is the speed function of the interval n, and ω4 (t), ω5 (t) are symmetric to ω1 (t), ω2 (t) about t = T /2, respectively. By substituting Eq. (9.30) in Eq. (9.12), the energy function of semi-S curve acceleration/deceleration can be obtained as E=

4 rJ 2 θ 2 · 3t1 (T − 2t1 )2 KM2

(9.21)

9.2 Parameter Optimization of Speedy Flexible Gripping [1]

423

Its minimum value can be obtained by Eq. (9.21) when t 1 = T /6 E=

18rJ 2 θ 2 KM2 T 3

(9.22)

That is, when the motion time of acceleration of acceleration, deceleration of acceleration, acceleration of deceleration, and deceleration of deceleration stage are all equal to T /6, and the motion time of constant-speed stage is T /3, the speed profile obtained is the semi-S speed curve profile with the least energy consumption, as shown in Fig. 9.16b. (4)

Energy consumption of S curve acceleration/deceleration algorithm

The motion time of the acceleration of acceleration, deceleration of acceleration, constant-speed, acceleration of deceleration, and deceleration of deceleration stage of the best S curve velocity profile is shown in Fig. 9.17a. Based on the S curve acceleration and deceleration control algorithm, the equation and the boundary conditions of the curve are derived. Assuming the acceleration curve equation in time [0, t 3 ] ⎧ ⎨ At + B1 0 < t ≤ t1 a = At1 + B1 t1 < t ≤ t2 ⎩ −At + B2 t2 < t ≤ t3

(9.23)

According to the boundary conditions, the acceleration curve equation of the acceleration stage can be obtained, and then the speed and its displacement equation can be obtained by integral: (1)

The acceleration curve equation ⎧ a1 ⎪ ⎨ t1 t a = a1 ⎪ ⎩ − a1 t + t1

(2)

(9.24)

The speed curve equation ⎧ v 2 3 ⎪ ⎨ 2t1 t2 t v3 v t v = t2 t − 2t3 21 ⎪ ⎩ v3 2 − 2t1 t2 t +

(3)

a1 t3 t1

0 < t ≤ t1 t 1 < t ≤ t2 t 2 < t ≤ t3

0 < t ≤ t1 t 1 < t ≤ t2 v3 t3 t t1 t2

+ v3 −

v3 t32 2t1 t2

(9.25)

t 2 < t ≤ t3

The displacement equation sa =

v3 (t1 + t2 ) 2

(9.26)

424

9 Control Optimization and Test Study

Fig. 9.17 The S curve speed profile

Suppose that the angular displacement of the motor required in the total planned displacement L is θ, v0 = 0, it can be found θ =2sa + sc v3 =2 × (t1 + t2 ) + v3 [T − 2(t1 + t2 )] 2 =v3 [T − (t1 + t2 )]

(9.27)

By substituting Eq. (9.27) in Eq. (9.25), the energy function of S curve acceleration/deceleration can be obtained as

9.2 Parameter Optimization of Speedy Flexible Gripping [1]

ω(t) =

⎧ θ · t2 ⎪ ⎪ ⎪ 2t1 t2 [Tθ−(t1 +t2 )] ⎪ t1 ⎪ ⎪ ⎪ t2 [T −(t1 +t2 )] · (t − 2 ) ⎪ 2 ⎪ θ ⎪ · [− t2 + (t1 + t2 )t − ⎨ t1 t2 [T −(t 1 +t2 )] θ T −(t1 +t2 )

⎪ ⎪ ⎪ ⎪ ω5 (t) ⎪ ⎪ ⎪ ⎪ ω6 (t) ⎪ ⎪ ⎩ ω7 (t)

425

0 < t ≤ t1 t 1 < t ≤ t2 t12 +t22 ] t2 2

< t ≤ t3 t 3 < t ≤ T − t3 T − t 3 < t ≤ T − t2 T − t 2 < t ≤ T − t1 T − t1 < t ≤ T

(9.28)

By substituting Eq. (9.28) in Eq. (9.12), the energy function of S curve acceleration/deceleration can be obtained as E=

2(3t2 − t1 ) rJ 2 θ 2 · 2 2 KM 3t2 [T − (t2 + t1 )]2

(9.29)

The graph of the energy function (taking time step of 0.01) drawn in Matlab is shown in Fig. 9.18. When x 2 = 0.32 and x 1 = 0.01, the energy consumption of the S curve acceleration/deceleration algorithm approaches the minimum. The minimum energy consumption of S curve acceleration/deceleration algorithm is found with MATLAB program of the function. The minimum value of M for different time steps is shown in Table 9.6. As seen from Table 9.6, the smaller the time step, the closer the calculated minimum M is to its actual minimum. Therefore, for the S curve acceleration/deceleration algorithm, when the values of t 1 and t 2 are around 0.01 T and

Fig. 9.18 The S curve energy function graph. Note The time coefficient x 1 = t 1 /T, x 2 = t 2 /T, E = M r J 2 θ 2 /K M , and M is the energy consumption coefficient

426

9 Control Optimization and Test Study

Table 9.6 Minimum value of M for different time step

Time step

x2

x1

M min

0.1

0.2

0.1

17.007

0.05

0.3

0.05

14.903

0.01

0.32

0.01

13.779

0.32 T, respectively, the S curve speed profile based on the optimal energy can be obtained. Its energy consumption is E=

13.779rJ 2 θ 2 KM2 T 3

(9.30)

That is when the motion time of acceleration of acceleration, deceleration of acceleration, acceleration of deceleration, and deceleration of deceleration stage is equal to 0.01 T, and the motion time of constant acceleration and constant deceleration stage is equal to 0.31 T, the speed profile obtained is S curve speed profile with the least energy consumption, as shown in Fig. 9.17b. 4.

Comparison of energy consumption of different acceleration/deceleration control algorithms

Based on the above deduction, the energy consumption of different acceleration/deceleration algorithms at different time settings is compared with that of the optimal speed profile, as shown in Table 9.7. As shown in Table 9.7 that (1)

(2)

For the same acceleration/deceleration control algorithm, when the acceleration of acceleration time and the constant acceleration time are set to different values, the energy consumption varies greatly. Especially for the semiS curve acceleration and deceleration algorithm, its energy consumption ratio in Table 9.7 is as high as 6:1. For different acceleration/deceleration control algorithms, the minimum energy consumption of the semi-S curve acceleration/deceleration algorithm is significantly higher than that of the trapezoidal and S curve acceleration/deceleration algorithm.

9.2.3 Speed Optimization of Speedy Flexible Gripping The speed in the acceleration/deceleration process is the key factor to determine the stability of the gripping efficiency. There is a greater instantaneous impact force of fingers on tomato fruits in the speedy gripping process, and the greater the possibility of damage. At the same time, the faster the finger movement, the greater the energy consumption of the motor. Taking the minimum energy consumption and the highest

9.2 Parameter Optimization of Speedy Flexible Gripping [1]

427

Table 9.7 Comparison of between energy consumption of different acceleration/deceleration algorithms and that of optimal speed profiles Parameters Minimum energy consumption

Higher energy consumption

Higher energy consumption

Optimal speed profile

Trapezoidal speed profile

Semi-S curve speed profile

S curve speed profile

(T /3,0)

(T /6,0)

(0.01 T, 0.32 T)

13.5

18

13.779

Percentage of energy consumption

12.5%

50%

14.83%

Time setting (t 1 , t 2 )

(T /7, 0)

(T /7, 0)

(T /7, 2T/7)

Energy consumption coefficient/M

19.06

18.29

17.86

Percentage of energy consumption

58.8%

52.44%

48.83%

Time setting (t 1 , t 2 )

(T /9, 0)

(T /9, 0)

(T /9, 2T/9)

Energy consumption coefficient/M

22.78

108

16.875

Percentage of energy consumption

89.8%

8

40.63%

Time setting (t 1 , t 2 ) Energy consumption coefficient/M

12

Note Percentage of energy consumption = (Energy consumption of acceleration/deceleration curveEnergy dissipation of optimal speed profile)/Energy dissipation of optimal speed profile

stable gripping efficiency as double indexes, the optimization of the constant speed stage can be realized. 1.

