Advanced Manufacturing and Automation X (Lecture Notes in Electrical Engineering, 737) [1st ed. 2021] 9813363177, 9789813363175

Citation preview

Lecture Notes in Electrical Engineering 737

Yi Wang Kristian Martinsen Tao Yu Kesheng Wang   Editors

Advanced Manufacturing and Automation X

Lecture Notes in Electrical Engineering Volume 737

Series Editors Leopoldo Angrisani, Department of Electrical and Information Technologies Engineering, University of Napoli Federico II, Naples, Italy Marco Arteaga, Departament de Control y Robótica, Universidad Nacional Autónoma de México, Coyoacán, Mexico Bijaya Ketan Panigrahi, Electrical Engineering, Indian Institute of Technology Delhi, New Delhi, Delhi, India Samarjit Chakraborty, Fakultät für Elektrotechnik und Informationstechnik, TU München, Munich, Germany Jiming Chen, Zhejiang University, Hangzhou, Zhejiang, China Shanben Chen, Materials Science and Engineering, Shanghai Jiao Tong University, Shanghai, China Tan Kay Chen, Department of Electrical and Computer Engineering, National University of Singapore, Singapore, Singapore Rüdiger Dillmann, Humanoids and Intelligent Systems Laboratory, Karlsruhe Institute for Technology, Karlsruhe, Germany Haibin Duan, Beijing University of Aeronautics and Astronautics, Beijing, China Gianluigi Ferrari, Università di Parma, Parma, Italy Manuel Ferre, Centre for Automation and Robotics CAR (UPM-CSIC), Universidad Politécnica de Madrid, Madrid, Spain Sandra Hirche, Department of Electrical Engineering and Information Science, Technische Universität München, Munich, Germany Faryar Jabbari, Department of Mechanical and Aerospace Engineering, University of California, Irvine, CA, USA Limin Jia, State Key Laboratory of Rail Traffic Control and Safety, Beijing Jiaotong University, Beijing, China Janusz Kacprzyk, Systems Research Institute, Polish Academy of Sciences, Warsaw, Poland Alaa Khamis, German University in Egypt El Tagamoa El Khames, New Cairo City, Egypt Torsten Kroeger, Stanford University, Stanford, CA, USA Qilian Liang, Department of Electrical Engineering, University of Texas at Arlington, Arlington, TX, USA Ferran Martín, Departament d’Enginyeria Electrònica, Universitat Autònoma de Barcelona, Bellaterra, Barcelona, Spain Tan Cher Ming, College of Engineering, Nanyang Technological University, Singapore, Singapore Wolfgang Minker, Institute of Information Technology, University of Ulm, Ulm, Germany Pradeep Misra, Department of Electrical Engineering, Wright State University, Dayton, OH, USA Sebastian Möller, Quality and Usability Laboratory, TU Berlin, Berlin, Germany Subhas Mukhopadhyay, School of Engineering & Advanced Technology, Massey University, Palmerston North, Manawatu-Wanganui, New Zealand Cun-Zheng Ning, Electrical Engineering, Arizona State University, Tempe, AZ, USA Toyoaki Nishida, Graduate School of Informatics, Kyoto University, Kyoto, Japan Federica Pascucci, Dipartimento di Ingegneria, Università degli Studi “Roma Tre”, Rome, Italy Yong Qin, State Key Laboratory of Rail Traffic Control and Safety, Beijing Jiaotong University, Beijing, China Gan Woon Seng, School of Electrical & Electronic Engineering, Nanyang Technological University, Singapore, Singapore Joachim Speidel, Institute of Telecommunications, Universität Stuttgart, Stuttgart, Germany Germano Veiga, Campus da FEUP, INESC Porto, Porto, Portugal Haitao Wu, Academy of Opto-electronics, Chinese Academy of Sciences, Beijing, China Junjie James Zhang, Charlotte, NC, USA

The book series Lecture Notes in Electrical Engineering (LNEE) publishes the latest developments in Electrical Engineering - quickly, informally and in high quality. While original research reported in proceedings and monographs has traditionally formed the core of LNEE, we also encourage authors to submit books devoted to supporting student education and professional training in the various fields and applications areas of electrical engineering. The series cover classical and emerging topics concerning:

• • • • • • • • • • • •

Communication Engineering, Information Theory and Networks Electronics Engineering and Microelectronics Signal, Image and Speech Processing Wireless and Mobile Communication Circuits and Systems Energy Systems, Power Electronics and Electrical Machines Electro-optical Engineering Instrumentation Engineering Avionics Engineering Control Systems Internet-of-Things and Cybersecurity Biomedical Devices, MEMS and NEMS

For general information about this book series, comments or suggestions, please contact leontina. [email protected]. To submit a proposal or request further information, please contact the Publishing Editor in your country: China Jasmine Dou, Editor ([email protected]) India, Japan, Rest of Asia Swati Meherishi, Editorial Director ([email protected]) Southeast Asia, Australia, New Zealand Ramesh Nath Premnath, Editor ([email protected]) USA, Canada: Michael Luby, Senior Editor ([email protected]) All other Countries: Leontina Di Cecco, Senior Editor ([email protected]) ** This series is indexed by EI Compendex and Scopus databases. **

More information about this series at http://www.springer.com/series/7818

Yi Wang Kristian Martinsen Tao Yu Kesheng Wang •

•

•

Editors

Advanced Manufacturing and Automation X

123

Editors Yi Wang School of Business Plymouth University Plymouth, UK Tao Yu Shanghai Second Polytechnic University Shanghai, Shanghai, China

Kristian Martinsen Department of Manufacturing and Civil Engineering Norwegian University of Science and Technology Gjøvik, Norway Kesheng Wang Department of Mechanical and Industrial Engineering Norwegian University of Science and Technology Trondheim, Norway

ISSN 1876-1100 ISSN 1876-1119 (electronic) Lecture Notes in Electrical Engineering ISBN 978-981-33-6317-5 ISBN 978-981-33-6318-2 (eBook) https://doi.org/10.1007/978-981-33-6318-2 © The Editor(s) (if applicable) and The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2021, corrected publication 2021 This work is subject to copyright. All rights are solely and exclusively licensed 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 publisher, 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 publisher nor the authors or the editors give a warranty, expressed or implied, with respect to the material contained herein or for any errors or omissions that may have been made. The publisher remains 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

Preface

IWAMA—International Workshop of Advanced Manufacturing and Automation— aims at providing a common platform for academics, researchers, practicing professionals, and experts from industries to interact, discuss current technology trends and advances, and share ideas and perspectives in the areas of manufacturing and automation. IWAMA began at the Shanghai University, 2010. In 2012 and 2013, it was held at the Norwegian University of Science and Technology, in 2014 at Shanghai University again, in 2015 at Shanghai Polytechnic University, in 2016 at Manchester University, in 2017 at Changshu Institute of Technology, in 2018 at Changzhou University, and in 2019 at Plymouth University. The sponsors organizing the IWAMA series have expanded to many universities throughout the world, including Plymouth University, Changzhou University, Norwegian University of Science and Technology, SINTEF, Manchester University, Shanghai University, Shanghai Polytechnic University, Changshu Institute of Technology, Xiamen University of Science and Technology, Tongji University, University of Malaga, University of Firenze, Stavanger University, The Arctic University of Norway, Shandong Agricultural University, China University of Mining and Technology, Indian National Institute of Technology, Donghua University, Shanghai Jiao Tong University, Changshu Institute of Technology, Dalian University, St. Petersburg Polytechnic University, Hong Kong Polytechnic University, Lingnan Normal University, Civil Aviation University of China, and China Instrument and Control Society, etc. As IWAMA becomes an annual event, we are expecting more sponsors from universities and industries, who will participate in the international workshop as co-organizers. Manufacturing and automation have assumed paramount importance and are vital for the economy of a nation and the quality of daily life. The field of manufacturing and automation is advancing at a rapid pace, and new technologies are also emerging. The main challenge faced by today’s engineers, scientists, and academics is to keep on top of the emerging trends through continuous research and development.

v

vi

Preface

IWAMA 2020 takes place in Lingnan Normal University, Zhangjiang, China, October 12–13, 2020, organized by Plymouth University, Norwegian University of Science and Technology, Shanghai University, and Lingnan Normal University. Because of the COVID-19 pandemic, we arranged IWAMA2020 as a virtual conference. All participants in the conference are able to present their papers and discuss their options in an online way. The program is designed to improve manufacturing and automation technologies for the next generation through discussion of the most recent advances and future perspectives and to engage the worldwide community in a collective effort to solve problems in manufacturing and automation. The main focus of the workshop is focused on the transformation of present factories, toward reusable, flexible, modular, intelligent, digital, virtual, affordable, easy-to-adapt, easy-to-operate, easy-to-maintain, and highly reliable “smart factories.” Therefore, IWAMA 2020 has mainly covered five topics in manufacturing engineering: 1. 2. 3. 4. 5.

Industry 4.0 Manufacturing Systems Manufacturing Technologies Production Management Design and Optimization

All papers submitted to the workshop have been subjected to strict peer review by at least two expert referees. Finally, 93 papers have been selected to be included in the proceedings after revision processes, which will be published in Lecture Notes of Electrical Engineering (LNEE) by Springer and will be indexed by EI. We hope that the proceedings will not only give the readers a broad overview of the latest advances, and a summary of the event, but also provides researchers with a valuable reference in this field. On behalf of the organization committee and the international scientific committee of IWAMA 2020, I would like to take this opportunity to express my appreciation for all the kind support, from the contributors of high-quality keynotes and papers, and all the participants. My thanks are extended to all the workshop organizers and paper reviewers, to Lingnan Normal University, Plymouth University, and NTNU for the financial support, and to co-sponsors for their generous contribution. Thanks are also given to Jian Wu, Bo Chen, and Shifeng Chen, for their hard editorial work of the proceedings and arrangement of the workshop. Yi Wang

Organization

Organized and Sponsored by LNU (Lingnan Normal University, China) PLYU (Plymouth University, UK) NTNU (Norwegian University of Science and Technology, Norway)

Co-organized by SHU (Shanghai University, China) SSPU (Shanghai Second Polytechnic University, China) TU (Tongji University, China) SJTU (Shanghai Jiao Tong University, China)

Honorary Chairs Minglun Fang Kesheng Wang Jan Ola Strandhagen

China Norway Norway

General Chairs Yi Wang Kristian Martinsen Tao Yu,

UK Norway China

Local Organizing Committee Bo Chen (Chair), China Yi Wang, UK Zhijun Zheng, China Junjie Yang, China Tao Wu, China vii

viii

Shifeng Chen, China Haitao Sang, China Xiue Gao, China Jian Wu, China Lilan Liu, China

International Program Committee Jan Ola Strandhagen, Norway Kesheng Wang, Norway Asbjørn Rolstadås, Norway Per Schjølberg, Norway Knut Sørby, Norway Erlend Alfnes, Norway Torgeir Welo, Norway Hirpa L. Gelgele, Norway Wei D. Solvang, Norway Yi Wang, UK Chris Parker, UK Jorge M. Fajardo, Spain Torsten Kjellberg, Sweden Fumihiko Kimura, Japan Gustav J. Olling, USA Michael Wozny, USA Wladimir Bodrow, Germany Guy Doumeingts, France Van Houten, Netherlands Peter Bernus, Australia Janis Grundspenkis, Latvia George L. Kovacs, Hungary Rinaldo Rinaldi, Italy Gaetano Aiello, Italy Romeo Bandinelli, Italy Yong Gan, China Binheng Lu, China Jinhui Yang, China Dawei Tu, China Minglun Fang, China Ming Chen, China Yun Chen, China Henry Xinguo Ming, China Keith C. Chan, China Xiaojing Wang, China Jin Yuan, China Bo Chen, China

Organization

Organization

Ming Li, China Cuilian Zhao, China Chuanhong Zhou, China Jianqing Cao, China Shirong Ge, China Jianjun Wu, China Guijuan Lin, China Shangming Luo, China Quiqing Li, China Zumin Wang, China Guohong Dai, China Sarbjit Singh, India Vishal S. Sharma, India

Secretariat Jian Wu, China Shifeng Chen, China

ix

Contents

Experimental Investigation of Viscosity Reduction of Heavy Oil via Hydrodynamic Cavitation in Laval Nozzle . . . . . . . . . . . . . . . . . . . . . . . Shichun Zhu, Xuedong Liu, and Zhihong Zhang Study on the Cutting Simulation and Experiment of Titanium Alloy . . . Li Hui, Gao Feng, and Li Yan

1 10

Effect of Backlash on Transmission Error and Time Varying Mesh Stiffness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Getachew A. Ambaye and Hirpa G. Lemu

18

Fatigue Life Study of False Banana/Glass Fiber Reinforced Composite for Wind Turbine Blade Application . . . . . . . . . . . . . . . . . . Temesgen Batu and Hirpa G. Lemu

29

Numerical Simulation of Separation Mechanism in V-Shaped Outlet Hydrocyclone for Coalbed Gas Fracturing Flow-Back Fluids . . . . . . . . Xiaolong Xiao, Mingxiu Yao, Chao Xie, Zhen Wu, Yaoyao Wei, Zhenjiang Zhao, Huajian Wang, Youle Liu, and Bing Liu Numerical Simulation of Flow Field and Separation Performance of a Two-Cone Hydrocyclone for Oil-Water Separation . . . . . . . . . . . . . Chao Xie, Zhen Wu, Zhenjiang Zhao, Yaoyao Wei, Jianliang Xue, Peishan Huang, and Bing Liu Parameter Optimization of Hydrocyclone at the Inlet of Spiral Tube for Offshore Oil Water Separation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Zhen Wu, Yaoyao Wei, Chao Xie, Zhenjiang Zhao, Wenyu Yang, Peishan Huang, and Bing Liu Sit-to-Stand Intention Recognition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Dayou Li, Hang Lu, Renxi Qiu, Carsten Maple, and Zuobin Wang

41

50

57

65

xi

xii

Contents

A Method of Bearing Fault Diagnosis Based on Transfer Learning Without Parameter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Yang Ge, Jiancong Qin, and Jianxin Ding

73

Optimal Design of Control System for Ground Source Heat Pump Unit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Xiaomei Jiang, Michael Namokel, Chaobin Hu, and Ran Tian

82

Design of Elevator Type Three-Dimensional Garage . . . . . . . . . . . . . . . Xiaomei Jiang, Michael Namokel, Chaobin Hu, and Ran Tian Research on Elevator Fault Information Extraction and Prediction Diagnosis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Xiaomei Jiang, Michael Namokel, Chaobin Hu, and Ran Tian

90

99

Static Performances of Misaligned Journal Bearing with Slip-Texture Surface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107 Jian Jin, Yinhui Chang, Guiqin Li, and Peter Mitrouchev An Algorithm of 3D Surface Reconstruction of Human Body Based on Millimeter Wave Sensor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115 Guiqing Li, Hanlin Wang, Tiancai Li, and Peter Mitrouchev Uniformity Analysis of Melt-Blown Flow Based on Simulated Annealing Algorithm Optimization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123 Feng Xiong, Jiaoyang Wang, and Guiqin Li Design and Simulation of Millimeter Wave Fresnel Antenna . . . . . . . . . 130 Guiqing Li, Xihang Li, Tiancai Li, and Peter Mitrouchev Reliability Assessment of Micro Grid Power Supply System Based on D-Vine Copula Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137 Feng Xiong, Xiangxin Ji, Dongfei Wei, and Guiqin Li High Redundancy Data Communication System Based on LwIP Protocol Stack . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 146 Lixin Lu, Xinyi Jiang, Guiqin Li, and Peter Mitrouchev Fault Diagnosis of Massage Chair Motor Based on Wavelet Packet Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153 Lixin Lu, Hui Li, Guiqin Li, and Peter Mitrouchev Management System of 3D Anthropometric Data Based on B/S Framework . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160 Guiqin Li, Jianping Lv, Tiancai Li, and Peter Mitrouchev Signal Detection and Noise Reduction Method of Massage Chair Based on EMD . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167 Lixin Lu, Huajie Yu, Guiqin Li, and Peter Mitrouchev

Contents

xiii

Study on Detection of Potato Starch Content by Optimum Hyperspectral Characteristic Wavelength Method . . . . . . . . . . . . . . . . . 174 Wei Jiang, Ming Li, Zhongyan Liu, Yao Liu, and Qichao Li Prediction of Torque Parameters in Automobile Rear Axle Assembly Based on Long Short-Term Memory . . . . . . . . . . . . . . . . . . . 183 Shouzheng Liu, Lilan Liu, Kai Guo, and Fang Wu Teaching Demonstration of Reducer Based on Augmented Reality Technology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 189 Kai Guo, Lilan Liu, Shibo Yuan, Muchen Yang, Shouzheng Liu, Tao Xu, and Fang Wu Temporal and Multi-level Context Attentive GRU Neural Networks . . . 196 Chuanhong Zhou, Haifeng Liu, and Xiaoyu Jiang Medical Image Recognition Based on Improved Convolutional Neural Network . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 204 Chuanhong Zhou, Yiyang Zhang, and Lihua Yang Research on Server Fault Diagnosis Based on Expert System . . . . . . . . 213 Chuanhong Zhou, Daohao Guo, and Chong Zhang Research on Multi-modal Data Cross-Media Perception Fusion Algorithm Based on Digital Twin . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 221 Xiang Wan, Lilan Liu, Bowen Feng, Chen Chen, and Yi Wang Digital Twin-Driven Surface Roughness Prediction Based on Multi-sensor Fusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 230 Xiangyu Zhang, Lilan Liu, Fang Wu, and Xiang Wan Dynamic Visualization Fluid Analysis of Rolling Furnace Based on Augmented Reality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 238 Bowen Feng, Lilan Liu, Xiang Wan, and Chen Chen Research on Operation Status Prediction of Production Equipment Based on Digital Twins and Multidimensional Time Series . . . . . . . . . . 246 Qiang Miao, Lilan Liu, Chen Chen, Xiang Wan, and Tao Xu Application of Digital Twin in the Assembly of Automobile Rear Axle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 255 Muchen Yang, Lilan Liu, Tao Xu, Kai Guo, Shibo Yuan, and Fang Wu Optimization of Digital Twins in the Workshop . . . . . . . . . . . . . . . . . . . 262 Tao Xu, Lilan Liu, Chen Chen, Kai Guo, Shibo Yuan, and Qiang Miao Research on Fault Diagnosis Algorithm Based on Bi-directional Long Short-Term Memory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 270 Bin Yin, Xiaolong Li, Lilan Liu, and Fang Wu

xiv

Contents

Construction and Exploration of Intelligent Manufacturing and Virtual Simulation Laboratory Based on Integration of Production and Education . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 280 Zhang Fei, Zhou Fengxu, Wang Zhenhua, and He Yafei Analysis of Wear Performance of Two Finger Seal Structure . . . . . . . . 286 Junjie Lei, Meihong Liu, Xiangping Hu, Junfeng Sun, and Yuchi Kang Design and Development of Intelligent Factory Information System Based on Digital Twin Technology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 295 Haiwei Zhou, Lilan Liu, Xiangyu Zhang, and Chen Chen Research on Magnetic Field Distribution of Magnetic Drug Nanoparticle Transport with Implant Assisted Magnetic Seed in Non-permeable Microvessel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 303 Jiejie Cao and Jian Wu The Application of a Lightweight Domain-Adversarial Neural Network in Bearing Fault Diagnosis . . . . . . . . . . . . . . . . . . . . . . . . . . . . 312 Jie Wu, Tang Tang, Ming Chen, and Kesheng Wang Lightweight Design of Riveter Based on Evolutionary Algorithm . . . . . 321 Te Li and Xinyong Li Lightweight Design of Riveting Machine Frame Based on Topology Optimization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 330 Xinyong Li and Te Li Effect of Lubricant Viscosity on the Friction Behavior of the mm-scale Specimen . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 339 Jian Wu and Knut Sørby Research on Identification of Chronic Disease Using Automated Machine Learning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 347 Hongxia Cai, Tianjie Shen, and Jian Xu A Predictive Workshop Material Distribution Model . . . . . . . . . . . . . . . 356 Hongxia Cai and Liangliang Wu Design of the Integrated Device New Speed Governor and Safety Gear . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 364 Ran Tian, Jiaxin Ma, Yang Ge, Junjun Liu, and Yong Ren Study of Wheel-Rail Contacts at Railway Turnout Using Multibody Dynamics Simulation Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 371 Yohanis D. Jelila and Hirpa G. Lemu FASHION 4.0: A Potential Solution to a More Sustainable Fashion Industry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 380 Chloe Vieira de Haro and Yi Wang

Contents

xv

Critically Evaluate How Industry 4.0 Can Be Implemented Within the Car Manufacturing Process . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 387 Danielle McLean and Yi Wang Understanding the Impact of Game Theory on Circular Economy Within the Apparel Industry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 394 Lillian Rogers and Yi Wang Implementing Vertical Integration in the Fashion Industry . . . . . . . . . . 401 William Dumbrill and Yi Wang Research Challenges in Off-Line Ancient Handwriting Recognition – A Deep Learning Approach . . . . . . . . . . . . . . . . . . . . . . . 408 Yi Wang, Chen Wang, and Bo Chen Status Monitoring for Packaging Equipment Based on Digital Twin System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 416 Nanyan Shen, Yingjie Xu, Jing Li, Zehui Ma, Tianqiang He, and Guiqin Li Research on Recognition of Rice Panicle Blast in Cold Region Based on UAV . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 424 Jianqing Yuan, Weiwei Wang, and Qiuju Zheng Singularity Analysis of the 3-DoF (P-4R) Parallel Mechanisms by Screw Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 431 Kunquan Li and Rui Wen Research on Sound Source Localization Algorithm Based on Multilayer Perceptron . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 438 Jinliang Zhang, Xianfeng Weng, Jing Zhou, and DongPing Li Soft Measurement Model of Passenger Vehicle Tire Mileage Based on GA-BP Neural Network . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 445 Chao Shi, Bo Wang, and Ning He Simulation Analysis of Harvesting Positioning of White Asparagus Harvesting Robot Based on Adams and Simulink . . . . . . . . . . . . . . . . . 451 Xiaoyu Zheng, Xuemei Liu, and Jin Yuan Coupling Analysis and Experimental Research of Discrete Element and Multibody Dynamics for Deep Loosening . . . . . . . . . . . . . . . . . . . . 460 HaoYu Ma, Gong Liu, and Jin Yuan Multi Target Threat Assessment Method Based on Improved Dual Variable Weight . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 469 Shaoming Qiu, Jianwei Wang, Xuecui Zhang, and Xiuli Du Designing Shoe Lasts Through 3D Feet Scans Clusterization Using Anthropometric Parameters of Population Groups . . . . . . . . . . . . . . . . 479 Evgeniy Pavlov, Ivan Selin, Pavel Drobintsev, Nikita Voinov, and Ilya Shemyakin

xvi

Contents

Evaluation of Human-Robot Collaborative Assembly Task Allocation Plan . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 487 Shibo Yuan, Zenggui Gao, Mengao Dong, Lilan Liu, Tao Xu, Kai Guo, and Fang Wu Optimal and Self-correcting Covariance Intersection Fusion Kalman Filters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 496 Peng Zhang and Jinfang Liu Optimal Design of Pelvic Weight Support System of Walking-Assistant Rehabilitation Robots . . . . . . . . . . . . . . . . . . . . . . 505 Yanhong Chen and Jie Zhang Image Texture Balancing in the Wavelet Domain Based on a Ripple Matrix Permutation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 513 Xiuli Du, Jinting Liu, Wei Zhang, and Ya’na Lv Research on Wireless Sensor Robot Lithium Battery SOC Estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 521 Xuehuan Jiang, Jinliang Zhang, Mingjie Dai, and Lei Zhang Storage Genetic Scheduling Algorithm Based on Leader Selection Operator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 528 Tianshu Zhang, Xiue Gao, Kesheng Wang, and Shifeng Chen Simulation Study on the Mechanical Influence of Different Diameters of Artificial Hip Joints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 536 Zikai Hua, Hui Liu, and Xiuling Huang Effect of 3D Printing Orthoses on Hand Edema Rehabilitation of Stroke Patients . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 544 Zikai Hua, Jiali Dai, Yikang Shen, and Xiuling Huang 4D Printed Self Assembling Module for Terraforming Environment . . . 552 Francesca Parotti, Lapo Chirici, and Yi Wang Tool Management System Based on Bar Code Intelligent Terminal . . . 561 Jie Chen and Jiahai Wang Multidimensional Network Analysis to Predict Correlations Between Offline and Online Interactions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 569 Lapo Chirici, Francesca Parotti, and Yi Wang An Unmanned Cluster Network Routing Protocol . . . . . . . . . . . . . . . . . 579 Shuo Zhang, Xianwu Chu, Yunming Wang, and Bo Chen Integrating Additive Manufacturing into a Virtual Industry 4.0 Factory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 587 Mohammad Azarian, Hao Yu, and Wei Deng Solvang

Contents

xvii

Industry 4.0 and Sustainable Supply Chain Management . . . . . . . . . . . 595 Xu Sun, Hao Yu, and Wei Deng Solvang Research on the Permanent Magnet Synchronous Motor Controller Development Based on Model Design . . . . . . . . . . . . . . . . . . . . . . . . . . . 605 Xuehuan Jiang, Rong Jia, Jinliang Zhang, Chao Chen, Mingjie Dai, and Lei Zhang Small Target Detection Algorithm in Remote Sensing Image Based on Improved Yolo . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 612 Chaoliang Peng, Xianwu Chu, Yunming Wang, and Xiue Gao Equipment Maintenance Task Scheduling Method Based on Improved PSO . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 621 Shengbin Zhou, Xiue Gao, Shifeng Chen, and Yannan Gao Additive Manufacturing and Spare Parts: Literature Review and Future Perspectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 629 Mirco Peron and Fabio Sgarbossa Best Practices of Just-in-Time 4.0: Multi Case Study Analysis . . . . . . . . 636 Mirco Peron, Erlend Alfnes, and Fabio Sgarbossa Digital Assembly Assistance System in Industry 4.0 Era: A Case Study with Projected Augmented Reality . . . . . . . . . . . . . . . . . . 644 Marco Simonetto, Mirco Peron, Giuseppe Fragapane, and Fabio Sgarbossa Communication Between Human and Robots Within a Collaborative Workspace . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 652 Théo Degeorges and Gabor Sziebig Fault Diagnosis of Industrial Internet of Things Equipment Based on Edge Intelligent Drive . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 658 Haitao Sang and Bo Chen A Variable Neighbourhood Search Algorithm for Solving Dynamic Vehicle Routing Problem Under Industry 4.0 . . . . . . . . . . . . . . . . . . . . 666 Shifeng Chen, Yanlan Yin, Bo Chen, Yannan Gao, and Junjie Yang Self-adjusting Information Fusion Wiener Filter for the Multisensor Signal Systems with Correlated Noise . . . . . . . . . . . . . . . . . . . . . . . . . . 674 Jinfang Liu, Lei Liu, Yannan Gao, Peng Zhang, and Yao Liu Kinematics Analysis and Trajectory Planning of SCARA Robot Based on Matlab . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 683 Yu Li, Chen Wang, Chao Liu, Xiufeng Zhang, and Changfeng Xiang Intelligent Detection Method for Welding Seam Defects of Automobile Wheel Hub Based on YOLO . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 693 Xiufeng Zhang, Chen Wang, Changfeng Xiang, Chao Liu, and Yu Li

xviii

Contents

A Review of Structure Optimization of Convolutional Neural Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 703 Liu Chao, Wang Chen, Li Yu, Xiang Changfeng, and Zhang Xiufeng A Review of Lane Line Detection Technology Based on Machine Vision . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 713 Xiang Changfeng, Chen Wang, Li Yu, Liu Chao, and Zhang Xiufeng Aerospace Device for Augmented Human Perception . . . . . . . . . . . . . . 718 Francesca Parotti, Lapo Chirici, and Kesheng Wang Research and Design of Multi-car Elevator Model with Inclined Rotation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 725 Yan Dou, Wenmeng Li, and Siyong Jiao Design of an Inspection Robot for Elevator Guide Rail . . . . . . . . . . . . . 730 Yan Dou, Wenmeng Li, and Jinglin Sang Research on Structural Parameters of Mono-tube Permanent Magnet Speed Governor Based on Magnet Simulation . . . . . . . . . . . . . 735 Zhiping Zhao, Xinyong Li, Zhenyang Feng, and Chao Wang The Power Up Principle and Fault Analysis of FANUC 0iD Servo . . . . 743 Xueping Chu Correction to: Integrating Additive Manufacturing into a Virtual Industry 4.0 Factory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Mohammad Azarian, Hao Yu, and Wei Deng Solvang

C1

Author Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 751

About the Editors

Dr. Yi Wang obtained his PhD from the Manufacturing Engineering Center, Cardiff University, in 2008. He is a lecturer in Business School, Plymouth University, UK. Previously, he worked in the department of computer science, Southampton University, and at the Business School, Nottingham Trent University. He holds various visiting lectureships in several universities worldwide. He has special research interests in supply chain management, logistics, operation management, culture management, information systems, game theory, data analysis, semantics and ontology analysis, and neuromarketing. He has published 98 technical peer-reviewed papers in international journals and conferences. He has authored three books, for example, Operations Management for Business, Fashion Supply Chain Management, and Data Mining for Zero-defect Manufacturing, etc., edited six books, and made five book chapters. Dr. Kristian Martinsen took his PhD at the Norwegian University for Science and Technology (NTNU) in 1995, with the topic “Vectorial Tolerancing in Manufacturing.” He has 15 years’ experience in the manufacturing industry. He is a professor at the faculty of engineering and department for manufacturing and civil engineering, the Norwegian University for Science and Technology (NTNU), and is the manager of the manufacturing engineering research group in this department. He is a corporate member of the international academy for production engineering and a member of the high-level group of the EU Technology Platform for manufacturing, MANUFUTURE. He is the manager of the Norwegian national infrastructure for manufacturing research laboratories, MANULAB, and is the international coordinator for the Norwegian Center for Research-based Innovation, SFI MANUFACTURING. He has published many papers in international journals and conferences. His major research area is within the field of measurement systems, variation/quality management, and tolerancing toward Industry 5.0: research challenges in human–machine systems.

xix

xx

About the Editors

Dr. Tao Yu is the president of the Shanghai Second Polytechnic University (SSPU), China, and professor of Shanghai University (SHU). He received his PhD from SHU in 1997. He is a member of the group of Shanghai manufacturing information and a committee member of the International Federation for Information Processing IFIP /TC5. He is also an executive vice president of the Shanghai Science Volunteer Association and executive director of the Shanghai Science and Art Institute of Execution. He managed and performed about 20 national, Shanghai, enterprises commissioned projects. He has published hundreds of academic papers, of which about thirty were indexed by SCI, EI. His research interests are mechatronics, computer-integrated manufacturing system (CIMS), and grid manufacturing. Dr. Kesheng Wang holds a PhD in production engineering from the Norwegian University of Science and Technology (NTNU), Norway. Since 1993, he has been an appointed professor at the department of mechanical and industrial engineering, NTNU. He is a director of the Knowledge Discovery Laboratory (KDL) at NTNU at present. He is also an active researcher and serves as a technical adviser in SINTEF. He was elected member of the Norwegian Academy of Technological Sciences in 2006. He has published 19 books, ten chapters, and over 240 technical peer-reviewed papers in international journals and conferences. His current areas of interest are intelligent manufacturing systems, applied computational intelligence, data mining and knowledge discovery, swarm intelligence, condition-based monitoring, and structured light systems for 3D measurements and RFID, predictive maintenance, and Industry 4.0.

Experimental Investigation of Viscosity Reduction of Heavy Oil via Hydrodynamic Cavitation in Laval Nozzle Shichun Zhu1,2(&), Xuedong Liu1,2, and Zhihong Zhang3 1

School of Mechanical Engineering, Changzhou University, Changzhou, China [email protected], [email protected] 2 Jiangsu Key Laboratory of Green Process Equipment, Changzhou, China 3 School of Petrochemical Engineering, Changzhou University, Changzhou, China [email protected]

Abstract. Hydrodynamic cavitation is rarely studied for viscosity reduction of heavy oil, although ultrasonic viscosity reduction technique has been widely reported. To evaluate the effect of hydrodynamic cavitation on the characters of heavy oil, a home-made setup equipped with a Laval nozzle was assembled. The density and viscosity of heavy oil were investigated. Function groups in heavy oil were determined by fourier transform infrared spectroscopy (FTIR). The results show that the effect of cavitation on the density can be negligible. The viscosity reduction can reach up to 3.8%. Longer circulation time, higher flow rate and upstream pressure lead to more reduction in viscosity continuously. More light compounds can be produced by higher flow rate, the ratio value of infrared absorbance of methylene (=CH2) to methyl (–CH3) can be decreased by 20.42%. However, without sufficient hydrogen donors, cracked compounds will be formed into heavy chains in the closed pipeline during the circulation. Besides of cracking of heavy chains by the energy wave of collapsing bubbles, resonant vibration and bulk nanobubbles (ultrafine bubbles) are introduced to the viscosity reduction of heavy oil. Keywords: Heavy oil


Viscosity
Hydrodynamic cavitation
Laval nozzle

1 Introduction Heavy oil reservoirs are important petroleum resource that is widely distributed worldwide. More than three-fourth of the world’s total oil resources is heavy oil including medium heavy oil, extra heavy oil and bitumen [1]. However, a difficult issue arises during the gathering and transportation of heavy oil due to its high viscosity. Heavy oil contains high molecular mass organic structures within which inorganic functionalities are inserted [2]. In the process of viscosity reduction of heavy oil, one of the main goals is cracking of these heavy compounds using additives and special techniques. Various existing methods [3], including thermal method, mixed with light oil method, hydrothermal catalytic cracking method, soluble viscosity reducing method, emulsification method and microbial method have been investigated for © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2021 Y. Wang et al. (Eds.): IWAMA 2020, LNEE 737, pp. 1–9, 2021. https://doi.org/10.1007/978-981-33-6318-2_1

2

S. Zhu et al.

reducing the viscosity. However, the problems of these methods are energy consumption, secondary pollution, high cost and other issues. Cavitation can produce conditions of very high localized temperature (up to 20,000 K) and pressure (several thousand bar) at ambient temperature in very short time [4]. The cavitation technique can be used for viscosity reduction of heavy oil by cracking the bonds of heavy molecules. Heavy chains in heavy oil subjected to acoustic cavitation can be cracked, releasing radical compounds and forming light compounds [5–7]. The viscosity reduction even reaches up to 63.95% under specified conditions because of those fresh light compounds [5]. However, the temperature of heavy oil rises rapidly due to the heating effect of acoustic irradiation [6]. Light compounds are evaporated form oil, which can make the viscosity of heavy oil increased [5–7]. There have only been a few investigations on upgrading heavy oil with hydrodynamic cavitation [8, 9]. Askariana et al. reported that adding 2 vol% gasoline as hydrogen donor into the heavy oil could reduce the viscosity of heavy oil by 33% [8]. In addition, hydrodynamic cavitation has advantages of scaling up and higher energy efficiency in comparison with acoustic cavitation. Avvaru et al. [10] reviewed current knowledge of cavitation technology for the petroleum industry in 2018. It is very clear that three typical characteristics exist in viscosity reduction of petroleum oil using cavitation. First of all, almost all researches focus on acoustic cavitation in field tests or laboratory experiments, although hydrodynamic cavitation has been studied deeply in many other areas. Secondly, mechanism of viscosity reduction is related to the cracking of heavy chains by the energy wave of collapsing bubbles, and the formation of light compounds. Thirdly, the rising temperature caused by acoustic irradiation affects viscosity significantly. To sum up previous studies, on the one hand the area of viscosity reduction of heavy oil via hydrodynamic cavitation should be explored further, on the other hand the mechanism of viscosity reduction which may be associated with resonant vibration and bulk nanobubbles should be analyzed in details. In this study, a home-made hydrodynamic cavitation setup equipped with a Laval nozzle was designed and erected. The effect of throat diameter and circulation time on the density and viscosity were investigated. Cracking of heavy compounds during the cavitation was analyzed with fourier transform infrared spectroscopy (FTIR). The purpose of this paper is to provide the experimental basis for the application of hydrodynamic cavitation on the viscosity reduction of heavy oil.

2 Experiment 2.1

Experimental Setup

The experimental setup equipped with a Laval nozzle is shown in Fig. 1. The setup contains a heavy oil tank, a buffer tank, a products tank, a gear pump with a 4.0 kW electric motor, a cavitation chamber equipped with nozzle, and a recycle pipeline. In this study, two nozzles of different throat were used.

Experimental Investigation of Viscosity Reduction of Heavy Oil

3

Fig. 1. Experimental setup used in this study

2.2

Experimental Setup

In this paper, base oil SN-500 (saturated hydrocarbon) supplied by Changrun Petroleum Ltd was used as heavy oil, the main properties are shown in Table 1. Table 1. Properties of base oil SN-500 Parameter Value Density (15 °C) 891.3 kg/m3 Kinematic viscosity (40 °C) 104.3 mm2/s Viscosity index 90 Pour point −15 °C Flash point 235 °C

2.3

Procedure

The arrangement of experiments in this study is shown in Table 2. Heavy oil samples were circulated in the setup. Two nozzles of different throat diameter were used respectively. The pressures (upstream pressure) before cavitation chamber were 1.0 and 3.0MPa. The circulation times were 20, 40, 60, 80 and 100 min respectively. In each experiment, the initial oil temperature was approximately 20 to 25 °C due to environment temperature. When each experiment was finished, oil sample was collected and kept for 24 h at room temperature to reach stability. The kinematic viscosity was measured by SYP1003-II. The relative density was determined by pycnometer. FTIR spectra of the samples for all bands was obtained from the Shimadzu FTIR-8400s device.

4

S. Zhu et al. Table 2. The arrangement of experiments No

Upstream pressure (MPa) Spec. 1 1.0 Spec. 2 1.0 Spec. 3 1.0 Spec. 4 1.0 Spec. 5 1.0 Spec. 6 1.0 Spec. 7 3.0 Spec. 8 3.0 Spec. 9 3.0 Spec. 10 3.0 Spec. 11 3.0

Nozzle Circulation diameter (mm) time (min) 1.0 20 1.0 40 1.0 60 2.0 20 2.0 40 2.0 60 1.0 20 1.0 40 1.0 60 1.0 80 1.0 100

3 Results and Discussion 3.1

Density

Figure 2 shows the effect of circulation time and throat diameter (d = 1.0 or 2.0 mm) on the density of heavy oil. When the circulation time is 20 min, densities decrease by 0.18% and 0.13% respectively. However, when the circulation time is prolonged to 40 or 60 min, the densities are recovering. The densities of oil samples almost return to the initial value after 60 min. The decrease value in density is so low that the effect of hydrodynamic cavitation on the density can be negligible.

Fig. 2. Density of heavy oil sample. Upstream pressure = 1.0 MPa

3.2

Viscosity

Figure 3 shows the viscosity of oil samples under different operating conditions. It is clear that the viscosity of heavy oil samples can be reduced continuously when the circulation time is prolonged. The reduction in viscosity can reach up to 3.8% when throat diameter is 1.0 mm and the upstream pressure is 3.0 MPa. Meanwhile, the viscosity of oil samples tends to be stable as the circulation time is extended over 80 min.

Experimental Investigation of Viscosity Reduction of Heavy Oil

5

Three curves in Fig. 3 show that higher flow rate leads to more reduction in viscosity. On the one hand, higher flow rate probably generates more intense cavitation, then more cracked compounds are produced and more fresh light compounds are formed. On the other hand, higher upstream pressure means drastic pressure variation of the stream in the nozzle, which affects the generation of cavitation bubbles, subsequent growth and collapse significantly. Figure 4 shows a sketch map of cavitation occurring in the heavy oil. When the oil is passing through the nozzle, a bubble grows up to a sufficiently large size and collapses instantaneously. Energy wave carrying high temperature and pressure is generated by this collapsing over a very small location, which can crack the bonds of heavy chains, and form light compounds in heavy oil. When a large number of bubbles are collapsing, the domino effect of collapsing process leads to mechanical shock. When the frequency of the shock is equal to the natural frequency of molecules in oil, resonant vibration happens, and existing bonds of the molecules can be broken. Then radical compounds are released and light compounds are formed. This resonant vibration is shown in Fig. 4 as cracking II. If the frequency of the shock does not meet the condition of resonant, bonds cannot be broken but to vibrate for a while.

(a) Upstream pressure = 1.0 MPa

(b) Upstream pressure = 3.0 MPa

Fig. 3. Viscosity of heavy oil sample

Fig. 4. Sketch map of cavitation occurring in heavy oil

6

S. Zhu et al.

Cavitation is mainly classified as transient and stable cavitation [11]. Transient cavitation connotes a relatively violent bubbles collapse. In contrast, the much less violent form of cavitation, stable cavitation is associated with vibrating gaseous bodies. Bulk nanobubbles (ultrafine bubbles) produced by cavitation, especially during the stable cavitation, affect viscosity by altering stress fields at a molecular level to a certain extent in a short or long period of time [12]. The effect of bulk nanobubbles on the viscosity should not be neglected. Bulk nonobubbles are generated during the cavitation, working as cavitation nuclei [13, 14]. After stopping the cavitation, microbubbles move upward by buoyancy and disappear at the liquid surface, but bulk nanobubbles dispersing in liquid are stable, even for more than a month after their generation [15]. The viscosity reduction caused by nanobubbles will contribute to the heavy oil transportation. When oil sample is circulated in the pipeline, oil temperature cannot rise too much. Even after 100 min in the experiment, the temperature of oil sample, measured immediately, was keeping around 40  C. So the effect of temperature on the viscosity is not taken into account during the experiments. The evaporation of light compounds from the oil can be greatly reduced because of the low temperature. At the same time, existing evaporated compounds can act as donors during the circulation. This operating condition is probably in favor of reorganizing molecular structures during the cavitation. 3.3

FTIR Analysis

Figure 5 shows the FTIR transmittance spectra of all oil samples under different operation conditions. On the transmittance spectra, the range 2500–3700 cm−1 is called the hydrogen-stretching zone, because the vibrations of C-H, N-H, and O-H appear in the frequencies in this region. The range 1000–1600 cm−1 is referred to as the fingerprint region, because various bonds such as C-O, C-N, C-C (single bonds), the C-H bending bond, and some bonds related to the benzene ring are located in this region, which is used to determine the type of functional group. The range 400–1000 cm−1 is aromatic region, it shows the aromatic bonds. In this study, peaks in the oil samples under different hydrodynamic conditions are in the same position, but the shapes are different. This phenomenon indicates that bonds in oil samples are cracked in various degree.

Fig. 5. FTIR transmittance spectra of the heavy oil samples.

Experimental Investigation of Viscosity Reduction of Heavy Oil

7

Oil sample used in this study is base oil SN-500, the characters are dominated by saturated hydrocarbon. Peaks at the frequencies of 2960 cm−1 and 2920 cm−1 represent the asymmetrical stretching of C-H bonds in methyl and methylene respectively. When the bonds are cracked during cavitation, one group of -CH2-CH2- can be divided into two groups of -CH3. Thus the change of chain length in heavy oil can be determined by the ratio value of infrared absorbance of methylene to methyl. Table 3 shows the infrared absorbance of methylene and methyl. It is clear that the ratio value of infrared absorbance can be decreased by 20.42% (Spec. 2), which indicates that bonds of oil sample are cracked, and light compounds are produced. However, the ratio also can be increased, even by 4.49% (Spec. 8). The increase of the value implies that long-chain compounds are formed during hydrodynamic cavitation. There is an original balance between methylene and methyl in oil samples (saturated hydrocarbon) before the experiments. During the cavitation, cracked compounds in oil are unstable due to the short of sufficient hydrogen donors. So cracked compounds in short-chain length tend to form long-chain compounds. The final outcome in the closed pipeline is the reorganization of molecular structures. This reorganization is affected by flow rate, upstream pressure and circulation time obviously. Table 3. Infrared absorbance of methylene and methyl No Before Spec. 1 Spec. 2 Spec. 3 Spec. 4 Spec. 5 Spec. 6 Spec. 7 Spec. 8 Spec. 9 Spec. 10 Spec. 11

Methylene @2920 cm−1 0.931036 1.559540 1.428777 1.480866 1.311640 1.557532 1.519898 1.928468 0.999241 1.154294 1.175657 1.072141

Methyl @2960 cm−1 0.674950 1.144570 1.301561 1.249880 0.946690 1.086848 1.120812 1.571751 0.693293 0.968088 0.954669 0.818713

Ratio (Methylene/Methyl) 1.379414 1.362556 1.097741 1.184807 1.385501 1.433073 1.356069 1.226955 1.441297 1.192344 1.231481 1.309544

Change of the ratio (%) – −1.22 −20.42 −14.11 0.44 3.89 −1.69 −11.05 4.49 −13.56 −10.72 −5.07

It shows in Table 3 that cracked compounds prefer to form heavy chains when the flow rate is low. With the circulation time increasing, those fresh light compounds acting as hydrogen donors can be absorbed into oil again. Thus a new balance between methylene and methyl will be achieved eventually. In other words, more light compounds can be formed by a relative higher flow rate and a relative shorter circulation time.

8

S. Zhu et al.

When the throat diameter is 1.0 mm, higher upstream pressure advances the formation of heavy chains. Although higher upstream pressure improves flow rate, it makes cracked compounds formed into heavy chains more easily. It seems that the formation of heavy chains is sensitive to upstream pressure. According to above analysis and inference, if hydrogen donors are provided sufficiently, a relative higher flow rate and upstream pressure can form more light compounds. While longer circulation time is not in favor of the formation of light compounds during hydrodynamic cavitation.

4 Conclusions Hydrodynamic cavitation has less influence over the density of saturated hydrocarbon in a closed pipeline. However, higher flow rate and upstream pressure lead to more viscosity reduction. As circulation time is prolonged, the viscosity of heavy oil can be reduced continuously. Molecular structures of heavy oil are reorganized by cracking heavy chains and forming fresh compounds during hydrodynamic cavitation. More light compounds can be achieved by a relative higher flow rate. However, cracked compounds in heavy oil prefer to form heavy chains as circulation time is prolonged. Higher upstream pressure also can make cracked compounds advanced into heavy chains. Without sufficient hydrogen donors, light compounds produced by cavitation would act as donors and be absorbed into heavy oil again during the circulation. If outside hydrogen donors are provided sufficiently, light compounds can be produced more.

References 1. Emadi, A., Sohrabi, M., Jamiolahmady, M., Ireland, S., Robertson, G.: Reducing heavy oil carbon footprint and enhancing production through CO injection. Chem. Eng. Res. Des. 89, 1783–1793 (2011) 2. Yoshio, M., Yuuto, S.: Effect of physical properties on micro-explosion occurrence in waterin-oil emulsion droplets. Energy Fuels 24, 1854–1859 (2010) 3. Kharisov, B.I., González, M.O., Quezada, T.S., Gomez, I., Rodriguez, F.E.L.: Materials and nanomaterials for the removal of heavy oil compounds. J. Pet. Sci. Eng. 156, 971–982 (2017) 4. Suslick, K.S., Flannigan, D.J.: Inside a collapsing bubble: sonoluminescence and the conditions during cavitation. Annu. Rev. Phys. Chem. 59, 659–683 (2008) 5. Huang, X., Zhou, C., Suo, Q., Zhang, L., Wang, S.: Experimental study on viscosity reduction for residual oil by ultrasonic. Ultrason. Sonochem. 41, 661–669 (2018) 6. Taheri, J., Shekarifarda, A., Naderic, H.: Analysis of the asphaltene properties of heavy crude oil under ultrasonic and microwave irradiation. J. Anal. Appl. Pyrolysis 129, 171–180 (2018) 7. Taheri, J., Shekarifarda, A., Naderic, H.: The experimental investigation of effect of microwave and ultrasonic waves on the key characteristics of heavy crude oil. J. Anal. Appl. Pyrolysis 128, 92–101 (2017)

Experimental Investigation of Viscosity Reduction of Heavy Oil

9

8. Askariana, M., Vatania, A., Edalata, M.: Heavy oil upgrading via hydrodynamic cavitation in the presence of an appropriate hydrogen donor. J. Pet. Sci. Eng. 151, 55–61 (2017) 9. Askarian, M., Vatani, A., Edalat, M.: Heavy oil upgrading in a hydrodynamic cavitation system: CFD modeling, effect of the presence of hydrogen donor and metal nanoparticles. Can. J. Chem. Eng. 95, 670–679 (2016) 10. Avvaru, B., Venkateswaran, N., Uppara, P., Iyengar, S.B., Katti, S.S.: Current knowledge and potential applications of cavitation technologies for the petroleum industry. Ultrason. Sonochem. 42, 493–507 (2018) 11. Mason, T.J., Joyce, E., Phull, S.S., Lorimer, J.P.: Potential uses of ultrasound in the biological decontamination of water. Ultrason. Sonochem. 10, 319–323 (2003) 12. Yasui, K., Tuziuti, T., Lee, J., Kozuka, T., Towata, A., Iida, Y.: Numerical simulations of acoustic cavitation noise with the temporal fluctuation in the number of bubbles. Ultrason. Sonochem. 17, 460–472 (2009) 13. Yasui, K., Tuziuti, T., Izu, N., Kanematsu, W.: Is surface tension reduced by nanobubbles (ultrafine bubbles) generated by cavitation? Ultrason. Sonochem. 52, 13–18 (2019) 14. Yasui, K., Tuziuti, T., Kanematsu, W.: Mysteries of bulk nanobubbles (ultrafine bubbles); stability and radical formation. Ultrason. Sonochem. 48, 259–266 (2018) 15. Sugano, K., Miyoshi, Y., Inazato, S.: Study of ultrafine bubble stabilization by organic material adhesion. Jpn. J. Multiphase Flow 31, 299–306 (2017)

Study on the Cutting Simulation and Experiment of Titanium Alloy Li Hui1,2, Gao Feng1(&), and Li Yan1 1

2

Faculty of Mechanical and Precision Instrument Engineering, Xi’an University of Technology, Xi’an 710048, China [email protected] Department of Mechanical, Henan University of Engineering, Zhengzhou 451191, China

Abstract. The finite element analysis software AdvantEdge is used to simulate the milling process of titanium alloy TB8 in two dimensions. Comparing with the experimental results, it can be concluded that simulation results are reasonable. Based on the model, the influences of the milling speed vc, the axial milling depth ap, the per-feed feed rate fz on the milling force are analyzed. The temperature field distribution at different milling speeds and the influence of milling speed on the rake face and relief face of temperature are studied. The results show that when milling titanium TB8, it is recommended to use higher milling speed, smaller axial milling depth and feed per tooth when conditions permit. The temperature distribution of the tool has a significant gradient change, the maximum temperature is located in the rake face of the main cutting edge near the tip of 0.01–0.02 mm position. Most of the cutting heat of Titanium alloy milling process is taken away, and only a small part of the heat is transferred to the tool and workpiece. Keywords: Finite element analysis Temperature field


Chip forming
Milling force


1 Introduction During the cutting process, cutting force, cutting temperature, chip formation and distribution of residual stress influence each other, which plays a decisive role in cutting life, surface quality and other properties of the processed workpiece. So within the scope of the high speed cutting, through the theoretical simulation of cutting process on the ultra-high strength TB8 titanium alloy, the investigation of the cutting mechanism using reasonable and effective finite element method can provide an important basis for calculating cutting power, designing and using machine tools, cutting tools and fixtures, and also provide theoretical guidance for optimizing the cutting parameters, cutting tool structure and processing technology, extending the cutting tool life, improving the workpiece processing quality, effectively increasing the efficiency of titanium alloy processing, and reducing the processing cost. Ozel et al. [1] used dynamic grid repartition technology for chip formation. Jianling Chen [2] constructed a 3D milling mechanical model, in which the edge plowing effect in titanium © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2021 Y. Wang et al. (Eds.): IWAMA 2020, LNEE 737, pp. 10–17, 2021. https://doi.org/10.1007/978-981-33-6318-2_2

Study on the Cutting Simulation and Experiment of Titanium Alloy

11

alloy milling is included, and built an empirical prediction model of milling force coefficient which is suitable for high-speed milling; CAI Yujun et al. [3] used the j-c model and the fracture criterion to simulate the serrated chip morphology of high-speed cutting hardened steel and discussed the influence of tool front Angle and serrated chip morphology on the cutting force. Cui Xiaobin et al. [4] analyzed the cutting force and cutting temperature of hardened ceramic tool cart; Chen Wuyi, Yuan Yuefeng et al. [5] established a simulation model for tool wear and prediction in titanium alloy processing, which comprehensively considered abrasive wear, bond wear and diffusion wear. From the current study situation, comprehensive and integrated research on titanium alloy cutting mechanism is still very few, but it is of great significance. Through the analysis of the application and cutting of titanium alloy, the efficient cutting of titanium alloy is the main trend of the future development and also remains a difficult problem to be solved in the field of wide application of titanium alloy. In order to study the high speed cutting mechanism of titanium alloy systematically and comprehensively, we carried out simulation research on cutting mechanism of titanium alloy milling process, which was verified with experiments, and made a comprehensive analysis of simulation result based on elastic-plastic theory of material deformation and material damage failure theory through the finite element method.

2 The Establishment of Finite Element Analysis Model In fact, the metal cutting process is the process that the workpiece material under the action of cutting and extrusion of the tool, firstly, occurs elastic deformation, then plastic deformation and strain hardening, and finally, tears and flows out along the front cutting surface to form chips [6]. Because the strength and hardness of the tool material is much larger than the workpiece material, therefore the author assumes that the tool is elastic material, thus it is calculated according to the elastic material; in the meantime, the workpiece material is calculated according to the elastoplastic material. The workpiece material is conducted by elastoplastic analysis, in which large strain elastoplastic element is adopted. It is basically consistent with the actual cutting condition. 2.1

The Establishment of Geometric Model

During metal cutting, plastic deformation of workpiece and friction between the surface of the chip and that of the tool are two main heat sources. In order to couple the interaction between the thermal load and the mechanical load, it is assumed that the material has the anisotropic strain hardening property. Prandtl-reuss' flow law and Von Mises' yield criterion can be used to derive the thermoelastic thermodynamic coupling constitutive equation. Next, according to the small incremental displacement in the large-deformation and large-strain theory, and the updated Lagrange's formula and the incremental variational principle, the following equation is derived, which shows thermal elastic-plastic large deformation coupling equation [7].

12

L. Hui et al.





 Kep þ ½KG  þ ½KC   Kf fvg Z Z T ep t ¼ ½Be  ½D fe gdv  ½Be T ½ReT dv þ fFng v

ð1Þ

v

In Eq. (1), v is the displacement of the node, and Fn is the normal loading rate of tool and workpiece contact (that is, the ratio of load to time). [Kf], [Kc], [KG] and [Kep] are friction correction matrix, load correction matrix, geometric stiffness matrix and elastic-plastic stiffness matrix. The cutting tool in this analysis uses high speed steel material, and only the workpiece part close to the processing surface is selected as the analysis object, because processing, stress, deformation, and friction heat actually mainly act on this part. In addition, this helps quickly simulate the actual processing to get the expected analysis result. The cutting schematic diagram and the established geometric model are shown in Fig. 1.

Fig. 1. The geometric model

2.2

The Establishment of Material Model

The constitutive model of materials is the flow stress-strain relation, which is the basis of reflecting the deformation of materials. Before performing a cutting simulation on the material, the constitutive behavior of the processed material in high temperature deformation needs to be determined first. Through the experiment, it can obtain the flow stress value with the change of strain rate, strain and temperature in the process of large deformation of metal materials. The data must be able to meet the plastic deformation condition for cutting, truly reflect high strain rate (sometimes as high as /s) and high temperature (above 1000 °C), and constitutive behavior of materials in the condition of large strain (above 4) [8]. The cutting process involves the application of plastic yield criterion, flow criterion and hardening criterion. Because of the influence of thermal coupling and rate, the Johnson-Cook's constitutive relationship is adopted in this study:   0     e T  Tr m re ¼ ½A þ Bðee Þ  1 þ C ln 1 e0 Tm  Tr n

ð2Þ

Study on the Cutting Simulation and Experiment of Titanium Alloy

13

The first part on the right hand side of the equation above shows the effect of the strain e on the flow stress r; the second part shows the effect of strain rate e0 on flow stress r; the last section shows the effect of temperature T on flow stress r. In the formula (2), A represents yield strength under quasi-static condition; B is the ultimate strength; n is the strain hardening index; C denotes the strain rate sensitivity coefficient; m is the temperature sensitivity coefficient; T r is the reference thermodynamic temperature; Tm is the thermodynamic temperature of melting point; re is the yield stress; 0 ee is the strain; e is the strain rate; e0 is the reference strain rate, and the coefficients A, B, C, n and m are constant parameters, which are determined by experiments [9], as shown in Table 1. Table 1. Constitutive equation constants of titanium alloys A(MPa) B(MPa) C m n 968.88 567.17 0.0349 1.30 0.375

3 Simulation Analysis of Simulated Cutting 3.1

Chip Forming Analysis

Through the observation of the chip formation process, the sawtooth shape of the chip gradually forms when the tool cuts into the workpiece. The deformation of the shear zone in the main deformation zone is the most serious, which is concentrated from the tool tip contact area, and then extends upward along a shear angle to the upper surface layer, forming an oblique strip, and finally concentrates near the tool tip contact. In the shear zone, the mesh is elongated and severely distorted to form a sawtooth. When the grid distribution meets the value of the grid repartition criterion, the grid is automatically remeshing. As the tool continues to move, serrated chips continue to form, with the chips leaning slightly forward as shown in Figs. 2 and 3.

Fig. 2. Chip forming process

3.2

Fig. 3. Shear stress distribution

Stress Distribution

As shown in Fig. 4, during the cutting process, thermal softening occurs due to the high temperature effect caused by plastic deformation of the workpiece as a result of great pressure on tool tip, and then plastic flow occurs. With the continuous cutting effect of

14

L. Hui et al.

the tool, the softening zone extends diagonally upward along the shear band, which reduces the carrying capacity of the first deformation zone, resulting in the plastic instability of the zone and the concentrated slip of the entire deformation zone, thus leading to the emergence of adiabatic shear zone. Furthermore, as the tool continues to squeeze, the section slides upward, resulting in chip formation.

Fig. 4. Tool and workpiece stress distribution

3.3

The Analysis of Cutting Force

Based on the establishment of finite element model, the simulation analysis of cutting mechanism is carried out, and the relationship between multiple factors and target variables is summarized. In order to obtain the change law of cutting force in the process of cutting simulation, the simulation should be carried out under multi-factor and multi-level conditions. Due to the complex relationship between various factors in cutting, it is necessary to design a reasonable and accurate simulation scheme in order to obtain more ideal or near real simulation results, and to minimize the number of simulations and reduce the workload. In order to study the effect of cutting speed, cutting depth, and feed on cutting force, this research adopts the principle of orthogonal experiment to design the simulation scheme, selecting three factors and four levels shown in Table 2. Table 2. Cutting simulation test factor level table Level

Level Level Level Level

1 2 3 4

Factors Milling speed Vc(m/min) 50 100 150 200

Axial milling depth Feed per tooth Ap(mm) fZ(mm/r) 0.5 0.05 1 0.1 1.5 0.15 2 0.2

4 Experimental Analysis The experiment was carried out on the Siemens 840D five-axis CNC machine tool. The milling force was measured using nose-c902 three-direction force sensor and HIOKI8423 data signal acquisition instrument, and the transient value and waveform of the

Study on the Cutting Simulation and Experiment of Titanium Alloy

15

output force were connected with the computer. The experimental equipment is shown in Fig. 5, and the experimental panorama is shown in Fig. 6.

Fig. 5. The laboratory equipment

4.1

Fig. 6. The experimental panorama

Chip Shape Comparison

The chip shape comparison between simulation and experiment at a cutting speed of 50 m/min is shown in Fig. 7. The chip shape formed in the cutting simulation is a thick micro-curl light sawtooth shape, which is basically consistent with the chip shape obtained in the experiment.

Fig.7. Comparison of simulation and experimental chip shape

4.2

Comparison of Cutting Forces

From the comparison of cutting forces in Fig. 8, it is found that when the cutting speed is the same, the fluctuation range of cutting forces in actual milling is basically the

16

L. Hui et al.

same as that in the simulation results. By comparing the cutting force peaks of the simulated cutting force and the actual machining feed direction, they are 300 and 260 respectively, and the relative error is only about 13%, indicating that the prediction of the cutting force of the model at the speed of 50 m/min is correct and the error is within the allowable range.

Fig. 8. Comparison of simulation and experimental force

4.3

Rational Analysis of Cutting Temperature

Figure 9 shows the simulated temperature curve. Referring to the Milling Test of Titanium Alloy by Jishi and Ni Junhui et al. [13] and the Milling Test of Ti6Al4V by Xiaotian and Wang Huaifeng et al. [14], it can be known that when the speed is 50 m/min, the axial milling depth is 2 mm, and when the feed per tooth is 0.4 mm/r, the milling temperature is between 450 °C and 500 °C. The simulation results are consistent with them.

Fig. 9. Simulation temperature

5 Conclusion In this paper, the author established a finite element model of milling titanium alloy, based on which the numerical simulation value of the cutting temperature and cutting force in the process of cutting are compared with the actual experiment data. Moreover, we designed a simulation test factors level table, conducted a simulated analysis of the influence of cutting factors on cutting force, and discussed the law of temperature change at different speeds. The research has reached the following conclusions: (1) By comparing with experiments and verifying the correctness of the whole model, it can be used as a guide for practical engineering.

Study on the Cutting Simulation and Experiment of Titanium Alloy

17

(2) In the process of titanium alloy cutting, Fx and Fy all gradually decreased with the increase of cutting speed, while Fx and Fy both showed an increasing trend with the increase of axial milling depth and feed per tooth. (3) The temperature distribution of the tool had obvious gradient variation. The highest temperature was near the main cutting edge of the front cutting surface, 0.01–0.02 mm away from the cutting tip. In the process of titanium alloy milling, most of the cutting heat was carried away by the chip, but only a few was passed into the tool and workpiece. Therefore, when the milling speed reached 200 m/min, the highest temperature in the cutting zone even reached 800 °C, which was close to the phase transition temperature of titanium alloy, but it was speculated that the material will not undergo phase transition.

References 1. Ozle, T.: Computational modeling of 3-D turning with variable edge design tooling: influence of micro-geometry on forces, stresses, friction and tool wear. J. Mater. Process. Technol. 11, 5167–5177 (2009) 2. Chen, J.: Study on the Mechanism of Parameters Optimization during High Speed Milling of Titanium alloys. Shandong University, Jinan (2009) 3. Cai, Y., Duan, C., Li, Y., et al.: FE simulation of chip formation during high speed machining based on ABAQUS. J. Mech. Strength 31(4), 693–696 (2009) 4. Cui, X., Zhaojun, J., Zheng, G., et al.: Numerical simulation of cutting force and cutting temperature in intermittent turning of hardened carbon steel with ceramic inserts. Modular Mach. Tool Autom. Manuf. Tech. 12, 18–20+24 (2010) 5. Chen, W., Yuan, Y.: Research development of cutting technology for titanium alloy. Aeronaut. Manuf. Technol. 15, 26–30 (2010) 6. Rongdi, H.: Principles and Tools of Metal Cutting. Harbin Institute of Technology Press, Harbin (2007) 7. Moriwaki, T., Sugimura, N., Luan, S.: Combined stress: material flow and heat analysis of orthogonal micromachining of copper. Ann. CIRP 42, 75–78 (1993) 8. Kal Pakjian, S.: Manufacturing Process for Engineering Materials, 3rd edn. Addison Wesley /Longman, Menlo Park (1997) 9. Deng, X.: The Simulation Research on the Machining Mechanism during High Speed Cutting of Titanium Alloys. Shenyang University of Technology (2012) 10 Hua, J., Shivpuri, R.: Prediction of chip morphology and segmentation during the machining of titanium alloys. J. Mater. Process. Technol. 150, 124–133 (2004) 11. Jianming, Z., et al.: Finite element simulation and analysis of the chip forming Ti6Al4V based on J-CModel. Mach. Tool Hydraul. 11, 7–10 (2009) 12. Lu, D., Cai, L., Yang, M.: Simulation research of micro-cutting process based on MSG theory. J. Syst. Simul. 25(12), 933–944 (2013) 13. Ji, S., Ni, J., Zhan, B., Yu, W.: Research on milling cuting temperature of TC4 titanium alloy based on orthogonal test method. Manuf. Technol. Mach. Tool 02, 91–93 (2016) 14. Xiao, T., Wang, H., Wu, W.: Finite element meshod simulation in high-speed milling of Ti6Al4V alloy based on AdvantEdge. Coal Mine Mach. 33(05), 138–140 (2012)

Effect of Backlash on Transmission Error and Time Varying Mesh Stiffness Getachew A. Ambaye1 and Hirpa G. Lemu2(&) 1

Department Mechanical Design Engineering, Jimma Institute of University, Jimma, Ethiopia [email protected] 2 Faculty of Science and Technology, University of Stavanger, Stavanger, Norway [email protected]

Abstract. Gears suffer by unavoidable time-varying mesh stiffness. Noise and vibrations amplitudes show the transmission error occurred within the components of power transmission error. The Gear Trax software has been used to model the gear with different amounts of backlash (from 0 mm to 1 mm by 0.2 mm increments) so that this gives a backlash due to the reduction of the thickness of the gear tooth. A plane strain analysis is used to predict the static transmission error, dynamic transmission error and time-varying mesh stiffness of spur gear with different backlash at different coefficients of friction using ABAQUS software. For finite element analysis, the mesh convergence has been done for maximum contact pressure and von Mises stress. The analytical mesh stiffness and the finite element mesh stiffness for gear without backlash have been compared, based on this the prediction time-varying mesh stiffness and transmission errors are continued to predict for the spur gear with different backlashes and coefficient of friction. Keywords: Backlash


Transmission error
Time varying mesh stiffness

1 Introduction Backlash is the gap between gear teeth at the pitch circles. This distance can be involved in the gears either deliberately or without any intention due to the manufacturing error and assembly fault. The basic purpose of introducing backlash is to prevent locking the gears, as well as to prevent contact on both sides of one tooth simultaneously. A small amount of backlash is desirable to make the necessary gap for lubrication and partially expansion of gears. Also, the amount of backlash should not be increased exceedingly, because the backlash causes noise and vibration [1]. Backlash can be introduced in two ways; (1) by reducing the thickness of the gear and (2) by modifying the center distance between the gear mates. Backlash due to tooth thickness change is typically measured along the pitch circle. In this case, either one or both of the gear mate thickness (tb) will be reduced from the designed value (ti). Figure 1(a) shows the left side of gear whose thickness is reduced. The backlash due to this effect can be expressed as Øt = ti – tb. The second way to introduce backlash © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2021 Y. Wang et al. (Eds.): IWAMA 2020, LNEE 737, pp. 18–28, 2021. https://doi.org/10.1007/978-981-33-6318-2_3

Effect of Backlash on Transmission Error and Time Varying Mesh Stiffness

19

in gear is by modifying the center distance of the gear, in which case there is no contact between pitch circles of the two gears. The contact of the gear teeth exists on the addendum. The backlash that is created due to modified center distance is shown in Fig. 1(b). The thickness of the gear tooth will decrease from pitch circle to the top face of the gear. So, there will be a gap between these teeth, expressed as ;c ¼ 2Dc tan a.

Fig. 1. Backlash due to (a) tooth thickness modification and (b) center distance modification.

Walha [2] investigated the dynamics of a two-stage gear system having backlash and time-dependent mesh stiffness. In this study, the periodically changing stiffness and a backlash of the gear contact was observed to result in loss of contact. A linearization technique was used in the study to investigate the nonlinear dynamic response of the system by decomposing the nonlinear system into a set of linear systems that can satisfy given conditions. For low speeds, the system is characterized by a discontinuity of movement which is due to the discontinuous transfer of the kinetic energy from the input wheel to the output wheel. Ji [3] studied the torsional vibrations of the gearbox of a wind turbine that has two planetary gear stages and one parallel gear stage. The nonlinear dynamic model developed considers factors such as time-varying mesh stiffness, damping, and static transmission error and gear backlash. Baumann and Bertsche [4] reached on a conclusion that the reduction of gear rattle noise level can be achieved by avoiding meshing impacts, e.g. by minimizing the traction coefficient of the gear oil or high lubrication film thickness at the gear mesh. In addition, Fernandez-Del-Rincon et al. [5] shown the effect of lubricant entrance in the contact area plays a decisive role in the dynamic behavior. An experiment conducted on helical gear pairs of automotive gearbox in the “idle gear rattle” condition by varying the lubrication mechanism by Russo [6] reveals the effect of lubrication on gear rattling. Transmission error has been defined in different ways. According to Harris [7], it is the difference between the angular position of the output shaft of the driven if the drive were perfect and the actual position of the output. The positional deviation pinion for any given position of the driving gear (as shown in Fig. 2), the actual angular position of the gear is not perfect, due to the deformation of gear tooth, tooth profile error and assembly and installation error of the gear pair, so the actual angular position of the output gear will not be exact with the theoretical and actual position.

20

G. A. Ambaye and H. G. Lemu

Fig. 2. Transmission error.

The transmission error (TE) would be zero if a pair of meshing gears with rigid, perfect, uniformly spaced involute teeth transmit exact and uniform angular motion [8]. Equations (1) and (2) give expressions for the angular TE and its linear form respectively: NG hG ð x Þ NP   DP NG 1 hP ðxÞ  TE ðlmÞ ¼ hG ðxÞ ¼ ½DP hP  DG hG  2 2 NP TE ðxÞ ¼ hP ðxÞ 

ð1Þ ð2Þ

Power and motion cannot be transmitted exactly basically due to the geometrical imperfection and deformation due to the applied torque, this is called the transmission error. If the gear mates are perfect geometrically and un-deformed then the torque and the motion will be transmitted without any deviation. As explained above, the source of the transmission error is geometrical imperfection and deflection due to the load applied on the gear. The geometry error mainly arises from manufacturing error and assembling the gear system. In addition, TE could occur due to different gear failure modes, such as pitting, scuffing and others. Minimization of thickness due to wear or manufacturing error can be considered as a backlash and this article reports the study conducted on the effect on the transmission error and time-varying mesh stiffness.

2 Methods of Estimating Time-Varying Mesh Stiffnes and Load Sharing The time-varying mesh stiffness (TVMS) can be estimated primarily in three main methods: (1) analytical, (2) finite element based and (3) approximate methods.

Effect of Backlash on Transmission Error and Time Varying Mesh Stiffness

2.1

21

Analytical Approach

Three contributing factors need to be considered when analytically calculating the tooth deformation (or deflection) in the line of action (LOA) at a contact point subjected to a certain mesh force and these are (1) local deformation caused by the Hertzian contact; (2) beam deflection caused by tooth considered as a cantilever beam, and (3) deflection caused by the flexibility of the foundation of the gear body. Hertzian Contact Deformation dh : The Hertzian contact deformation dh between the meshing tooth surfaces of a mating tooth pair is nonlinear and various approximate formulas for the calculation of dh have been developed by Kiekbusch et al. [9]. A simplified nonlinear contact deformation based on the semi-empirical equation (Eq. (3)) developed by Palmgren [10] has been adopted by many such as Ma et al. [11]. dh ¼

1:275F 0:9 E0:9 l0:8

ð3Þ

Tooth Beam Induced Deformation: The tooth beam induced deformation can be determined by considering it as a non-uniform cantilever beam and is considered as it is rigidly fixed at its foundation. The potential energy method is widely used to derive the tooth beam deflection under a contact load. In general, when a tooth beam is under the action of the mesh force F at point P (as shown in Fig. 3), there will be potential energy stored in the tooth beam due to the bending moment M, the shear force and the axial compressive force. Therefore, the total tooth deformation consists of bending moment, shear force and axial compressive force induced. Expressions that can be used to calculate these deformations are given in [12].

Fig. 3. Geometric parameters for gear body.

Tooth Foundation Induced Deformation: Foundation induced deflection has been reported by Sainsot et al. [13] who studied the effect of fillet foundation deflection on the gear mesh stiffness, derived this deflection, and applied it for a gear body. The article also presents methods of calculating fillet foundation deflection.