The experiment of a relation between the input motor current (output torque) and the gripping force

According to the results of plate compression experiment, the force is linear with the deformation in the range of elastic deformation of light red tomato (Fig. 3.20). In order to achieve accurate flexible gripping, it is necessary to establish an accurate mathematical model between the output torque (current) and gripping force of the motor. Theoretically, the mathematical relation model between the motor output torque and the gripping force can be realized by the force analysis of the mechanical transmission system, but because the dynamic and static friction coefficients of the transmission system are difficult to accurately obtain, the mathematical model

428

9 Control Optimization and Test Study

between motor output torque and finger clamping force is established by experimental measurement. (1)

Experimental materials and methods (1)

(2)

(3)

Experimental materials The Fenguan906 light-red tomato fruits, which were manually picked from a tomato planting base in Dantu District, Zhenjiang City, were used as experimental materials. The transverse and longitudinal diameters of the fruits were 70.28–75.65 mm and 69.84–75.01 mm, respectively. The weight was 177–211 g. Experimental device The experimental device is shown in Fig. 9.19. The finger is mounted on the end-effector, whose open/close is controlled by a DC motor. The maximum output torque of the motor can be changed by setting different current values through the interface program of the motor, thus changing the gripping force of the finger. When the finger holds the tomato, the force sensor mounted on the finger can detect the change of the gripping force in real time. The analog voltage signal of the sensor collected by the data acquisition card is converted into a digital signal and transmitted to the PC, and the output is processed. The range of the force sensor is 50 N, and the accuracy is 0.01 N. Initial current measurement Figure 9.20 shows the required motor current to move the fingers at a constant speed without load, which is up to 420 mA and is used to overcome the transmission friction of the gripper mechanism. Therefore, in the mathematical relation model between motor current and finger gripping force, the setting value of the current should be more than 420 mA.

1. Finger 2. Force sensor 3. Battery 4. AD data acquisition card 5. Wire 6.Support platform 7. PC 8. Fruit

Fig. 9.19 Experimental device for output torque-gripping force of the motor

9.2 Parameter Optimization of Speedy Flexible Gripping [1]

429

Fig. 9.20 Change of the motor current in constant speed stage of fingers

(4)

(2)

Experimental method Ten randomly selected tomato fruits were divided into two groups and numbered. Through the interface program of the motor, set the maximum motor current value of 500–2100 mA with 100 mA interval to grip the fruit, respectively. The force sensor records the force change at each moment in the process of gripping in real time, and finally, the data is saved in PC through AD data acquisition card. At the same time, in order to observe the effect of different speeds on the gripping force at low speed, the motor speed for the first and second groups is set as 25 rpm and 50 rpm, respectively.

Data processing and analysis As shown in Fig. 9.21, for the same current (output torque), the higher the motor speed, and the greater the gripping force. When the motor speed is 25 rpm, the finger speed is calculated to be 0.09 mm/s, which can be regarded as a quasi-static state.

2.

Experiment of steady gripping force

There are some differences in size, weight, and mechanical parameters among tomato fruit individuals, but it is unrealistic to select input motor current for gripping according to the above characteristics of each individual in actual harvesting operation. Since more than 80% of tomato fruits are distributed between 140 and 250 g in weight, the error is less than ±2%, the same stable gripping force may be applied to ensure that the deformation of each body is within the elastic deformation range.

430

9 Control Optimization and Test Study

Fig. 9.21 Experimental results of input motor current-gripping force

(1)

Experimental materials and methods (1)

(2)

(2)

Experimental materials The Fenguan906 light-red tomato fruits, which were manually picked from a tomato planting base in Dantu District, Zhenjiang City, were used as experimental materials. The transverse and longitudinal diameters of the fruits were 65.38–82.04 mm and 64.31–81.08 mm, respectively. The weight was 143–247 g. Experimental methods 10 randomly selected tomato fruits were divided into two groups and numbered. Through the interface program of the motor, set the maximum input motor current starting from 400 mA to increase with 100 mA interval with a set speed of 25 rpm, respectively, to grip the tomato fruit. Record the minimum current Imin required for stable gripping of tomato fruit and the current Imax required for the motor when the tomato fruit is deformed for 5 mm (at the elastic stage).

Experimental results and analysis As shown in Table 9.8, there is a common band [650, 1600] mA of the motor current needed to stable gripping of light-red tomato fruit with the weight distribution of 140–250 g. According to Fig. 9.21, the stable clamping force is within the range of [7.7, 28.9] N. Therefore, in fruit gripping control, same force can be applied on different tomato fruit to realize the stable gripping.

9.2 Parameter Optimization of Speedy Flexible Gripping [1]

431

Table 9.8 Experimental results of stable finger gripping force No.

Transverse diameter (mm)

Longitudinal diameter (mm)

Mass(g)

I min (mA)

I max (mA)

1

74.52

74.61

175

500

1800

2

71.58

71.24

180

520

1850

3

65.81

66.4

143

450

1600

4

77.52

73.68

194

550

1900

5

82.04

81.08

247

650

2100

6

76.28

75.72

200

580

1900

7

74.05

76.32

219

630

1950

8

65.38

64.31

145

460

1600

9

74.11

74.62

201

580

1900

10

72.13

70.46

182

510

1850

3. (1)

Experiment on a relation between finger speed and gripping force Experimental materials and methods (1)

(2)

(2)

Experimental materials The Fenguan906 light-red tomato fruits, which were manually picked from a tomato planting base in Dantu District, Zhenjiang City, were used as experimental materials. The transverse and longitudinal diameters of the fruits were 65.38–82.04 mm and 64.31–81.08 mm, respectively. The weight was 143–247 g. Experimental methods In the experiment, 15 tomato fruits were randomly selected and divided into three groups and numbered. The maximum current of the motor was set to 800 mA, 1000 mA, 1200 mA, 1400 mA, and 1600 mA, respectively. The gripping speed was set to 25 rpm, 50 rpm, 250 rpm, 500 rpm, 750 rpm, 1000 rpm, 1500 rpm, 2000 rpm, 2500 rpm, and 3000 rpm, respectively. The force sensor records the force change in the gripping in real time, and finally, the data is saved in PC through AD data acquisition card.

Experimental results and analysis

The average value of each group of data is obtained, and the experimental results are shown in Fig. 9.22. As can be seen from Fig. 9.22, different current inputs should be applied at different speeds to stabilize the finger gripping of tomato fruit.

432

9 Control Optimization and Test Study

Fig. 9.22 Experiment results of a relationship between the finger speed and gripping force. Note F1 is the minimum stable gripping force and F2 is the maximum stable gripping force

⎧ 800 mA v ≥ 520 ⎪ ⎪ ⎨ 1000 mA 25 ≤ v ≤ 3000 I= ⎪ 1200 mA 25 ≤ v ≤ 1400 ⎪ ⎩ 1400 mA 25 ≤ v ≤ 250

(9.31)

Considering the harvesting efficiency, when 650 mA < I < 1000 mA, the speed of the motor can be higher than 3000 rpm, and the time to complete a stable gripping action is up to 1 s. Thus, efficiency similar to that of manual gripping can be achieved, so as to meet the actual requirements of harvesting operations.