22

2.2

G. A. Ambaye and H. G. Lemu

Finite Element Approach

FE models are the primary tools used to obtain gear mesh stiffness due to their significant advantage in representing the crucial tooth contact behavior. Two-dimensional (2D) models and three dimensional (3D) models are both common in the literature. Time-varying mesh stiffness of the gear with different backlash will be examined using plane strain FE analysis in Sect. 3. 2.3

Approximate Mesh Stiffness Analysis

Philippe [14] developed a hypothesis stating that the minimum elastic potential energy and the mesh stiffness of a couple of teeth in contact at a certain point approximately. The first two methods (analytically and finite element) have been used for the analysis of time-varying mesh stiffness of the gear with zero mm backlash for a gear whose specification is shown in Table 1. Table 1. Geometric, material properties and operating conditions of the meshing gear pair [15]. Parameter Value Module, m 4 mm Pressure angle, a 20° 14 and 22 Number of teeth, ZP and ZG Face width, B 6 mm Normal tooth load per unit face width, w 75.5 N/mm Angular velocity of pinion, xP RPM Mass density 7:86  106 Kg=mm3 Modulus of elasticity, EP ; EG 207  103 N/mm2 Poisson ratio, m 0.3

3 Numerical Analysis of TE For this analysis, the sketch model of a single tooth was imported to ABAQUS and then develop a geometry of gear as a plane object and the material property of the gear were applied. The analysis type was defined as a general static analysis in order to determine the mesh convergence. Then the boundary conditions were applied and the system was subjected to a torque of TG ¼ 3322 N mm (Fig. 4). To develop the mesh model, CPE4 (A 4-node bilinear plane strain quadrilateral) element type was used. Then, the mesh convergence is done by fixing all linear and angular degrees of freedom of both centers of pinion and gear except for the rotational degree of freedom of gear (Fig. 4). The mesh convergence analysis was done at the pitch point, i.e. the teeth of the gear pair’s contact at the pitch point. Figs. 5(a) and (b) show the von Mises and contact stresses for 0.6 mm mesh size respectively. In this article, the contact stress and von Mises stress are used as an output to find a mesh convergence. As Fig. 6 shows, from 1.1 mm to 0.2 mm mesh size the contact pressure and von Mises stress increase slowly and then increase rapidly until 0.06 mm

Effect of Backlash on Transmission Error and Time Varying Mesh Stiffness

23

mesh size. And after 0.06 mm mesh size, the contact pressure and von Mises stress slowly increase. So, for the rest of the analysis 0.06 mm mesh size will be taken.

Fig. 4. Boundary condition and load.

Fig. 5. (a) von Mises stress, (b) Contact pressure.

Fig. 6. Mesh convergence.

4 Static Transmission Error Analysis with Backlashes In static transmission error (STE) analysis, the gear is free to rotate about its axis and the pinion is fixed in all degrees of freedom. The static general step is applied for this analysis. The torque is applied on the gear and then the STE is determined in ABAQUS using the following three steps: (1) initial step: setting initial conditions before the gear sets are subjected to any external load, (2) contact step: applying the torque on the center of the gear to ensure the gear tooth surfaces are in contact and (3) rotation step: granting rotation of the pinion. The amount of rotation should be equal or greater than   role angle and this is to get a full mesh cycle. In this case, a role angle of 32 is 32 is

24

G. A. Ambaye and H. G. Lemu

used. The number of elements and nodes is given in Table 2 and the obtained STE for a different amount of backlash and friction are shown in Fig. 5. For contact without backlash, the TE is maximum (48 lm) at a single tooth contact region and minimum (43 lm) at double tooth contact region for a coefficient of friction (COF) of l = 0.15. For l = 0.2, the STE in the single tooth contact region is around 51 lm and 34.5 lm in the double tooth contact region. Table 2. Finite element analysis summary Backlash of gear Number of elements Number of nodes Total CPU time (hrs) 0 mm 62830 190488 10.84194 0.2 mm 62692 190048 8.161389 0.4 mm 65187 197565 10.81167 0.6 mm 62107 188319 11.99361 0.8 mm 64747 196209 15.19306 1 mm 4048 194148 11.43111

Fig. 7. STE of gear with: (a) 0, (b) 0.2, (c) 0.6, and (d) 1 mm backlash.

As depicted in Fig. 5(a), the maximum TE for l = 0.3 is 54 lm and the minimum value is 36 lm. As the backlash increases, the STE also increases (shown Fig. 5(b)– (d)). In general, backlash and friction of the gear material seriously affect the STE of

Effect of Backlash on Transmission Error and Time Varying Mesh Stiffness

25

the gear. The vibration of the transmission system is the result of TE, therefore, the backlash will seriously affect it.

5 Dynamic TE Analysis of Gear with Backlashes The boundary condition in this case is the same as that applied for the STE analysis. The only difference is that dynamic implicit step is used and there is only a single step beside the initial step. This step can be renamed as rotation, and in this case, both the torque and the rotations are applied at the same time. DTE with different coefficient of friction is examined and results for 0, 0.2, 0.6 and 1 mm are plotted in Fig. 6(a)–(d). As shown in Fig. 6(a), the DTE of spur gear without backlash is lower than those with backlash (Fig. 6(b)–(d)), and generally, the DTE increases dramatically with increasing backlash and COF. The DTE in the double tooth contact region can be minimized by modifying the gear involute profile.

Fig. 8. DTE of gear with: (a) 0 mm, (b) 0.2 mm, (c) 0.6 mm, and (d) 1 mm backlash.

26

G. A. Ambaye and H. G. Lemu

6 Influence of Backlash on Time-Varying Mesh Stiffness An effective time-varying mesh stiffness (TVMS) model is a basic condition to conduct a dynamic analysis, especially with regard to the mechanism of the noise. The main internal excitation source of gear dynamics is time-varying gear mesh stiffness. There are two types of stiffness for a pair of gears: (1) rectilinear mesh stiffness and (2) torsional mesh stiffness. Rectilinear mesh stiffness is an equivalent mesh stiffness of a pair of gears along the action line, while torsional mesh stiffness is the ratio between a torque applied on a gear (pinion fixed) and the corresponding angular displacement of the gear body. These two types of mesh stiffness are related to each other, as given in [16]. Rectilinear mesh stiffness can be calculated using the experimental, FEM and potential energy methods, while torsional mesh stiffness is commonly evaluated using FEM. In this study, analytical and FEA of a gear without backlash and with zero COF has been done to determine the TVMS (Fig. 7). Minimum values of 36 MN/mm and 70 MN/mm were obtained in single and double contact, resp. The prediction of TVMS for the spur gear with a backlash from 0 mm to 1 mm by 0.2 mm increment and with different COF was conducted using FEA (plots of 0, 0.2, 0.6 and 1 mm are shown in Fig. 8). The stiffness of the gear varies between 36 and 70 MN/mm with time or roll angle. This is due to the load variation along the line of action. In general, as the backlash increases the mesh stiffness decreases drastically (Figs. 9 and 10).

Fig. 9. Time varying mesh stiffness gear with zero backlash and without friction.

Effect of Backlash on Transmission Error and Time Varying Mesh Stiffness

27

Fig. 10. TVMS of gear with: (a) 0 mm, (b) 0.2 mm, (c) 0.6 mm and (d) 1 mm backlas

7 Conclusion The reduction of gear tooth thickness due to operational wear or design and manufacturing error can be modeled as backlash. In this article, the effect such wear (backlash) on the transmission error and time-varying mesh stiffness has been analyzed using finite element analysis. From this analysis, the following conclusions are drawn: • Backlash affects both dynamic and static transmission error. As the backlash increases, the transmission error increases drastically. This revealed that a little thickness reduction will lead to high noise and amplitude of vibration. • The mesh stiffness of the gear is decreased as the backlash is increased. The gear will suffer due to a maximum mesh stiffness variation as a result of small tooth thickness modification. • The COF of the gear material will affect the transmission error, specifically, its effect is maximizing the STE when compared to the DTE. In addition, the TVMS is highly affected by the COF. As the friction between the mating gears increases the time-varying mesh stiffness of the gear will decrease.

28

G. A. Ambaye and H. G. Lemu

References 1. Papageorgiou, D., Blanke, M., Niemann, H.H., Richter, J.H.: Backlash estimation for industrial drive-train systems. IFACPapers OnLine 50(1), 3281–3286 (2017) 2. Walha, L., Fakhfakh, T., Haddar, M.: Nonlinear dynamics of a twostage gear system with mesh stiffness fluctuation, bearing flexibility and backlash. Mech. Mach. Theory 44(5), 1058–1069 (2009) 3. Zhao, M., Ji, J.C.: Nonlinear torsional vibrations of a wind turbine gearbox. Appl. Math. Model. 39(16), 4928–4950 (2015) 4. Baumann, A., Bertsche, B.: Experimental study on transmission rattle noise behaviour with particular regard to lubricating oil. J. Sound Vib. 341, 195–205 (2015) 5. Fernandez-Del-Rincon, A., Diez-Ibarbia, A., Theodossiades, S.: Gear transmission rattle: assessment of meshing forces under hydrodynamic lubrication. Appl. Acoust. 144, 85–95 (2019) 6. Russo, R., Brancati, R., Rocca, E.: Experimental investigations about the influence of oil lubricant between teeth on the gear rattle phenomenon. J. Sound Vib. 321(3–5), 647–661 (2009) 7. Harris, S.L.: Dynamic loads on the teeth of spur gears. Proc. Inst. Mech. Eng. 172(1), 87– 112 (1958) 8. Mark, W.D.: Analysis of the vibratory excitation of gear systems: basic theory. J. Acoust. Soc. Am. 63(5), 1409–1430 (1978) 9. Kiekbusch, T., Sappok, D., Sauer, B., Howard, I.: Calculation of the combined torsional mesh stiffness of spur gears with two and three-dimensional parametrical FE models. SV J. Mech. Eng. 57(11), 810–818 (2011) 10. Palmgren, A.: Ball and Roller Bearing Engineering. SKF Industries Inc., Philadelphia (1959) 11. Ma, H., Zeng, J., Feng, R., Pang, X., Wen, B.: An improved analytical method for mesh stiffness calculation of spur gears with tip relief. Mech. Mach. Theory 98, 64–80 (2016) 12. Yu, W., Shao, Y., Mechefske, C.K.: The effects of spur gear tooth spatial crack propagation on gear mesh stiffness. Eng. Fail. Anal. 54, 103–119 (2015) 13. Sainsot, P., Velex, P., Duverger, O.: Contribution of gear body to tooth deflections—a new bidimensional analytical formula. J. Mech. Des. 126(4), 748–752 (2004) 14. Philippe, V.: On the modelling of spur and helical gear dynamic behavior. In: Mechanical Engineering. InTech, Croatia (2012) 15. Taburdagitan, M., Akkok, M.: Determination of surface temperature rise with thermo elastic analysis of spur gears. Wear 261(5–6), 656–665 (2006) 16. Lin, T., Ou, H., Li, R.: A finite element method for 3D static and dynamic contact/impact analysis of gear drives. Comput. Methods Appl. Mech. Eng. 196(9–12), 1716–1728 (2007)

Fatigue Life Study of False Banana/Glass Fiber Reinforced Composite for Wind Turbine Blade Application Temesgen Batu1 and Hirpa G. Lemu2(&) 1

2

School of Mechanical and Industrial Engineering, Wollo University, Dessie, Ethiopia [email protected] Faculty of Science and Technology, University of Stavanger, Stavanger, Norway [email protected]

Abstract. Using the hybridization of natural fibers and synthetic fibers reinforced composite material particularly for wind turbine blade applications reduces the problems exist in wind energy sectors i.e. higher cost, higher density, and effects in polluting the environment. This paper focuses on the study of potential of false banana/glass reinforced hybrid composites for wind turbine blade under fatigue load. The physical properties of the hybrid composite were obtained experimentally. Based on obtained material properties, structural analyses and fatigue life predictions have been performed for the blade in ANSYS for specific wind speed. Von Mises stress, total deformation, fatigue life, fatigue damage, fatigue safety factor and weight of materials per blade were compared with randomly oriented short synthetic fiberglass-reinforced epoxy. The results obtained from designed hybrid composite for turbine blade indicate reduction of weight and improvement of fatigue life compared to synthetic fiber glass- reinforced epoxy composite for turbine blade. Keywords: False banana


Wind turbine blade
Composite
Fatigue life

1 Introduction Nowadays, an increasing demand for clean energy to replace fossil fuel to reduce the negative effects of fossil fuels. The energy obtained from the wind is the most costeffective and feasible energy resource by using wind turbines [1]. In a wind turbine system, the most critical component is the turbine blade since the manufacturing cost of the blades is approx. a fifth of the total cost for all components and replacing them during operation is not only difficult but also expensive [2]. Because the wind turbine blade operates under dynamic loads, the effect of aerodynamic loads acting on the blades may cause unpredictable failure. To date, the predominant number of accidents in wind farms has been due to blade failure; a failure rate of 1:184 per year [3]. Fatigue failure can have a direct or indirect effect on the overall performance of wind turbine blades, accelerating their degradation process and decreasing their energy production efficiency. There are three design drivers of the blades that rely on fatigue [4]; © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2021 Y. Wang et al. (Eds.): IWAMA 2020, LNEE 737, pp. 29–40, 2021. https://doi.org/10.1007/978-981-33-6318-2_4

30

T. Batu and H. G. Lemu

(1) aerodynamic performance (blade shape), (2) power performance (power efficiency, power curves, noise) and (3) loading performance. This makes fatigue analysis of the blade an important factor in evaluating its performance. Failure and life of wind turbine blade is directly related with the material it is constructed from. The composites are selected for this sector due to superior mechanical properties, higher strength and lower weight, as compared to many metals and alloys. For small wind turbines, Kumar et al. [5] established the fatigue behavior of small blades and predicted the working lifespan and also compared the suitability of the blade with different composite materials such as Kevlar, Glass Fiber Reinforced Plastic (GFRP) and Carbon Fiber Reinforced Plastic (CFRP) by simulating the displacement and stress using ANSYS. They concluded that GFRP material is safer and secure. The great problems facing the blades of a modern turbine are that the used materials have a limitation of recyclability. Alternative materials with better properties such as low density, longer life, higher performance, ease of processing, capability to recycle, less expensive than the current thermoset technology is needed for wind turbine blade. Natural fiber-reinforced composites are getting popularity to solve the challenge in this sector because they have many advantages compared to glass fiber. This reinforcement can replace conventional fibers such as glass as an alternative [6]. Literature indicates that natural fibers such as bamboo, banana, flax, jute, hemp, sisal, and pineapple, etc. have potentials for turbine blade application since they have significant advantages compared with synthetic fibers [7]. Ensete (Ensete ventricosum), also known as false banana [8], which is abundantly available Sothern part of Ethiopia, is also another suitable plant with a great potential to produce natural fibers. The fiber is traditionally used to make ropes, mat and for other purposes. The main challenges associated with natural fiber reinforcements include severe moisture absorption, fire resistance, mechanical properties and durability, variability, and manufacturing/ processing of natural fiber reinforced plastic composites [6]. Hybridizing natural fibers with synthetic fibers can reduce the limitations of both natural and synthetic fiber- reinforced composites. Thus, this paper intends to study the performance of false banana and glass fiber reinforced epoxy hybrid composite for small wind turbine blade under fatigue loads in ANSYS software.

2 Materials and Methods 2.1

Material Property Characterization of False Bana/Fiberglass Hybride

Materials: For this experiment, short randomly oriented glass fibers of false banana and epoxy resin with its hardener are obtained from local market. False banana fibers were extracted from the Enset plant manually by scraping the layers on plane wood. For bleaching and cleaning the surface of fibers, to maximize the efficiency of the fiber as reinforcement, the chemical treatment process was preceded by NaOH. This removes moisture content from the fibers thereby increasing its strength. Furthermore, it enhances the flexural rigidity of the fibers, clears all the impurities that are adjoining the fiber material and stabilizes the molecular orientation. For this study 500 gm of NaOH in pellets form was used (brand name: RANKEM and code, S0290).

Fatigue Life Study of False Banana/Glass Fiber Reinforced Composite

31

Epoxy is the most commonly used resin in polymer matrix composites. It is one of the advanced thermosetting resin types which do not give off reaction products when cured and has low shrinkage. It also has good adhesion to other materials, good chemical and environmental resistance, good chemical properties and insulating properties [9]. Due to its superiority advantages it is commonly used for wind turbine blade. For this study, epoxy resin with a brand name of EPOXY MAS RESIN was used. The epoxy resin is cured by adding a catalyst, which causes a chemical reaction without changing its own composition. The curing agent has a brand name HARDNER MAS with the ratio of 2:1 for epoxy and hardener respectively as recommended. Finally, proper amounts of epoxy and hardener were mixed and stirred for few minutes. 2.2

Composite Material Fabrication and Characterization

Experimental Design: For fabrication and characterization of composites, first the experiment was designed by considering two factors that affect the output; namely (1) fiber orientation and (2) fiber volume fractions. The volume fraction of the fiber used for wind turbine blades ranges from 50–60% [10], while in some other literatures, for instance in [11], 30–50% volume fraction of natural fibers are also reported. Besides, as fiber volume increases, more delamination takes place between the fibers. Thus, considering these cases, 50% volume fraction (glass and false banana) fibers and 50% matrix were used. Thus, varying the volume fraction of false banana fiber and glass fiber, different orientations were considered. As different researchers indicated, 0° direction orientation gives higher tensile strengths. Thus, for orientation of false banana fiber 0° direction was considered. Using the density of the, false banana fiber, glass fiber and epoxy resin of 1.4 g/cm3, 2.57 g/cm3, 1.2 g/cm3 respectively, the mould of composite was prepared with a size (mm) of 300  300  3. Hand Layup Method for Fabrication Process: Due to its economical nature and simplicity, hand lay-up was used to apply the reinforcement material into the mold. The reinforcing material (i.e. the natural fibers) is placed in the mold and then saturated with epoxy resin using a brush or a two-component spray system. First lamina was prepared for both glass and false banana fibers based on experimental design. Then, the matrix was added based on the determined percent. At the end, the molded composites were compressed under compression machine with the pressure of 5 MPa. After manufacturing of the laminated hybrid composite materials, the specimen for tensile test was cut into smaller pieces of testing specimen according to ASTM D 3039 [12] standard. Then, a universal testing machine was used for the test. 2.3

Blade Modelling and Stress and Fatigue Life Analysis Using FEM

Solid Works was used to develop the 3D model (Fig. 1) of the blade that has a blade length of 4.953 m. The geometry distribution data was obtained from the wind turbines used by the Department of Aerospace and Mechanical Engineering at the University Notre Dame in Notre Dame, Indiana [13]. The 3D model of the blade was generated in three coordinates (x, y, and z) taking each blade element from the airfoils tools site [14]

32

T. Batu and H. G. Lemu

for NACA 4415 airfoil. Then, by importing the coordinates of each element of the blade, the profile curve of the blade element was drawn (Fig. 1(a)). The figure shows the profile curves of blade elements along the spanwise direction, and the chords of each blade element changes along the spanwise direction. Figure 1(b) shows the 3D model of turbine a blade.

Fig. 1. (a) Profile curves and (b) 3D model of the blade.

Defining Material Properties in FEM Software: Based on the experimentally determined values, other values such as Young’s moduli (E1 and E2) were computed from the stress-strain curve. Micromechanics formula (Eq. (1)) was used to compute the shear moduli (G) and the Poisson’s ratios m12 and m21 were calculated using Eq. (2a, 2b) because of the lack of relevant universal testing machine with the strain gauge.

G ¼

E 2ð 1 þ v Þ

v12 ¼ VG vG þ VFB vFB þ Vm vm

v21 ¼ v12

E2 E1

ð1Þ ð2aÞ

ð2bÞ

where VG, VFB & Vm are volume ratios of the glass fiber, false banana and matrix resp. Christoforo et al. [15] investigated the effect of the Poisson’s ratio on mechanical behavior of the composites using different levels of the Poisson ratio and concluded that variation of Poisson’s ratio of the fibers does not affect the FEA results. Thus, when the micromechanics model was used, it was observed that the value of the poison’s ratio has no significant effect on the computed results. For this study, the average m = 0.175 for false banana fibers, 0.22 for glass fibers and 0.4 for epoxy resin were used.

Fatigue Life Study of False Banana/Glass Fiber Reinforced Composite

33

S-N curves are widely used to characterize the fatigue behavior of materials. The stress life method is particularly good for high cycle fatigue. To assess the damage of composite blades by the S-N curve, it is necessary to know the fatigue life cycles of the materials. Fatigue properties of materials are usually expressed in the form of SN curves based on experimental data on stress amplitude versus the number of cycles to failure relation. There are different methods to model S-N curve if there is no experimental data. A typical S-N curve model for composite materials can be expressed as Eq. (3) [16]. N¼

rult  rmax a0 ðrmax Þ1:6

!1=b þ1

ð3Þ

Where rult is the ultimate strength, rmax is the peak stress, and a0 ¼ 0:0075, b ¼ 0:048, R = 10 are parameters used for the capability of curve fitting and applicability to different stress ratios. The relationship between the stress ratio R = 10 and the S-N curve by the model is used to calculate the fatigue damage. Transforming the material properties into engineering data of ANSYS workbench; the material of the blade was defined as shown in Fig. 2.

Fig. 2. Workbench material properties of false banana/ glass fiber reinforced epoxy resin hybrid composite material.

Attaching the Material Model and Meshing: To facilitate the analysis in ANSYS, the 3D model of the wind turbine blade was modeled by Solid Works and saved in a neutral file format such (IGES format), which was then imported into the ANSYS workbench software. The imported 3D model is shown in Fig. 3.

34

T. Batu and H. G. Lemu

Fig. 3. The browsed 3D model of wind turbine blade.

By applying the meshing step, an adequate mesh density is then provided on contact surfaces to allow contact stresses to be distributed smoothly. Likewise, mesh densities that are adequate for resolving stresses in areas where stresses or strains are of interest require a relatively fine mesh compared to that needed for displacement or nonlinearity resolution. The quality of meshing and accuracy of the work is evaluated using the procedure; Mesh ! Details of “Mesh” ! Statistics ! Mesh Metric ! Skewness OR Orthogonal Quality. A good meshing can be sensed through a maximum skewness that is lower than 0.95 and minimum orthogonality greater than 0.15 as shown by quality criteria in Table 1 [17]. Table 1. Mesh quality evaluation for the model

Skewness Orthogonal quality

Outstanding Very Good Sufficient Bad Inappropriate good 0–0.25 0.25–0.50 0.50–0.80 0.80–0.95 0.95–0.98 0.98–1.00 0.95–1.00 0.70–0.95 0.20–0.70 0.15–0.20 0.001–0.15 0–0.001

In this case, the quality of meshing obtained has 0.844 a minimum orthogonal quality and 0.36055 a maximum skewness which is categorized under very good in

Fig. 4. Mesh qualities for the wind turbine blade.

Fatigue Life Study of False Banana/Glass Fiber Reinforced Composite

35

both cases as observed in Fig. 4. The grid can be automatically divided and the model after meshing has 81 216 nodes and 81 120 elements. Applying Loads and Boundary Conditions: The blades are generally exposed to three different load sources: wind load, gravity, and centrifugal load. Since centrifugal force is relatively low, the influence is often neglected and will not be considered. The wind loads on the wind turbine can be decomposed into normal and tangential force components. Due to the flat section of the blade, the impact of tangential force on the bending and strength of the blade is small, and thus it can also be neglected. So, among the loads of the wind turbine, only the normal force component of the wind load should be considered. For this study, the critical wind loads were obtained by analyzing one year data of Adama I wind farm. The normal wind load as a pressure load was applied based an average wind speed of the daily report from July 08, 2016–July 07, 2017 of Adama wind farm I. This critical load was computed from the average of 365 days wind speeds of 7.684 m/s, which was used for fatigue analysis of the wind turbine blade. Fatigue Analysis: Fatigue analysis was done to study if the material can survive the number of cycles the blade experiences during its lifetime. Stress life approach was used based on Stress-Cycle (S-N) curves and then modified by a variety of factors, such as loading type, mean stress effects, multiaxial stress correction, and fatigue modification factor. Stress life approach is concerned with the blade’s total life and it does not distinguish between initiation and propagation. Ratio based constant amplitude that considers proportional loading with a load ratio of 10 was used since the model used for computing stress-cycle (S-N) curves indicate this loading ratio. The solution is generated based on the input parameters. The total deformation and equivalent (von Mises) stress, fatigue life, fatigue damage, fatigue safety and fatigue sensitivity were solved by the software.

3 Discussion of Results In this section, the results of designed false banana/glass fiber reinforced polymer (FBGFRP) and glass fiber reinforced polymer (GFRP) blades obtained from the analysis as well as results of hybrid and glass/epoxy composite materials are discussed. 3.1

Equivalent (Von Misses) Stress and Deformation

The von Mises stress values of both the laminated false banana/glass epoxy hybrid and the glass epoxy composite turbine blades from the FEA are shown in Fig. 5.

36

T. Batu and H. G. Lemu

Fig. 5. Equivalent (von Misses) stress of FBGFRP and GFRP composite blade model.

As can be observed, the stress levels are directly related with the material type. The deformation levels (figure omitted) show similar condition. For a good and safe design of wind turbine blade, a low value of deformation (i.e. high stiffness) and stress level are desirable. Lower von-Misses stress improves the life span of the blade. By comparing the stress and deformations, it is observed that false banana /glass fiber reinforced hybrid composite reduces von-Misses stress by 6.23% and deformation by 7.2% of the wind turbine blade. Thus, it is concluded that the false banana /glass fiber reinforced hybrid has better performance than glass/epoxy. 3.2

Weight Reduction for the Blade

As can be observed from Fig. 6, weights of the blade for both materials are displayed. The mass of false banana/glass epoxy hybrid reinforced composite wind turbine blade is 9.8058 kg, while that of glass/epoxy composite wind turbine blade is 11.935 kg. That implies that the mass ratio becomes 0.822 and the percentage reduction of mass becomes 17.84%. The weight of the blade is reduced by about 17.84% by replacing glass/epoxy with a false banana/glass epoxy hybrid composite material for a wind turbine blade. The smaller mass of the laminated false banana/glass epoxy hybrid composite makes the blade lightweight, so the blade can be operated at a lower wind speed and the turbine speed can be increased leading to improved output power. The wind blade also overcomes friction and begins to rotate at a very lower cut-in wind speed. 3.3

Fatigue Life

The fatigue life analysis results were obtained based on the range from contour plots of a specific result over the whole model or the most damaged point in the scope of the result. Figure 7(a) and (b) show the fatigue lives of false banana and glass fiber reinforced polymer composite simulation results respectively.

Fatigue Life Study of False Banana/Glass Fiber Reinforced Composite

37

Fig. 6. Total blade mass of FBGFRP and GFRP composite blade model.

Fig. 7. Fatigue life simulation of FBGFRP and GFRP composite blade model.

Based on the analysis, the fatigue life of the wind turbine blade using ANSYS tool indicates the margin of possible life available for the present work. For both blades root area of the blade is the most critical possible location which undergoes stress reversals. The fatigue life shows the available life for the given fatigue analysis which represents the number of cycles until the part fails due to fatigue. The number of cycles expected for wind turbine blades are estimated to the number of load cycles are up to 108 or 109 load cycles. It is observed that the minimum fatigue life of wind turbine blades are 6.0298e + 009 and 2.5913e + 009 cycles for glass fiber reinforced hybrid composite and false banana /glass fiber reinforced hybrid composite blade model respectively at 7.684 m/s wind speed. These values meet the requirement expected from materials and reached at the location where maximum stress is expected to occur. Fatigue Damage: Most wind turbine rotor blades are designed for economical lifetime more than 20 years. Estimates of number of load cycles are up to 108 or 109. Figure 8 shows the fatigue damage at 109 cycles. Fatigue damage values greater than 1 indicate failure before design life, while values less than one indicate failure is not happen before design life. As can be observed from figure, fatigue damage of both FBGFRP and GFRP composite blade model is less than one. So, both materials are safe and FBGFRP is safer than GFRP. Another key parameter in fatigue life analysis is the factor of safety with respect to the given design life. For the wind turbine blade at 109 cycles, the minimum fatigue

38

T. Batu and H. G. Lemu

safety factor for both FBGFRP and GFRP composite blade model is 1.7048 and 1.2917 respectively. This means both materials are safe and FBGFRP is safer than GFRP.

Fig. 8. Damage of FBGFRP and GFRP composite blade model.

Fatigue Sensitivity: The fatigue sensitivity the hybrid composite that shows how the fatigue results change as a function of the loading at critical locations is given in Fig. 9. Using fatigue sensitivity plots, it’s possible to observe the influence on fatigue life if the load changes, for example from 50% of the current load up to 150% of the current load. As shown in the plots, it is observed that when the load is increased up to 150%, the life decreases to 2.95e + 10 cycles for glass fiber epoxy reinforced composite and to 1.551e + 10 for false banana /glass fiber epoxy reinforced hybrid composite.

Fig. 9. Fatigue sensitivity curves of (a) FBGFRP and (b) GFRP composite blade models.

Fatigue Life Study of False Banana/Glass Fiber Reinforced Composite

39

4 Conclusion In this study, the performance of false banana/glass reinforced hybrid composites has been investigated when used for wind turbine blade applications under fatigue load. The mechanical properties such as elastic modulus, ultimate strength and density of the constituent materials were obtained from experimental tensile test of the material. The problem under study was modelled and simulated using Solid Works and ANSYS respectively. The stress distribution, deformation and weight of false banana/glass reinforced hybrid composites for the selected blade profiles were analyzed. The fatigue damage during the designed service life of the turbine is then assessed as fatigue life and sensitivity of the blade for failure were computed by the software and discussed. From the numerical simulation results, it has been concluded that the replacement of randomly oriented glass/epoxy composite based wind turbine blades by designed false banana/glass reinforced hybrid composites leads to a weight reduction per blade by 17.84%. In addition, the minimum fatigue life cycle of the hybrid composite is higher than the designed 109 cycles.

References 1. Talbot, J., Wang, Q., Brady, N., Holden, R.: Offshore wind turbine blades measurement using Coherent Laser Radar. Measurement 79, 53–56 (2016) 2. Auger, D., Wang, Q., Trevelyan, J., Huang, S., Zhao, W.: Investigating the quality inspection process of offshore wind turbine blades using B-spline surfaces. Measurement 115, 162–172 (2017) 3. Lee, H.G., Lee, J.: Measurement theory of test bending moments for resonance-type fatigue testing of a full-scale wind turbine blade. Compos. Struct. 200, 306–312 (2018) 4. Brøndsted, P.B., Nijssen, P.L.: Advances in Wind Turbine Blade Design and Materials. Woodhead Publishing Limited, Cambridge (2013) 5. Kumar, M.S., Krishnan, A.S., Vijayanandh, R.: Vibrational fatigue analysis of NACA 63215 small horizontal axis wind turbine blade. Mater. Today Proc. 5, 6665–6674 (2018) 6. Alene, A.: Design and analysis of bamboo and E-glass fiber reinforced epoxy hybrid composite for wind turbine blade shell. Master thesis, Addis Ababa University, Ethiopia (2013) 7. Kalagi, G., Patil, R., Nayak, N.: Natural fiber reinforced polymer composite materials for wind turbine blade applications. Int. J. Sci. Dev. Res. 1(9), 2455–2631 (2016) 8. Tsehaye, Y., Kebebew, F.: Diversity and cultural use of Enset (Ensete ventricosum (Welw.) Cheesman) in Bonga in situ conservation site. Ethnobotany Res. Appl. 4, 147–157 (2006) 9. Alqahtani, R.A.: Flexural properties of sisal fibre/Epoxy composites. Bachelor Project, Faculty of health, Engineering Science, University of Southern Queens Land (2014) 10. Vieira, P., Romao, C., Marques, A.T., Esteves, J.L.: Mechanical characterisation of natural fibre reinforced plastics. In: II International Materials Symposium, 14–16 April 2003 (2003) 11. Gupta, M.K., Srivastava, R.K.: Tensile and flexural properties of sisal fibre reinforced epoxy composite: a comparison between unidirectional and mat form of fibres. Procedia Mater. Sci. 5, 2434–2439 (2014) 12. ASTM standard, D3039: Standard test method for tensile properties of polymer matrix composite materials. ASTM international, USA (2010)

40

T. Batu and H. G. Lemu

13. Corke, T., Nelson, R.: Wind Energy Design. CRC Press Taylor & Francis Group, Boca Raton (2018) 14. Airfoil Tools. https://airfoiltools.com/airfoil/details?airfoil=naca4415-il. Accessed 8 May 2020 15. Silva, L.J., Panzera, T.H., Christoforo, A.L., Durão, L.M., Lahr, F.A.R.: Numerical and experimental analyses of biocomposites reinforced with natural fibres. Int. J. Mater. Eng. 2 (4), 43–49 (2012) 16. Burhan, I., Kim, H.S.: S-N curve models for composite materials characterisation: an evaluative review. J. Compos. Sci. 2(3), 38 (2018) 17. Lachance-Barrett, S.: FLUENT - Wind Turbine Blade FSI (Part 1) - SimCafe – Dashboard, Confluence.cornell.edu (2016). https://confluence.cornell.edu/pages/viewpage.action? pageId=262012971. Accessed 20 May 2020

Numerical Simulation of Separation Mechanism in V-Shaped Outlet Hydrocyclone for Coalbed Gas Fracturing Flow-Back Fluids Xiaolong Xiao1,2, Mingxiu Yao1,2, Chao Xie3, Zhen Wu3, Yaoyao Wei3, Zhenjiang Zhao3, Huajian Wang3, Youle Liu3, and Bing Liu3(&) 1

Shandong Provincial Key Laboratory of Oilfield Produced Water Treatment and Environmental Pollution Control, Dongying 257026, China 2 Petroleum Engineering Co. Ltd., SINOPEC, Dongying 257026, China 3 College of Mechanical and Electronic Engineering, Shandong University of Science and Technology, Qingdao 266590, China [email protected]

Abstract. In order to reduce the fine particles in the underflow when the hydrocyclone is used to treat coalbed gas fracturing Flow-back fluids, a Vshaped outlet hydrocyclone was designed. Internal flow field of two hydrocyclones was simulated by RSM (Reynolds stress model) and Mixture model. The study shows that the V-shape outlet hydrocyclone can reduce concentration of the underflow particles with a diameter of 2.5 lm to 4.1%, which is 5.29 times lower than that of the ordinary hydrocyclone, and the particle concentration of 10 lm is reduced by 55.17%. In the internal flow field, the maximum pressure value of the V-shaped outlet hydrocyclone is 59.61% lower than that of the ordinary one. Compared with the ordinary hydrocyclone, the maximum negative pressure of V-shaped outlet hydrocyclone is reduced by a factor of 2.1. The maximum value of the tangential velocity in the semi-freedom vortex zone at the V-shaped outlet is 28.32% lower than that of the ordinary structure; the maximum of the axial velocity in the forced vortex zone is 40.03% higher than that of the ordinary one; the radial velocity of the V-shaped structure is reduced by 1.5 times as compared with the ordinary structure. Keywords: Hydrocyclone
Reduce underflow fine particles simulation
Fracturing flow-back fluids
Internal flow field


Fluent

1 Introduction In order to increase the production of coalbed methane wells, various measures need to be taken for production wells. Fracturing technology is one of the most important methods [1]. However, after the fracture treatment, a large number of counterdischarges are discharged to the ground, and improper handling will cause environmental pollution [2]. In fracturing Flow-back fluids, there are many kinds of organic matter, mud, coal powder and sand. After the chemical treatment, the impurities need to be removed from the fracturing Flow-back fluids. In China, the most important method © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2021 Y. Wang et al. (Eds.): IWAMA 2020, LNEE 737, pp. 41–49, 2021. https://doi.org/10.1007/978-981-33-6318-2_5

42

X. Xiao et al.

for removing impurities is gravitational settlement, but this method is time-consuming, labor-intensive and unsafe [3]. At present, the method for reducing underflow fine particles is to reduce the diameter of the outlet [4] but this method can easily cause plugging of the outlet. Decreasing the feed rate of the mixture at the inlet to reduce the centrifugal force in the internal flow field can reduce the fine particles in the underflow [5]. Increasing the diameter of the hydrocyclone can reduce the fine particles of the underflow effectively [6]. Increasing the size of the hydrocyclone can also reduce the fine particles in the underflow [7]. By reducing the height of the hydrocyclone, the residence time of the particles in the hydrocyclone becomes shorter, so that the purpose of reducing the fine particles can be achieved [8], but this method causes energy waste. In this work, after improvement of the straight tube + cone tube, a V-shaped outlet hydrocyclone was designed. The FLUENT 16.0 software was used for numerical simulation, by simulating of the flow field in the V-shaped outlet hydrocyclone and ordinary hydrocyclone, the pressure field and velocity field were obtained, and the simulation results were analyzed, which lays the foundation for hydrocyclone design.

2 Model Establishment and Verification A reasonable geometric model was established by using the hydrodynamic mass, energy and momentum conservation equations. The continuity equation is expressed as follows: @q @qu @qv @qw þ þ þ ¼0 @t @x @y @z

ð1Þ

Where, q is the density of fluid; t is time; u, v, w represent velocity components in three directions of x, y, z. This article uses incompressible fluids with constant viscosity, so the N-S (NavierStokes) equation is expressed as follows: q

 2  dux @P @ ux @ 2 uy @ 2 uz þl ¼qFx  þ þ @x dt @x2 @y2 @z2

ð2Þ

q

 2  duy @P @ uy @ 2 uy @ 2 uy þl ¼qFy  þ þ @y dt @x2 @y2 @z2

ð3Þ

 2  duz @P @ uz @ 2 uz @ 2 uz þl ¼qFz  q þ þ @z dt @x2 @y2 @z2

ð4Þ

Where, qFx, qFy, qFz are the components of the mass force acting on the unit area in the corresponding direction, l is the dynamic viscosity, and P is the compressive stress acting on the fluid.

Numerical Simulation of Separation Mechanism in V-Shaped Outlet Hydrocyclone

43

The energy conservation equation is expressed as follows: @ ðqT Þ @ ðquT Þ @ ðqvT Þ @ ðqwT Þ þ þ þ @t @x @y @z       @ k @T @ k @T @ k @T ¼ þ þ þ ST @x cp @x @y cp @y @z cp @z

ð5Þ

Where, Cp is the specific heat capacity; T is the thermodynamic temperature; k is the fluid heat transfer coefficient; ST is the viscous dissipation term. The hydrocyclone uses a velocity inlet. The treatment quantity Q is 2.6 m3
h−1, the main phase of the inlet is water; quartz sand volume fraction is 10%. In this study, the component of fracturing flow-back fluids was analyzed. The particle diameter of 2.5 lm, 10 lm, 35 lm, 45 lm were selected, the density of the particles are 2650 kg/m3.

(a) Ordinary type

(b) V type

(c) Ordinary type

(d) V type

Fig. 1. Structure and meshes of hydrocyclone

The difference of the V-shaped outlet hydrocyclone and the ordinary hydrocyclone is outlet structures, the other structures and dimensions are the same. The structure shown in Fig. 1, column diameter D is 50 mm, column height H1 is 160 mm, the hydraulic diameter of overflow pipe Do is 17 mm, insertion depth of overflow pipe H3 is 90 mm, the hydraulic diameter of the inlet Di is 12 mm, cone height H2 is 200 mm, and the hydraulic diameter of the outlet is 6.6 mm. 1) Mesh independence detection In order to avoid the calculation error caused by the problem of grid accuracy, this paper carries out 6 grid divisions on the same model and simulates the six models with different grid numbers. The results of the division are shown in Fig. 1(c, d).

44

X. Xiao et al.

It can be seen from the Fig. 2(a), that when the number of grids reaches 100000, with the increase of the number of grids, the values of simulated static pressure and axial velocity tend to be stable. In order to save the simulation time and the consumption of computer resources, this paper selects the grid number of 110000 to simulate.

(a) Number of grid

(b) Dimensionless diameter

Fig. 2. Simulation reliability verification

2) Experimental verification To verify the reliability of CFD software, the PIV test experiment was conducted in the internal flow field of the hydrocyclone. Pick a line at the same axial position to compare the axial velocity. Figure 2(b) compares the axial velocity distribution at Z = 200 of the hydrocyclone. The axial velocity discontinuity at the PIV experiment is due to the air column. The good agreement between the experimental measurement and CFD simulation demonstrates the validity of using the numerical method to study the likely effect of different parameters on the hydrocyclone hydrodynamics.

3 Results and Discussion In order to compare and analyze the flow field changes of the two hydrocyclones accurately, the same axial position line was taken during the study, the interception position shown in Fig. 1(a, b). As can be seen from Fig. 3(a, b), the distribution of the position of the column section and the cone section in the two hydrocyclones presents a similar pattern, which is basically the same as the pressure distribution in the combined vortex flow field. The strong swirling flow leads to higher pressure gradient along the radius direction. Negative pressure occurs in the air core region. Particles with different size and density will have different response to the flow velocity and pressure gradient, which achieves particles’ separation.

Numerical Simulation of Separation Mechanism in V-Shaped Outlet Hydrocyclone

(a) Ordinary type

Dimensionless diameter

45

(b) V type

Dimensionless diameter

(c) Z=352

(d) Z=360

Fig. 3. Pressure distribution

From Fig. 3(c, d), the pressure drop and pressure of the V-shaped outlet hydrocyclone in the outlet structure region are smaller than that of the ordinary hydrocyclone obviously. The maximum pressure difference is up to about 10000 Pa and the maximum pressure drop is around 15000 Pa, the air entering the V-shaped structure hydrocyclone is relatively small, reducing the diameter of the air column, making the actual participation in the separation area increase, improving the separation efficiency. We can also see that the lowest pressure point of the V-shaped outlet hydrocyclone is closer to the center of the axis. This is due to the small pressure drop that reduces the entry of air. In turn, the impact of air on the internal flow field is reduced, so the Vshaped outlet structure can play a role in stabilizing the flow field to some extent.

46

X. Xiao et al.

Dimensionless diameter (a) Z=352

Dimensionless diameter (b) Z=360

Fig. 4. Tangential velocity distribution

The tangential velocity is the precondition for the generation of centrifugal force. Figure 4(a, b) is the area of the outlet. In the semi-free vortex zone, the maximum of tangential velocity of the two hydrocyclones is almost the same, the tangential velocity of the V-shaped structure is slightly smaller than that of the ordinary hydrocyclone, and the maximum difference is not more than 0.2 m/s. The point where the velocity is 0 occurs due to the V-shaped structure. In the forced vortex zone, the tangential velocity of the V-shaped structure is obviously greater than that of the ordinary one, and the maximum difference is about 0.5 m/s, which makes the panning effect of the V-shaped outlet hydrocyclone at the underflow strengthen and more fine particles enter the forced vortex zone and flow out through the overflow port. Decreasing of the tangential velocity in the semi-freedom vortex reduces the wear of the outlet.

Dimensionless diameter (a) Z=352

Dimensionless diameter (b) Z=360

Fig. 5. Axial velocity distribution

From Fig. 5(a, b), it can be seen that near the outlet at the forced vortex zone, the axial velocity of the V-shaped structure is greater than that of the ordinary hydrocyclone significantly, and the maximum velocity difference is 1.2 m/s. The fine particles

Numerical Simulation of Separation Mechanism in V-Shaped Outlet Hydrocyclone

47

quickly become overflow products, which are discharged through the overflow port, and reduce the fineness in the underflow. In the semi-free vortex zone, the axial velocity of the V-shaped structure is also larger than the ordinary structure. When the maximum velocity difference reaches 2.1 m/s, the high-density large particles can be quickly discharged from the underflow port and can prevent the underflow port from clogging. By analyzing Fig. 6(a, b), In the area of outlet at the lower part of the cone, the radial velocity of the V-shaped outlet structure decreases significantly, especially in the forced vortex zone, and the maximum velocity difference can reach 2.5 m/s. This is due to obstruction of the flow field by the V-shaped structure.

Dimensionless diameter (a) Z=352

Dimensionless diameter (b) Z=360

Fig. 6. Radial velocity distribution

As the radial velocity slows down, the particle's internal migration velocity in the flow field slows down, and more fine particles become internal swirling flow, which is discharged through the overflow port. The content of fine particles in the underflow is reduced.

Table 1. Underflow concentration of each particle size Particle sized V(%) 2.5 4.1 10 11.2 35 44.8 45 39.9

Ordinary(%) 25.8 20.3 28.2 25.9

It can be seen from Table 1 that the particle diameter of the 2.5 lm particles in the underflow of the V-shaped outlet hydrocyclone is 5.29 times less than that of the ordinary hydrocyclone. For particles with a particle diameter of 10 lm, the flow rate of the V-shaped outlet hydrocyclone is 0.81 times lower than that of ordinary hydrocyclone (Fig. 7).

48

X. Xiao et al.

2.5μm

10μm 35μm 45μm (a) Ordinary type

2.5μm

10μm 35μm (b) V type

45μm

Fig. 7. Particle concentration distribution

4 Conclusion Seen from the velocity of each type, the maximum value of the tangential velocity in the semi-freedom vortex zone of the V-shaped structure is 28.32% less than that of the ordinary structure, which strengthens the function of internal elutriation to allow more fine particles to enter the internal rotational flow, and the particles will be discharged through the overflow. The axial maximum velocity in the forced vortex is 40.03% higher than that of ordinary hydrocyclone, which reduces the residence time of fine particles when they are at the underflow port; the V-shaped structure’s radial velocity is reduced by 1.5 times as compared to ordinary one, making the movement of the particles in the V-shaped outlet separator slower, so that the elutriation is more complete. By analyzing of the underflow particle concentration, in case the particles with the size of 2.5 lm in the underflow are reduced to 15.89% of the ordinary hydrocyclone and the particles with size of the 10 lm are reduced to 55.17% of the ordinary hydrocyclone, the fine particles in the underflow are reduced significantly. Acknowledgment. This study was supported by open fund of Shandong Provincial Key Laboratory of Oilfield Produced Water Treatment and Environmental Pollution Control.

References 1. He, M., Lai, X., Li, N., Xiao, Y., Shen, L., Liu, X.: Recovery and treatment of fracturing flowback fluids in the Sulige Gasfield, Ordos Basin. Nat. Gas Ind. B 2(5), 467 (2015) 2. Michalski, R., Ficek, A.: Environmental pollution by chemical substances used in the shale gas extraction—a review. Desalin. Water Treat. 57(3), 1336 (2016) 3. Liu, Y., Chang, Q., Cao, Y., Hao, Y.: Research progress on treatment of fracturing waste fluid. Petrochem. Ind. Appl. 32(9), 5 (2013)

Numerical Simulation of Separation Mechanism in V-Shaped Outlet Hydrocyclone

49

4. Lyu, X.-L.: A study of the effect of apex diameter of hydrocyclone and physical properties of feed material on motion of solid particles. Coal Prep. Technol. (5), 5 (2017) 5. Yang, B.W., Zheng, X.T., Zhou, H.H.W., Wang, H.Y., Huang, S.: Numerical simulation of liquid-solid separation in cyclone separator. Chem. Equip. Technol. (2), 16 (2017) 6. Liu, Y., Yang, Q., Qian, P., Wang, H.L.: Experimental study of circulation flow in a light dispersion hydrocyclone. Sep. Purif. Technol. 137(7), 66 (2014) 7. Acherman, S.R., Silva, L.B., Villanueva, L.A.: The hydrocyclone as an alternative for the partial recovering of filter aids in the sugar refining process. Dyna 77(164), 292 (2010) 8. Zeng, S.X.: The design and selection of liquid-solid cyclone separator. Chlor-Alkali Ind. (11), 23 (2001)

Numerical Simulation of Flow Field and Separation Performance of a Two-Cone Hydrocyclone for Oil-Water Separation Chao Xie1, Zhen Wu1, Zhenjiang Zhao1, Yaoyao Wei1, Jianliang Xue2, Peishan Huang3(&), and Bing Liu1 1 College of Mechanical and Electronic Engineering, Shandong University of Science and Technology, Qingdao 266590, China 2 College of Safety and Environmental Engineering, Shandong University of Science and Technology, Qingdao 266590, China 3 Qingdao Deepblue Subsea Engineering Technology Co., Ltd., Qingdao 266500, China [email protected]

Abstract. In order to study the internal flow field and oil-water separation mechanism of double cone hydrocyclone, based on fluid dynamics software Fluent, the effects of oil phase volume distribution, tangential velocity distribution, feed flow rate at different oil droplet sizes and fractional ratio at different oil concentration on the separation efficiency of the hydrocyclone were analyzed. The results show that the feed flow rate should be less than or equal to 3.5 m3/h to reduce the energy loss; The separation efficiency is the best when the oil droplet size is 1 mm–1.5 mm and the feed flow rate is in the range of 2.5 m3/h–3.5 m3/h; In order to achieve the highest separation efficiency, the split ratio should be 1.2–1.3 times of the inlet oil concentration. The research can provide reference for the separation mechanism and flow field characteristics of hydrocyclone. Keywords: Hydrocyclone
Oil-water separation Internal flow field
Separation efficiency


Multiphase flow
CFD


1 Introduction With the increase of the years of oilfield exploitation, the comprehensive water content of produced fluid is constantly rising, and the water content of some oil wells is even as high as 98%, leading to high production and treatment costs. Therefore, oil-water separation is the first problem to be solved [1]. Because there are many parameters affecting the separation efficiency of the hydrocyclone, and there are quite complex flow and separation processes, it is very important to clarify the influence of relevant parameters on the flow field and separation efficiency in the hydrocyclone. Jaseer E [2] studied the flow field in a hydrocyclone by changing the feed flow rate; Liu [3] studied and analyzed the influence of the split ratio on the oil-water separation of the cylindrical hydrocyclone, and found that the oil core rose with the increase of the overflow ratio; Zhang [4] studied the influence of different oil droplet sizes on oil-water © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2021 Y. Wang et al. (Eds.): IWAMA 2020, LNEE 737, pp. 50–56, 2021. https://doi.org/10.1007/978-981-33-6318-2_6

Numerical Simulation of Flow Field and Separation Performance

51

separation in a hydrocyclone, and found that the larger the particle size was, the easier the separation would be; Kharoua, N [5] inspected the separation efficiency of hydrocyclone by changing the oil concentration at the inlet. However, the influence of the change of a single parameter on the separation efficiency of hydrocyclone has been studied. Therefore, the study on the separation performance of hydrocyclone under multi-factor conditions has become an urgent problem to be solved. In order to improve the separation performance of hydrocyclone, this paper uses the Reynolds stress model and the mixture model, based on the previous single-factor simulation analysis of feed flow rate, oil droplet size, split ratio and oil concentration, the influence of feed flow rate at different oil droplet sizes and split ratio at different oil concentration on the separation efficiency of the hydrocyclone was analyzed comprehensively, and the optimal parameter range was determined. The research results further reveal the separation mechanism of hydrocyclone and provide reference for the optimal design of hydrocyclone.

2 Numerical Simulation 2.1

Geometric Model

At present, the oil-water hydrocyclone separation structure mostly adopts double-cone type. Due to its reliable operation and good separation effect, it has been widely used. The hydrocyclone designed in this paper conforms to Martin Thew standard [6], its nominal diameter D is 20 mm, and the specific structure size is shown in Fig. 1. In order to improve the simulation accuracy and make the grid boundary and the flow direction of the fluid as vertical or coincident as possible, the hexahedral grid is used in the model [7]. In order to better capture the boundary features, the wall surface is locally encrypted. The grid division is shown in Fig. 2.

Fig. 1. Structure of hydrocyclone

2.2

Boundary Conditions

The simulated medium is oil-water two-phase mixture, the water is treated as a continuous phase, which has a density of 998 kg/m3 and a viscosity coefficient of 0.001003 kg/(m
s); The discrete phase medium is oil with a density of 834 kg/m3 and viscosity coefficient of 0.008 kg/(m
s). The inlet velocity is used as the boundary condition, the out-flow condition is used at both underflow and overflow. PRESTO format is regarded as pressure interpolation, SIMPLEC algorithm was used for

52

C. Xie et al.

Fig. 2. Grid structure

pressure-velocity coupling, and QUICK format is considered as discrete convectiondiffusion equation [8]. The wall surface is non-leakage and has no slip conditions [9].

3 Simulation Results 3.1

Concentration of Oil Phase

When the oil-water mixture enters the hydrocyclone along the normal direction of the inlet, a high-speed rotating flow is generated. Because of different density, oil and water are separated by different centrifugal force in hydrocyclone [10]. Figure 3 shows the volume distribution of oil phase in axial section. Oil and water are separated under centrifugal force, and the oil phase forms an oil core in the center of the hydrocyclone. Its diameter is almost the same as that of the vortex finder, the oil is discharged from the vortex finder with the inner hydrocyclone, while the water flows to the bottom outlet with external flow field. On the cross sections with different axial position, along the radius, the oil phase volume concentration on the walls of both sides of the cylinder was almost zero along the radius. As it moved toward the center, the oil phase volume concentration gradually increased. With the downward flow of the hydrocyclone, the shape of the oil core became thinner, and it was mainly concentrated near the vortex finder and at the center of large and small cone segments.

Fig. 3. Oil-water concentration distribution map

3.2

Tangential Velocity Field

Figure 4 shows that the distribution of tangential velocity presents a typical Rankine vortex, and the tangential velocity increases with the increase of the feed flow rate. According to Bernoulli’s equation, this is caused by the increase of feed flow rate, which leads to the increase of inlet pressure, the increase of tangential velocity increases the ability of the oil phase to move towards the wall, thus increasing the

Numerical Simulation of Flow Field and Separation Performance

53

concentration of the oil phase. However, when the inlet velocity is greater than 3.5 m3/ h, the increase amplitude of tangential velocity decreases. Through the study of the internal flow field, it is found that the increase of inlet velocity increases the turbulence intensity, thus increasing the energy consumption of the internal flow field and reducing the conversion between pressure and kinetic energy. Therefore, in order to reduce energy loss, the feed flow rate should be less than or equal to 3.5 m3/h. 50

Flow rate (m3/h) 1.5 2.0 2.5 3.0 3.5 4.0

Tangential Speed (m/s)

40

30

20

10

0 -20

-10

0

10

20

Radius(mm)

Fig. 4. Tangential velocity at different feed flow rate

3.3

Separation Efficiency

The separation efficiency is the most important index to evaluate the separation performance of hydrocyclone [11], there are many factors affecting the separation efficiency, among which the feed flow rate, oil droplet size, oil concentration and split ratio are the important parameters affecting the separation efficiency. Therefore, it is very important to make clear the exact relationship between these parameters and the separation efficiency. (1) Relationship between feed flow rate and separation efficiency Under the condition of 9% inlet oil concentration and 10% split ratio, the relationship between flow rate and separation efficiency under different oil droplet sizes in oil-water two-phase flow field was simulated numerically. It can be seen from Fig. 5 that when the oil droplet size is less than or equal to 0.2, the separation efficiency increases gradually with the increase of the feed flow rate, and the separation efficiency increases significantly when the feed flow rate is 3.5 m3/h–4 m3/h; However, when the oil droplet diameter was greater than 0.2, the separation efficiency showed a trend of first increasing and then decreasing, the peak value is 3.5 m3/h. When the feed flow rate exceeds 3.5 m3/h, because the feed flow is too high, the rotation speed of the mixed

54

C. Xie et al. 100

Separation Efficiency(%)

90 80

Droplet diameter(mm)

70

0.1 0.2 1.0 1.5 2.0

60 50 40 30

1.5

2.0

2.5

3.0

3.5

4.0

3

Flow Rate (m /h)

Fig. 5. Flow-separation efficiency diagram for different particle sizes

liquid in the hydrocyclone is accelerated, so that the separated oil phase droplets are broken into small droplets which are difficult to be effectively separated under the action of excessive shear stress. The adverse effect caused by the fragmentation of the oil droplets exceeds the favorable condition brought by the centrifugal force difference brought by the greater tangential velocity, so that the separation efficiency is reduced. From the overall analysis, the separation efficiency showed a trend of first increasing and then decreasing with the increase of oil droplet size at the same feed flow rate, and the separation efficiency changed little when the particle size was 1 mm–1.5 mm. In conclusion, the separation effect is the best when the diameter of the oil droplet is 1 mm–1.5 mm and the feed flow rate is in the range of 2.5 m3/h–3.5 m3/h. (2) Relationship between split ratio and separation efficiency Under the condition that the feed flow rate is 3 m3/h and the droplet diameter is 1 mm, the relationship between the split ratio and the separation efficiency is shown in Fig. 6. When the oil concentration remains unchanged, the separation efficiency increases first and then decreases with the increase of the fractional flow ratio. The reason for this phenomenon is that the increase of the fractional flow ratio leads to the increase of the overflow flow rate and the probability of the oil phase flowing out of the vortex finder increases; However, too large split ratio leads to too small bottom flow and too large overflow flow, part of the water phase is re-involved in the internal flow field and discharged by the vortex finder, which further affects the stability of the oil core and leads to the reduction of separation efficiency. For the hydrocyclone with fixed structure, there is an optimal split ratio range to achieve the optimal separation efficiency [12]. When the oil concentration is 3% and 5%, the optimal split ratio range is 3%–4% and 6%–8% respectively; When the oil concentration is 7% and 9%, the optimal split ratio range is 8%–9% and 10%–12% respectively. Based on the above

Numerical Simulation of Flow Field and Separation Performance

55

data, the separation efficiency is the highest when the split ratio is usually 1.2–1.3 times of the inlet oil concentration.

100

Separation Efficiency (%)

80

60

40

Oil concentration 3% 5% 7% 9%

20

0

0

2

4

6

8

10

12

14

Split Ratio(%)

Fig. 6. Distributary ratio - efficiency curve at different oil concentration

4 Conclusion Computational fluid dynamics software was used to study the internal flow field and separation efficiency of oil-water separation hydrocyclone, and the following conclusions were drawn within the research scope: (1) In the process of oil-water separation, the oil phase is mainly concentrated near the vortex finder and the central region of the cone section; As the feed flow rate increases, the turbulence intensity increases and the energy required to form a stable flow field increases, so that the tangential velocity increases to the maximum at 3.5 m3/h. (2) When the oil droplet diameter is less than 0.2, the separation efficiency is positively correlated with the feed flow rate; When the oil droplet size is greater than or equal to 0.2, the separation efficiency first increases and then decreases due to the increase of overflow flow rate and the decrease of bottom flow rate, and the separation effect is the best when the feed flow rate ranges from 2.5 m3/h–3.5 m3/h. (3) When the oil volume fraction is constant, the separation efficiency increases at first and then decreases with the increase of split ratio; When the oil concentration is 3% and 5%, the optimal split ratios ranged from 3% to 4% and 6% to 8%, respectively; When the oil concentration is 7% and 9%, the optimal split ratios ranged from 8% to 9% and 10% to 12%, respectively. In conclusion, the separation efficiency is the highest when the split ratio is usually 1.2–1.3 times of the inlet oil concentration.

56

C. Xie et al.

Acknowledgment. The work described in this article has been conducted as part of the National Natural Science Foundation of China (Grant No. 51408347).

References 1. Cai, X.L.: Study on theory and engineering application of Compact Flotation Unit. Beijing University of Technology, pp. 1–2 (2017) 2. Hamza, J.E., Al-Kayiem, H.H., Lemma, T.A.: Experimental investigation of the separation performance of oil/water mixture by compact conical axial hydrocyclone. Therm. Sci. Eng. Progress 17, 100358 (2020) 3. Liu, H.F., Zhong, X.F., Xu, J.S.: Numerical study of the characteristic oil-water separation in cylindrical hydrocyclone. Shipbuilding China 050(A11), 369–374 (2009) 4. Zhang, J.S., Feng, S.C.: The study on flow mechanism and application of hydrocyclone. J. Filtr. Sep. (03), 17–20 (2001) 5. Kharoua, N., Khezzar, L., Nemouchi, Z.: Computational fluid dynamics study of the parameters affecting oil–water hydrocyclone performance. Proc. Inst. Mech. Eng. Part E J. Process Mech. Eng. 224(2), 119–128 (2010) 6. Colman, D.A., Thew, M.T.: Correlation of separation results from light dispersion hydrocyclone. Chem. Eng. Res. Des. 61(4), 233–240 (1983) 7. Li, F., Xiong, F., Liu, C.Y.: Effect of oil droplet coalescence and breakup behavior on separation performance of hydrocyclone. China Petrol. Mach. 047(006), 73–78 (2019) 8. Shang, Q., Zhang, J., Zhou, L.X.: Application of the QUICK scheme to the simulation of swirling turbulent flow. J. Comput. Phys. 04, 4–10 (2004) 9. Hwang, K.J., Hwang, Y.W., Yoshida, H.: Design of novel hydrocyclone for improving fine particle separation using computational fluid dynamic. Chem. Eng. Sci. 85(1), 62–68 (2013) 10. Ma, Y., Yang, Y., Li, B.S.: Numerical simulation of flow field and study on experimental equipment in hydrocyclone for oil-water separation. Environ. Eng. (2017) 11. Liu, Y., Wang, Z.B.: Research progress in the factors influence the efficiency of hydrocyclone separation. Fluid Mach. 44524(02), 57–60 (2016) 12. Shu, C.H., Chen, W.M., Xiao, X.C.: Discussion on the relationship between the flow ratio and the basic performance of the deoiling cyclone. Fluid Mach. 05, 12–15 (2001)

Parameter Optimization of Hydrocyclone at the Inlet of Spiral Tube for Offshore Oil Water Separation Zhen Wu1, Yaoyao Wei1, Chao Xie1, Zhenjiang Zhao1, Wenyu Yang2(&), Peishan Huang3, and Bing Liu1 1 College of Mechanical and Electronic Engineering, Shandong University of Science and Technology, Qingdao 266590, China 2 College of Safety and Environmental Engineering, Shandong University of Science and Technology, Qingdao 266590, China [email protected] 3 Qingdao Deepblue Subsea Engineering Technology Co., Ltd., Qingdao 266500, China

Abstract. Existing design equations can hardly provide reliable guidance in hydrocyclone designs and give explanations for the selected optimum range of spiral tube inlet. In this paper, the spiral tube inlet hydrocyclone for offshore oilwater separation was optimized by using the response surface method and multiobjective optimization method. With optimization results which were verified by numerical simulation and experiments discussed and analyzed, the optimal entrance structural parameters were obtained: d = 7.45 mm; D = 32.50 mm; h = 7.80 mm; C = 2.00, and the separation efficiency of the optimized hydrocyclone was increased by 8.34% and Euler number reduces 4.12. The optimization process show that the response surface method is reliable for the optimization of the inlet structure of the spiral tube, which provides technical guidance for the development and optimization of the sea oil produced water treatment equipment. Keywords: Oil-water separation
Hydrocyclone
Numerical simulation Spiral tube inlet
Parameter optimization
Response surface method




1 Introduction Oil production fluid treatment offshore has special requirements for sea conditions, production equipment and other facilities [1]. Compact oil-water separators are essential for treatment of the production streams of oil fields at sea [2]. As the representative of the compact oil-water separator, hydrocyclone has become the main treatment device of oil produced water on offshore [3, 4]. Generally, modifying the inlet section size is the simplest way to improve the performance of hydrocyclone [5–7]. A new inlet hydrocyclone designed to reduce erosion-induced wear in mineral dewatering processes [8], which was proved to greatly improve the performance of the hydrocyclone. Noroozi [9] changed the shape of the inlet chamber of the hydrocyclone to increase its separation efficiency by 8%, while © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2021 Y. Wang et al. (Eds.): IWAMA 2020, LNEE 737, pp. 57–64, 2021. https://doi.org/10.1007/978-981-33-6318-2_7

58

Z. Wu et al.

replaced the ordinary inlet passage with involute double inlet passage, the separation performance had been improved [10]. However, those researches only focus on different inlet structures on the performance of hydrocyclone, less on the optimization of hydrocyclone inlet structure parameters to improve the separation performance. In this paper, response surface method and multi-objective optimization method are used to optimize structural parameters of the spiral tube inlet of hydrocyclone, the influence of structural parameters on separation performance is discussed, and the optimal entrance structural parameters are obtained. With numerical simulation and experiment carried, the optimal entrance structure parameters which are determined by the optimization method are verified, which has a certain guiding significance for the design and optimization of the hydrocyclone for oil-water separation offshore.

2 Numerical Simulation and Experiment 2.1

Structure

The geometry of the hydrocyclone for oil-water separation studied in this work is shown in Fig. 1, whose inlet combines a spiral tube and an additional guide tube. The main structural parameters of the spiral tube inlet including: pipe diameter d; nominal diameter D; pitch h; number of turns C.

Fig. 1. Structure of oil-water hydrocyclone at the inlet of spiral pipe

The main structural dimensions of hydrocyclone for oil-water separation used for simulation are shown in Fig. 2, the geometric proportions of the hydrocyclone used in this work satisfy Martin Thew’s criteria [11].

Parameter Optimization of Hydrocyclone at the Inlet of Spiral Tube

59

Fig. 2. Structural computational

2.2

Simulation Conditions

Research on Oil-water mixture of a certain oil field in Bohai, China, the density of oil in wastewater was 880 kg/m3, the viscosity of oil was 11.2 mPa
s, and the viscosity of water was 1.003 mPa
s. The primary phase is water and the second phase is oil. The volume fraction of oil 2%, and the initial velocity of oil and water is set at 10 m/s. The Mixture model and the Reynolds stress model are chosen in this simulation, and the particle diameter of oil is set to 10 lm. A “velocity inlet” boundary condition is used at the inlet, and the “outflow” condition at both overflow and underflow, a “no-slip” boundary condition is set at the wall. 2.3

Experiments for Simulation Results

Fig. 3. Oil-water separation experiment flow

The experimental device as shown in the Fig. 3 is designed to verify the reliability of the numerical simulation. Five-hole probe is used to measure the flow field and pressure distribution in the hydrocyclone. Sampling points A, B and C are set at the inlet, overflow and underflow of the hydrocyclone to calculate the separation efficiency of the hydrocyclone.

60

Z. Wu et al. 9

0.14

8

0.12

7

Pressure(MPa)

0.10

Velocity (m/s)

6

0.08

5 4

0.06

3

0.04

2

0

-20

-16

-12

-8

-4

Position

0

4

8

12

16

Simulation data Experimental data

0.02

Simulation data Experimental data

1

0 20 22

-20

-15

-10

(a) Tangential velocity distribution

-5

0

5

10

15

20 22

Position(r/mm)

r/mm

(b) Static pressure intensity distribution

Fig. 4. Comparison of experimental results and numerical simulation results

It can be seen from the picture (a) and (b) in Fig. 4 that within the premise of excluding the experimental error and the simulation error, the simulation results are consistent with the experimental results, indicating that the numerical simulation method can better analyse the internal flow field, with high reliability. Therefore, it is reliable to use numerical simulation method to complete the subsequent optimization design experiment.

3 Optimization and Analysis 3.1

Response Surface Method

Response surface method is a kind of parameter optimization method combining mathematical method and statistical analysis, which was first proposed in 1951 by Box and Wilson [12]. The procedure of Response Surface Method is to firstly estimate the corresponding coefficients according to the least square method and get the initial response surface equation then analyse the influence of each factor on the response surface according to the significance test [13]. In this paper, the second-order response design is selected and the second-order model is fitted: y ¼ b0 þ

k X i¼1

bi xi þ

k X i¼1

bii x2i þ

k X

bij xi xj þ e

ð1Þ

i\j

Here, y is the approximate function of unknown function; xi stands for independent variable; bi, bii and bij are the mean of Regression coefficients representing primary, secondary and interaction, respectively; k represents number of influencing factors; e is the meaning of error, which including experimental error and experimental fitting error. In this paper, optimization design the inlet structure of spiral tube of hydrocyclone is carried by center combination design method [13]. Considering that the oil-water hydrocyclone with a certain inlet flow, keep its structural parameters unchanged, select d, D, h and C as input variable, and select Euler number Eu and separation efficiency η

Parameter Optimization of Hydrocyclone at the Inlet of Spiral Tube

61

as output variable. Comprehensively analyse the influence relationship of four input variables on output variables, and take the best value obtained by single variable method as the canter point, and determine the upper and lower limits of each factor level as shown in Table 1. Table 1. Design factor level Design factor The lower limit The upper limit Center point Pipe diameter d/mm 6.3 7.7 7.0 Nominal diameter D/mm 27.0 33.0 30.0 Pitch h/mm 7.2 8.8 8.0 Number of turns C 2.0 4.0 3.0

With the level of each factor was input into the software of design-export according to Table 1, the experimental scheme was formed and a series of numerical simulations are carried out. By calculating different groups of output variables, the response surface equation of Eu and g were fitted as follows: Eu ¼ 30:8182  5:4464d  1:0929D  3:0908h  6:1231C þ 0:1013Dd þ 0:4071dh þ 0:7947dC þ 0:1176Dh þ 0:1334DC þ 0:1013hC þ 0:0081d 2 þ 0:0012D2

ð2Þ

þ 0:0937h2 þ 0:4914C2 g ¼ 0:7923 þ 0:0089d  0:0035D  0:0042h  0:0003C þ 0:0003Dd þ 0:001dh þ 0:0021dC þ 0:0001Dh þ 0:00015DC þ 0:0:0015hC þ 0:0001d 2

ð3Þ

þ 0:000017D2 þ 0:00012h2 þ 0:00067C 2

3.2

Analysis of Optimization

By using Eq. 2, kept the two variables at it’s centre level and used the software of design-export, the Euler number response surface are as shown in Fig. 5. With the increase of d and D, the Euler number decreases significantly obtained by picture (a) Fig. 5, this means that when the flow rate is fixed, the velocity of the fluid decreases, the intensity of vortex flow decreases, the internal energy consumption of the fluid decreases, resulting in the decrease of the pressure drop and the Euler number. The influence of C and h on Euler number is that the bigger value of C and h, the bigger the Euler number, which is drawn from picture (b), (c) and (d) Fig. 5. The reason for this pattern is that increasing value of C and h, increasing the flow path equivalently, makes the energy consumption of internal flow field increase, pressure drop increases,

62

Z. Wu et al.

(a) h=8 mm C=3

(c) D=30 mm h=8 mm

(b) D=30 mm C=3

(d) d=7 mm D=30 mm

Fig. 5. The influence of different parameters on Euler number

and the Euler number will increase as well. Beside this, d and the D have the greatest influence on Euler number, the second is h, and the least is C in global perspective. 3.3

Optimizal Results

Combined with the comprehensive qualitative analysis of the two output factors, the multi-objective optimization is carried out by using MATLAB software, and the optimal value of the structural parameters for the spiral inlet obtained is that: d = 7.45 mm; D = 32.50 mm; h = 7.80 mm; C = 2.00. The spiral tube structure of hydrocyclone is remodelled according to the optimized size, and the optimization results are verified by numerical simulation. The comparison between the optimized value of Euler number and separation efficiency and the simulation value is shown in Table 2. It can be seen from Table 2 that the Euler number error between the optimized value and the simulation value is 3.63%, and separation efficiency error is 4.96%. Errors of them are within 10%, which shows that the response surface method can better optimize the structural parameters of hydrocyclone.

Parameter Optimization of Hydrocyclone at the Inlet of Spiral Tube

63

Table 2. Comparison of performance for hydrocyclones before and after optimization Parameter Optimization value Before After 12.35 8.23 Eu η 83.37% 94.36%

Simulation value Before After 15.28 8.54 81.34% 89.68%

Experiment value Before After 16.56 8.16 78.67% 90.82%

4 Conclusion The response surface method and multi-objective optimization are used to optimize the structure of the spiral tube inlet of the hydrocyclone, and the conclusions are as follows: (1) Pipe diameter d and nominal diameter D are the main factors that affect the separation performance of the hydrocyclone for oil-water separation. Specifically, when the value of d and D increase in the range of factor level, the fluid speed decreases, while the intensity of vortex flow increases, resulting in the increase of flow pressure drop and centrifugal force, Euler number and separation efficiency will increase then. This regular pattern provides a basis to search for best optimal structural parameters for the spiral tube inlet. (2) Other factors such as the number of turns C and pitch h have little effect on the separation performance of hydrocyclone. When the value of C and h increase in it’s factor level, the flow path increases equivalently, makes the energy consumption of internal flow field increase, pressure drop increases, and the Euler number will increase as well as separation efficiency. But the influence of C and h for the separation performance is not as strong as that of d and D, which can give the guidance to search for best optimal structural parameters for the spiral tube inlet. (3) Compromised between high separation efficiency and low pressure drop, response surface method and multi-objective optimization method are used to optimize the value of d, D, h and C, optimal structural parameters are that: d = 7.45 mm; D = 32.50 mm; h = 7.80 mm; C = 2.00. After remodelling the spiral tube inlet of hydrocyclone with the optimization results, the simulation results show that the separation efficiency is 8.34% higher, Euler number reduces 4.12 and the separation performance is much better than before. It illustrated that the optimization of inlet spiral tube of hydrocyclone had certain reliability, which has certain guiding significance for the design and optimization of hydrocyclone for oil water separation offshore. Acknowledgment. This study was financially supported by the National Natural Science Foundation of China (Grant No. 51804183).