9.3 Control Optimization of Vacuum Sucked Pulling [7] 9.3.1 The Relationship Between Maximum Pulling Speed and Displacement in Acceleration Stage 1.

Theoretical maximum continuous speed of the rack

According to the structure of a feeding mechanism of the suction pad in the damagefree harvesting end-effector shown in Chap. 4, the motor drives the rack forward and backward through the gearbox and the gear-rack meshing transmission. Determined by the speed of the motor, the theoretical maximum continuous speed of the rack (vr0 ) is as

9.3 Control Optimization of Vacuum Sucked Pulling [7]

433

Fig. 9.23 The relationship between the displacement in acceleration stage and the value of acceleration at different speeds

vr0 =

nb1 π m0 z 60ib

(9.32)

where nb1 is the rated speed of the motor in the feeding mechanism of suction pad (rpm), m0 is the modulus of gear and rack in the feeding mechanism of suction pad, z is the tooth number of the gear in the feeding mechanism of suction pad, ib is the reduction of the gearbox in the feeding mechanism of the suction pad as known to be 24. Substituting all known parameters into Eq. (9.32), the result is vr0 = 401.4 mm/s. 2.

Determination of maximum sucked pulling speed

For the law of linear acceleration, there are s0 =

v02 2a0

(9.33)

The relationship between the displacement in the acceleration stage and the value of acceleration at different speeds is shown in Fig. 9.23. It is found from Eq. (9.33) and Fig. 9.23, the displacement in acceleration stage (s0 ) is directly proportional to the square of speed v0 . Therefore, too high pulling speed will make displacement in the acceleration stage (s0 ) too large. According to the pulling force of the high-speed sucked pulling operation, Eq. (9.33), the changing process of the sucked pulling force under dynamic conditions is shown in Fig. 9.24. Combination Eq. (9.33) and Fig. 9.24 show that if the displacement in the acceleration stage (s0 ) is increased, the peak sucked pulling force will increase, even the acceleration–constant–deceleration process cannot be completed within the range of the pulling distance. But at the same time, the low pulling speed also leads to the reduction of efficiency. By synthetically consideration of these two factors, the pulling speed is determined to be 200 mm/s.

434

9 Control Optimization and Test Study

Fig. 9.24 The change process of sucked pulling force under dynamic condition

3. (1)

Relationship between maximum acceleration and static sucked pulling force Equivalent moment of inertia of the feeding mechanism of suction pad

The equivalent moment of inertia of the feeding mechanism of suction pad can be obtained from the following equation: Jbr = Jb1 + Jb

1 vr 2 + mr ( ) wb1 ib2

(9.34)

where J br is the equivalent moment of inertia of the feeding mechanism of the suction pad, gmm2 ; J b1 is the moment of inertia of the motor in the feeding mechanism of the suction pad, it is known from the product catalog as 417 gmm2 ; J b2 is the moment of inertia of the gearbox in the feeding mechanism suction pad, it is known from the product catalog as 40 gmm2 ; mr is the mass of both the rack and the guide rod in the feeding mechanism of the suction pad, it is 80.7 g by measurement; vr is the speed of both the rack and the suction pad, mm/s; ωb1 is the angular speed of the motor in the feeding mechanism of the suction pad, rad/s. Since vr =

ωb1 mz 2ib

(9.35)

So, it is known as   vr ωb1 = mz 2ib

(9.36)

Substituting the known parameters m0 = 0.8, z = 30 into Eq. (9.36) can result vr /ωb1 = 0.5 mm/rad. As a result of Eq. (9.34), J br = 437.2 gmm2 .

9.3 Control Optimization of Vacuum Sucked Pulling [7]

435

Fig. 9.25 The no-load current of the motor in the feeding mechanism of the suction pad

(2)

Friction torque of the feeding mechanism of the suction pad

During the operation of the feeding mechanism of the suction pad, the friction torque acts on it, and its value can be measured and calculated under the condition of no-load. In the EPOS controller interface program, the motor is set to move at a uniform speed of 10 rpm in the constant motion mode under the no-load condition, and the current change during the constant-speed motion is recorded by the data recording function of the interface program (Fig. 9.25). Thus, the friction torque of the feeding mechanism of the suction pad is obtained from the following Equation: Mbf = γb M · Ibf /1000

(9.37)

where M bf is the friction torque of the feeding mechanism of the suction pad, mNm; γ bM is the torque constant of motor in the feeding mechanism of the suction pad, it is known from the product catalog as 24.4 mNm/A; I bf is the current of the motor required to overcome the friction torque in the feeding mechanism of the suction pad, it is known from Fig. 9.24 as 90 mA. So, the friction torque of the feeding mechanism of the suction pad can be obtained from Eq. (9.37) as 2.20 mNm. (3)

The load torque of the motor for sucked pulling

During the process of sucking fruit and pulling it back, the rack must provide the pulling force F p needed for the sucked pulling, the load torque of the motor is Mbw = Fp

vr = 0.5Fp ωb1

(9.38)

436

9 Control Optimization and Test Study

where M bw is the load torque of the motor in the feeding mechanism of the suction pad. (4)

Maximum acceleration

During the process of sucking fruit and pulling it back, the feeding mechanism of the suction pad is driven by the motor torque to overcome the effect of load torque and friction torque and accelerate the rack. The theoretical maximum acceleration is β0 =

Mb − Mbf − Mbw0 × 106 Jbr

(9.39)

where β 0 is the theoretical maximum angular acceleration of the rack in sucked pulling, rad/s2 ; M b is the rated torque of the motor in the feeding mechanism of the suction pad, it is known from the product catalog as 12.1 mNm; M bw0 is the load torque required to supply the pulling force F p0 in constant-speed motion, mNm, M bw0 = 0.5F p0 . And the maximum acceleration of the rack is a00 =

Mb − Mbf − Mbw0 mz β0 = × 106 2ib2 2Jbr

(9.40)

where a00 is the maximum acceleration of both the rack and the suction pad, mm/s2 . It is also the maximum acceleration of fruit in sucked pulling when calculating the threshold of vacuum degree. As shown in Fig. 9.26, when the pulling distance is less than 10 mm, the required static pulling force is small, and the success rate of sucked pulling exceeds 98.6% when the pulling force is 2.0 N. For F p0 (s0 ) = 2.0 N, it is obtained from Eq. (9.39) that a0 = 10.18 m/s2 Fig. 9.26 Probability distribution of static pulling force at different sucked pulling distances

9.3 Control Optimization of Vacuum Sucked Pulling [7]

437

Substituting it into Eq. (9.33), it is solved that s0 = 1.96 mm

9.3.2 The Relationship Between the Dynamic Pulling Force and the Threshold of Vacuum Degree By substituting the known data into the equation of dynamic pulling force in highspeed sucked pulling of fruit, Eq. (7.34), the normal probability distribution of the instantaneous pulling force at the end of the acceleration stage was obtained (Fig. 9.27). When the pulling force F p (s0 ) reached 5.0 N, the probability of successful pulling reached 99.8%. The threshold of the vacuum degree is [ pu ] = −

1000 Fp (s0 ) = −24.9 KPa 64π

From Fig. 9.24, except for the peak pulling force at the end of the acceleration stage, there will be peaks at the beginning of the deceleration stage and at the end of the sucked pulling motion, respectively. The peak value at the beginning of the deceleration stage is equal to the value of the static analysis Fp (x0 − s0 ) = Fp0 (x0 − s0 )

(9.41)

While the peak value at the end of the sucked pulling motion is equal to the difference between the static analysis value and the deceleration term Fp (s0 ) = Fp0 (s0 ) − 10−6 ma0 Fig. 9.27 The normal distribution curve of the dynamic pulling force at the end of the acceleration stag

(9.42)

438

9 Control Optimization and Test Study

Fig. 9.28 The normal distribution curve of the dynamic pulling force at the end of sucked pulling motion

From Fig. 7.18, the static threshold of vacuum degree is 16 kPa when the suction pull is 30 mm, so the threshold of vacuum degree at the beginning of deceleration and the sucked pulling motion are lower than 16 kPa. The normal probability distribution curve of the pulling force at the end of sucked pulling motion was obtained by substituting the known data and fruit mass as normal distribution series into Eq. (9.42) (Fig. 9.28). It could be found that if the fruit size did not restrict the pulling distance, only 2.5 N was needed to realize the 30 mm pulling distance for almost all fruit.