64

Z. Wu et al.

References 1. Das, T., Preben, F.T., et al.: Modelling and optimization of compact subsea liquid-liquid separation system. In: 26th European Symposium on Computer Aided Process Engineering, pp. 1255–1260 (2016) 2. Das, T., Heggheim, S.J., et al.: Optimal operation of a subsea separation system including a coalescence based gravity separator model and a produced water treatment section. Ind. Eng. Chem. Res. 58, 4168–4185 (2019) 3. Bing, L., Huajian, W., et al.: Study of GRL and inlet velocity on hydrocyclone for fracturing flow-back fluids. Math. Probl. Eng. 12, 1–2 (2019) 4. Changjun, L., Qian, H.: Analysis of droplet behavior in a de-oiling hydrocyclone. J. Dispersion Sci. Technol. 317–327 (2016) 5. Nenu, R., Yoshida, H.: Comparison of separation performance between single and two inlets hydrocyclones. Adv. Powder Technol. 20(2), 195–202 (2009) 6. Bing, L., Luncao, L., Huajian, W., et al.: Numerical simulation and experimental study on internal and external characteristics of novel Hydrocyclones. Heat Mass Transf. (17) (2020) 7. Peikun, L., Huajian, W., et al.: Effect of gas–liquid ratio on the performance of hydrocyclones for desanding flowback fracturing fluids. Nat. Gas. Ind. 39(11), 44–54 (2019) 8. Xu, P., Wu, Z., Mujumdar, A.S., et al.: Innovative hydrocyclone inlet designs to reduce erosion-induced wear in mineral dewatering processes. Drying Technol. 27(2), 201–211 (2009) 9. Noroozi, S., Hashemabadi, S.: CFD analysis of inlet chamber body profile effects on deoiling hydrocyclone efficiency. Chem. Eng. Res. Des. 89(7), 968–977 (2011) 10. Qiaoduo, Y.: Numerical simulation of oil-water separation and droplet breakage in Hydrocyclone. Huazhong University of science and technology, Wuhan, pp. 26–50 (2014) 11. Abdul, M.: Theoretical and numerical study of swirling flow separation devices for oil-water mixtures. Michigan State University, Michigan State, pp. 24–30 (2015) 12 Box, G., Wilson, K.: On the experimental attainment of optimum conditions. J. R. Stat. Soc. Ser. B 13, 1–45 (1951) 13. Tang, B., Xu, Y., et al.: Numerical study on the relationship between high sharpness and configurations of the vortex finder of a hydrocyclone by central composite design. Chem. Eng. J. 278, 504–516 (2015)

Sit-to-Stand Intention Recognition Dayou Li1,2(&), Hang Lu1, Renxi Qiu1, Carsten Maple3, and Zuobin Wang1,3 1

School of Computer Science and Technology, University of Bedfordshire, Luton LU1 3JU, UK [email protected] 2 Centre for Nano Metrology and Manufacturing Technologies, Changchun University of Science and Technology, Changchun 130022, China 3 Advanced Manufacturing Technique Centre, Warwick University, Coventry CV4 7AL, UK

Abstract. Sit-to-stand (STS) difficulties are common among elderly because of the decline of their cognitive capabilities and motor functions. The way to help is to encourage them to practice their own functions and to assist only at the point where they need during STS processes. The provision of such support requires the elderly’s intention of standing up to be recognised and the amount of support as well as the moment when the support would be needed to be predicted. The research presented in this paper focuses on intention recognition as it is difficult due to uncertainties existing in STS processes and differences in individual’s biomechanical features. This paper presents fuzzy logic based selfadaptive approach to the recognition of standing up intention from sensor signals that contain the uncertainties. Keywords: Uncertainty handling
Intention recognition
Prediction
Robot
Fuzzy logic
Neural networks
STS

1 Introduction The aging generation is more likely to be confronted with difficulties in sit-to-stand (STS) movement, due to the decline of their cognitive capabilities and motor functions. Various STS assistive devices become market available over the decades. However, these devices often operate like elevators without virtual interaction with the user. The users, on the other hand, can become overly dependent on the lifting function of the chairs in STS, which can causes the further decline of their cognitive capabilities and motor functions [1, 2]. Encouraging the elderly to use their strength in STS and providing support only when it becomes necessary would help them to maintain their cognitive capabilities and motor functions. It is desirable for the elderly to have such assistance provision built in with assistive devices. The recognition of standing up intention of the users is essential to the implementation of such assistive devices. The followings need to be taken into account in the intention recognition: 1) Safety is primarily critical to the intention recognition. Actuating supporting mechanism with mis-interpreting the users’ behaviours could drive the elderly users into panic situations and potentially cause injuries, 2) Robustness is another concern as humans do not © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2021 Y. Wang et al. (Eds.): IWAMA 2020, LNEE 737, pp. 65–72, 2021. https://doi.org/10.1007/978-981-33-6318-2_8

66

D. Li et al.

normally follow the same pattern when performing STS, and 3) Timing also plays a critical role. Intention recognition has to be completed within a limited time scale. STS process commonly takes from 1.91 s to 2.30 s for completely able-bodied healthy adults [3–5]. Although STS duration is proved to be longer for able-bodied elderly users, some will not resist more than 3.00 s until he/she sits back to seated position. This paper presents a self-adaptive intention recognition approach (SAIR) which is a part of the development of an STS-assistive robot chair. The chair consists of a seat that can rise up and decline, a pressure sensor array embedded in a footmat and a controller to automate the seat movements. Safety-critical, robust and timely intention recognition from sensory data collected from the footmat encounters uncertainties in the sense of blurry boundary between intended movements the users performed when they intended to stand up, known as STS intended movements, and other unintended random movements called non-STS intended movements, because of the limited capacity of sensing systems and the difference in individuals’ biomechanical features. SAIR tackles the uncertainties using fuzzy sets to model the blurry boundary and performing fuzzy reasoning to classify the STS intended and non-STS intended movements. SAIR also allows the fuzzy inference system to be self-adaptive to different users’ biomechanical features with a neural-fuzzy network that is able to identify the difference.

2 Problem Statement An STS process consists of 6 distinctive stages through the measurement on ground reaction force (GRF). They are: 1) Initiation refers to the period from subject being informed to initiate the STS, 2) Counter is mainly caused by early lifting of the thighs from the seat by contracting the hip flexor muscles while upper torso mostly remains in its original position, 3) Seat-off is signified by the moment where subject’s buttocks separate from seat, 4) Peak is the stage where the captured GRF reaches its maximum, 5) Rebound is represented by GRF observed after the peak stage due to the tendency of both feet leaving the ground, and 6) Standing. These are shown in Fig. 1. Referring to this human biomechanical model for STS movements, GRF cpuld be useful for STS intention recognition. However, GRF can also be generated when the users perform movements for the purposes other than standing up, for example, changing sit postures. Figure 2 illustrates the similarity of the STS intended movements and a random movement at the duration of the first 1.10 s that covers the early events of STS, namely, initiation, counter and seat-off. The study of the Centre of Pressure (COP) [6, 7] discovered COP shifting patterns during STS. COP transparently shifts backward after the initiation stage and then forward for the sake of the dominating upward movement afterwards. Before the completion of STS, COP again shifts slightly backward. A smooth transition of COP in longitudinal directions is supposed to be observed from successful STS movements [3], given the leaning forward motion of upper torso after the users applying force to ground mainly vertically to generate upwards motion and the upper torso straightening in the first four STS stages. However, clear differences between STS intended movements and non-STS intended movements still cannot be seen from both the magnitude

Sit-to-Stand Intention Recognition

67

of change in COP and the longitudinal shift (LOS). The possible reason for this is that both STS intended movements and changing sit posture involve torso swinging forward and backward with small magnitudes. Lateral COP shift (LAS) is more complex than magnitude of change in COP and LOS. The comparison of LAS of one non-STS movement with that of STS intended movements shows more “violent” LAS of the former than the later. Although this can be confirmed in general when looking at more datasets, there are still some cases where STS intended have larger LAS and non-STS have smaller LAS, causing blue boundary between the two in terms of LAS.

Fig. 1. Distinctive events in STS through normalised GRF [5]

successful STS (No.2) unintended (No.12) unsuccessful STS (No.26)

GRF/ Body weight

1

0.8

0.6

0.4

0.2

0

0

1

2

3

4

5 6 Time elapsed

7

8

9

10

11

Fig. 2. GRF of STS intended movements and other movements over the first 1.10 s period

Different users in the same household can have different biomechanical features such as body weights, heights (measured with shoes) and thigh lengths. Because of health conditions, the same user’s biomechanical features and sitting habits can change along with time. The differences and the changes introduce further uncertainties into intention recognition. For example, users with different thigh lengths can have different buttock movements when changing sitting posture, resulting in wither squeezed or

68

D. Li et al.

enlarged LAS. LAS will also change when the users rest his buttock on different areas of the seat or sits with different postures.

3 Max-Difference Based STS Intention Recognition Max-difference based intention recognition (MDLS) takes into account of the difference between the maximum and the minimum change in LS of COP recorded in the first 1.1 s of each sample movement. As seen from the previously extracted LS of COP from STS process, a substantial increase of magnitude could be spotted at the beginning of movements, followed by a gentle rebound at the following 3 time intervals. When approaching the last few tenths of seconds, typically from 0.7 to 1.1 s, the magnitude tends to decrease and stabilise. On the other hand, non-STS intended movements shows little alignment with certain phenomenon. Figure 3 shows the distribution of the maximal LS of COP (MDLS) calculated from COP of every samples of STS intended and non-STS Intended movements. The majority of STS intended movements show MDLS ranging from 0 to 1. On the contrary, non-STS intended movements mainly have MDLS greater than 1. Such separation suggests two fuzzy sets A and B as shown in Fig. 4. Fuzzy rules and their truth values are given as follows: Rule 1: If MDLS = “A” then “Unintended”, with a truth degree of 0.9 (9/10). Rule 2: If MDLS = “B” then “Intended”, with a truth degree of 0.95 (19/20). Rule 3: If MDLS = “A” then “Intended”, with a truth degree of 0.1 (1/10). Rule 4: If MDLS = “B” then “Unintended”, with a truth degree of 0.05 (1/20). The followings are the criteria for evaluating intention recognition approaches: 1) False matching rate of non-STS intended movements to STS intended (FMR) is the measure for the safety criterion. The rate should be minimised as the false matching will cause the assist actions of the robot chair to take place, which may force the user who has no intention to stand up and lead to a fall. 2) True matching rate of STS intended to STS intended (TMR) is the measure for the accuracy. This should be maximised as the aim of the classification is to find out the user’s need for assistance in STS process. 3) Variety in sample data (VSD) in conjunction with FMR and TMR are used to measure the robustness. High variety means high uncertainty level in sample data. Together with FMR and TMR, VSD reflects the capabilities of handling the uncertainty. 4) Duration of intention recognition (DIR) reflects the time criterion. Intention recognition must complete by the end of first 1.1 s period. Confusion matrices corresponding to tests with 2-fold testing data and with pure testing data are given in Fig. 5. The top-right corner of each confusion matrix shows the FMRs,which were recorded as 0%, for both 2-fold data and pure testing data. The bottom-right corner of the above confusion matrices shows TMR. TMRs with 2-fold and pure testing data were recorded as 70% and 65%, respectively. The reason this poor performance is that TMR takes the registered maximum and minimum over the period of first 1.1 s, which is sensitive to the uncertainties such as randomness in reposition of arms while standing up and difficulties when stabilising body motion in a significant torque. All of these change the captured LS in a rather short time period.

Sit-to-Stand Intention Recognition

Fig. 3. Distribution of MDLS of the respecting non-STS and STS intended movements

0.8

1.4

Fig. 4. Fuzzy sets of A and B with respect to MDLS distribution

(a) with 2-fold testing data,

(b) with pure testing data

Fig. 5. Confusion matrices of MDLS intention recognition

69

70

D. Li et al.

VSD is the measure of robustness of the intention recognition approaches. LS of COP is related to the sitting postures such as positions of feet and arms and buttock position on seat. These positions determine what strategy the user adopts to stabilise his/her body when standing up. Humans tend to adapt their STS strategies to environments and conditions during STS. According to [8], ankle positioning and the users’ functional capability are correlated to the strategies used for postural control and upper torso stabilisation, proved by the extraction of joint torques. The users appear to use ankle strategy when upper torso movements are restricted, which will affect the captured LS appearing on footmat. In contrast, when STS is performed based on a rather wide stance the stabilisation motion will require less muscle activation hence results in different shift of COP. When generating data for designing and testing the intention recognition approaches, the users were advised to avoid retaining the same foot, arm or buttock positions. It can be summarised that VSD is high in the datasets as movements were executed naturally to introduce realistic uncertainties. DIR is considered through two aspects, namely, input data feeding time and processing time of fuzzy inference. The duration of feeding data into the fuzzy logic based intention recognition is limited to 1.10 s. In addition to this, the average processing times of fuzzy inference was recorded as 0.001 s. With the processing time of MDLS classifications rather negligible comparing to the 1.10 s of input data capturing.

4 Self-adaptive Intention Recognition 4.1

Bio-Features and Sitting Posture Acquisition

The magnitude of LAS recorded with different users varies, partially because of the difference in their biomechanical features, such as body weight, height, and thigh length. However, those features are difficult to acquire and to classify given the robot chair settings. The width of shoes reflecting the size of shoes and has a correlation with the biomechanical features suggested by [9–11]. The analysis on shoe geometry is essentially an affecting factor for the lateral balancing during STS process [12]. It offers an opportunity for the robot chair acquire the biomechanical features of the users in an indirectly manner. The width of shoes can be measured through the numbers of pressure sensors activated by users’ feet while stepping on the footmat. Figure 6, demonstrates the difference in number of activated cells as well as GRF patterns (cell-wise) generated by two users, namely A and B, who are very different in the sense of biomechanical features as mentioned early. Taking into account that the number of rows and the total number of cells activated by the same subject are rather fixed, the shoe widths ws of individuals can be normalised and converted to a computational feature in the way such No: of FSRs as ws ¼ No: of Rows. However, due to ununiformed geometry of shoe soles and the limited number of FSRs adopted on footmat, it would be unrealistic to acquire an exact match between the numbers of activated presser sensors and the widths of shoes. Therefore, Ws has to be fuzzified.

Sit-to-Stand Intention Recognition

71

3D mesh plot of FSR activation grid of Subject B

3D mesh plot of FSR activation grid of Subject A

5

5 4.5

4

GRF (Unit: kg)

GRF (Unit: kg)

4 3.5 3 2.5 2 1.5

3

2

1

1 0.5

0 7

0 7

6

6

8

5

5

7

Longitudinal direction

4

6

4

5

3

3

4 3

2 1

Longitudinal direction

Lateral direction

2 1

(a)

2 1

1

2

3

5

4

6

7

8

Lateral direction

(b)

Fig. 6. 3D-mesh plot of activated FSRs with height being cell-wise GRF values generated by user A (a) and user B (b)

4.2

Self-adaptive Neural Fuzzy Classification

Introducing fuzzified shoe width, ws, that implies individual user’s biomechanical features into the previously developed MDLS intention recognition makes it selfadaptable to different users. To realise the adaptive MDLS, neural-fuzzy networks with as an additional input were implemented, as illustrated in Fig. 7. Input Layer

Rule Layer

Output Layer

Input 1: Accumulated change in LAT Or Max-difference in LAT

2 fuzzy sets of Input 1 (FuzzificaƟon process)

IntenƟon classificaƟon output Unintended Movements

Intended STS Movements

Input 2: Normalised shoe width

Weights (Truth degrees)

5 fuzzy sets of Input 2 (FuzzificaƟon process)

Logical OperaƟons: “And”

Fig. 7. Neural-fuzzy networks implementing adaptive MDLS

It can be seen that the self-adaptive MDLS generates slightly less TMR (90% against 95%) but significant less FMR (0% against 15%), as shown in Fig. 8. The latter was reduced in a good manner because it relates to the safety concerns of the robot chair and was discussed as the primary measure of criteria. Classification Unintended

Results Intended

Unintended Actual

True Positive

False Negative

Datasets

Intended False Positive

True Negative

Fig. 8. Confusion matrix of self-adaptive MDLS tested with two subjects

72

D. Li et al.

5 Conclusions STS intention recognition is difficult given uncertainties existing in STS processes and limited signals that can be collected in real-world domestic environment. The presented fuzzy logic based classifiers address the problem in the sense of successful differentiation of STS intentions (both able to stand up and unable to do so) and non-STS intentions. The neural-fuzzy network enables the classifiers to be adaptable to individual users whose biomechanical features are different from one another. The approaches to STS intention recognition represented is a step forward to robotic chair developments. It will facilitate robots’ decision-making on whether assistance actions needed to STS processes. The classifiers employ pressure data generated by a footmat where the users rest their feet when sitting in a chair, which can be available in smart home environments where smart floor is in place.

References 1. Galumbeck, M.H., Buschbacher, R.M., Wilder, R.P., Winters, K.L., Hudson, M.A., Edlich, R.: The sit & standTM chair. A revolutionary advance in adaptive seating systems. J. LongTerm Effects Med. Implants 14(6), 535–544 (2004) 2. Lu, H., Li, D., Oyekan, J., Maple, C.: Integrated sensing techniques for assistive chairs: a survey towards sit-to-stand problems. Intell. Mechatron. Rob. 3(4), 58–70 (2013) 3. Kerr, K.M., White, J.A., Barr, D.A., Mollan, R.A.B.: Analysis of the sit-stand-sit movement cycle in normal subjects. Clin. Biomech. 12(4), 236–245 (1997) 4. Schlicht, J., Camaione, D.N., Owen, S.V.: Effect of intense strength training on standing balance, walking speed, and sit-to-stand performance in older adults. J. Gerontol. Ser. A Biol. Sci. Med. Sci. 56(5), 281–286 (2001) 5. Etnyre, B., Thomas, D.Q.: Event standardization of sit-to-stand movements. Phys. Ther. 87 (12), 1651–1666 (2007) 6. Hughes, M.A., Schenkman, M.L.: Chair rise strategy in the functionally impaired elderly. J. Rehabil. Res. Dev. 33(4), 409 (1996) 7. Bae, J.H., Moon, I.: Design of electric assist-standing chair for persons with disability design of electric chair to assist person with disability in stand up and sitting down. In: 2010 International Conference onControl Automation and Systems (ICCAS), pp. 574–575 (2010) 8. Kuo, A.D., Zajac, F.E.: Human standing posture: multi-joint movement strategies based on biomechanical constraints. Prog. Brain Res. 97, 349–358 (1992) 9. Robbins, L.M.: Estimating height and weight from size of footprints. J. Forensic Sci. 31(1), 143–152 (1986) 10. Siminoski, K., Bain, J.: The relationships among height, penile length, and foot size. Ann. Sex Res. 6(3), 231–235 (1993) 11. Ozden, H., Balci, Y., Demirüstü, C., Turgut, A., Ertugrul, M.: Stature and sex estimate using foot and shoe dimensions. Forensic Sci. Int. 147(2), 181–184 (2005) 12. Lord, S.R., Sherrington, C., Menz, H.B., Close, J.C.: Falls in Older People: Risk Factors and Strategies for Prevention. Cambridge University Press, Cambridge (2007)

A Method of Bearing Fault Diagnosis Based on Transfer Learning Without Parameter Yang Ge1,2,3(&), Jiancong Qin3, and Jianxin Ding3 1

School of Mechanical and Electrical Engineering, Soochow University, Suzhou, China [email protected] 2 Jiangsu Key Laboratory for Elevator Intelligent Safety, Changshu Institute of Technology, Changshu, Jiangsu 215500, PR China 3 Southeast Elevator Co. LTD, Changzhou, China

Abstract. In this paper, a bearing fault diagnosis method based on transfer learning is proposed to solve the problem that the traditional fault diagnosis method is not satisfactory under multi-working conditions. First, the Transfer Component Analysis method is employed to transform the source domain and the target domain into the same space. Then annotation probability matrix is proposed for fault diagnosis. Finally, the proposed method is verified on the bearing data set of CWRU university, and the recognition accuracy is obviously higher than the traditional methods. It is worth noting that the proposed method does not need parameters tuning and is very simple. Keywords: Fault diagnosis
Transfer learning
Annotation probability matrix

1 Introduction Rolling bearing is one of the key parts in many large rotating electromechanical machineries. It is particularly helpful to prevent the occurrence of equipment accidents that to propose a reasonable fault diagnosis for bearing. The working condition is often changed in the actual work of bearings. Even with the same type of bearing, the same fault may behave differently under different working conditions. Moreover, it is impossible to obtain fault samples for all types of bearings. In many cases, we may have only a few bearing failure sample data. Therefore, transfer learning method has likewise become a hot research topic in bearing fault diagnosis, which transfers the existing calibration data to the target test samples for fault diagnosis. In recent years, the transfer learning method has been introduced into various fields from the earliest application in image recognition. Because of the wide application of transfer learning, we only summarize some transfer learning methods used in fault diagnosis, which mainly includes the following aspects. (1) Instance transfer. It is assumed that some parts of the data in the source domain can be reused by reweighted learning in the target domain. In [1], a fault diagnosis method based on variational mode decomposition (VMD) multi-scale permutation entropy (MPE) and feature-based transfer learning (FTL) was proposed, which has higher accuracy in multi-state classification of rolling bearings under variable working © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2021 Y. Wang et al. (Eds.): IWAMA 2020, LNEE 737, pp. 73–81, 2021. https://doi.org/10.1007/978-981-33-6318-2_9

74

Y. Ge et al.

conditions. In [2], a fault diagnosis model of distribution transformer transfer learning considering multi-factor state evolution was proposed. In [3], the mass source data with different distribution is added to the target data as training data. (2) Mapping transfer. Instances in the source and target domains are mapped to new data spaces to reduce differences between the source and target domains. In [4], a multi-core balanced distribution adaptation method was proposed. In [5], a depth transfer learning method for fault diagnosis of rolling bearings under variable operating conditions was proposed. An improved joint distribution adaptive (IJDA) transfer learning method was proposed in [6]. (3) Network transfer. In [7], convolutional neural network (CNN) and multi-layer perceptron (MLP) are both used for fault diagnosis. Some on-line fault detection methods based on deep transfer learning were proposed based on the network model of image data, such as VGG-16 [8], AlexNet [9], ShuffleNet V2 [10]. In [11], a new semisupervised deep learning framework was proposed, which integrates manifold regularization for parameter optimization, and the automatic encoder with softmax regression model is used to encode label information. The major contributions of this paper can be summarized as follows: 1) Annotation probability matrix is proposed for fault diagnosis, the probability attribute of this method coincides with the idea of transfer learning. 2) Compared with other traditional method, the proposed method has a higher accuracy of fault identification in the case of multi working condition. 3) The proposed method is a very simple transfer learning method, which does not need parameter tuning and is more suitable for engineering applications.

2 Basic Theory 2.1

Transfer Component Analysis

Due to the manufacturing process, working conditions and other reasons, even the same type of bearings may have different features under the same fault mode. In fault pattern recognition, sometimes it is difficult to find the training samples which are consistent with the features of the test samples, and sometimes there is no training samples of the same type of bearing. In this case, the transfer learning method is a good choice, which converts the features of the training sample and the test sample to the same subspace, as shown in Fig. 1. Because of different working conditions, the edge distribution of source domain and target domain is often different, i.e. PðXs Þ 6¼ PðXt Þ. In this case, the recognition results of using the traditional machine learning classification method directly may not 0 0 be ideal. Assume / is a feature mapping, and /ðXs Þ ¼ Xs , /ðXt Þ ¼ Xt , which makes
0
0 P Xs  P Xt . With the mapping, the difference between the source and target domain edge distributions can be reduced. Transfer component analysis (TCA) is a mapping method based on the idea. It uses the maximum mean difference (MMD) as the metric, and the MMD of the source domain and the target domain after mapping can be calculated by the following formula.

A Method of Bearing Fault Diagnosis Based on Transfer Learning

  0 0   1 Xns  1 Xnt  D Xs ; Xt ¼  /ðxsi Þ  /ðxti Þ  i¼1 i¼1 ns nt

75

ð1Þ

Source Target

Fig. 1. Spatial mapping

In order to facilitate the solution of the above equation, the author of TCA transforms it into the following form: trðKLÞ  k trðK Þ

ð2Þ

Where, K is the kernel matrix,  K¼

Ks;s Kt;s

Ks;t Kt;t



2 Rðns þ nt Þðns þ nt Þ

ð3Þ

L is the distribution difference matrix, and the specific calculation is as follows:

Lij ¼

8 >
:  1 ; otherwise ns nt

ð4Þ

~ 2 Rðns þ nt Þm , a new kernel matrix can Introduce a low rank matrix W ¼ K1=2 W be got:    ~ ¼ KK 1=2 W ~ 1=2 K ¼ KWW T K ~ WK K

ð5Þ

At this point, the final optimization goal of TCA is:

 min tr W T KLKW þ ltr W T W

ð6Þ

s.t:W T KHKW ¼ Im

ð7Þ

W

Where H is a central matrix, H ¼ Ins þ nt  1=ðns þ nt Þ11T . Where 1 2 Rðns þ nt Þ is a column vectors whose entries are all one, and Im is the identity matrix. Solve the first m eigenvalues of ðKLK þ lI Þ1 KHK, that is W.

76

Y. Ge et al.

Let

X ¼ XsT ; XtT

ð8Þ

The above formula denotes that the two matrices XsT and XtT are spliced horizontally. Z ¼ WT X

ð9Þ

X s ¼ ð Z ð  ; 1 : ns Þ Þ T

ð10Þ

0

Where Z ð  ; 1 : ns Þ means to collect 1 to ns columns of matrix Z. Then 0

Xt ¼ ðZ ð  ; ns þ 1 : end ÞÞT 0

ð11Þ

0

After obtaining Xs and Xt , traditional machine learning methods can be used for recognition. 2.2

Classifier Based on Annotation Probability Matrix

Annotation probability matrix (APM) is introduced to build a nonparametric classifier. Note that the classification label set of the sample is Y 2 f1;


; Lg. Assume that there is an APM as shown in Fig. 2. There are 4 types of samples in the matrix, where the probability of X1t belonging to the four classes is 0.1, 0.1, 0.6 and 0.2 respectively. The maximum probability is 0.6, which means that X1t is the most likely to belong to class 3. Similarly, we can get the classification of target samples such as X2t and X3t . The purpose of this paper is to get this probability matrix to classify the target samples.

Fig. 2. Annotation probability matrix

It should be noted that the target domain and the source domain must have the same label space (i.e. Ys ¼ Yt ), when using this probability matrix to classify the samples in the target domain. If pij denotes the probability of Xit belonging to class j of Xs , the sum of the probabilities that Xit belongs to all categories of Xs should be 1, i.e.

A Method of Bearing Fault Diagnosis Based on Transfer Learning

XL j

77

ð12Þ

pij ¼ 1; 8i 2 f1;


; nt g

In fact, it is certain which category the sample belongs to. That is to say, the value of pij is only 0 or 1. When pij ¼ 1, Xit belongs to class j, otherwise, pij ¼ 0 means that Xit don’t belongs to class j. Since each label has at least one sample of the target domain that matches it, we can also get Xnt i

pij  1; 8j 2 f1;


; Lg

ð13Þ

Let cj be the center of class j sample in the source domain, which can be obtained by the following formula. cj ¼

1 Xns Nsj

i



Xis yis ¼ j

ð14Þ

Where Nsj denotes the number of class j samples in the source domain, ½½
 is the indicator function, the value is 1 when the condition is true, otherwise, the value is 0. Let dij be the distance from the target sample to the center of the j-th class samples in the source domain. We calculate it by Euclidean distance as the follow formula.     dij ¼ Xit  c2j  ð15Þ In conclusion, we can build the following cost function:

s:t:

P P min g ¼ ni t Lj dij pij 8 0 pij 1 > > > L > < P p ¼ 1; 8i 2 f1;


; n g ij

t

ð16Þ

j > nt > P > > : pij  1; 8j 2 f1;


; Lg i

The probability matrix can be obtained by formula (16), and the classification label of sample Xit can be obtained by the following formula. pim yit ¼ arg max PL m j pij

ð17Þ

Now that the classification is completed, from the whole classification process, it can be seen that there are no parameters need to be tuned. Some traditional classification methods, such as neural networks, support vector machines and deep learning methods, need to set many parameters, such as the number of layers, learning rate and so on. These parameters often have a great impact on the classification result. In some case, it is difficult to select an appropriate parameter.

78

Y. Ge et al.

3 Experiment and Result Analysis The experimental data is from the bearing data center of Case Western Reserve University. We select the vibration data from the drive end, whose sampling frequency is 12k. Four working conditions are selected to verify the presented method above. As the original data is long, for all failure modes, the original data is segmented by every 1000 consecutive points in the experiment. Each section is set as a new sample, and the specific working condition attributes are shown in Table 1. There are four failure modes in condition A, B and C, i.e. normal, inner ring failure, rolling element failure and outer ring failure. And there is no outer ring failure mode in condition D. The failure mode labels are shown in Table 2. Table 1. Bearing vibration acceleration data Working conditions A B C D

Motor speed (rpm) 1797 1772 1750 1730

Motor load (HP) 0 1 2 3

Fault diameter (inches) 0.007 0.014 0.021 0.028

Samples of each conditions 975 974 974 967

Table 2. The labels of each failure mode Label 1 2 3 4

3.1

Failure mode Normal Inner ring failure Rolling element failure Outer ring failure

Feature Extraction

Twenty-four features are extracted from each sample, including 8 time domain features, 8 frequency domain features and 8 wavelet decomposition energy features [8]. The features of time domain and frequency domain are shown in Table 3. Table 3. The time and frequency domain features extracted from the original signal Time domain features qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi P P F1 ¼ N1 Ni¼1 xi F2 ¼ N1 Ni¼1 x2i P P pffiffiffiffiffiffi 2 F4 ¼ N1 Ni¼1 jxi j F3 ¼ N1 Ni¼1 jxi j F5 ¼ N1

3 i¼1 xi

maxð xÞ

F7 ¼ 1 PN N

qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi P

PN

i¼1

jxi j

F6 ¼

N

1 N

x2 i¼1 i

F4

Frequency domain features P 2 P
P F9 ¼ N1 Ni¼1 Si F10 ¼ N1 Nj¼1 Sj  N1 Ni¼1 Si PN PN
1 PN 3 1 fi Si Sj N Si N j¼1 F11 ¼ Pi¼1 N
pffiffiffiffiffiffi3 i¼1 F ¼ 12 S j¼1 j F10 sP qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ffi ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi PN N 2 2S 1 f F ¼ s ð f  F Þ i i 14 i i 12 i¼1 i¼1 N PN F13 ¼ j¼1

Sj

sP ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi PN 4 PN 2 N x f s f 4 S F ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi i¼1 i i¼1 i i  F8 ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi i¼1 i i 4 16 P PN PN N F15 ¼ PN N 2 1 N

1 N

x2 j¼1 j

f j¼1 i

Sj

s j¼1 j

f S k¼1 k¼1 k

A Method of Bearing Fault Diagnosis Based on Transfer Learning

79

Where Si is the Fourier transform amplitude of xi , and fi is the frequency. 3.2

Experimental Setup

In order to verify the effectiveness of the methods proposed in this paper, we choose some common classification methods for comparison, including transfer methods and non-transfer methods. Including support vector machine (SVM) [1], Balanced Distribution Adaptation (BDA) [5], Joint Distribution Adaptation (JDA) [6], Subspace Alignment (SA) [6], Correlation Alignment (CORAL) [6], Manifold Embedded Distribution Adaptation (MEDA) [7], Geodesic Flow Kernel (GFK) [8], Domainadversarial Neural Networks (DANN) [9], and Conditional Adversarial Adaptation Networks (CDAN) [10]. It should be noticed that two approaches are adopted when using the APM method. The first approach is to input the 24 extracted features into APM for classification without mapping. The second approach to use TCA to reduce the 24 features into 16 mapping features and then input them into APM for classification. The settings of the source domain and target domain of the experiment are divided into four forms: single source to single target, single source to multi-target, multi-source to single target and multi-source to multi-target, as shown in Table 4 and Table 5. 3.3

Experimental Results and Analysis

In order to distinguish the recognition effect of different methods, this paper intentionally increases the difficulty, and the motor speed, motor load and fault diameter are different under each working condition. Table 4. Accuracy of single source domain to single target domain (%) ID

(S)

(T)

SVM

SASVM

TCASVM

BDA

JDA

CORAL

MEDA

GFK

DANN

CDAN

1

A

B

29

50

50.5

27.25

25

6.75

25

37

71.25

73.75

2

B

A

30.25

49.75

77.75

25

25

24.75

59.5

33.75

70.5

74.25

3

B

C

49.25

40

49.25

25

25

24.25

25

24.25

72.25

78

4

C

B

40

50

41.75

26.25

27

24.25

25.25

30

59.75

5

A

C

27.25

77

83.75

88.75

90.25

25.25

50.25

90.25

89.75

6

C

A

27

75

96.75

98

99

25.5

99

81

AVG

33.79

56.96

66.63

48.38

48.54

21.79

46.75

Avg rank

12

6

4

10

9

13

11

APM

SAAPM

TCAAPM

41.75

55

74.25

25.25

52

74.5

25.25

45.25

79.25

61.5

34

52

62.25

91

88.75

80

91.25

97.5

98.75

99

77.25

99.25

49.38

76.83

79.54

52.33

60.25

80.13

8

3

2

7

5

1

Where TCA-APM means to use TCA method to reduce the dimension of features and then use APM method to classify, and the same to SA-SVM, TCA-SVM and SAAPM. As can be seen from Table 4, compared with SVM, APM performs well in both cases with and without mapping, and the classification effect is obviously better than SVM. Compared with other transfer methods, APM also has a very good effect.

80

Y. Ge et al. Table 5. Accuracy of mixed domains (%)

ID

(S)

(T)

SVM BDA JDA

CORAL MEDA GFK

DANN CDAN APM TCAAPM

1 2 3 4 5 6 AVG

B AB A AD CD BCD -

AD CD BCD B AB A -

44.14 42 25.45 36 38.5 49.5 39.27

28.71 28.57 19 25.25 25.12 25.25 25.32

69 67.25 80.25 49.75 68.25 99 72.25

0.57 51.57 64.91 29.75 40.25 45.75 38.80

71.71 50.29 53 28.25 47 74.75 54.17

57.14 43 72.09 25.5 54 100 58.62

37.29 46.86 73.09 35.5 51.5 86 55.04

70.5 65.75 84 49.75 72 99.25 73.54

42.57 60.43 74 37.75 59.5 80 59.04

73.25 69.75 85 50 71.25 99.5 74.79

Table 5 shows the experimental results in the mixed domains. There are three mixed domains situations: single source domain to multi-target domains, multi-source domains to multi-target domains and multi-source domains to single-target domain. It can be seen that APM still performs well in the mixed domains situations. Table 6. Comparison of different methods Method CORAL Avg rank 13 Number of parameter 2 Time (s) 128.5

GFK DANN CDAN 8 3 2 3 1 1 135.2 >1000 >1000

JDA 9 5 110

APM 7 0 25.3

TCA-APM 1 3 80.4

The average rank, parameter numbers and running time of several methods are shown in Table 6. Note that except APM, other methods all need parameter tuning, the three parameters of the first ranked TCA-APM are all from TCA. In terms of running time, APM consumes the shortest time among the seven methods. Some insight can be found. 1) Compared to deep transfer learning methods (DANN and CDAN), TCA-APM achieves almost the same recognition accuracy. Although the method needs the mapped features by TCA as input, the running time is much less than the deep methods. 2) In practical applications, it is often difficult to find appropriate tuning parameters, so APM without parameters tuning may be a good choice. 3) The experiments imply that the two-stage method (TCA-APM) is batter than some onestage deep transfer methods. Of course, each method has its advantages and disadvantages. 4) It can be seen that the experimental recognition accuracy is not high, which may be related to feature selection and parameter tuning. In this paper, we didn’t specifically optimize for feature selection and parameter tuning.

4 Conclusion and Future Work A fault identification method based on transfer learning is presented in this paper. We proposed APM as the transfer learning approach which is easy, efficient, and accurate. In particular, APM does not need to tune any parameters explicitly. Experiment

A Method of Bearing Fault Diagnosis Based on Transfer Learning

81

demonstrate the superiority of APM in accuracy and efficiency over several popular methods. In collocation with some mapping methods, such as TCA, APM can achieve better results. In this paper, there is no special optimization in fault feature extraction, nor deep network structure, which will be our future research work. Beyond that, the proposed method only focuses on how to transfer. But in practice, what to transfer is also important. We will explore this by extending APM in the future work.