9.3.3 Optimization of Displacement/Position Parameters for Sucked Pulling of Fruit 1.

The relationship between rack stroke and pulling distance (1)

The theoretical relationship between rack stroke and pulling distance In the actual operation of the vacuum sucked pulling of on-tree fruit, the sucked pulling motion is realized by driving the rack movement. The rack displacement s is not consistent with the actual pulling displacement x of the fruit due to the existence of the deformation of the suction pad, z. As shown in Fig. 9.29 x + z − z = s + z

(9.43)

z = x − s( z ≥ 0)

(9.44)

That is

(2)

The moments of peak pulling force According to Eq. (7.4), the compressed deformation of the suction pad depends on the sucked pulling force and the level of vacuum degree, that is, the difference between vacuum suction force and pulling force leads

9.3 Control Optimization of Vacuum Sucked Pulling [7]

439

Fig. 9.29 Relationship between rack stroke and pulling distance

(3)

to the compression deformation of the suction pad. At the same time, the vacuum degree maintenance property of the check valve decides that the vacuum degree will gradually decrease in the process of sucked pulling. The three moments of peak pulling force are as follows: t(s0 ) = av00 = 0.02 s; 0 t(x0 − s0 ) = av00 + H0 −2s = 0.15 s; v0 2v0 H0 −2s0 t(x0 ) = a0 + v0 = 0.17 s. The peak level of vacuum degree and sucking force The check valve is a directional control valve which makes use of the pressure difference on both sides to make the airflow only one way but not reverse flow. The check valve has the ability to maintain a vacuum degree under the condition of interruption or misoperation of the vacuum system. At the same time, reasonable use of the vacuum degree maintenance capacity of the check valve can effectively save the consumption of compressed air and energy. The rational use of check valves is very

440

9 Control Optimization and Test Study

Fig. 9.30 Curve of vacuum degree maintenance of the check valve

important for the energy-saving and efficiency of the vacuum suction system. It is found that the check valve can maintain a vacuum state of about 7 s after closing the air supply solenoid valve and stopping the air supply of compressed air under the condition of sucking fruit, and can maintain a maximum vacuum of over 63% for up to 2.1 s (Fig. 9.30). The vacuum degree is approximately linearly with time when it exceeds 15%. If taking the moment that vacuum degree starts to fall as the 0 points of time pu = 16.48 t − 89.4

(9.45)

Its goodness of fit (R2 ) reaches 0.9999. According to the drop curve equation of vacuum degree maintained by check valve, Eq. (9.45), the negative vacuum pressure at three moments of peak suction force in sucked pulling of on-tree fruit are as follows: (1) (2) (3)

pu (t(s0 )) = −89.1 KPa; pu (t(x0 − s0 )) = −86.9 KPa; pu (t(x0 )) = −86.6 KPa

According to experimental results of the influence of vacuum degree on pull-off force and Eq. (7.17), all the peak pulling force at the above three moments in sucked pulling reaches more than 10 N. After the sucked pulling is completed, the suction pad will continue to suck the fruit until the fingers finish the fruit gripping. The time of finger gripping is predicted to be 2 s, and the negative vacuum pressure is − 53.6 kPa according to Eq. (9.45). The vacuum suction is still above 10 N according to experimental results of the influence of the vacuum degree on pull-off force and Eq. (7.17).

9.3 Control Optimization of Vacuum Sucked Pulling [7]

(4)

441

The actual relation between rack stroke and pulling distance From the analysis above, it can be seen that the maximum pulling force is only 5.0 N during the sucked pulling process, so the difference between vacuum suction force and pulling force is more than 5.0 N. According to the compression test of suction pad and Eq. (7.27), the difference will lead to the complete compression of the suction pad.

Therefore, in the process of sucked pulling of on-tree fruit, the suction pad first contacts the fruit first and is successfully sucked, and then the suction pad reaches full compression of z0 = 8 mm and keeps the state until the fruit is successfully gripped. If the suction pad stops at the moment it contacts the fruit, the actual stoke of the rack in the sucked pulling is only 22 mm. In practice, the travel allowance of 8 mm is added and the rack stroke H0 is still set as 30 mm to ensure the full contact between the suction pad and the fruit surface to form a closed space, so as to ensure the success rate of sucked pulling operation. 2. (1)

Initial position of the end-effector and the suction pad Initial position of the end-effector

During operation, the end-effector is first transported to the nearby position of the fruit by the manipulator, and then the suction pad moves forward from the initial position to suck and pull the fruit to the gripping center. The initial position of the end-effector and the suction pad is the beginning of the sucked pulling operation. Guarantee of the reasonable initial position of the end-effector and the suction pad is the foundation of a successful sucked pulling operation. Because the size of different fruits varies greatly, as shown in Fig. 9.31, the distance from the gripping center to the fruit center can be set as 30 mm before sucked pulling operation with the help of the vision system of the harvesting robot.

Fig. 9.31 Initial position of the end-effector and the suction pad in the sucked pulling operation

442

(2)

9 Control Optimization and Test Study

Initial position of the end-effector

At the beginning of motion of the suction pad, the suction pad is in a natural state. After moving forward H0 , the suction pad sucks the fruit and it is fully compressed. Therefore, in the beginning, the distance between the lip of the suction pad and the nearest point of the fruit is x r = H 0 − 8, that is, 22 mm (Fig. 9.31). During operation, the distance is measured by a vision system or a distance sensor, and then it is adjusted by the control system to a predetermined distance to determine the initial position of the suction pad.

9.3.4 Optimization of Control Mode for Motion Coordination 1. (1)

Coordination between suction and rack movement Different modes of coordination between suction and rack movement

The sucked pulling of tomato fruit by the vacuum suction system needs the coordinated action of the vacuum system and mechanical system. In each cycle of operation, the vacuum generation/release, sucking/blowing, and moving forward stop and backward must be closely coordinated. The effective action coordination of these two systems is the key to ensure the success of sucked pulling. The effects of different coordination modes are also significantly different. In order to effectively save energy, the vacuum suction system adopts intermittent air supply mode. In order to suck fruit successfully, the time to start the vacuum is related to the time when the suction pad reaches the nearest point of the fruit. In theory, the optimal mode of energy and efficiency is to produce the vacuum while touching the fruit. But in practice, because of the error of fruit distance information and the error of mechanical movement, the vacuum starting, and the fruit touching cannot realize at the same time. Therefore, there are two types of coordinated control methods: (1)

(2)

Supply compressed air, then touch the fruit Firstly, the air supply solenoid valve is opened to produce a vacuum, and then the rack drives the suction pad to reach the fruit and contact it so that the vacuum degree rises rapidly and the fruit is sucked. Touch the fruit, then supply compressed air Firstly, the rack drives the suction pad to move toward the fruit and contacts with it, and then the air supply solenoid valve opens to produce a vacuum rapidly, realizing the suction of the fruit. These two modes of coordination have their own characteristics:

(1)

Full contact between the fruit and the suction pad to form a closed space is the key to successful suction and also the key to determining the feasibility of the

9.3 Control Optimization of Vacuum Sucked Pulling [7]

(2)

(2)

443

two coordination models. For the “air supply then contact” mode, according to the distance information of the fruit and the rack movement speed, the opening time of the air supply solenoid valve can be set to make it produce vacuum before contacting the fruit. The jump of the vacuum when contacting the fruit can be used as the criterion of contacting and successfully suction, which provides the judgment condition to decide moving stop of the rack. For the “contact then air supply” mode, the touch between the suction pad and the fruit can be determined by visual or contact sensing information to achieve judgment and feedback control. By contrast, the senseless “contact then air supply” mode makes the suction pad fully compressed by the aforementioned enlarged rack stroke to ensure reliable contact with the fruit, and opens the air supply solenoid valve then to judge whether it has touched and sucked the fruit successfully by the rapid rise of vacuum. If the answer is “no”, it will continue to move forward and to be judged again. For energy consumption, firstly, the timing of air supply for “air supply then contact” mode should be set manually, causing waste of compressed air before reaching the fruit. Secondly, the consumption of compressed air in each cycle is different and cannot be accurately controlled. By contrast, the “contact then air supply” mode has not any waste compressed air, which is conducive to energy-saving operations. Optimization of motion parameters in different modes (1)

“Air supply then contact” mode

When the control system detects the jumping edge of the vacuum degree, it sends a stop command, then the motor slows down and stops moving. The deceleration rate of the motor can be set in the I215 variable of PMAC. The time and displacement of the motor in the deceleration stage are as follows after the stop command is issued tc = sc =

vp ac

(9.46)

1 vp2 2 ac

(9.47)

where t c is the time needed for the deceleration of the rack after the stop command is issued, s; sc is the displacement of the rack after stopping command is issued, mm; vp is the maximum speed of rack moving forward, mm/s; ac is the deceleration rate of the rack, mm/s2 . The maximum deceleration rate of the rack moving forward is ac =

Mb + Mbf × 106 2Jbr

(9.48)

444

9 Control Optimization and Test Study

The ac is calculated out to be 16.4 m/s2 by substituting the known parameters into Eq. (9.48). By Eqs. (9.46) and (9.47), when the rack is moving forward at the theoretical maximum sustained speed, the time and displacement of the deceleration stage are 0.025 s and 4.91 mm, respectively. Similarly, the maximum acceleration, time, and displacement of the rack without load can be calculated under the symmetrical acceleration/deceleration algorithm, M −M respectively ad = b2Jbr bf × 106 = 11.3 m/ s2 td = 0.036 s sd = 7.13 mm (2)

Senseless “contact then air supply” mode

For “contact then air supply” mode without sensing, the rack moves forward and then stops according to the predetermined stroke, and then the supply solenoid valve is open. If the vacuum degree jumps rapidly to a certain value, it means the successful suction; otherwise, the suction pad moves forward furtherly at a certain distance and the vacuum degree is detected to judge again. 2.

Coordination of vacuum release and mechanical gripping

The sucked pulling of fruits creates favorable conditions for carrying out the gripping action. However, it is necessary to achieve good coordination between sucked pulling and subsequent gripping actions in order to ensure that the fruit is harvested smoothly. The key of coordination is how to release the fruit by the suction pad and the timing of both the release and the finger gripping. The interference of release and gripping may lead to fruit abscission, bruise, or failure of gripping. There may be several coordination modes: (1)

(2)

Release then grip The suction pad first releases the fruit and then the finger grips it. As the suction pad releases the fruit, the fruit will return to its original position before the sucked pulling, the gripping will not be completed and the sucked pulling doesn’t mean anything. Grip then actively release After reaching the predetermined pulling distance, the suction pad still maintains a certain suction force on the fruit due to the vacuum maintenance effect of the check valve. Therefore, after the finger completes the gripping of the fruit, the suction pad actively releases the fruit by opening the air-blowing solenoid valve to blow the fruit off.

9.3 Control Optimization of Vacuum Sucked Pulling [7]

(3)

445

Grip then automatically release Due to the limited vacuum maintenance capacity of the check valve, after the finger completes the gripping of the fruit, the vacuum degree continues to decline until the fruit is released automatically.

The difference between the second and third modes is that the active release is achieved by blowing, which adds both the control complexity and additional compressed air consumption. In fact, the process of peduncle detaching and fruit transporting by manipulator exist between the gripping and the fruit placing into the box. Under the present technical conditions, the time of this process is sufficient to meet the needs of the automatic release of the fruit with the vacuum degree drop under the action of the check valve. Therefore, “grip then automatically release” is regarded as our most ideal choice.

9.4 Hand–Arm Coordination Control for Speedy Flexible Harvesting [8] 9.4.1 Hand–Arm Coordinative Control Modes 1.

Theoretical hand–arm coordinative control modes

Harvesting robots have to fulfill the fruit harvesting task relying on a hand-arm coordinative motion. And, there are different coordinative motion modes (Table 9.9): (1)

(2)

(3) (4)

Ordinal mode One of the end-effector and the manipulator will act when the other is stopped, and end-effector and manipulator will act one by one; Alternating mode Both the manipulator and the end-effector have to act and stop several times alternately; Parallel mode The end-effector and manipulator may act simultaneously; Composite mode There may be mixing of the above modes in the operation cycle. In harvesting, possible motion sequence is as follows:

Table 9.9 Different coordinative control mode of manipulator and end-effector Control mode

Ordinal

Alternating

Parallel

Composite

Hand-arm motion sequence

●★

●★●★

● ★

●★ ● ★★ ●

● Motion of manipulator; ★ Motion of end-effector

446

9 Control Optimization and Test Study

Difficulty and complexity increase gradually from ordinal mode to composite mode. In practical, fruit harvesting operation, usually, manipulator sends end-effector to harvesting position, and after the fruit is detached from the plant, manipulator sends end-effector with fruit back to crate and releases it, so both manipulator and endeffector have to act and stop several times in one harvesting cycle, and an alternating or composite mode is necessary. Specific to this end-effector that has been developed, several mechanical motions and solenoid actions will occur in one harvesting cycle. To achieve a satisfactory success rate and efficiency of robotic harvesting, it is necessary to select an optimal motion sequence and to realize it by constructing a corresponding control mode. 2.

Flow charts of robotic harvesting for this hand-arm system related to different coordinative control modes

Based on basic harvesting method and hardware structure, two motion sequences were put forward and compared by experiment. Figures 9.32 and 9.33 show task sequences of the harvesting procedure that were used during the experiments, which are related to alternating mode and composite mode, respectively. The basic flow charts of robotic harvesting actions are as follows: (1)

(2)

(3)

(4)

(5)

The control system performs path planning firstly and moves the manipulator to send the end-effector to harvesting position near the object tomato fruit. The position of end-effector from object fruit was detected by distance sensors. Then the air supply solenoid valve is open to producing the vacuum and the suction pad moved forward to suck the fruit. The success of sucking was judged by negative pressure. And then the needed displacement of the suction pad moving back from sucking position to gripping position is calculated by a control system and the fruit is pulled back. The fingers start to close, at once force sensors on fingertips detect that the gripping force had reached the threshold, the blowing solenoid valve is open to blow the fruit off, so the fruit is able to be bent off from peduncle by wrist joint rotation of manipulator. At last, the manipulator sends the end-effector back with fruit and the fingers open to release fruit into the crate.