References 1. Ren, H., Liu, W.Y., Shan, M.C., Wang, X.: A new wind turbine health condition monitoring method based on VMD-MPE and feature-based transfer learning. Measurement 148, 8 (2019) 2. Yang, Z.C., Shen, Y., Zhou, R.F., Yang, F., Wan, Z.L., Zhou, Z.Q.: A transfer learning fault diagnosis model of distribution transformer considering multi-factor situation evolution. IEEJ Trans. Electr. Electron. Eng. 15, 30–39 (2020) 3. Xiao, D.Y., Huang, Y.X., Qin, C.J., Liu, Z.Y., Li, Y.M., Liu, C.L.: Transfer learning with convolutional neural networks for small sample size problem in machinery fault diagnosis. Proc. Inst. Mech. Eng. Part C J Eng. Mech. Eng. Sci. 233(14), 5131–5143 (2019) 4. Cao, N., Jiang, Z.N., Gao, J.J., Cui, B.: Bearing state recognition method based on transfer learning under different working conditions. Sensors 20(1), 12 (2020) 5. Che, C.C., Wang, H.W., Fu, Q., Ni, X.M.: Deep transfer learning for rolling bearing fault diagnosis under variable operating conditions. Adv. Mech. Eng. 11(12), 11 (2019) 6. Qian, W.W., Li, S.M., Yi, P.X., Zhang, K.C.: A novel transfer learning method for robust fault diagnosis of rotating machines under variable working conditions. Measurement 138, 514–525 (2019) 7. Li, X.D., Hu, Y., Li, M.T., Zheng, J.H.: Fault diagnostics between different type of components: a transfer learning approach. Appl. Soft. Comput. 86, 11 (2020) 8. Mao, W.T., Ding, L., Tian, S.Y., Liang, X.H.: Online detection for bearing incipient fault based on deep transfer learning. Measurement 152, 9 (2020) 9. Ma, P., Zhang, H.L., Fan, W.H., Wang, C., Wen, G.R., Zhang, X.N.: A novel bearing fault diagnosis method based on 2D image representation and transfer learning-convolutional neural network. Meas. Sci. Technol. 30(5), 16 (2019) 10. Liu, H.C., Yao, D.C., Yang, J.W., Li, X.: Lightweight convolutional neural network and its application in rolling bearing fault diagnosis under variable working conditions. Sensors 19 (22), 20 (2019)

Optimal Design of Control System for Ground Source Heat Pump Unit Xiaomei Jiang1(&), Michael Namokel2, Chaobin Hu1, and Ran Tian1 1

Jiangsu Key Laboratory of Elevator Intelligent Safety, School of Mechanical Engineering, Changshu Institute of Technology, Changshu 215500, Jiangsu, China [email protected] 2 Nordhessen University of Applied Sciences, Kassel, Germany [email protected]

Abstract. To realize the efficient and energy-saving operation of ground-source heat pump for refrigeration and heat supply, it is necessary to consider the realtime change of meteorological parameters and terminal load. The design of the electronic control system adopts PWM rectification and frequency conversion control technology to improve the system energy efficiency and reduce the interference of the power supply network, at the same time the cascade control technique is adopted to incorporate the operating conditions, load rate, energy efficiency and other parameters into the control system, each control loop circulating fluid is kept in the reasonable range of the flow rate, this method not only improves the overall efficiency of the system, and solves the problem of the deteriorating efficiency of ground source heat exchanger year by year due to the unequal total heat amount and the total discharge amount. The practical application results show that the optimization design has higher system energy efficiency level, and some advantages in energy saving, economic efficiency and environmental benefits. Keywords: Ground source heat pump technology PWM rectifier


Cascade control
DSP


1 Introduction Ground source heat pump (GSHP) central air conditioning system is a renewable energy technology, which takes the earth as the cold and heat source and uses the shallow layer ground cold and heat resources on the earth surface for energy conversion. In winter, the low-level heat energy in the earth is increased by heat pump to heat the building; in summer, the heat in the building is transferred to the underground by heat pump to cool the building, so as to realize the constant temperature regulation of the building space [1]. Ground source heat pump (GSHP) energy is derived from low enthalpy heat energy of earth surface shallow layer soil. Its temperature is relatively stable throughout the year, and it is not limited by region and resources. As a cold and heat source, its control system adopts closed circulation [2, 3]. The underground heat exchanger is buried underground and makes full use of green space, road, parking lot and other places. © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2021 Y. Wang et al. (Eds.): IWAMA 2020, LNEE 737, pp. 82–89, 2021. https://doi.org/10.1007/978-981-33-6318-2_10

Optimal Design of Control System for Ground Source Heat Pump Unit

83

2 Ground Source Heat Pump Detection and Control System The main factor affecting the current use of ground-source heat pump is that the efficiency of the system decreases with the increase of the service life of the system. The key reason lies in the single control of the control system, which causes the imbalance of land heat use and inevitably poor energy storage. Because the ground and underground heat exchangers undergo unstable heat transfer after heat exchange, after the system runs for a certain number of years, the initial temperature of the soil becomes overheated or undercooled because of its heat transfer, and the direct consequence is the instability of the operation of the ground-source heat pump system, which ultimately leads to the reduction of the system efficiency and even not working properly [4, 5]. If the load difference between winter and summer is large, the system efficiency will also be reduced, resulting in the waste. It is very important to investigate the engineering site condition and carry out the data acquisition in the early stage of the design of the ground source heat pump system, including the thermal property of soil, the thermal property of backfill material, the number of pipe wells, the distribution form of pipe well, the form of buried pipe, etc. On the basis of the analysis and research, the ground heat exchanger model of the ground source heat pump control system is established. The heat exchanger model should fully consider the attenuation characteristics of soil heat storage with the use of years, and adopt a reasonable control system to realize the optimal control of the ground source heat pump system [6]. The control system consists of geothermal circulating water pump loop, indoor circulating water pump loop and compressor refrigerant circulating loop. Geothermal circulating pump loop and indoor circulating pump loop respectively provide the power needed for their water circulation, while compressor refrigerant circulating loop provides the main motor refrigeration power. Figure 1 is the general diagram of the heat pump control system.

Fig. 1. General diagram of heat pump control system.

84

X. Jiang et al.

3 Mathematical Analysis In order to design conveniently, it is assumed that the working performance of the underground heat exchanger and the indoor loop heat exchanger in the geothermal loop are basically the same, and the circulating water pump power of the geothermal loop and the indoor loop of the system are respectively WP1 and WP2 , The compressor (main motor power) is assumed WCP , the inlet and outlet temperature of each buried pipe of the circulating water in the geothermal loop are respectively tGI and tGO , tRI and tRO are the inlet and outlet temperature of each loop heat exchanger in the room. According to the principle of energy balance, the model between the energy efficiency ratio and its relationship of the system is derived as follows [7]: Cp qGM ðtGI  tGO Þ þ Wp1 þ Wp2 þ Wcp ¼ Cp qRM ðtRI  tRO Þ

ð1Þ

Cp qRM ðtRI  tRO Þ ¼ COP Cp qGM ðtGI  tGO Þ

ð2Þ

In formula: CP : constant pressure specific heat of water. qGM : mass flow of circulating water in geothermal loop. qRM : mass flow of circulating water in indoor loop. COP: System Energy Efficiency Ratio: The equipment parameters of the heat pump system are determined according to the model, and the system design is optimized. As can be seen from the formula 1, the actual load of the buried heat pump system is not only related to the water supply flow, but also the influence of temperature difference between supply and return water on the system cannot be neglected [8]. In order to ensure that the water flow of the water system is changed when the indoor load is changed, the energy delivered to the load is more reasonable. We need not only to change water flow to track the change of system load, but also stable temperature difference between supply and return water to ensure the reasonable water flow meet the demand of the system load change and reduce the power consumption of the system to the maximum extent. When the indoor load decreases, the water flow of the water system decreases accordingly, this will reduce the energy conveyed by the system to the load, so as to meet the requirements of load reduction [9]. Because the reduction of water flow can greatly reduce the energy consumption of the water system, the impact of the ground source system on soil is minimized, and the service period and maintenance cost of the ground source system are greatly prolonged. Therefore, the design of the system has remarkable economic and social effects.

4 Optimum Design of Control System In traditional central air conditioning systems, electric valves are often used to control the flow, resulting in a large amount of energy consumption in throttling devices. Ground source heat pump system exchanges heat with soil through buried pipe soil

Optimal Design of Control System for Ground Source Heat Pump Unit

85

heat exchanger system. If the heat use is not balanced, it will inevitably lead to the deterioration of soil thermal storage. In recent years, the technology of frequency conversion is becoming more and more mature, which makes the application of frequency conversion technology to the ground source system become a trend. The optimization control system of ground source heat pump fully considers the heat and cold balance of ground source heat pump system, reasonably distributes the heat absorbed in winter and released in summer by ground source heat pump and maintains the heat balance of soil.

Fig. 2. Heat pump system control main circuit

In this design, the frequency inversion control technology is used for the circulating pump in the ground source heat pump system, and the speed of the circulating pump is intelligently controlled by DSP, so as to reduce the influence of the circulating water quantity at the ground source side on the heat transfer efficiency, on the energy consumption of the heat pump unit and the energy consumption of the ground source side water pump itself, and determine the water flow rate of the circulating water at the ground source side, so as to further achieve the purpose of energy saving of the ground source heat pump system. Figure 2 is the main circuit diagram of heat pump system control.

86

4.1

X. Jiang et al.

Key Point Parameter Design

In order to further improve the efficiency of heat pump control system and reduce the impact of its control system on the power grid, the main control circuit of this design abandons the traditional phase-controlled rectifier and uses PWM rectifier which is becoming more and more mature because of the drastic reduction in cost. In the control system, the sampling and calculation of input voltage, output voltage and three-phase current, 3/2 rotation conversion, PI regulation, 2/3 reverse rotation conversion, sector judgment, control algorithm and output of PWM signal are realized by TMS320F240. The program is programmed by special assembly language of TMS320F240 DSP, and the communication interface between DSP and PC is designed by VB software. Figure 3 is the structure diagram of PWM rectifier voltage and current double closed-loop control system. The design of three-phase VSR network inductance will affect the three-phase VSR system comprehensively. The value of network inductance will not only affect the dynamic and static performance of the system, but also have a great impact on other factors such as the rated output power of three-phase VSR [10, 11]. Therefore, the design of inductance on the network side is very important. Therefore, two key technical indicators of rated power and input current pulsation of three-phase VSR are emphasized to design inductance on the network side. When both fast current tracking and harmonic current suppression are needed, the inductance range of three-phase VSR is determined. pffiffiffi 2Ua þ 23 UO L fia max sinðxtÞ f Dia max 2 3 UO

ð3Þ

In formula: ia max —fundamental wave peak of power supply current. Ua —effective value of a phase voltage. Uo —DC side voltage. x —angular speed of three-phase input power supply at grid side. Dia max —20% ia max . Figure 3 is an analog current input conversion circuit and Fig. 4 is an analog voltage input conversion circuit.

Fig. 3. Analog current input conversion circuit Fig. 4. Analog voltage input conversion circuit.

Optimal Design of Control System for Ground Source Heat Pump Unit

4.2

87

Realization of Improved Main Control Circuit

As a new type of technology, the research and application of ground source heat pump in control is not perfect. The existing ground source heat pump control technology cannot adapt to the dynamic change of load and ambient temperature in buildings. It is easy to cause circulating fluid to deviate from the reasonable flow velocity range, and seriously make the heat transfer state of fluid in pipe enter the transition state zone, resulting in the decrease of system efficiency. In this design, the frequency conversion control technology is used and the environmental temperature and load changes are included. Due to the large number of disturbance sources in the ground source control loop and the lag of temperature transmission, the conventional single loop control system used in the early stage has some problems such as not timely, large deviation and poor control quality [12]. In the improved cascade control scheme, the output of the temperature controller is used as the set point of the flow controller, and the output of the flow controller is used to control the frequency converter pump. Figure 5 is the schematic diagram. The main and secondary controllers work together to make the temperature quickly return to the set value.

Fig. 5. Schematic diagram of ground source control loop

Figure 6 is the schematic diagram of the end loop control of the system. The design adapts the change of the end load of the air conditioning system by adjusting the temperature difference between the supply and return water. The temperature difference between supply and return water can keep unchanged, even if the water flow changes with the change of load. When the end load decreases, the water flow of the water system decreases accordingly, this reduces the energy conveyed by the system to the load, so as to meet the requirements of load reduction. Because the reduction of water flow can greatly reduce the energy consumption of the water system transportation, the energy-saving effect of the system has been further improved.

Fig. 6. Schematic diagram of end loop control

88

X. Jiang et al.

In the design of indoor control system, 8 bit MCU 89C52 is used as main control chip and three-phase PWM waveform is generated by chip SA8281. LM334 integrated temperature sensor is used to collect temperature data and ADC0809 chip is used to complete analog-to-digital conversion. MAX485 serial communication interface chip is used to communicate with the host computer. Indoor display uses 4-bit 8-segment LED, which is driven by SAA1064 in a dynamic way. Three of them are used to display temperature and related parameters, and the other one is used to drive six states of LED indicator. Keyboard is used for indoor system switch and refrigeration, heating, temperature setting. E2PROM24C01 is used to save setting parameters. Figure 7 shows the circuit schematic diagram of the indoor computer control system.

Fig. 7. Schematic diagram of indoor control system.

5 Practical Verification Using this design scheme, the air conditioning system of a large hotel is optimized and renovated. The total building area of the hotel is 30,000 m2. The cold load of the hotel is 3687 KW, the heat load is 2950 KW, hot water of the room wash, the restaurant kitchen, the bathing center and the indoor swimming pool is supplied by the waste heat of the ground source end circulation system. Figure 8 is a comparison chart of key power consumption under monthly average temperature before and after the renovation of the hotel. The annual operation data curve shows that the outdoor temperature significantly affects the power consumption of the compressor and water pump. Because the compressor and pump adopt frequency conversion technology, they can run at low speed and low power consumption when the outdoor temperature is suitable. The curve shows that the power output of the control system can be adjusted in time under different meteorological conditions and load changes. And the control energy-saving effect after optimization has been greatly improved compared with that before optimization, which greatly improves the energy-saving of the whole system.

Optimal Design of Control System for Ground Source Heat Pump Unit

89

Fig. 8. Comparison of key power consumption before and after renovation.

Acknowledgment. This work was supported by Natural Science Research Major Project of higher education institution of Jiangsu Province (grant numbers: 17KJA460001). Special thanks to all those who helped me during the writing of this paper.

References 1. Chen, Y., Ding, G.C., Yang, M.: Experimental investigation on thermal responsive characteristics and shift operation of heat pump. Acta Energiae Solaris Sinica 32(2), 257– 261 (2011) 2. Marcotte, D., Pasquier, P.: Fast fluid and ground temperature computation for geothermal ground-loop heat exchanger systems. Geothermics 37(6), 651–665 (2008) 3. Rivera, J.A., Blum, P., Bayer, P.: A finite line source model with Cauchy-type top boundary conditions for simulating near surface effects on borehole heat exchangers. Energy 31(2), 321–332 (2016) 4. Rivera, J.A., Blum, P., Bayer, P.: Analytical simulation of groundwater flow and land surface effects on thermal plumes of borehole heat exchangers. Appl. Energy 35(2), 421–433 (2015) 5. Stauffer, F., Bayer, P., Blum, P., Giraldo, N.M., Kinzelbach, W.: Thermal Use of Shallow Groundwater”. Taylor & Francis Group, CRC Press, Boca Raton (2014) 6. Liu, Q., Jiang, Q.Q., Jiang, L.L.: Energy saving research of constant temperature and humidity air conditioning system driven by ground source heat pump with condensing heat recovery. Fluid Mach. 43(4), 70–74 (2015) 7. Lei, F., Hu, P.F., Huang, S.Y.: Energy and exergy analysis of a ground water heat pump system. Fluid Mach. 40(2), 57–62 (2012) 8. Hu, T., Zhu, J.L., Zhang, W.: Energy-exergy analysis of a ground-source. Acta Energiae Solaris Sinica 35(2), 271–277 (2014) 9. Tolooiyan, A., Hemmingway, P.: A preliminary study of the effect of groundwater flow on the thermal front created by borehole heat exchangers. Int. J. Low-Carbon Technol. 9, 284– 295 (2014) 10. Bouafia, A., Gaubert, J.P., Krim, F.: Predictive direct power control of three -phase pulse width modulation (PWM) rectifier using space-vector modulation (SVM). IEEE Trans. Power Electron. 25(1), 228–236 (2010) 11. Pei, S.P., Wu, B.R.: The determination for the main circuit parameters of the three-phase PWM rectifier. Electr. Measur. Instrum. 51(13), 88–97 (2014) 12. Jiang, X.M.: Design and application of control system for ground source heat pump unitround source heat pump unit. Chin. J. Electron Devices 41(8), 256–260 (2018)

Design of Elevator Type Three-Dimensional Garage Xiaomei Jiang1(&), Michael Namokel2, Chaobin Hu1, and Ran Tian1 1

Jiangsu Key Laboratory of Elevator Intelligent Safety, School of Mechanical Engineering, Changshu Institute of Technology, Changshu 215500, Jiangsu, China [email protected] 2 Nordhessen University of Applied Sciences, Kassel, Germany [email protected]

Abstract. With the rapid development of China's economy, the number of household cars has increased dramatically, so parking problems are common in cities. The traditional form of garage can not meet the needs of the development of modern cities, so it is necessary to use modern automated three-dimensional garage to solve the parking problem. In this paper, the elevator type threedimensional garage is studied, the lifting mechanism, the car carrier plate, the whole steel structure bracket are designed, the important parts are checked, the model of steel structure bracket 3D garage is established, the lifting structure of steel wire rope is selected, and the traction force is analyzed and checked. The mechanical analysis of the car carrier bottom plate through punch process is selected and compared with the finite element analysis. The improvement of the three-dimensional garage makes the utilization ratio and efficiency increase, which mainly reduces the manufacturing cost, improves the safety performance. Keywords: Elevator type 3D garage


Lifting mechanism
Load analysis

1 Introduction With the development of three-dimensional parking garage in recent years, it has developed in China, but the development is not good enough. This is related to stereo garage itself and the immaturity of modern technology. The technical reason is that the garage access time is relatively long, the operation is not convenient. Because of the above disadvantages, it is difficult to popularize this parking mode. But on the other hand, with the rapid growth of cars in China, the number of cars in the market is increasing year by year. Some cities even have parking problems, and parking spaces are sold at a high price. With the increase of automobiles and the shortage of construction land, there are fewer parking spaces in the city. How to solve the problem of parking and car storage has become the most prominent problems in major cities [1, 2]. The form of the three-dimensional parking garage is various today, but no matter how it changes, it is composed of basic support, vehicle access, electrical control, power operation, etc. The definition of support is more extensive. It is mainly made of different materials [3]. It is useful for concrete structure, steel structure of various types © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2021 Y. Wang et al. (Eds.): IWAMA 2020, LNEE 737, pp. 90–98, 2021. https://doi.org/10.1007/978-981-33-6318-2_11

Design of Elevator Type Three-Dimensional Garage

91

and mixed materials. No matter what materials are used, it is mainly used to support and frame. In the support part, the main supporting structures are column, support girder, cross brace and so on. Some of the bottom plates of the car storage are movable and some are fixed; the movable ones are divided into up and down lifting activities and side direction sliding activities [4]. The elevator type three-dimensional garage is simply summarized as the elevator to finish the positioning and stop the car layer by layer, realize the multi-layer parking function of the car, realize the lifting function of the middle component as the elevator is steel wire rope, need less maintenance personnel, low cost, occupy less land, and more actual parking layers, so that more cars can be parked. The overall support adopts steel structure, which makes the overall frame have higher strength and stronger seismic resistance. Finally, the reliability is enhanced. The steel structure has high modulus of elasticity, high compression and bending resistance. The overall design weight is much lighter than other concrete designs. The steel structure has high toughness and plasticity, and it will not break suddenly, and its stress relaxation performance is better, and its vibration absorption performance is better with low cost and convenient installation [5]. This structure is easy to process in the factory, with strong production capacity and high automation of processing equipment, making the cost relatively cheap. This kind of bracket adopts welding, bolt and thread connection to assemble quickly.

2 Operating Principle Elevator type lift parking garage is shown in Fig. 1. The storage and access me chanism of three-dimensional garage uses belt drive, gear meshing and lateral movement to realize vehicle storage and access. There are two layers of belt driven traverse device, the upper layer is with traverse mechanism, which is to store the vehicle in the garage parking space. The lower layer is a baffle mechanism, which is used to fill the space between the car carrier and parking space. When the vehicle is stored, the baffle is in the horizontal position, driving the traverse mechanism and the parking belt wheel to move. When the car is moved over the gap between the car carrying plate and the

Fig. 1. Elevator type 3D garage structural diagram.

92

X. Jiang et al.

parking space, and the parking is completed;, When the car is taken, the motor will rotate in the opposite direction, so that the drive belt will rotate in the direction of the vehicle plate, and the moving vehicle will face the vehicle plate. By reasonably arranging the motor position and selecting the appropriate traction motor, the operation accuracy of the lifting mechanism is improved and the energyconsumption is reduced. The structural design of the carrier plate can avoid the inclination when the vehicle moves horizontally. The garage is made of section steel by welding and bolt connection. The triangle structure design and channel steel are used in combination.

3 Lifting Mechanism In this paper, the traction drive mechanism with traction ratio 1:1 is selected as the lifting mechanism of this stereo garage. For driving connector, steel wire rope selected as the conversion device of intermediate force and the lifting tool of vehicle plate. The overall traction elevator structure structure is shown in Fig. 2, which is composed of driving motor, traction wheel, steel wire rope, pulley block and counterweight. The motor rotation drives the traction wheel to rotate, which moves through friction, thus completing the lifting function of the car. When a car is put in, the control system controls the motor to rotate, the motor drives the roller to move, and the roller drives the wire rope to move, so as to drive the car carrying board to rise and complete the car rising function. The effective traction force produced by the rotation of the traction wheel is related to the friction coefficient and the wrap angle. The larger the value of the two (generally, the angle is not more than 180° and it can be more than 180° if the compound winding is used) the greater the tractive force is.

Fig. 2. Schematic diagram of traction elevator structure (1:1)

In order to achieve the lifting of the vehicle, the traction force generated by the friction between the traction pulley and the traction rope must be greater than the load generated by the car carrier plate and the maximum load in the traction rope. The force on the traction rope is generated by the dynamic friction between the wire rope and the groove of the traction pulley. In order to ensure the safety and reliability of the lifting

Design of Elevator Type Three-Dimensional Garage

93

process, ensure that there is no slip between the wire rope and the pulley groove [6]. The main methods of increasing drag force are as follows: make a circle around the traction rope, guide pulley and traction pulley, and then connect them to the car; avoid excessive slip of wire rope to reduce friction coefficient; the traction pulley is made of high-quality materials with wear resistance and large friction coefficient; increase the wrap angle of the wire rope at the traction pulley; select the appropriate shape of the concave groove rope traction pulley.

4 Load Analysis and Calculation The design of the garage is mainly used for the parking of ordinary cars, so the selection of model is not large. Generally, the length of family car is about 4–5 m (L); the width of car is about 1.8 m–2 m (W), the height is 1.3 m–1.6 m (H), and the wheelbase is 2.2 m–3 m (WB). In order to meet the most extensive parking demand, the selection of the length, width and height is based on the largest model size of the family car, with more margin. In this way, in addition to increasing the application scope of the car carrier plate, the wider space reduces the parking difficulty of the car. The main body of the carrier plate is made of carbon structural steel Q235, and the surface is made of zinc and other anticorrosion materials to prevent corrosion caused by rainy days or humid environment. The shape of the bottom plate is easy to punch and manufacture, wear-resistant, rust proof and good overall performance. As shown in Fig. 3, the bottom of the carrier plate is similar to the groove shape, with the middle high and both ends sunken, so it is easier to install the conveyor belt, roller and bearing. The two ends of the car carrier plate are equipped with car stopping devices, which are controlled by the motor and can realize 90° turning. When the stop lever is raised, it is used to prevent the car from sliding back and forth. When the gear lever is lowered and in the horizontal position, the gap between the loading plate and the parking space can be filled, the bottom motor and gear transmission convey the power of car moving.

Fig. 3. 3D modeling drawing of car carrier chassis

The loading object of the car elevator is the car, and the weight of the car itself is very large, so it is necessary to carry out a certain mechanical analysis of the car board. Calculate the vehicle mass with 125% of the rated load, take 1500 kg, and the self weight is 1500 N. According to the general car, the front and rear wheel axles are 3000mm apart, and the ratio of the stress on two tires of the common car is about 6:4.

94

X. Jiang et al.

Therefore, the pressure of the front and rear tires on the carrier plate is proposed to be 9000 N and 6000 N respectively. The force on the carrier plate and the force analysis of static load are shown in Fig. 4. (1) Thus, the reaction force of support can be obtained Fa ,Fb P Ma ¼ 0 F1  1000 þ F2  4200Fb  5200 ¼ 0

Fig. 4. Force analysis on the carrier plate.

so Fb ¼ 6576:9 N X F¼0 F1 þ F2  Fa  Fb ¼ 0 so Fa ¼ 8423.1 N: Calculation of shear force by section method: Qa ¼ Fa ¼ 8423.1 N: Shear force at the left supporting point: Q1 ¼ Fa  F1 ¼ 8423.1 N  9000 N ¼ 576.9 N Shear force in the middle section: Qb ¼ Fb ¼ 8423.1 N. Shear force at right end fulcrum: Qb ¼ Fb ¼ 8423.1 N. Maximum bending moment at front tire: M1 max ¼ Fa  1000 mm ¼ 8423:1 N
m. Maximum bending moment at rear tire: M2 max ¼ Fb  1000 mm ¼ 6576:9 N
m. As shown in the Fig. 5 , taking the fulcrum Fa at the left end of the carrier plate as the origin, the corresponding shear diagram (left) and bending moment diagram (right) can be drawn.

Design of Elevator Type Three-Dimensional Garage

95

Fig. 5. Stress analysis diagram of carrier plate

Cut off the section of the chassis part of the carrier plate, and calculate the modulus of the bending section of the chassis centroid. Because the carrier plate itself is symmetrical, the left half of the chassis section of the carrier plate is taken for calculation. Take the Z axis of the center as the vertical coordinate, and the Y axis of the bottom as the horizontal coordinate. So the abscissa of centroid is. The interface of the intercepted carrier plate can be divided into four parts, A1 A2 A3 A4 , as shown in Fig. 6 (Table 1).

Fig. 6. Chassis section of carrier plate.

Table 1. Data of chassis section of carrier plate Rectangular section A1 A2 A3 A4

Width (mm) 20.0 600.0 20.0 500.0

Height (mm) 330.0 20.0 330.0 20.0

Area (mm2) Ai 6600.0 12000.0 6600.0 10000.0

Centroid ordinate (mm) Zci 185.0 10.0 185.0 340.0

The vertical coordinate formula of the combined section centroid is [7]: Pn 6600  185 þ 12000  10 þ 660  185 þ 10000  340 i ¼ 1 Ai Zci Zc ¼ P ¼ n 6600 þ 12000 þ 6600 þ 10000 A i i¼1

96

X. Jiang et al.

In the formula:- the section consists of several parts, here is 4; ——Area size of each part, mm2; ——The centroid ordinate of each component section. ——Longitudinal coordinates of the centroid of the chassis section; The moment of inertia formula of each section (centroid axis) is: Z Iyci ¼ A

Zi2 dAi þ b2i Ai

Iyc1 ¼

20  3303 þ ð185  169:375Þ2  6600 ¼ 61506328:124 mm4 12

Iyc2 ¼

600  203 þ ð10  169:375Þ2  12000 ¼ 305204687.4 mm4 12

Iyc3 ¼

20  3303 þ ð185  169:375Þ2  6600 ¼ 61506328:124 mm4 12

Iyc4 ¼

500  203 þ ð340  169:375Þ2  10000 ¼ 291462239.582 mm4 12

The moment of inertia of the section to the centroid; ——The longitudinal coordinate of the section centroid; ——The section area; ——The relative distance between the section and the longitudinal coordinate of the center of gravity of the vehicle plate section. Iyc ¼ ðIyc1 þ Iyc2 þ Iyc3 þ Iyc4 Þ  2 ¼ 1439359166:666 mm4 Modulus of flexural section: Wyc ¼

Iyc 1439359166:66 ¼ 796877:036 mm3 ¼ 350  169:376 Zmax

Maximum bending stress of the carrier plate: rmax ¼

Mmax 8423:1 N
m ¼ ¼ 10:57 MPa 796877:046 mm3 Wyc

rmax ¼ 10:57 MPa\\½r ¼ 235 MPa It can be seen that the yield limit of Q235 fully meets the requirements and the service conditions of the car carrier plate. The car carrying plate is the load-bearing component with sufficient strength, and the mechanical calculation shall be carried out to ensure that the car carrying plate has sufficient strength and safety coefficient, which is necessary for design. The maximum stress of the car carrier plate under dynamic impact load is 49.8 MPa. The safety factor

Design of Elevator Type Three-Dimensional Garage

97

(5.12) under this working condition can be obtained. The steel frame can meet the requirements of working condition. From the checking calculation of car carrier plate under various working conditions, it can be concluded that the car carrier plate meets the requirements of use. The finite element method has great advantages in checking calculation of structure, and has higher calculation accuracy than traditional simplified calculation [8]. Composite stress nephogram of car carrier plate is shown in Fig. 7, Compared with the safety factor of each working condition, some structure still can be optimized, and the targeted structural optimization can be carried out according to the calculated results, which can reduce the self-weight of the car carrier plate itself and thus reduce the production cost.

Fig. 7. Composite stress nephogram of car carrier plate.

5 Conclusion The overall support of the elevator type three-dimensional parking garage is a steel structure frame. In the overall design, the load and other factors are taken into account, and the strength design are increased. The overall structure design is more solid through welding. The overall height of each floor can be divided into several types. The bottom floor is designed with vehicle inlet and outlet, and the height of different vehicles should be considered in the design, so the height of the bottom floor is relatively high. Here, 2.6 m is selected. The middle layer is the car placement layer, so the height can be reduced a little, and the design is about 2.2 m. The top part of the overall support is the placement part of the motor, which will not be affected by the force of the top, so it is required to reduce the weight as much as possible, but also consider the reasons for future maintenance and repair. The design here is about 1.5 m. The following problems need to be solved further. The garage should be more intelligent and convenient, so the vehicle access time is short to meet the market needs. remote management, convenient for vehicle access. Although the application space of the 3D garage is relatively large, the overall intelligent level is relatively low. In the future design, intelligent design should be added to realize the function of face recognition control, automatic timing car pick-up, etc.

98

X. Jiang et al.

Acknowledgment. This work was supported by Natural Science Research Major Project of higher education institution of Jiangsu Province (grant numbers: 17KJA460001). Special thanks to all those who helped me during the writing of this paper.

References 1. Ren, B.M.: Mechanical three-dimensional parking garage. Ocean Press, Beijing (2001) 2. Wang, W.Q.: Design Manual of Urban Car Parking. China Construction Industry Press, Beijing (2002) 3. Chen, L.Y.: Safety Code for Elevator Manufacturing and Installation - GB 7588 Understanding and Application. China Quality Inspection press, Beijing (2017) 4. Hao, C., Xiao, S.M.: Structural design and operation optimization of car loading plate in elevator type three-dimensional garage. Mech. Res. Appl. (6), 76–81 (2014) 5. Jing, Y.L.: Research and Design of Elevator Lift Parking Garage. Shandong University of Science and Technology, Qingdao (2003) 6. Jiang, X.M., Namokel, M.: Stress analysis of hoist auto lift car frame. In: International Workshop of Advanced Manufacturing and Automation, Manchester, UK, pp. 151–159 (2019) 7. Liu, H.W.: Material Strength, 5th edn. Higher Education Press, Beijing (2010) 8. Jiang, X.M., Guo, L.Z.: Stress analysis of machine supporting beam system for large. In: 8th Symposium on Lift & Escalator Technologies, Hong Kong, China, pp. 124-135 (2018)

Research on Elevator Fault Information Extraction and Prediction Diagnosis Xiaomei Jiang1(&), Michael Namokel2, Chaobin Hu1, and Ran Tian1 1

Jiangsu Key Laboratory of Elevator Intelligent Safety and School of Mechanical Engineering, Changshu Institute of Technology, Changshu 215500, Jiangsu, China [email protected] 2 Michael Namokel, Nordhessen University of Applied Sciences, Kassel, Germany [email protected]

Abstract. Nowadays, most of the fault diagnosis methods are based on the collected data, which can not realize the timely prediction of fault diagnosis. By extracting the vibration information that can reflect the different operation states of elevator, the signal is preprocessed by the multi-threshold denoising method of wavelet packet, and the predicted data is extracted by feature information. The vibration acceleration signals of elevator in X, Y and Z directions are measured by the elevator vibration tester EVA-625, and preprocessed by the method above. Then, the maximal peak value, peak-peak value and A95 value of the signals are calculated to judge whether the elevator has faults or not and the comfort of elevator ride according to the relevant national standards, combined with the historical vibration state of elevator operation, the analysis and comparison are carried out to realize prediction of mechanical faults and the prediction diagnosis. Keywords: Fault diagnosis signal


Prediction
Data fusion
Elevator
Vibration

1 Introduction At present, elevators have become an indispensable part of people’s lives. With the increase of the total elevator volume, the number of elevator accidents is also increasing. Therefore, how to improve the safety and comfort of the elevator, how to predict the elevator fault more accurately and efficiently, has become the focus of attention in today’s society. At the same time, these problems have become the focus and difficulty of the elevator industry [1, 2]. I Vibration signal generally exists in the running mechanical and electrical equipment. When the mechanical and electrical equipment is abnormal, the vibration will increase and the working condition will change, and most of the mechanical equipment will be manifested in the form of vibration when the mechanical equipment fails [3, 4]. Vibration condition monitoring technology is a powerful means to monitor the vibration state of mechanical equipment by synthesizing modern scientific and © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2021 Y. Wang et al. (Eds.): IWAMA 2020, LNEE 737, pp. 99–106, 2021. https://doi.org/10.1007/978-981-33-6318-2_12

100

X. Jiang et al.

technological research. Combined with the fault history and present situation of the equipment, considering the environmental factors, the collected data are analyzed and processed in order to find out the root cause of the equipment operation fault in time and provide the decision basis for fault diagnosis and prediction [5, 6]. Based on sensor and communication technology, the real-time operation status of each important mechanical node of elevator is perceived. Real-time feature of elevator mechanical parts collected by sensors is the most valuable characteristic value, according to the criteria for discarding elevator main parts, the elevator characteristic values can be collected by sensor as shown in Fig. 1. The characteristic values can be collected at the same time, how to remove the correlation features with small influence factors from a large number of complex original features and extract the key features is essential to improve the detection accuracy and efficiency [7].

Fig. 1. Real-time elevator characteristics

There are two kinds of common failures of elevator, namely, mechanical system failure and electrical control system failure. Usually the consequences of the former are more serious. In this paper, the relationship between elevator mechanical system fault and car vibration characteristic parameters is studied. Among the many reasons leading to mechanical system failure, serious wear of guide boots, poor lubrication of guide rails, excessive fatigue of wire ropes and loosening of fastening bolts are common. These fault reasons are related to elevator vibration. Elevator is a non-linear electromechanical coupling system, and its operation process is often accompanied by mechanical vibration, and the vibration signal is an effective information carrier

Research on Elevator Fault Information Extraction and Prediction Diagnosis

101

reflecting the operation status of elevator equipment [8]. Therefore, the mechanical components of elevator can be monitored by vibration acceleration sensor to reflect its operation status, and the most sensitive information reflecting the change of working conditions can be further extracted as a characteristic signal for recognition and fault diagnosis of early safety hidden dangers in elevators. The cause of elevator failure is complex, and the elevator vibration contains all the dynamic factors in the mechanical equipment. Using the vibration signal of the elevator car to diagnose the elevator mechanical fault, extracting the characteristic information related to the elevator mechanical fault from the measured vibration signal is the key of the fault prediction and diagnosis process.

2 Vibration Signal Processing and Feature Extraction In order to facilitate the analysis, the actual complex vibration testing system is simplified to a quadratic differential equation of motion, that is: pffiffiffiffiffiffiffi m€x þ kx_ þ 2n mk x ¼ f ðtÞ

ð1Þ

In “(1)”, m: the equivalent mass of system, k: the equivalent stiffness of system, n: _ €x: the displacement, velocity and acceleration the equivalent damping coefficients, x; x; of the system output, f(t): various vibration excitation sources within the system. It can be seen from “(1)” that the vibration acceleration is not only related to the internal excitation source but also related to the equivalent stiffness and the equivalent damping coefficient of the system, so that the running state of the elevator can be characterized by the acceleration vibration curve and the its characteristic parameters. Because of the randomness and the fluctuation of the vibration signal of the elevator, it is difficult to find the inherent variation rule of the system directly from the vibration curve, and it is necessary to process the vibration signal. At present, the most widely used elevator vibration testing field is the EVA625 Vibrator of PMT Company of USA, which tests the elevator speed, running distance, acceleration/deceleration, jump and noise data [9], combined with GB/T 10059– 2009 Elevator Test Method [10], GB/T 10058–2009 Elevator Technical Conditions [11], and GB/T 24474–2009 Elevator Ride Quality Measurement [12], the technical parameters of these three standards are used to evaluate the comfort of elevator passengers and the quality performance of elevator transportation during the operation of elevator car. Because the wavelet packet decomposition method can decompose not only the low frequency part but also the high frequency part, and the corresponding frequency band can be adaptively selected to match the signal spectrum according to the signal characteristics and analysis requirements, it is a more refined decomposition method than the wavelet decomposition method [13]. The principle of wavelet packet comes from tower algorithm of multi-scale analysis. Wavelet packet decomposition is realized by solving Cj,l,2n and Cj,l,2n+1 through Cj+1,k,n. The decomposition algorithm is as follows:

102

X. Jiang et al.

8 P 1 < Cj;l;2n ¼ p2 hk2l Cj þ 1;k;n k P : Cj;l;2n þ 1 ¼ p12 gk2l Cj þ 1;k;n

ð2Þ

k

In “(2)”, Cj,l,2n and Cj,l,2n+1: next-level wavelet packet decomposition results, Cj+1,k, : upper-level decomposition results, j: scale coefficients, l: position coefficients, n: n frequencies, k: variables, h, g: orthogonal conjugate low-pass and high-pass filters. The vibration signal of elevator is evaluated by peak value. Vibration peak value refers to the arithmetic sum of two signal peaks in the opposite direction separated by each intersection zero on the time history curve of elevator vibration acceleration. The maximum peak value refers to the maximum of all peak values found within the defined boundaries. A95 refers to the value whose peak value is equal to or less than 95% of the defined limit. The calculation of vibration peak-peak value, maximum peakpeak value and A95 value can be calculated according to the following procedures: Step 1: Find the first, second, third, and…zero intersections. Step 2: Find the maximum positive and negative signal values between the first and third intersection zeros. Step 3: Calculate the sum of the absolute values of the two quantities, and write it down as P1;3 ; Step 4: Between zero intersections 2 to 4, 3 to 5, 4 to 6,… Repeat steps 2 and 3 to get P2;4 , P3;5 , P4;6 , ……; Step 4: Statistical analysis of P1;3 , P2;4 , P3;5 , P4;6 ……, shown that the maximum is the maximum peak-peak value, and the 0.95 confidence upper limit with zero as the confidence lower limit is the A95 value.

3 Signal Denoising Processing Signal analysis is an effective method for fault diagnosis, but because of the complexity of the field environment and the serious noise, the reliability of the diagnosis results is greatly reduced due to the signal interference. Wavelet has many advantages, such as multi-resolution, de-correlation, low entropy, and flexibility of base selection, which determines its strong superiority in dealing with non-stationary signals and removing signal noise [13]. The main methods of wavelet denoising are modulus maxima denoising, wavelet threshold denoising and translation invariant method. However, the single threshold denoising method does not fully consider the signal and its noise distribution, so the denoising results are either incomplete noise removal or excessive denoising, and the useful signal is also removed. The multi-threshold denoising method based on wavelet packet is a good solution to this problem. The inherent vibration signal of elevator car is low-frequency, and the multi-threshold denoising method of wavelet packet is very suitable for extracting low frequency signal. There are four threshold estimation methods available for multiple thresholds, as follows: (1) Fixed threshold (sqtwolog) threshold ‘‘k = 2ln(M)’’, M is the length of the signal.

Research on Elevator Fault Information Extraction and Prediction Diagnosis

103

(2) An adaptive threshold selection (SURE) based on stein unbiased likelihood estimation principle (STDN) is used to obtain its likelihood estimation for a given value T. Then the non-likelihood T is minimized to obtain the selected threshold. (3) heuristic threshold(heursure) is the combination of the two values and the choice of the most predictive variable threshold. (4) minimax threshold (minimax) also adopts a fixed threshold, which is to find the minimum value of the maximum mean square error under the worst conditions. The multi-threshold denoising method can be selected according to the denoising advantages of different threshold values. For example, sqtwolog and heusure rules have high denoising intensity, so they are suitable for denoising in high frequency part, while minimaxi and rigrsure thresholding rules are relatively conservative and can effectively extract weak useful signals, so they are suitable for processing in low and medium frequency part. Then, the threshold of wavelet packet analysis coefficients is selected according to the frequency band. The threshold rules are in the Table below (Table 1): Table 1. Threshold selection Threshold estimation Frequency band Low frequency Medium frequency High frequency Methods Rigrsure Rigrsure Sqtwolog Methods Minimaxi Minimaxi Heusure

4 Fault Prediction and Diagnosis Method After the elevator is put into use, the greatest impact is the operation status of the elevator guide rail. Whether the elevator meets the relevant requirements has a great relationship with the failure rate and safety of the elevator. There are many kinds of rail faults, which are mainly due to the large errors between the guide rail installation and the relevant standards requirements and the rail faults caused by the long-term use of elevators (such as rail surface wear). It is impossible to complete the comprehensive detection of multiple items by relying on a single instrument and equipment. Therefore, by measuring the vibration signal of the car, observing the sudden change of the signal, and comprehensively analyzing the fault information of the components of the guidance system contained in the signal, the fault prediction and diagnosis of the elevator guide rail can be realized. The elevator vibration acceleration signals in X, Y and Z directions contain abundant dynamic factors in the elevator mechanical system. By analyzing the vibration signals of elevator car, the dynamic characteristics of elevator operation can be more accurately and effectively reflected, so that the mechanical fault diagnosis of elevator can be more accurate and effective. There are a lot of complex background noises, some random peak interference, and also random noises with zero mean value in the car vibration signals collected directly from the scene. The ideal acceleration curve of elevator car vibration is the lower

104

X. Jiang et al.

frequency signal, and the interference is usually the higher frequency noise. Firstly, the car acceleration signals collected in X, Y and Z directions are de-noised by wavelet packet multi-threshold denoising method, and the real signals are obtained by preprocessing to remove the vibration signal noises of the elevator, which makes good preparation for the later signal analysis. Then, by calculating the peak value, peak-topeak value, kurtosis and other characteristic values of the signal, according to the relevant national standards to determine whether there is a fault in the elevator. If the car runs on a smooth vertical rail, it can be considered that the excitation to the car is a stationary random process. However, if there are defects in the guide rail or a large deviation in the joint position of the guide rail, it is equivalent to adding a shock excitation to the car suddenly, which results in an increase in the horizontal vibration of the car and a large fluctuation in the vibration curve of the car. At this time point, the signal vibration amplitude increases and the energy increases, thus realizing the fault prediction and diagnosis of elevator guide rail.

5 Experiments and Results Analysis In this paper, several vibration signal feature extraction methods introduced above are used to test and experiment an elevator with abnormal sound in the field. The vibration acceleration signals collected from the mechanical vibration system of elevator car in X, Y, Z directions (left and right, front and back, vertical direction) are processed by the wavelet packet multi-threshold processing method, and the results before and after processing are shown in Fig. 2, Fig. 3 and Fig. 4.

Fig. 2. X direction before and after processing

From the car vibration signal in the Y direction in the figure, it can be observed that the elevator suffered a big impact at the elevator running to 2.1 s, 2.5 s, 9.3 s and 10.1 s, and the amplitude increased. Therefore, it can be calculated that there may be

Research on Elevator Fault Information Extraction and Prediction Diagnosis

105

rail faults at 3.9 m, 4.7 m, 18.3 m and 20.3 m of the guide rail. According to the relevant regulations on car vibration in “GBT10058–2009 Elevator Technical Conditions” [10], through Matlab analysis, the peak value of A95 in X direction is calculated to be 0.1484, which is less than 0.15 m/s2, the peak value of A95 in Y direction is 0.2452 > 0.15 m/s2, and the peak value of A95 in Z direction is 0.3133, which is larger than 0.2 m/s2 required by the standard. Therefore, through analysis, it can be concluded that the elevator does not meet the relevant national standards and is a faulty elevator.

Fig. 3. Y direction before and after processing

Fig. 4. Z direction before and after processing

To sum up, the experiments in this paper show that the vibration acceleration signals in X, Y and Z three directions contain a wealth of dynamic factors in elevator

106

X. Jiang et al.

mechanical system. By using the wavelet packet multi-threshold denoising method, the car vibration signals can be smoothed close to the real acceleration signal, and the dynamic characteristics of elevator running can be reflected more comprehensively, accurately and effectively, so that the fault prediction and diagnosis of elevator machinery can be more accurate and effective. Therefore, it is of great social significance and economic benefit to monitor the running state of elevators online, detect faults in time and realize safe operation of elevators. Acknowledgment. This work was supported by Natural Science Research Major Project of higher education institution of Jiangsu Province (grant numbers: 17KJA460001). Special thanks to all those who helped me during the writing of this paper.

References 1. Wei, S.H.: Elevator Fault Diagnosis and Maintenance, 1st edn. Soochow University Press, Suzhou (2013) 2. Wu, Y.: (2015) Analysis and maintenance of elevator common faults. China Mach. 8, 140– 141 (2015) 3. Ren, M.Z.: Analysis and Control and the Calculation Methods of Mechanical Vibration. Mechanical industry press, Beijing (2011) 4. ISO 8041: Human response to vibration – Measuring instrumentation (2005) 5. Song, Z.B.: Common Fault Analysis and Condition Monitoring System Design of Traction Elevator. Hebei University of Science and Technology, Shijiazhuang (2017) 6. Jiang, X.M., Rui, Y.N.: Research on vibration Control of traction elevator. In: International Industrial Informatics and Computer Engineering Conference, Xian, China, pp. 2144–2147 (2015) 7. Jiang, X.M., Namokel, M.:Research on lift fault prediction and diagnosis based on multisensor information fusion. In: International Workshop of Advanced Manufacturing and Automation, Manchester, UK, pp.160–168 (2019) 8. Chang, N., He, W., Li, Z.H.: Strategies for vertical vibration reduction of elevator system. Noise Vibr. Control 2(3), 117–120, 126 (2017) 9. Zhan, H.L.: Elevator vibration performance detection and application of EVA-625 comprehensive analyzer. Electromech. Technol. 2(2), 124–127 (2016) 10. GB/T 10059–2009 Elevator Test Method. China Standards Press, Beijing (2009) 11. GB/T 10058–2009 Elevator Technical Conditions. China Standards Press, Beijing (2009) 12. GB/T 24474–2009 Elevator Ride Quality Measurement. China Standards Press, Beijing (2009) 13. Zhang, Z.T., Zhu, J.J., Kuang, C.L.: Wavelet packet multi-threshold denoising method and its application in deformation analysis. J. Surveying Mapp. 1(4), 13–20 (2014)

Static Performances of Misaligned Journal Bearing with Slip-Texture Surface Jian Jin1, Yinhui Chang1, Guiqin Li2(&), and Peter Mitrouchev3 1

School of Mechatronic Engineering and Automation, Shanghai University, Shanghai 200072, China 2 Shanghai Key Laboratory of Intelligent Manufacturing and Robotics, Shanghai University, Shanghai 200072, China [email protected] 3 University Grenoble Alpes, G-SCOP, 38031 Grenoble, France

Abstract. Hydrodynamic journal bearings will be misaligned due to external loads during operating. It is necessary to realize the effect of slip-texture area on the performances of misaligned hydrodynamic journal bearing. The geometric model of a misaligned bearing is established. Quadratic programming is used to study the influence of wall slip on the hydrodynamic lubrication performance of a two-dimensional journal bearing. The numerical simulation method is used to investigate the influence of the limit shear stress and misaligned angle on the static performances of the bearing. It is found that the slip-texture surface can increase the load carrying capacity and decrease the friction force under low limit shear stress. And the improvement is more obviously with the increase of the misalignment of the journal. Keywords: Hydrodynamic lubrication Texture
Wall slip


Journal bearing
Misalignment


1 Introduction Surface texture were introduced by Hamilton et al. [1]. The convergence of the texture forms many small wedges, which can increase fluid dynamic pressure. While the formation of cavitation is due to the divergence gap at the textured area, which can damage the dynamic pressure of the fluid. With the convergence effect and cavitation, the surface texture will generate additional dynamic pressure on the surface of the friction pair, which can improve the load carrying capacity. Galda analysed the texture area ratio and depth-diameter ratio as the most important geometric parameters [2]. The effect of texture geometric parameters on the Stribeck curve of journal bearing was studied experimentally. In high speed condition, the fluid will slip on the solid surface when shear stress exceed the limit shear stress. Kristian investigated the load carrying capacity and friction force of infinite width bearings [3]. They found that texturing at the inlet of oil flow could increase the oil intake, thereby increasing the load carrying capacity. By using a multiphase flow cavitation model, Cupillard et al. researched the influence of texture distribution. They found that texturing downstream of the maximum oil film © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2021 Y. Wang et al. (Eds.): IWAMA 2020, LNEE 737, pp. 107–114, 2021. https://doi.org/10.1007/978-981-33-6318-2_13

108

J. Jin et al.

could reduce friction force and increase the load carrying capacity [4]. Adatep H studied the tribological properties of groove less and micro-grooved bearings under dynamic loads [5]. Lin Qinyin et al. studied the influence of groove on bearing performance at high speed using journal center locus at different groove depths [6]. In this work, a model of a misaligned journal bearing with silp-texture surface is established. The oil film thickness expression is obtained according to the geometric relationship of the model. Reynolds equation is modified with the boundary slip. The influence of misalignment is investigated to explore the combined effects of surface texturing and wall slip on tribological performance of journal bearings.

2 Theoretical Model 2.1

Geometric Model

The geometric model of textured bearing is illustrated in Fig. 1. The schematic diagram of texture is shown in Fig. 2. It is necessary to judge whether the calculated position is at textured area when modelling the film. The film thickness and texture depth need to be considered in textured aera.

Fig. 1. Geometric model of a textured bearing

Fig. 2. Spherical texture and cylindrical texture

Rp is the dimensionless radius of the cross section, Hp is the maximum depth, Ra is the radius of curvature of the spherical texture. The relationship between Ra, Rp and Hp can be expressed by the following equation:

Static Performances of Misaligned Journal Bearing with Slip-Texture Surface

Ra ¼

R2p þ Hp2 2Hp

109

ð1Þ

The depth of spherical texture can be obtained as follows:   8 qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi   ffi qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi  2ffi > < Hp  Ra  ðX0  Cx Þ2 þ Y0  Cy 2 ðX0  Cx Þ2 þ Y0  Cy  Rp DH ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi  2ffi > : 0 ðX0  Cx Þ2 þ Y0  Cy [ Rp ð2Þ The depth of the cylindrical texture can be obtained as follows: ffi 8 qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi < Hp ðX0  Cx Þ2 þ Y0  Cy 2  Rp qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi DH ¼ : 0 ðX  C Þ2 þ  Y  C  2 [ R 0 x 0 y p

ð3Þ

The deformation of the shaft, when acted by a heavy external load, can result in the misalignment of the journal bearing hole. Figure 3(a) shows the schematic geometry of a misaligned journal bearing. Figure 3(b) shows the position relationship of the center point in projection plane. Tilt gradient em is the dimensionless projection length of OO1, which represents the misaligned rate of the journal. hm is the misaligned angle of shaft and y-axis.

Fig. 3. Misaligned journal bearing

According to the geometric relationship in Fig. 3, the film thickness at different em , hm can be derived: hð;; kÞ ¼ cð1  ðx þ DxÞ  sin;  ðy þ DyÞ  cos;Þ where

ð4Þ

110

J. Jin et al.

Dx ¼

2.2

k  2l l 2

 em  sinðhm Þ; Dy ¼

k  2l l 2

 em  cosðhm Þ

ð5Þ

Governing Equations

The Reynolds equation is modified with boundary slip condition:         @ qh3 @p @ qh3 @p @ qUh @ qðuas  ubs Þh þ ¼ þ @x 12l @x @z 12l @z @x 2 @x 2

ð6Þ

where uas ; ubs are the slip velocities of the shaft and bearing, respectively. According to the variational principle, the solution of the lubrication problem is to find the distribution of film pressure and fluid velocity that make a specific functional take extreme value or stationary value [7]. The corresponding function of Reynolds equation can be expressed as follows: Z

 h3 h J ð pÞ ¼ rp  ðU þ Uas þ Ubs Þ~i þ ðwas þ wbs Þ~j rpgdA þ f½ 2 24l A

Z ps qs dS ð7Þ Sq

s:t: jsa j  sla ; and p  0

ð8Þ

where A is the area of the fluid film domain, ~i and ~j are respectively the unit vector in x and y directions, qs is boundary volume flow, and ps is corresponding film pressure. Sq is the boundary of the fluid domain. For a bearing with a finite width, the fluid film forces are given below: Z Fx ¼ 

Z

u1

Z Fy ¼ 

u2

u2

u1

2

 Pdk sinudu

ð9Þ

2

 Pdk cosudu

ð10Þ

0

Z 0

where u1 and u2 are the initial boundary of circumferential convergence for the oil film and the rupture boundary of the oil film, respectively. The dimensionless oil film friction resistance can be obtained by integrating the pressure distribution: 1 Ft ¼ 2

Z 0

2

Z

up u1

! ! Z Z Z Z u2 2 u2 dudk 1 2 dudk @P dudk Hb H þ þ 2 H 2 0 H @u up u1 0

ð11Þ

Static Performances of Misaligned Journal Bearing with Slip-Texture Surface

111

3 Simulation Results The basic parameters of the bearing are presented in Table 1. Table 1. Bearing parameters Parameter Bearing length Bearing diameter Radius clearance Eccentricity

3.1

Symbol L D C e

Value 80 mm 100 mm 0.145 mm 0.4

Parameter Texture area ratio Texture depth Texture radius Tilt gradient

Symbol Sp hp rp em

Value 20% 0.145 mm 2.9 mm 0.2

Film Thickness

The journal rotation speed is 6000 r/min. The oil film thickness distributions are shown in Fig. 4.

Fig. 4. Oil film thickness

The distribution of slip velocity and bearing pressure are solved, under the condition of non-texture, cylindrical texture, spherical texture between the limiting shear stress s0a = 1000–10000 pa, and the pressure distribution is integrated to obtain the load carrying capacity and friction force. 3.2

Static Characteristics at Different Limiting Shear Stress

As shown in Fig. 5, the texture has a positive effect on the load carrying capacity in low limiting shear stress, while with the increase of the limiting shear stress, the load carrying capacity becomes significantly lower than that of the non-textured bearing. Under the low limit shear stress, the depth of the texture increases the oil film thickness which reduces the gradient of the fluid velocity in texture, so that the surface shearing stress becomes smaller and the general slip is less than the non-textured surface. But the effect of the general slip strength on the load carrying capacity is negative. In addition, the textured surface has a wider slip distribution range than the non-textured surface. Under high shear stress, the load carrying capacity of non-textured structure is

112

J. Jin et al.

bigger than that of the textured structures. It shows that the full texture distribution has a negative effect on the load carrying capacity under high shear stress.

Fig. 5. Load carrying capacity at different limiting shear stress

As shown in Fig. 6, as the limiting shear stress increases, the area where slip decreases, which leads to an increase in friction force. The friction force of the texture bearing is always smaller than that of the non-textured one, and the cylindrical texture case is smaller than the spherical texture case. Slippage can reduce friction force, because the shear flow resistance is proportional to the fluid velocity gradient. Comparing the textured and non-textured conditions, the average velocity gradient of textured bearing bush is higher than that of the nontextured bearing, because the average slip strength is slightly weaker (the slip in the textured pit is weaker). But on the other hand, the velocity gradient of the whole area is decreasing, which is due to the increase of the average oil film thickness at textured area. The effect of the latter is greater than the former, which leads to a decrease in the shear flow resistance of the textured surface, while in the pressure flow area, the textured surface has a stronger load carrying capacity and a greater resistance to pressure flow. Therefore, it is speculated that the low friction resistance of the textured surface is mainly caused by the shear flow.

Fig. 6. Friction force at different ultimate shear stresses

Static Performances of Misaligned Journal Bearing with Slip-Texture Surface

3.3

113

Static Characteristics of a Misaligned Journal Bearing

The tilt gradient em is 0.2–0.7. Misaligned angle hm is 0. The limiting shear stress is 1000 Pa and 4000 Pa, respectively. As shown in Fig. 7, at low shear stress, with the increase of the misalignment of the journal, the load carrying capacity of the texture bearing increases more obviously, which is significantly higher than that of the non-texture journal bearing, and the friction force is significantly reduced. In the case of high limit shear stress, with the tilt gradient increases, the raise in the load carrying capacity of the non-textured bearing is greater than that of the textured bearings. The friction force of textured bearings is lower than that of non-textured bearing. The texture has a certain positive effect on the static characteristics of the bearing, especially in the case of lower limiting shear stress.

Fig. 7. Static characteristics of bearing at different tilt gradient

4 Conclusions The combined effects of surface texturing and wall slip of a misaligned journal bearing are investigated in the study. (1) The existent of boundary slip reduces the velocity gradient, thereby reducing the friction. It improves the inlet flow rate and allows a better supply of fluid if the slip zone is arranged at the circumferential inlet. (2) Texturing has a positive effect on the load carrying capacity at low limiting shear stress. But with the increase of the limiting shear stress, the load capacity becomes significantly lower than that of the non-texture bearing.

114

J. Jin et al.

(3) At low shear stress, with the increase of the misaligned angle of the journal, the load capacity of the texture bearing increases more obviously, which is significantly higher than that of the non-texture journal bearing. And the friction is also reduced. However, under the condition of higher limit shear stress, the misaligned angle reduces the beneficial effect of texture on the load carrying capacity.

References 1. Hamilton, D.B., Walowit. : A theory of lubrication by micro-irregularities. J. Basic Eng. 97 (2), 117–185 (1966) 2. Galda, L., Pawlus, P., Sep, J.: Dimples shape and distribution effect on characteristics of stribeck cureve. Tribol. Int. 42(10), 1505–1512 (2009) 3. Tonder, K.: Inlet roughness tribodevices: dynamic coefficients and leakage. Tribol. Int. 34 (12), 847–852 (2001) 4. Cupillard, S., Glavatskih, S., Cervantes, M.J.: Computational fluid dynamics analysis of a journal bearing with surface texturing. J. Eng. Tribol. 222(2), 97–107 (2008) 5. Adatepe, H., Bykloglu, A., Sofuoglu, H.: An investigation of tribological behaviors of dynamically loaded non-grooved and micro-grooved journal bearings. Tribol. Int. 58, 12–19 (2013) 6. Lin, Q., Wei, Z., Wang, N., Zhang, Y.: Effect of recess configuration on the performances of high speed hybrid journal bearing. Ind. Lubr. Tribol. 68(3), 301–307 (2016) 7. Zhang, Z.: Hydrodynamic Lubrication Theory of Sliding Bearings. Higher Education Press, Beijing (1986)

An Algorithm of 3D Surface Reconstruction of Human Body Based on Millimeter Wave Sensor Guiqing Li1(&), Hanlin Wang1, Tiancai Li2, and Peter Mitrouchev3 1

2

Shanghai Key Laboratory of Intelligent Manufacturing and Robotics, Shanghai University, Shanghai 200072, China [email protected] Shanghai Jingji Photoelectric Technology Co., Ltd., Shanghai, China 3 University Grenoble Alpes, G-SCOP, 38031 Grenoble, France

Abstract. The algorithm of 3D surface reconstruction is proposed to realize the measurement of the dressed person based on millimeter wave sensor, which has the characteristic of being able to penetrate non-conductive materials and can’t penetrate conductive materials. Compared with the manual measurement, the method based on millimeter wave sensor improves the measuring speed and accuracy, and reduces the error caused by human factors, meanwhile millimeter wave is harmless to the human body. A more accurate surface model of human body is obtained by using image recognition and it can be used in clothing design, rapid security inspection and other fields as well. Keywords: Millimeter wave reconstruction


Image recognition
Anthropometry
3D

1 Introduction During the process of 3D reconstruction of the human body surface, how to obtain 3D point cloud data of the human body is related to the accuracy of the reconstruction result. Yuan Lin [1] uses the method of anatomical medicine to obtain the standardized measurement results of Chinese virtual human. Qin Ke [2] studies the contour slice sampling algorithm of 3D human body to obtain the human body contour. Yang Feng [3] studies how to use image recognition to reconstruct the three-dimensional motion of human body. Zhan Yujian [4] uses moire fringe interference to obtain the threedimensional lattice data of the human body model, and uses Java 3D technology to build the three-dimensional human body model. Lin Qinqin [5] uses the twodimensional image of human body to extract the contour, and uses the deformation technology based on the viewpoint to deform the standard human body model to obtain the three-dimensional model of the actual measured human body; Wang Li [6] proposes a human body feature point extraction algorithm based on the human body proportion relationship and Freeman chain code, and constructs the human body contour through the orthogonal two-dimensional image measurement data as the

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2021 Y. Wang et al. (Eds.): IWAMA 2020, LNEE 737, pp. 115–122, 2021. https://doi.org/10.1007/978-981-33-6318-2_14

116

G. Li et al.

measurement point; Wang Xiaoxia [7] studies the reconstruction of three dimensional human body with slice image based on image recognition. The non-contact measurement algorithm based on millimeter wave sensor is analyzed in this paper. The algorithm uses image recognition to eliminate the influence of position of person during the measuring. Data point assembly and slice measurement are used to convert the distance data measured by sensors into three-dimensional data points of human body contour. Finally, the classic Delaunay triangulation algorithm [8] is used to realize the three-dimensional reconstruction of human body. The algorithm is simple, has certain precision, and can achieve the measurement without taking off clothes.

2 Measurement Device The measurement device is designed as shown in Fig. 1, in which 1 is a binocular camera, which is used to obtain the position of the person on the measuring base; 2 is the measuring base, which is used to build a world coordinate system; 3 is a narrow beam millimeter wave sensor, which has multiple evenly distributed in the vertical columns.

Fig. 1. Measurement device

Before measurement starts, the person is required to stand on the base and face the column. The tested person rotates itself clockwise for one cycle. Millimeter wave continuously scan the human body to generated three-dimensional point cloud in order to reconstruct the human body surface.

3 Algorithm The distance between the point on the surface of human the body to the millimeter wave sensor can be obtained by millimeter wave can sensor, however it can’t be used before it is converted from world coordinate system to human body coordinate system. The world coordinate system and the human space as well as the transformation equation must be established. In order to make it, it must be calibrated in space. The calibration is divided into two parts, camera calibration and sensor calibration.

An Algorithm of 3D Surface Reconstruction of Human Body

117

The camera is used to take photos of the person, and the images are used to identify the position where the person stands. In order to construct the equation that converts the person position from image coordinate system to world coordinate system accurately, the calibration needs to be done by several steps shown as follows: (1) A picture of checkerboard is printed and stuck on the base of the as a calibration object. (2) Checkerboard corners are extracted from pictures taken from different directions by the camera. (3) Five internal and six external parameters are estimated in the case of ideal distortion free. (4) The distortion coefficient under the actual radial distortion is estimated by using the least square method. (5) The maximum likelihood method is used to optimize the estimation and improve the estimation accuracy. The calibration of sensors takes the left-bottom of the base plate as the origin of the world coordinate system to establish the X-axis and Y-axis, and to establish the Z-axis in the direction of the column through the origin. The installation position of the sensor can be represented by a unique coordinate point in the world coordinate system. The sensor measurement plane determined by calibration is shown as follows: 8 z ¼ z1 > > < z ¼ z2 ð1Þ . . .. . . > > : z ¼ zn Where n refers to the index of sensors in the vertical direction. zi(i 2 ½1; n) is the ordinate of the sensor labeled i, whose value is obtained by calibration after the completion of equipment design, and remains unchanged. The two-dimensional representation of sensor coordinates can be obtained in any sensor plane, as shown in Fig. 2: pos ¼ ða; 0Þ

ð2Þ

Among them, pos is the two-dimensional coordinate representation of the sensor in the plane, and a is the fixed value of the device design, which is a constant and will not be changed after the determination.

Fig. 2. Diagram of measurement process

118

G. Li et al.

(1) Image processing The boundary between the human body and the base needs to be determined in order to define the position of the person by using image recognition. The boundary can be regarded as an oval. The origin of the person O; is defined as the intersection of the long axis (the connection between the two shoulders) and the short axis (the connection between the front chest and the back). The direction of human facing the positive Y-axis. The data points in the world coordinate system measured by the sensor will eventually be converted to the human body coordinate system O; . The origin of the human body coordinate system is expressed as: O; ðx; yÞ

ð3Þ

In Fig. 2 a is the angle of the human coordinate system relative to the world coordinate system. The angle a can be calculated by several steps: 1. In order to eliminate the influence of the initial position to the test, the test starts when the x-axis of the human body is parallel to the world X-axis for the first time. The rotation angle is set a Set to 0°. h is defined as the angle between the world coordinate X-axis and the human body coordinate Y-axis. The angle a can be calculated by the following formula: a ¼ 0þh

ð4Þ

2. When the X-axis of the human body is parallel to the world Y-axis for the first time, h is defined as the angle between the world coordinate Y-axis and the human body coordinate X-axis. The angle a can be calculated by the following formula: a ¼ 90 þ h

ð5Þ

3. When the X-axis of the human the body coincides with the world X-axis for the second time, h is defined as the angle between the world coordinate X-axis and the human body coordinate X-axis. The angle a can be calculated by the following formula: a ¼ 180 þ h

ð6Þ

4. When the X-axis of the human body coincides with the world Y-axis for the second time, h is defined as the angle between the world coordinate Y-axis and the human body coordinate X-axis. The angle a can be calculated by the following formula: a ¼ 270 þ h

ð7Þ

An Algorithm of 3D Surface Reconstruction of Human Body

119

5. When a ¼ 360 ; test is finished. (2) Data preprocessing The conversion from the world coordinate system to the human body coordinate system can be obtained from the coordinate transformation matrix. From Fig. 3, the translation vector of the two coordinate systems can be expressed as: OO; ¼ ðx; yÞ

ð8Þ

Defined a ¼ 0 when two coordinate systems coincide, so a½0; 360Þ. OP ¼ ða; pÞ

ð9Þ

Where p is the distance measured by the sensor.

Fig. 3. Coordinate data conversion

(3) Data calculation Point P in the human body coordinate system O; can be obtained by multiplying the coordinate value of measurement point P in the world coordinate system by the transformation matrix from the world coordinate system to the human body coordinate system: OP ¼ ða; p; 1Þ 2 3 2 cos h sin h 0 1 O; P ¼ 4 sinh cosh 0 5
4 0 0 0 1 0

ð10Þ 0 1 0

3 x y 5
OP 1

ð11Þ

The resulting vector O; P is the coordinate representation of the measured point P in the human body coordinate system, which is defined as P; , namely: P; ¼ ðP; x ; P; y ; 1Þ

ð12Þ

P; x ; P; y represent the abscissa and ordinate in human coordinate system respectively.

120

G. Li et al.

(4) Point cloud data processing Point to be measured P; The coordinate points in the two-dimensional space are converted and assembled with the Z-axis generated by the sensor plane, the 3D point cloud data can be described as follow:  ;  ; P; 11x ; P; 11y ; z1  P 12x ; P 12y ; z2  ; . . ..; . .  P; 1nx ; P; 1ny ; zn  P; 21x ; P; 21y ; z1  > > P 22x ; P 22y ; z2 > > > > > . . .. . . > > > > . . .. . . >  : ; P mnx ; P; mny ; zn 8 > > > > > > > > > > > > > > > Bk ¼ gCk Tr k > > > > h Q i > > : C ¼ @V ¼ @V @V k @Uc @SOC @X

ð7Þ

@R1 @E @V @E Where, @U ¼1,@SOC ¼ @SOC þ I  @SOC . Ck is the proportional coefficient between c the systematic observation values and the target values.

2.2

The Battery Parameters Identification

In order to obtain accurate battery parameters at the different SOC value, according to the literature [4], we adopt the test method in the paper, and it measures the battery parameters respectively. Then obtain battery model parameters R1, R2 and Cr values, and the results are shown in the Table 1. Table 1. Battery parameters identification value. SOC R1(mX) Discharge Charge R2(mX) Discharge Charge Cr(F) Discharge Charge

0

0.1 0.2 0.4 0.6 0.8 1.0

556 556 804 804 198 208

320 354 659 641 182 152

219 232 302 316 204 215

212 262 214 232 183 176

295 253 245 274 172 178

292 274 275 241 197 182

288 223 298 263 187 147

524

2.3

X. Jiang et al.

Error Analysis and Correction

(1) The effect on SOC caused by electromotive force Electromotive force is an important parameter for Kalman estimation algorithm, the accuracy of electromotive force directly affects the battery SOC estimation results. Generally, in the case of the same battery SOC value, the higher the temperature is, the greater the electromotive force is [5]. In order to reduce the effect on the battery SOC caused by electromotive force, the compensation value Et(T) is proposed, where, T is the ambient temperature. Below the room temperature (25 °C), the value of Et is 0 V, at the other temperatures, the value of Et should subtract the electromotive force value at the 25 °C. So the actual battery electromotive force is E ¼ E þ Et ðT Þ

ð8Þ

As for the Et(T) function, the electromotive forces are measured respectively at different temperatures, and then subtract the battery electromotive forces at 25 °C, the fitting curves can be obtained from the former data. According to measuring result, in the case of same temperature at different SOC, the electromotive force value is almost same as the case at 25 °C. Therefore, in order to simplify the model, we just take the electromotive force compensation value Et(T) into consideration which has the relationship with the ambient temperature. (2) The effect on SOC estimation caused by battery capacity Battery capacity Q is one of the basic parameters of the SOC state equation. Battery cycle life and working conditions can affect the battery capacity in some extent. In addition, the battery capacity Q is a important basis for the Kalman estimation algorithm. With the increase of the cycle number, the battery capacity would decrease [6]. The capacity compensation value is used for correcting the actual battery capacity on different charge-discharge cycles. N is a battery chargedischarge cycle number. The actual battery capacity Q is: Q ¼ Q0  Qn ð N Þ

ð9Þ

Where, Q0 is the rated capacity. (3) The effect on SOC estimation caused by the battery charge-discharge efficiency The charge-discharge efficiency η is an important parameter for the Ah integration, which is included in the Kalman filter matrix B, η is mainly influenced by the battery temperature and charge-discharge current. The battery has efficiency loss during the discharge process, the function η(T) of the battery charge-discharge efficiency is obtained as follows [7]. gðT Þ ¼ QD ðT Þ=QC ðT Þ Where, QD is the discharge capacity, QC is the charge capacity.

ð10Þ

Research on Wireless Sensor Robot Lithium Battery SOC Estimation

525

3 System Application Platform System is mainly composed of mobile robot and handle. System application platform structure diagram is shown as Fig. 2. The system uses the LPC1768 MCU to control the movement of the mobile robot, and collects the data of the sensors module, then sent the information by the Zigbee communication module, at the same time, calculate the battery SOC values through all kinds of information fusion. The handle controller collects the keyboard input information, sends control instructions to the mobile robot by the ZigBee communication module, at the same time, receive mobile robot sensing information and state information, and finally upload the data to the PC through the USB interface.

Power Supply

LCD Steering Gear

Voltage Module Current Module

MCU LPC1768

Temperature Module Battery

Electrically Controller

Motor

Zigbee Module

Sensors Module

Handle Controller

Keyboard and Joystick

Zigbee Module Handle

Mobile Robot

USB

Power Supply

Battery

Fig. 2. System application platform

4 System Test and Error Analysis 4.1

The Test of Battery Capacity Compensation Correction Effect on SOC

After 200 charge-discharge cycles, the battery SOC estimation is simulated in the Matlab environment. The compassion curves of capacity compensation during the battery discharge process are shown as Fig. 3. As is shown in the Fig. 3, at the initial discharge stage, the error is up to 12% without capacity compensation, after capacity compensation, the error is less than 7%. At the end discharge stage, the SOC estimation value almost the same. Because the change of battery voltage is very large at this stage, the value of correction is also large, it has very strong correction effect to SOC. 4.2

The Test of Battery Charging-Discharging Efficiency Correction Effect on SOC

In the case of charge-discharge efficiency errors exist, the battery SOC estimation is simulated environment. The compassion curves of Discharge efficiency compensation value are shown as Fig. 4, without discharge efficiency compensation, error would become larger slowly during the discharge. After discharging efficiency compensation, the error remains at 7%.

526

4.3

X. Jiang et al.

System SOC Estimation Test

The battery would be discharged at different current in the system test, during the test, the battery voltage, current, temperature and other information are collected, after the integrated processing, the battery SOC value is estimated. During the discharge process, the discharge is stopped at scheduled time intervals, until the battery voltage stability, the system measures the electromotive force, and calculate the SOC value as a standard reference value at the open circuit. Finally, the comparison between the calculating SOC values and the standard reference SOC values during discharge is plotted as Fig. 5, at the complex condition, system can accurately estimate the value of SOC; it can effectively correct initial value error, and ensure the system error within 5%.

Fig. 3. Capacity compensation

Fig. 4. Efficiency compensation

Fig. 5. SOC estimation value

5 Conclusion As for the SOC estimation of the lithium batteries, the Thevenin equivalent circuit model structure is easy to implement in the embedded environment. The Kalman filter can ensure a good accuracy during the estimating process; it can effectively correct the initial value error and has a strong suppression effect on the system noise. This method has played a key role in the accurate SOC estimation of the mobile robot platform battery pack and has a great help for its control decision. Acknowledgment. This work is partially supported by Science and Technology Research Project of Hubei Provincial Department of Education under Grant Q20161805.