There is a difference at step [9] “suck and pull back” between the above two control flows. In Fig. 9.32 which is an alternating mode, suction pad pulls the fruit back to the gripping position. While in Fig. 9.33 which is a composite mode, the suction pad pulls the fruit back, and the manipulator sends the whole end-effector forward at the same time. In alternating mode, with the help of the relay-based hand– arm communication circuit (Figs. 4.27, 4.28), the manipulator and the end-effector act and stop alternately by setting and changing the digital input status of PMAC and JRC several times one after another. In composite mode, by setting the digital input variables of PMAC as “0” and JRC as “on” at the same time, the manipulator motion [9] and the end-effector motion [9a] execute simultaneously in programs of WINSCAPS and PMAC.

9.4 Hand–Arm Coordination Control for Speedy Flexible Harvesting [8] [1] Initialize manipulator and end-effector

447

[17] Open fingers to home position

[2] Path planning

[16] Move manipulator to send end-effector with fruit to crate and release it

[3] Move manipulator to send endeffector to harvesting position

[15] Path planning Yes

[4] Has end-effector arrive at harvesting position?

No

No

[14] Has fruit been detached?

Yes [5] Open air supply solenoid valve

[13] Wrist joint rotation to bend peduncle

[6] Move suction pad forward to suck fruit

[12] Open solenoid valve to blow fruit off Yes

[7] Has negative pressure reach threshold?

No

[11] Has gripping collision force reach threashold?

No

Yes [8] Calculate displacement needed of suction pad back to gripping position

[9] Suck and pull fruit back to gripping positon

[10] Close fingers to grip fruit

Fig. 9.32 Flow chart of the harvest operation under alternating model

9.4.2 Hand–Arm Coordinated Harvesting Experiments 1. Experimental materials and methods Jinpeng 5 tomato plants were randomly collected from a vegetable base in Zhenjiang City, Jiangsu Province, China, and transported to the Laboratory of Modern Agricultural Equipment and Technology in Jiangsu University. All experiments were carried out within one day at a temperature range of 15–25 °C. The trunk and root of tomato plant were mounted on a vertical stand. All fruits were numbered, whose diameters and length of corresponding peduncle length from the main stem to calyx were measured using a micrometer caliper with a sensitivity of 0.01 mm. Two control flows as shown in Figs. 9.32 and 9.33 were executed 30 times with fruit samples

448

9 Control Optimization and Test Study [1] Initialize manipulator and end-effector

[17] Open fingers to home position

[2] Path planning

[16] Move manipulator to send end-effector with fruit to crate and release it

[3] Move manipulator to send endeffector to harvesting position

[15] Path planning Yes

[4] Has end-effector arrive at harvesting position?

No

No

[14] Has fruit been detached?

Yes [5] Open air supply solenoid valve

[13] Wrist joint rotation to bend peduncle

[6] Move suction pad forward to suck fruit

[12] Open solenoid valve to blow fruit off Yes

[7] Has negative pressure reach threshold?

[11] Has gripping collision force reach threashold?

No

Yes [8] Calculate displacement needed of suction pad back to gripping position

[9] Suck and pull fruit back for ½ displacement [9a] Move manipulator to send end-effector forward for ½ displacement

No

[10] Close fingers to grip fruit

Fig. 9.33 Flow chart of the harvest operation under composite mode

selected randomly, respectively. In the experiments, the fruit was gripped in the horizontal direction and bent off by upward rotation motion of angle 40°, respectively (Fig. 9.34). 2. (1)

Results and discussion Feasibility of robotic harvesting for two coordinative control modes

Experiment results proved that the self-designed end-effector and commercial manipulator were able to be integrated successfully, and communication between JRC and PMAC controller for either alternating or composite coordinative mode was feasible. In harvesting, JRC performed path plan and inverse kinematics calculation for the

9.4 Hand–Arm Coordination Control for Speedy Flexible Harvesting [8]

449

Fig. 9.34 Hand-arm coordinated harvesting experiments

end-effector automatically when the coordinate position of fruit and crate were given, which was convenient and valuable. However, the total control system had to be composed of two independent controllers and a PC, which were still far from an ideal highly integrated structure. (2)

Success rate of different coordinative control modes

There are several steps from the end-effector reaches harvesting position to detaching fruit from the plant, including sucking, pulling, gripping, and detaching, so the success rate of harvesting is a composite result of every step. Experimental results are shown in Table 9.10. It is found that 25 fruits were harvested successfully for composite mode, while only 21 fruits were harvested successfully for an alternating mode. The main difference appeared at the step of pulling back. (3)

Source of failure for different coordinative control modes

The results showed that the success rate of fruit harvesting reached 100% by bending up to 40°, and the failure occurred mainly in the sucked pulling process, which may also have adverse effects on the gripping: (1)

A number of fruits were not able to be sucked by the suction pad. It is difficult to produce enclosure space and then to produce a vacuum between the surface

Table 9.10 Experiment result of the success rate of harvesting for different coordinative control modes Control mode

Number of Samples

Number of Failure Sucking

Pulling

Gripping

Detaching

Success rate of harvesting/%

Alternating

30

2

7

0

0

70.0

Composite

30

2

2

1

0

83.3

450

9 Control Optimization and Test Study

(a) Compressed deformation

(b) Flexural deformation

Fig. 9.35 Deformation compensation ability of the multi-bellow suction pad

of the suction pad and some fruit, attributed to the irregular shape of the fruit surface. The success rate of sucking is mainly related to the difficulty of forming the enclosed space between the suction lip and the fruit surface. For fruits with a larger diameter, better shape, and surface quality, the suction can be easily realized without excessive stroke allowance. For smaller fruit, irregular surface shape, or worse relative position between the suction pad and the fruit, to make full use of the compensation and adaptability of the suction pad, the success rate of sucking can be improved by increasing stroke allowance. As shown in Fig. 9.35a, the compressed deformation of the tip of the suction pad can realize a certain degree of compensation for a certain angle deviation between the tip and the surface of the fruit or for the irregular surface. As shown in Fig. 9.35b, for the relative positional deviation between the suction pad and the fruit, the flexural deformation of the suction pad can also supply a certain degree of compensation. Although the vacuum suction pad shows good compensation and adaptability, it is still difficult to achieve the suction even if the stroke continues to be increased for a few too worse conditions. (2)

Main factors that lead to failure in pulling back step are displacement of suction pad, diameter, and length of peduncle. It is easy to understand that fruits with a shorter length and smaller diameter are more difficult to be pulled for a certain distance, while fruit movement for shorter displacement may be realized more easily.

During the sucked pulling process, if the pulling force Fp reaches the suction force Fs , that is Fs − Fp = 0, the deformation of the suction pad is completely restored, resulting in the fruit detaching from the suction pad and unwilling failure (Fig. 9.36a).

9.4 Hand–Arm Coordination Control for Speedy Flexible Harvesting [8]

(a) Unwilling detachment of fruit from the suction pad

451

(b) Unwilling detachment of fruit from the plant

Fig. 9.36 Different phenomena of sucked pulling failure

At the same time, both the bending moment and radial pulling force are applied to the abscission layer, which also increases with the pulling distance. When it exceeds the bend-off moment or pull-off force of the abscission layer, the fruit will be accidentally bent off or pulled off (Fig. 9.36b). The higher the vacuum degree, the lower the probability of fruit detachment from the suction pad, but the higher the probability of fruit detachment from the plant. (3)

The results showed that without the sucked pulling operation or with insufficient pulling distance, the gripping might cannot be completed, or the fruit deviated from the gripping center causing a bruise to adjacent fruit (Fig. 9.37). It indicated that the sucked pulling has a significant effect on avoiding the gripping

(a) Gripping deviation Fig. 9.37 Interference phenomena in fruit gripping

(b) Gripping bruise

452

9 Control Optimization and Test Study

For the composite control mode, the success rate of sucked pulling is obviously higher than that of the alternating control mode because of the decrease of the pulling distance. However, the probability of gripping interference will increase accordingly. But generally speaking, for the reasonable pulling distance of suction, the comprehensive success rate of the composite control mode has a prominent advantage. 3.