References 1. Huang, H.L., Savkin, A.V., Ding, M., Huang, C.: Mobile robots in wireless sensor networks: a survey on tasks. Comput. Netw. 148, 1–19 (2019) 2. Xia, C.Y., Zhang, S., Sun, H.T.: A strategy of estimating state of charge based on extended Kalman filter. Chin. J. Power Sources 31(5), 414–417 (2007) 3. Fan, B., Tian, X.H., Ma, J.W.: EKF-based estimation of lithium-ion traction-battery SOC. Chin. J. Power Sources 34(8), 797–799 (2010)

Research on Wireless Sensor Robot Lithium Battery SOC Estimation

527

4. Wang, L.F., Hu, Y.F., Liao, C.L.: Research on SOC estimation of NiMH batteries in hybrid electric vehicles based on improved Thevenin model. Chin. High Technol. Lett. 18(9), 948– 952 (2008) 5. Wen, J.P., Jiang, J.C., Wen, F., Zhang, W.G.: Error analysis of Kalman algorithm in its application to SOC estimation in PEV. Autom. Eng. 32(3), 188–192 (2010) 6. Pop, V.: Accuracy analysis of the state-of charge and remaining runtime determination for lithium-ion batterie. Measurement 10, 1016–1022 (2008) 7. Li, B.S.: Study on Lithium-Ion Battery Temperature Dependent Modeling and State of Charge Estimation in Electric Vehicles. Jilin University, Jilin (2017)

Storage Genetic Scheduling Algorithm Based on Leader Selection Operator Tianshu Zhang1, Xiue Gao1,2(&), Kesheng Wang3, and Shifeng Chen2 1

College of Information Engineering, Dalian University, Dalian, China [email protected] 2 College of Information Engineering, Lingnan Normal University, Zhanjiang, China 3 Department of Mechanical and Industrial Engineering, NTNU, Trondheim, Norway

Abstract. In order to meet the needs of warehouse cooperative scheduling and the problems of slow convergence and prematurity of existing genetic algorithms, we propose a warehouse genetic scheduling algorithm based on leader selection operator. Firstly, the population was divided into male subgroups and female subgroups, and the optimal solution was selected as the female leader and the male leader respectively in two subgroups. Secondly, female leader and male subgroup, male leader and female subgroup are staggered to form a waiting crossover sequence. Finally, an example simulation of the storage genetic scheduling algorithm is designed. The simulation results show that the leader selection operator proposed in this paper is superior to the roulette selection operator and the traversal selection operator in terms of convergence speed and global optimization ability, and improves the efficiency of coordinated storage scheduling. Keywords: Warehouse cooperative scheduling Selection operator


Genetic algorithms


1 Introduction Modern warehouse management relies more and more on the intelligent coordinated scheduling strategy, and genetic algorithm is gradually applied to the field of warehouse automation because of its characteristics such as independent gradient information, strong adaptability and self-learning habit. However, the convergence rate of traditional genetic algorithms is slow and is easily limited to local optima, some improved genetic algorithms have been proposed in succession, mainly from the three aspects of selection, crossover and mutation operation. Improvement of selection operator. Literature [1] proposed a non-dominant sequencing genetic algorithm, which mixed offspring with contemporary populations and selected n individuals with the highest individual fitness value to form a waiting crossover sequence. The sequence crossed to produce the next generation population, and quickly approached the optimal solution. On the basis of literature [1], literature [2] and literature [3] stratify the mixed population according to individual fitness value. At © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2021 Y. Wang et al. (Eds.): IWAMA 2020, LNEE 737, pp. 528–535, 2021. https://doi.org/10.1007/978-981-33-6318-2_66

Storage Genetic Scheduling Algorithm

529

the same time, literature [3] selected different numbers of individuals from different levels to form a waiting cross sequence to produce offspring, which slowed down the search speed and improved the population diversity at the same time. Improvement of crossover operator. Literature [4] proposed difference crossover operator on the basis of multi-parent crossover operator. On the basis of literature [4], literature [5] extends the parent selection method of difference crossover operator to further improve the convergence rate of genetic algorithm. Literature [6] proposed that crossover operators at different evolutionary stages exchange different gene numbers. Literature [7] proposed that the number of genes in cross exchange and genetic algebra were normally distributed. Literature [8] proposed that the number of cross-exchanging genes should be supplemented in the early stage of evolution to improve the global optimization ability, while the number of cross-exchanging genes should be reduced in the late stage of evolution to protect good genes. Improvement of mutation operator. Literature [9] proposed a strategy to control mutation points, which correspond to the pre-middle and late stages of evolution. Mutation points occur at high, middle and low positions of chromosomes. Literature [10,11] proposed a linear adaptive mutation operator to improve the convergence rate of genetic algorithm. Literature [12] proposes a curvilinear adaptive mutation operator, which assigns individual mutation probability according to hyperbolic tangent function, and the evolution of excellent individuals does not stagnate, which further improves the convergence speed and global optimization ability of genetic algorithm. The selection operator is the most important step in genetic algorithm, which produces the population waiting crossover sequence and directly affects the sufficiency of gene exchange and the diversity of offspring. Quasi to the problem that the existing selection operator is difficult to preserve the diversity of the descendant population, we proposed a leader selection operator in the light of animal evolution, in order to ensure the sufficiency of gene exchange between the population and the descendant diversity, so as to improve the global optimization ability and convergence speed of the genetic algorithm.

2 Genetic Algorithm Based on Leader Selection Operator 2.1

Leader Selection Operator

The process of leader selection operator is described as follows: ①The fitness values of all individuals in the population are calculated and stored. ② The population was divided into two subgroups according to odd and even rows, the odd behavior male subgroup and the even behavior female subgroup. The initial population diagram is shown in Fig. 1. Red represents the male subgroup, blue represents the female subgroup, and AA, BB, CC, DD and EE are some of the optimal solution modules. The more modules an individual contains, the greater the fitness value will be. ③ Referring to the stored individual fitness value, the individual with the highest fitness value in the male subgroup is selected as the male leader, and the individual with the highest fitness value in the female subgroup is selected as the female leader. In Fig. 1, S5 is the male leader and S10 is the female leader. ④ Copy the male leader and all females in a

530

T. Zhang et al.

staggered arrangement, copy the female leader and all males in a staggered arrangement, forming a waiting crossover sequence. The waiting crossover sequence is shown in Fig. 2. S1–S16 is the male leader and all female individuals staggered, while S17– S32 is the female leader and all male individuals staggered.

Fig. 1. Initial population

2.2

Fig. 2. Waiting crossover sequence

The Flow of Genetic Algorithm Based on Leader Selection Operator

The overall process of the algorithm is shown in Fig. 3, which is described as follows: Step 1: Initialize the population. The initial population containing n individuals was randomly generated according to the constraint conditions, and the fitness of the individuals was calculated and stored. Step 2: Form a waiting cross sequence. The leader selection operator is used to select the parent generation in the current population and form a waiting crossover sequence. Step 3: Produce offspring. Two - point crossover operator and one - point mutation operator are used to cross and mutate the parent generation to generate new offspring population. Step 4: Produce the next generation. The offspring population will be recombined with the current population to produce the next generation population in a certain proportion according to fitness ordering. Step 5: Judge algorithm terminates. If the current algebra is greater than the maximum genetic algebra, the algorithm is terminated and the optimal solution is output; otherwise, Step 2 is returned.

Storage Genetic Scheduling Algorithm

531

Fig. 3. The flow of genetic algorithm based on leader selection operator

3 Example Analysis 3.1

The Simulation Scene

The simulation scenario is shown in Fig. 4. When the warehouse robot is working, it starts from the shelf and moves the shelf to the designated picking station. After waiting for the workers to pick up the goods, the robot carries the goods shelf back to the original position, and then goes to the next shelf to perform the task according to the task sequence. In order to achieve the goal of collaborative scheduling, the task sequence with the minimum total moving steps of the robot should be solved. The task information is shown in Table 1, in which the coordinates of no. 1, 2, 3, 4 and 5 picking stations are (0,10), (0,30), (0,50), (0,70) and (0,90) respectively. Table 1. Task information Task number 1 2 3 4 5 6 7 8 9 10

Shelves coordinates (20,30) (32,15) (38,23) (29,48) (64,44) (70,45) (80,43) (90,01) (99,10) (97,34)

Picking station number 2 1 3 2 4 5 4 2 1 5

Task number 11 12 13 14 15 16 17 18 19 20

Shelves coordinates (31,100) (32,97) (36,82) (42,97) (46,87) (72,75) (77,98) (90,82) (95,68) (76,74)

Picking station number 2 4 3 1 5 3 4 2 1 3

532

T. Zhang et al.

Fig. 4. The simulation scene

3.2

Related Parameters

Population individuals adopt decimal coding, consisting of task numbers in the order of task execution, represents a possible sequence of tasks. Relevant parameters of simulation are shown in Table 2. Table 2. Related parameters Computer information 2.80 GHz CPU, 8.00 GB RAM

3.3

The simulation software Matlab 2018

The length of the individual 20

Population size 120

Maximum genetic algebra 300

Mutation Crossover probability probability 0.1

1

Analysis of Simulation Results

This paper compares the leader selection operator with the roulette selection operator and the traversal selection operator in terms of the optimal solution value, the standard deviation of offspring, the standard deviation of population, the mode value and the mean value. Optimal solution value analysis. The global optimal solution value is shown in Fig. 5, where the abscissa represents algebra and the ordinate represents the optimal solution value. It can be seen from Fig. 5 that all the three operators show a convergence trend. The leader selection operator has the fastest convergence speed, converges 40 generations ago and has the lowest optimal solution value. The convergence speed of roulette selection operator and ergodic selection is slow, both of which converge after more than 120 generations, and the optimal solution value is higher than that of the leader selection operator. This shows that the operator in this paper is superior to the roulette selection and traversal operator in terms of global optimal solution optimization ability and convergence speed.

Storage Genetic Scheduling Algorithm

533

Fig. 5. Optimal solution value

Standard deviation analysis of offspring and population. The standard deviation of offspring and population are shown in Fig. 6 and Fig. 7, where the abscissa represents algebra and the ordinate represents the standard deviation value. As can be seen from Fig. 6 and Fig. 7, the standard deviation of the offspring and the standard deviation of the population show a converge trend as a whole. The standard deviation of the offspring converges to a certain value due to the action of mutation operator, and the standard deviation of the population gradually converges to zero. The trend of standard deviation of roulette selection operator and traversal selection operator is roughly the same. The standard deviation of the operator in this paper is higher than that of the roulette selection operator and the traversal selection operator at the initial stage of evolution, and lower than that of the roulette selection operator and the traversal selection operator at the later stage of evolution. This shows that in the early stage of evolution, the diversity of offspring and population of the operators in this paper is higher, which is conducive to global optimization. At the later stage of evolution, the offspring and population changes of the operators in this paper are small, which is beneficial to the protection of the optimal solution.

Fig. 6. Standard deviation of offspring

Fig. 7. Standard deviation of population

534

T. Zhang et al.

Mode and mean analysis. The mode values and mean values of the populations are shown in Fig. 8 and Fig. 9, where the abscissa represents algebra and the ordinate represents the values of the modes and means. As can be seen from Fig. 8 and Fig. 9, the mode value and the mean value of the population show a convergence trend as a whole. The leader selection operator has the minimum value and the convergence speed is obviously faster than the roulette selection operator and the traversal selection operator. This shows that the overall performance of the operator in this paper is substantially better than the roulette selection operator and the traversal selection operator.

Fig. 8. Mode

Fig. 9. Mean

4 Conclusion In view of the synergetic nature and task planning of warehouse scheduling, we propose a warehouse genetic scheduling algorithm based on leader selection operator, which improves the convergence speed and global optimization ability of warehouse genetic scheduling algorithm, and improves the efficiency of warehouse collaborative scheduling. In the next step, we will further study the adaptability of selection operators and cluster scheduling strategy to improve the overall performance of the warehouse genetic scheduling algorithm.

References 1. Deb, K., Pratap, A., et al.: A Fast and Elitist Multiobjective Genetic Algorithm:NSGA-II. IEEE Trans. Evol. Comput. 6(2), 182–197 (2002) 2. Ahmadi, V., Jalili, A., Khorramizadeh, S., et al.: A hybrid NSGA-II for solving multiobjective controller placement in SDN. In: International Conference on Knowledgebased Engineering & Innovation. IEEE (2016) 3. Wang, Q., Xie, X., Zhou, G.: An improved genetic algorithm for nondominant sorting. Inf. Tech. Netw. Secur. 38(05), 28–32 (2019)

Storage Genetic Scheduling Algorithm

535

4. Ali, M., Awad, N., Suganthan, P., et al.: An improved class of real-coded genetic algorithms for numerical optimization. Neurocomputing 275(1), 155–166 (2018) 5. Liu, Z., Wang, X., Xue, L., et al.: The improvement and parallelization of real coding genetic algorithm. J. Hebei Univ. (Nat. Sci. Edn.) 39(1), 86–92 (2019) 6. Pasandideh, S., Niaki, S., Asadi, K.: Bi-objective optimization of a multi-product multiperiod three-echelon supply chain problem under uncertain environments: NSGA-II and NRGA. Inf. Sci. 292(1), 57–74 (2015) 7. Lu, Y., Zhao, C., Wu, X., et al.: Multi-objective optimization method based on improved the NSGA - II research. Comput. Appl. Res. 35(06), 139–143 (2018) 8. Shi, S., Zhang, X., Wang, Z.: An improved NSGA - II algorithm based on DCD and a tDX. Comput. Simul. 36(12), 257–262 (2019) 9. Wang, C.: Improvement of genetic algorithm mutation operator. J. Shandong Agric. Univ. (Nat. Sci. Edn.) 50(5), 898–901 (2019) 10. Bai, X., Deng, G.: Research on an improved multi-objective Genetic algorithm. Light Ind. Sci. Tech. 33(9), 67–68 (2017) 11. Yang, C., Qian, Q., Wang, F., et al.: An improved adaptive genetic algorithm for function optimization. In: IEEE International Conference on Information and Automation (ICIA). IEEE (2016) 12. Feng, J., Yin, F., Huang, G.: A heteromorphic improved adaptive genetic algorithm is presented. J. Southwest Univ. National. (Nat. Sci. Edn.) 45(5), 511–516 (2019)

Simulation Study on the Mechanical Influence of Different Diameters of Artificial Hip Joints Zikai Hua(&), Hui Liu, and Xiuling Huang School of Mechatronics Engineering and Automation, Shanghai University, Nanchen Road, Shanghai 200072, P.R. China [email protected]

Abstract. With the increase of life expectancy of the aging population and more and more young patients need surgical treatment of arthritis, there is an urgent need to improve the durability and clinical life expectancy of artificial hip joint. In this study, the range of motion (ROM), tribological performance and mechanical properties of artificial hip joints of different sizes were studied and compared by means of finite element analysis and computer aided design. We found that the larger femoral head has greater ROM and better mechanical properties. However, a larger head can cause serious wear problems, which can lead to aseptic loosening. Considering these three factors, a systematic analysis method of artificial hip joints is set up, and the range of 38 mm to 42 mm is believed to have better performance. In future work, more detailed factors, such as lubrication conditions, need to be considered in order to get a more accurate analysis. Keywords: Artificial hip joint FEA


ROM
Mechanical property
Biotribology


1 Introduction With the development of hip prostheses, aseptic loosening [1] and dislocation [2] are regarded as the two main problems of the artificial hip joints. Aseptic loosening is mainly due to occurrence of osteolysis. It is well known that the immune response of the human body to the wear debris of the prosthesis is the root cause of osteolysis, such as ultra high molecular weight polyethylene (UHMWPE). As a result, the lifetime of the most artificial joints has been shortened a lot [3–5]. As for dislocation, the incidence rate in hip arthroplasty is 2%–3%, which is the second most common complication [6] following a primary THA and 10% in revision arthroplasty [7]. The feasible range of motion (ROM) of the implant is a function of the effective location of the implant (affected by the preoperative and intraoperative environment) and the technical ROM (affected by the manufacturer's implant design). Obviously, large ROM provides stability and convenience to the THA patients. It is well recognized that larger femoral head and acetabular will provide a relatively large motion range, so that the stability of the artificial hip joints will be improved. However, the friction torque will increase with the increase of femoral head diameter, resulting in more serious acetabular wear. Therefore, these two problems relate to each other and can not be considered separatedly. © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2021 Y. Wang et al. (Eds.): IWAMA 2020, LNEE 737, pp. 536–543, 2021. https://doi.org/10.1007/978-981-33-6318-2_67

Simulation Study on the Mechanical Influence

537

In this study, ROM, tribological performance and mechanical property of artificial hip joints in different size are investigated and compared via finite element analysis (FEA) and CAD program. The ultimate goal of this study is to analyze the relationship between these three factors and to provide a relatively ideal range of artificial hip joint size. Better performance can be obtained by referring to this range when designing the prosthesis.

2 Mechanical Analysis 2.1

FEA Model

The mechanical properties of artificial hip joints with different sizes (diameter 28– 56 mm) are simulated by finite element method.It is assumed that the femoral head and the acetabulum in the artificial hip joint are concentric (Fig. 1), in which the material of the femoral head is CoCrMo, elastic modulus of 2.3e+5 MPa, Poisson's ratio of 0.3, elasticity, while the acetabular material is ultra-high molecular weight polyethylene, the elastic modulus of 1.4e+3 MPa, Poisson's ratio of 0.46, elastic-plastic. ANSYS® software is used for finite element analysis. In order to model more accurately, SOILD185 element is used to mesh the whole model, and CONTA174 and TARGE170 elements are used on the contact surface.

Fig. 1. Finite element model

2.2

Boundary and Loading Condition

According to the forces condition on the natural hip joint [8], the simplified load is expressed by three vectors in Cartesian coordinate system. In this study, based on the right hip joint, the forces on the hip joint were systematically analyzed by analyzing the movements of four different hip joints, including standing, heel down, toe off, and rising [9]. The specific parameters are shown in Table 1 and Fig. 2. 0 1 0 x 1 B C B0  ByC B RotðX; 45 Þ@ A ¼ @ z 0 1 0

0 cos 45 sin 45 0

0 sin 45 cos 45 0

10 1 0 x B C 0C CB y C 0 A@ z A 1 1

ð1Þ

538

Z. Hua et al. Table 1. Typical contact force of hip in different movements in human daily life Motion Standing Heel down Toe off Ascending

(a)

Load 1/3 BW 4.64 BW 4.33 BW 7.7 BW

Directions Z 24.4° to XY plane, 30° to YZ plane 12.1° to XY plane, 28.7° to YZ plane

(b)

Fig. 2. Schematic diagram of peak load in gait movement (a) Heel down, (b) Toe off

For the boundary conditions of the finite element model, according to the installation of the femoral handle of the hip joint in the human body, we completely fixed the neck of the hip joint.

3 Results and Discussion 3.1

Mechanical Property

In this study, the finite element analysis method is used to analyze and compare the stress of artificial hip joint under different sizes and different motion conditions. The results are shown in Fig. 3, Fig. 4 and Fig. 5. From the picture, we can see that in these four motion states, the mechanical parameters of the femoral head decrease as the diameter of the femoral head increases. In addition, one thing that deserves our attention is that we find that the mechanical properties of the artificial hip joint are not proportional to the size of the femoral head. Therefore, taking into account the economy-related problems, choosing a range of the size of the femoral head is valuable for the design of the prosthesis.

Simulation Study on the Mechanical Influence

539

By analyzing the finite element results, we can predict that as the size of the femoral head increases, the improvement of its mechanical properties may cause bottlenecks. Studies have found that femoral heads with a diameter of 36 mm to 52 mm can effectively improve mechanical properties.

Fig. 3. Maximum von Mise stresses of the UHMWPE cup

Fig. 4. Maximum von Mise strain of the UHMWPE cup

540

Z. Hua et al.

Fig. 5. Maximum deformation of the UHMWPE cup

3.2

Kinematic Performance

In the hip prosthesis model, the range of motion of the impact is determined by 3DCAD simulation. In Fig. 6, it is found that the ROM is enlarged with the decreasing of the neck diameter. In this study, considering the hip prosthesis used in clinical, the femoral heads ranging from 28 mm−56 mm is simulated, combining with 6mm or 11mm neck diameters. A safety coefficient at 1.4 is used in the simulation, which provides a relatively safe ROM and prevents the ultra motion. Therefore, the final ROM shown below is 1.4 times less than the simulation results. In Fig. 7, it is found that the ROM increases 15.67% totally, if the head/neck ratio increases from mininum ratio, 2.55 (28 mm/11 mm head/neck), to maxinum ratio, 4.67 (56 mm/6 mm head/neck). As to the model with 6mm neck diameter, it is obtained that ROM increases about 8.39% through the range of 28 mm to 56 mm, while 17.16% increasing with 11 mm neck diameter. However, in the two cases, ROM has just increased only 0.6% and 0.64% when the head diameter changes from 52 mm to 56 mm. In Fig. 7, the femoral head which is in 28 mm–44 mm diameter may give efficient improvement of kinematic performance. In the kinematic study, it is also found that ROM of the artificial hip joints is not proportion to the dimension of the femoral head, or the head/neck ratio, which somehow agrees to the results of the mechanical study. And it indicates that there may be an optimization of the structure and dimension of the artificial hip joints, considering the functionality.

Simulation Study on the Mechanical Influence

541

(a)

(b) Fig. 6. Relationship of the neck size and the kinetic motion range (a)-28 mm femoral head, (b)56 mm femoral head

Fig. 7. Improvements in ROM with different collocation of femoral head and neck 1– 6 mm neck diameter, 2–11 mm neck diameter

3.3

Tribological Analysis

The triobological property may determine the lifetime of the artificial hip joints in long terms. Based on the overview of the tribological study on the artificial hip joints, combining with the mechanical and kinematic properties discussed above, the tribological performance of the artificial joints in different size is analyzed. In Fig. 8, it is found that the wear amount of the UHMWPE acetabular cup increases with the diameter of the femoral head of the air lubricant [11]. Comparing

542

Z. Hua et al.

with wear volume of the 22 mm femoral heads, the wear volume of the 28 mm femoral heads is 1.5 times as 22 mm, while the 56 mm gives 3.5 times. There seems to be a linear relationship between the wear volume and the diameter of the femoral heads in Fig. 8. And the femoral heads ranging from 38 mm to 42 mm may give low increasing in wear volume. But it should be pointed out that the lubrication of the artificial hip joints can not be ignored, because proper lubrication will provide a remarkable antiwear performance of the joints [12, 13]. Furthermore, the limitation of the wear volume which will cause the osteolysis still need further investigation.

Fig. 8. Wear performance of the UHMWPE acetabulum cup in different sizes [10]

Therefore, considering the mechanical and kinematic performance of the artificial hip joints, the artificial hip joints range from 38 mm to 42 mm may provide better functional performance.

4 Conclusion In this study, the range of motion (ROM) and mechanical property of artificial hip joints in different size are investigated and compared. According to the FEM and CAD results, it is found that larger femoral heads provide more ROM and perform better mechanical property. However, it is found that the improvement of the properties is not proportion to the increasing of diameter of the femoral heads. The size ranging from 36 mm–52 mm and 28 mm–44 mm are believed to have efficient improvement of mechanical performance and kinematic performance. Furthermore, due to the severe wear problem caused by the larger heads, the tribological performance of the artificial hip joints is also compared. Considering these three factors, a systematic analysis method of artificial hip joints is set up, and the range of 38 mm to 42 mm is believed to have better functional performance. In the future work, more detailed factors such as lubrication conditions, bionic lubricant should be considered in order to get more precise analysis.

Simulation Study on the Mechanical Influence

543

References 1. Mckellop, H.A., Campbell, P., Park, S.H., Schmalzried, T.P., Sarmiento, A.: The origin of submicron polyethylene wear debris in total hip arthroplasty. Clin. Orthop. Relat. Res. 311 (311), 3–20 (1995) 2. D’Lima, D.D., Urquhart, A.G., Buehler, K.O., Walker, R.H., Colwell, C.W.: The effect of the orientation of the acetabular and femoral components on the range of motion of the hip at different head-neck ratios*. J. Bone Joint Surg. 82(3), 315–321 (2000) 3. Essner, A., Schmidig, G., Wang, A.: The clinical relevance of hip joint simulator testing: in vitro and in vivo comparisons. Wear 259(7), 882–886 (2005) 4. Harris, W.H.: The problem is osteolysis. Clin. Orthop. Relat. Res. 311(311), 46–53 (1995) 5. Su, S.H., Hua, Z.K., Zhang, J.H.: Design and mechanics simulation of bionic lubrication system of artificial joints. J. Bionic Eng. 3, 155–160 (2006) 6. Giurea, A., Zehetgruber, H., Funovics, P., Grampp, S., Gottsauner-Wolf, F.: Risk factors for dislocation of a cementless total hip endoprosthesis–a statistical analysis. Zeitschrift für Orthopädie 139(3), 194–199 (2001) 7. Morrey, B.F.: Difficult complications after hip joint replacement dislocation. Clin. Orthopaed. Relat. Res. 344(344), 179–187 (1997) 8. Hurwitz, D.E., Foucher, K.C., Andriacchi, T.P.: A new parametric approach for modeling hip forces during gait. J. Biomech. 36(1), 113–119 (2003) 9. Costigan, P.A., Deluzio, K.J., Wyss, U.P.: Knee and hip kinetics during normal stair climbing. Gait Post. 16(1), 31–37 (2002) 10. Yoshimine, F.: The influence of the oscillation angle and the neck anteversion of the prosthesis on the cup safe-zone that fulfills the criteria for range of motion in total hip replacements. The required oscillation angle for an acceptable cup safe-zone. J. Biomech. 38(1), 125–132 (2005) 11. Ebramzadeh, E., Sangiorgio, S.N., Lattuada, F., Kang, J.S., Chiesa, R., Mckellop, H.A., et al.: Accuracy of measurement of polyethylene wear with use of radiographs of total hip replacements. J. Bone Joint Surg. Am. 85A(12), 2378–2384 (2003) 12. Xie, X.L., Tang, C.Y., Chan, K.Y.Y., Wu, X.C., Tsui, C.P., Cheung, C.Y.: Wear performance of ultrahigh molecular weight polyethylene/quartz composites. Biomaterials 24(11), 1889–1896 (2003) 13. Wang, A., Essner, A., Polineni, V.K., Stark, C., Dumbleton, J.H.: Lubrication and wear of ultra-high molecular weight polyethylene in total joint replacements. Tribol. Int. 31(1–3), 17–33 (1998)

Effect of 3D Printing Orthoses on Hand Edema Rehabilitation of Stroke Patients Zikai Hua, Jiali Dai, Yikang Shen, and Xiuling Huang(&) School of Mechatronic Engineering and Automation, Shanghai University, Shanghai, China [email protected]

Abstract. This paper provides an optimized rehabilitation orthosis for wrist and hand made by 3D printing. A trial group of stroke patients with hand edema was involved in this study on the effect of the orthosis for the 6-week duration. The effect of the orthoses on the hand edema was studied by measuring seven section sizes with a 3D scanner before and after wearing orthoses, including the thickness of the palm, the perimeter of the wrist, interphalangeal joint of the thumb and proximal interphalangeal of four fingers. The results showed that there were significant differences in various dimensions of the hand before and after wearing orthoses (P  0.005). This indicates that 3D printing orthoses have a positive effect on reducing hand edema. Keywords: 3D printing


Orthoses
Edema
Stroke
Measurement

1 Introduction Stroke is one of the most serious neurological diseases worldwide. Upper extremity dysfunction is one of the complications of stroke patients. Consequently, hand edema is a common problem in upper extremity dysfunction. The cause of hand edema remains unclear. If patients are not well treated in time, they may suffer the loss of hand function, and finally irreversible disability [1]. There are many treatments for hand edema, such as splints, compression therapy, cryotherapy, neuromuscular electrical stimulation [2], et al. However, stroke patients still require more effective and convenient ways to help them to rehabilitate in their daily life [3]. Moreover, in the clinical evaluation of the stroke-induced limb edema needs more accurate methods. Currently, there are some methods used by rehabilitation medical men in limb edema measurement. Anthropometric [4] measurements are acceptable for changes of macroscopic limbs, but it has poor repeatability and low measurement accuracy. Water displacement [5] has high stability, but cannot be used when there is an open wound; Magnetic Resonance Imaging (MRI), ultrasound measurement and computed tomography (CT) can also be used to measure limbs, but they are expensive and time-consuming. In recent years, Hameeteman [6] and Lu [7] conducted related research, indicating that the 3D optical measurement can be used for measuring stroke patients safely. This paper provides optimized rehabilitation orthoses for wrist and hand made by 3D printing and a group of stroke patients with hand edema was tested for 6 weeks. © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2021 Y. Wang et al. (Eds.): IWAMA 2020, LNEE 737, pp. 544–551, 2021. https://doi.org/10.1007/978-981-33-6318-2_68

Effect of 3D Printing Orthoses on Hand Edema Rehabilitation

545

A 3D optical scanner is used to scan their hands before and after wearing their orthoses by comparing seven section sizes to evaluate the therapeutic effect of orthoses on the hand edema.

2 Method 2.1

Subjects

A total of 10 stroke subjects, aged between 35 and 80 years, were enrolled in the trial, including 7 males and 3 females. All subjects met the diagnostic criteria of this trial, including having the onset of stroke from 3 to 6 months, limb hemiplegia from the first-ever stroke, hand edema, and the ability to understand command actions. Participants will be excluded if they had a history of motor dysfunction, history of surgical correction of wrist hemiplegia, pressure sores, and enrolment in any other clinical trials or drug studies. 2.2

3D Printing Orthoses

Under the circumstances of the morbidity, rehabilitation orthosis is a kind of device which can achieve the desired effect by adjuvant therapy. The wear varies from person to person and engineering means are used for correction and fixation to relieve the pain.

Fig. 1. 3D printing orthosis

Fig. 2. 3D optical measurement

In this study, optimized orthoses for the wrist and hand made by stereolithography (SLA) technology were designed for all subjects. SLA is one of the most common 3D printing methods [8] which greatly simplifies the process from design to physical products [9]. Orthoses keep patients’ wrists and hands in a resting position. Contrary to conventional treatments, orthoses have the advantages of a high fit, excellent strength, ease of wear, and lighter weight, so it can replace rehabilitation products such as plasters. What’s more, visiting time is reduced and safe distances are maintained, and the ultraviolet curable resin used as a material of orthoses can be sterilized and reused. The final appearance of the orthosis is shown in Fig. 1.

546

2.3

Z. Hua et al.

Measurement System

The scanning system used in this study was provided by Creaform Co. Ltd., including a scanner and the software. This portable scanner with a precision of 0.05 millimeters (mm) and a resolution of 0.1 mm can fast scan objects from 30 cm to 3 m in size. Compared with other methods, 3D optical measurement has the advantages of high precision, easy usability, short time, and high repeatability, and it can be used in most cases. Even if patients have trauma and cannot be immersed in water, they can be measured by 3D optical scanners. 2.4

Trial Design

After enrolment, each subject was customized with 3D printing rehabilitation orthoses that completely fit the shape of their hands and wrists. Before the trial (week 0), the data of hands and forearms without wearing orthoses was acquired by the scanner. During the trial, according to their doctors' suggestion, participants wore orthoses for about 6 h every day for 6 weeks. After the trial (week 6), the same data was obtained after wearing orthoses.

Screening and Enrolment

Scan

Data Repairing

Modeling

3D printing

No

Data Analysis

Scan Week 6

Scan Week 0

Yes Wear

Postprocessing

Fig. 3. Flow chart of the trial

The scanner should be calibrated prior to the scanning. At the same time, subjects should stretch their hands to relieve the flexion and spasm so that hands can be extended as far as possible, with the help of a rehabilitation trainer if necessary. During the scanning, the patient sat in the chair and put the forearm on a scanning auxiliary device with the palm is downward. The whole scanning can be completed within 2 min. The process of scanning is shown in Fig. 2. The flow chart of the trial is shown in Fig. 3.

3 Results 3.1

Data Processing

Due to wrists, palms and fingers of the hemiplegia [10] after stroke will have more obvious edema, so seven section sizes are chosen to measure, including the thickness of

Effect of 3D Printing Orthoses on Hand Edema Rehabilitation

547

the palm, the perimeter of the wrist, the perimeter of interphalangeal joint of the thumb and the perimeters of proximal interphalangeal of four fingers, as shown on the left of Fig. 4. The measurement methods of each part are based on GB/T 16252-1996. The data at week 0 and week 6 is imported into the reverse engineering software. In order to reduce the measurement error, each selected part is measured three times with an interval of 3 mm. Then the average value of the three measurements is taken as the final value. The right of Fig. 4 shows the degree of edema of a subject’s hand. The data of the bar chart shows distances of the hemiplegic upper extremity before and after wearing the orthosis. From the wrist to fingers, the color changes from green to red, indicating that the degree of resolution of edema is gradually increasing.

Fig. 4. Measurement and degree of edema

3.2

Fig. 5. Outlier test

Statistical Analysis

The measurement data of each part before and after wearing orthoses was analyzed by the statistical software, and the method was performed by paired sample t-test. A significance level of a = 0.05 was taken, and P < 0.05 was considered to be statistically significant. Before the paired sample t-test, a hypothesis test of the normality and outliers is performed on the difference between the paired data of two groups to determine whether the data satisfies a precondition of the paired sample t-test. Outliers refer to some values that differ greatly from other values in the difference, which affect the mean and standard deviation of the group of the difference. This may have a great negative impact on the final statistical results. For small samples, the impact of outliers is particularly significant, so it is necessary to check for significant outliers in the group of the difference. The results of the hypothesis normality test of difference are shown in Table 1. The numbers 1 to 7 respectively represent the wrist, thumb, index finger, middle finger, ring finger, little finger, and palm.

548

Z. Hua et al. Table 1. Difference normality test Group of difference Shapiro-Wilk normality test Statistics Degree of freedom 1 0.767 10 2 0.848 10 3 0.944 10 4 0.942 10 5 0.987 10 6 0.980 10 7 0.772 10

Significance 0.006 0.056 0.602 0.580 0.992 0.966 0.007

The results of the outlier test are shown in Fig. 5, in which the data points more than 1.5 times the length of the box body from the edge of the box are abnormal values, which are represented by circles (°). The data points more than 3 times the length of the box body from the edge of the box are the extreme abnormal values, represented by asterisks (*). It can be seen from Table 1 that the significance level for the difference of wrist is P = 0.006, and that of the palm is P = 0.007, both of which are less than 0.05, so the difference between the wrist and the palm does not conform to normal distribution. Besides, it can be seen from Fig. 5 that there are abnormal values in the corresponding differences of the wrist, thumb, and palm. To sum up, the nonparametric Wilcoxon signed-rank test was used for the measurement corresponding to the difference between the wrist, palm, and thumb. The statistic was Z-value. The paired sample t-test was conducted for other measurement sizes that met the hypothesis test. The statistical analysis results of the measured dimensions of each part are shown in Table 2, 3, 4 and 5.

Table 2. Wilcoxon signed-rank test for measurements of the wrist, thumb, and palm Parts Time N M/mm Perimeter of 1 Week 0 10 183.15 Week 6 10 180.31 Perimeter of 2 Week 0 10 72.68 Week 6 10 65.59 Thickness of 7 Week 0 10 31.06 Week 6 10 29.16 * N: Number, M: Mean, SD: Standard

SD Z-value 15.78 −2.803 15.78 6.66 −2.803 4.33 3.75 −2.803 3.73 deviation

P-value 0.005 0.005 0.005

Effect of 3D Printing Orthoses on Hand Edema Rehabilitation

549

Table 3. Paired sample statistics of the measurements of four fingers Parts Time Perimeter of 3 Week 0 Week 6 Perimeter of 4 Week 0 Week 6 Perimeter of 5 Week 0 Week 6 Perimeter of 6 Week 0 Week 6 * SEM: Standard error of

N M/mm 10 68.17 10 62.46 10 69.51 10 62.02 10 66.17 10 59.32 10 59.26 10 52.89 mean

SD 6.13 5.53 6.44 6.28 6.69 5.88 6.23 3.12

SEM 1.94 1.75 2.03 1.99 2.12 1.86 1.97 0.99

Table 4. Paired sample correlation of four fingers Paired samples (Week 0 & 6) N Correlation P-value Perimeter of 3 10 0.808 0.005 Perimeter of 4 10 0.814 0.004 Perimeter of 5 10 0.898