Execution time of different coordinative control modes

The time required from the end-effector reaches harvesting position to detaching fruit from the plant is different for every harvesting cycle, related to a difference of fruit diameter and control mode. For different diameters of fruit, distance from endeffector at harvesting position to the centroid of fruit is different, so displacement needed of suction pad back to the gripping position is different, too. Meanwhile, the displacement needed of the finger to grip fruit is also different for different fruit diameters. Average cycle time for alternative mode and for composite mode was 6.2 s and 6.0 s, respectively. Harvesting for composite mode was a little faster than for alternative mode since only 1/2 of pulling back displacement is needed of suction pad to pull back fruit to harvesting position. However, a difference in cycle time between these two control modes is very small, in view of the fast and short motion of pulling back.

References 1. Li Z (2009) Acc/Dec Process of the Grip System on Tomato Havesting Robot[D]. Jiangsu University 2. Tau D (1999) Pmac2a-PC/104 Hardware Reference Manual[Z]. DELTA TAU Data System. Inc, USA 3. Xiao A (2007) Automatic control system and its application[M]. Tsinghua University Press 4. Luo T (2000) Signal system and automatic control principle[M]. China Machine Press 5. Motor M (2007) Epos 24/1—Positioning Controller[Z] 6. Cong S, Shang W (2005) Performance study of point to point control curve in motion control[J]. Mach Electron 7:16–19 7. Liu, J (2010) Analysis and optimal control of vacuum suction system for tomato harvesting robot[D], Jiangsu University 8. Liu J, Li Z, Wang F et al (2013) Hand-arm coordination for a tomato harvesting robot based on commercial manipulator[C]. In: Proceedings of the IEEE international conference on robotics and biomimetics, pp 2715–2720

Nomenclature

Symbol

Meaning

Units

a(t)

Acceleration of fruit in the horizontal movement direction

mm/s2

a

Distance from the centroid of the fruit to the nearest point of the adjacent fruit after sucked pulling

mm

a

Stiffness coefficient

N.mm2 . rad-1

Aa

Absorptivity of materials to laser energy

%

ad

Thermal diffusivity of materials

mm2 /s

A

The actual area of focal spot for certain incident angle

mm2

a0

Absolute value of the constant acceleration

mm/s2

A0

The area of focal spot when the incident angle is 0°

mm2

a 00

Maximum acceleration of both the rack and the suction pad

mm/s2

A2

Outer sectional area of suction pad

mm

ac

Deceleration rate of the rack

mm/s2

Ac

Specimen’s cross-sectional area

mm2

Ac

Compressed cross-sectional area

mm2

Ae

Effective area of suction pad

mm2

As

Suction area

mm2

at

Torsion stiffness coefficient

N.mm2 .rad-1

At

Tensile cross-sectional area

mm2

AV (n)

Actual speed in the servo cycle n

c

Distance from the centroid of the fruit to the nearest point of the adjacent fruit before sucked pulling

mm

c1

Span of the beam

mm

CA(n)

Instruction acceleration in the servo cycle n

C

c

Distance from the neutral axis to the compression edge of the beam mm

C

i

Relaxation coefficient of shear/bulk modulus

Coefficient

C

t

Distance from the neutral axis to the tensile edge of the beam

mm (continued)

© Science Press, Beijing and Springer Nature Singapore Pte Ltd. 2021 J. Liu et al., Rapid Damage-Free Robotic Harvesting of Tomatoes, Springer Tracts in Mechanical Engineering, https://doi.org/10.1007/978-981-16-1284-8

453

454

Nomenclature

(continued) Symbol

Meaning

CV (n)

Instruction speed in the servo cycle n

Units

D

Fruit deformation

Ds

Peduncle diameter

mm

d

Specimen’s cross-sectional thickness

mm

D0

Constant deformation of the model

mm

d

mm

Diameter of the abscission layer

mm

Da

Arithmetic mean diameter

mm

d

Diameter of focal spot

mm

a

f

Dg

Geometric mean diameter

mm

D max

Maximum integrated displacement of the material

mm

e(k)

Position tracking error

%

E(t)

Relaxation modulus

MPa

E

Equilibrium modulus of elasticity

MPa

0

e0

Adjustable Parameter

Constant

E

1

Instantaneous elastic coefficient

N/mm

E

2

Delayed elastic coefficient

N/mm

E

c

Compression elastic modulus

MPa

E

l

Tension elastic modulus

MPa

E

Mi

Modulus of elasticity of the ith Maxwell body

MPa

E

p

Plastic strain energy

J

E

r

Rupture energy

J

E

rec

Failure energy during compression testing

kJ/m3

E

rel

Failure energy during tension testing

kJ/m3

E

t

Tension elastic modulus

MPa

Force loading on the beam

N

f

Friction between robot finger and tomato

N

F(t)

Force applied on the model

N

F

0

Maximum static gripping force

N

F

a

Tension limit of the abscission layer

N

F

c

Normal contact force between the target and the suction pad

N

F

cmax

Elastic peak force

N

F

cw

Radial wall tissue of three-locular tomato

F

d

F

Sliding friction force

N

FE(n)

Error in the servo cycle n

F

max

Ultimate force when material breaks

F

n

Externally force applied to nipple of the suction pad in the positive N axial direction

N

(continued)

Nomenclature

455

(continued) Symbol

Meaning

Units

F

A

Axial compression force

N

F

N

Normal force between the surfaces

N

F

p

[F F

p

]

Pulling force

N

Pull-off force applied by the end-effector

N

p

(a)

Extra tensile force required to obtain this acceleration a

N

F

p

 (x)

Tensile force in high-speed operation

N

F

r

Radial force applied to the stem

N

[F s ]

Pull-off force of suction pad

N

F

s

Suction force between the target and the suction pad

N

F

t

Tangential force applied to the stem

N

F

x

Horizontal force applied to the fruit-stem system

N

F

y

Vertical force applied to the fruit-stem system

N

G

Weight of the fruit

N

G(t)

Shear relaxation kernel function

Function

G0

Initial shear/bulk modulus

MPa

G0 &K

Transient moduli of the viscoelastic material

0

G 0 or G ∞ Final shear/bulk modulus

MPa

G∞&Gi

Shear moduli

G b (z)

Transfer function of the proportional term

G D (z)

Transfer function of differential terms

Function

G P (z)

Transfer function of the PI term in the position loop

Function

IAE

Integral Absolute Error

%

ib

Reduction of the gearbox in the feeding mechanism of the suction pad

I

Current of the motor required to overcome the friction torque in the mA feeding mechanism of the suction pad

bf

Function

IE(n)

Integral of the following error in the servo cycle n

I

Moment of inertia of the cross-section about the neutral axis

mm4

Total moment of inertia

kg·m2

Moment of inertia of the motor in the feeding mechanism of the suction pad

gmm2

z

J J

b1

J

b2

J

br

Moment of inertia of the gearbox in the feeding mechanism suction gmm2 pad Equivalent moment of inertia of the feeding mechanism of the suction pad

gmm2

K(t)

Volume relaxation kernel function

Function

k0

equivalent bending stiffness coefficient of the entire stem

m.Nm/rad

K

∞

&K

i

Bulk moduli (continued)

456

Nomenclature

(continued) Symbol

Meaning

Units

ka

Laser absorptivity on the irradiated surface

Constant

ka

Bending stiffness coefficient of the join point b/w the peduncle and mNm/rad stalk

K

Function of acceleration feed-forward gain

Function

kb

Bending stiffness coefficient of the join point b/w the rachis and peduncle

m.Nm/rad

kc

Bending stiffness coefficient of the join point b/w the pedicel and rachis

m.Nm/rad

kd

Communication efficiency ratio in air of the output optical power of Constant a laser

K

aff

d

Function of differential gain function

Function

K

f

Beam parameter product

mm.rad

K

I

Function of integration gain

K

M

Torque constant of the motor

Constant

kp

Compression stiffness coefficient of the suction pad

N/mm

K

Function of proportional gain

Function

kt

The transmissivity of a focusing lens

Constant

kt

Bending stiffness coefficient

m.Nm/rad

K

p

Function of velocity feed-forward gain

Function

L

Length of the entire stem

mm

Lp

Actual length of drilling through path

mm

Ls

Initial length of specimens

mm

M

Torque applied to the stem by the virtual spring hinge

m.Nm

vff

m M

p

Mass of material

g

Bending moment at any point of the beam

N·mm

m0

Modulus of gear and rack in the feeding mechanism of suction pad

M

0

The fresh weight

g

M

A

Moment applied to the stem

m.Nm

M

b

Rated torque of the motor in the feeding mechanism of the suction pad

M

bf

Friction torque of the feeding mechanism of the suction pad

m.Nm

M

bw

Load torque of the motor in the feeding mechanism of the suction pad

m.Nm

M

bw0

Load torque required to supply the pulling force

m.Nm

M

d

The dry weight

g

Equivalent mass of the electro-mechanical gripper system to the finger

kg

me M

L

Mmax

The loss weight

%

Maximum bending moment of the beam

N·mm (continued)

Nomenclature

457

(continued) Symbol

Meaning

Units

M

The weight of day n

g

n

MNF

Model Neutral File

mr

Mass of both the rack and the guide rod in the feeding mechanism of the suction pad

g

N

Grabbing force (normal force) applied by the robot finger to the tomato

N

n b1

Rated speed of the motor in the feeding mechanism of suction pad

rpm

P

Optical output power of laser

W

pc

Normal contact pressure per area

KPa

PI

Proportional term and integral term

PID

Proportional-integral-differential

RH

Relative humidity

r

Loading slope

k

%

s

Movement distance of the suction pad

mm

Sa

Displacement in the acceleration section

mm

Sc

Displacement in constant speed stage

mm

sc

Displacement of the rack after stopping command

mm

sg(i)

Total rate of gripping interference after the vacuum sucked pulling in ith harvesting round

%

sm

Failure rate of sucked pulling

%

s r(i)

Actual success rate of gripping in the ith harvesting round

%

ta

Time of peak deformation of creep

s

T

Temperature rise at the center of the laser focal spot

K

T

t

Drilling through time

s

T

0

Drilling through time when the incident angle is 0°

s

tc

Time needed for the deceleration of the rack after the stop command s

T

cw

Radial wall tissue of four-locular tomato

T

f

Torsion applied on the abscission layer

m.Nm

[T f ]

Fatigue limit for torsion of the abscission layer

m.Nm

T

Locular tissue of four-locular tomato

L

V

Total mass of tomato

g

v0

Constant speed of the fingers

mm/s

vp

Maximum speed of rack moving forward

mm/s

V

Volume of mesocarp

cm3

p

vr

Speed of both the rack and the suction pad

mm/s

v r0

Theoretical maximum continuous speed of the rack

mm

w

Specimen’s cross- sectional width

mm

wc

Water content of tomato

% (continued)

458

Nomenclature

(continued) Symbol

Meaning

xg

Geometric permissible maximum movement distance of the fruit in mm the horizontal direction

Units

x min

Theoretical minimum sucked pulling distance

mm

yc

The distance from the neutral axis to the compression side of the beam

mm

y max

Permissible maximum movement distance of the fruit in the vertical direction

mm

yt

Distance from the neutral axis to the tensile side of the beam

mm

Z

Tooth number of the gear in the feeding mechanism of suction pad

n

Defocusing distance

mm

Z

d

z Rf

Depth of focus

mm

α

Angle between the stem and negative vertical direction

rad

α0

Initial angle between the stem and negative vertical direction

rad

β

0

Theoretical maximum angular acceleration of the rack in sucked pulling

rad/s2

γ

bM

Torque constant of motor in the feeding mechanism of the suction pad

m.Nm/A



Mean relative fitting error

%

L

Length different before and after test

mm

ε

Normal strain during compression

%

ε 0.5σ c

Strain at 50% of the yielding point stress on the stress–strain curves %

εc

Failure strain during compression testing

εc

Total compressive strain of the beam

ε cmax

Strain of the edge of the beam under compression

εl

Failure strain during tension testing

εT

Compressibility of the plate to tomato

εt

Total tensile strain of the beam

ε tmax

Strain of the edge of the beam under tension

ε yc

Strain of the compression side from the neutral axis yc

ε yt

Strain of the tensile side from the neutral axis yt

%

MPa

η1

Viscosity coefficient of the series viscous element

N·s/mm

η 1 (t)

Modified viscosity coefficient of the series viscous element in Burger’s model

N·s/mm

η2

Viscosity coefficient of the parallel viscous element

N·s/mm

ηM

Viscosity coefficient

Pa.s

Θ

Actual torsion angle of the whole peduncle

rad

θ

Angular displacement of the required motor

Λ

Thermal conductivity of materials

W/m·K

λ

Incident angle of laser beam

° (continued)

Nomenclature

459

(continued) Symbol

Meaning

μ

Static friction coefficient between robot finger and tomato

μd

Sliding coefficient of friction

μs

Static coefficient of friction

ρ

Thermal power density of focal spot

W/mm2

P avg

Average power density of focal spot for certain incident angle

W/mm2

ρ

g

Density of the gel

g/cm3

ρ

p

Density of mesocarp

g/cm3

σ

c

Failure stress during compression testing

MPa

σ

cmax

Maximum compressive strength of the beam during bending

MPa

σ

l

Failure stress during tension testing

MPa

σ

max

Maximum Von Mises stress

MPa

σ

min

Minimum Von Mises stress of the material

kPa

σ

tmax

Maximum compressive strength of the beam during bending

MPa

σ

yc

Compressive stress from the neutral axis yc

MPa

σ

yt

Tensile stress from the neutral axis yt

MPa

τ

i

Relaxation time of each component

s

τ

k

Creep delay time

s

τ

K

Elastic lag time

s

τ

M

Stress relaxation time

s

τ

s

Shear strength during shear testing

MPa

0

Units

ϕ

Sphericity

ϕ

Angle of projection of the ring-shaped structure

1

Inner diameter of the suction pad

mm

2

Outer diameter of suction pad

mm

e

Effective diameter of suction pad, which is defined as diameter of the maximum suction area

mm

s

Diameter of actual sucking area

mm

ω b1

Angular speed of the motor in the feeding mechanism of the suction pad

rad/s

τ

Mi τiG & τik

Relaxation time of the ith Maxwell body

s

Relative time of each Prony series component

s

P u

Negative pressure per area applied normally to object surface

KPa

P u (x)

Threshold of the negative pressure per area relative to the horizontal movement distance x of the fruit

KPa

z

Axial compression deformation

mm (continued)

460

Nomenclature

(continued) Symbol

Meaning

Units

z 0

Maximum axial compression deformation of the suction pad

mm

δ

Deformation of the peduncle

mm or rad

δ b

Deformation of the whole peduncle to break the abscission layer

mm or rad