Internet Computing and Internet of Things [1 ed.] 9781683925736, 9781601325037

Citation preview

WORLDCOMP’19

PROCEEDINGS OF THE 2019 INTERNATIONAL CONFERENCE ON PROCEEDINGS OF THE 2019 INTERNATIONAL CONFERENCE ON INTERNET COMPUTING & INTERNET OF THINGS INTERNET COMPUTING & INTERNET OF THINGS

Internet Computing and Internet of Things

ICOMP’19 Editors Hamid R. Arabnia Leonidas Deligiannidis, Fernando G. Tinetti

U.S. $109.95 ISBN 9781601325037

10995

EMBD-ICOMP19_Full-Cover.indd All Pages

Arabnia

9 781601 325037

Publication of the 2019 World Congress in Computer Science, Computer Engineering, & Applied Computing (CSCE’19) July 29 - August 01, 2019 | Las Vegas, Nevada, USA https://americancse.org/events/csce2019

Copyright © 2019 CSREA Press

18-Feb-20 5:34:46 PM

This volume contains papers presented at the 2019 International Conference on Internet Computing & Internet of Things. Their inclusion in this publication does not necessarily constitute endorsements by editors or by the publisher.

Copyright and Reprint Permission Copying without a fee is permitted provided that the copies are not made or distributed for direct commercial advantage, and credit to source is given. Abstracting is permitted with credit to the source. Please contact the publisher for other copying, reprint, or republication permission.

American Council on Science and Education (ACSE)

Copyright © 2019 CSREA Press ISBN: 1-60132-503-7 Printed in the United States of America https://americancse.org/events/csce2019/proceedings

Foreword It gives us great pleasure to introduce this collection of papers to be presented at the 2019 International Conference on Internet Computing and IoT (ICOMP’19), July 29 – August 1, 2019, at Luxor Hotel (a property of MGM Resorts International), Las Vegas, USA. The preliminary edition of this book (available in July 2019 for distribution on site at the conference) includes only a small subset of the accepted research articles. The final edition (available in August 2019) will include all accepted research articles. This is due to deadline extension requests received from most authors who wished to continue enhancing the write-up of their papers (by incorporating the referees’ suggestions). The final edition of the proceedings will be made available at https://americancse.org/events/csce2019/proceedings . An important mission of the World Congress in Computer Science, Computer Engineering, and Applied Computing, CSCE (a federated congress to which this conference is affiliated with) includes "Providing a unique platform for a diverse community of constituents composed of scholars, researchers, developers, educators, and practitioners. The Congress makes concerted effort to reach out to participants affiliated with diverse entities (such as: universities, institutions, corporations, government agencies, and research centers/labs) from all over the world. The congress also attempts to connect participants from institutions that have teaching as their main mission with those who are affiliated with institutions that have research as their main mission. The congress uses a quota system to achieve its institution and geography diversity objectives." By any definition of diversity, this congress is among the most diverse scientific meeting in USA. We are proud to report that this federated congress has authors and participants from 57 different nations representing variety of personal and scientific experiences that arise from differences in culture and values. As can be seen (see below), the program committee of this conference as well as the program committee of all other tracks of the federated congress are as diverse as its authors and participants. The program committee would like to thank all those who submitted papers for consideration. About 60% of the submissions were from outside the United States. Each submitted paper was peer-reviewed by two experts in the field for originality, significance, clarity, impact, and soundness. In cases of contradictory recommendations, a member of the conference program committee was charged to make the final decision; often, this involved seeking help from additional referees. In addition, papers whose authors included a member of the conference program committee were evaluated using the double-blinded review process. One exception to the above evaluation process was for papers that were submitted directly to chairs/organizers of pre-approved sessions/workshops; in these cases, the chairs/organizers were responsible for the evaluation of such submissions. The overall paper acceptance rate for regular papers was 21%; 15% of the remaining papers were accepted as poster papers (at the time of this writing, we had not yet received the acceptance rate for a couple of individual tracks.) We are very grateful to the many colleagues who offered their services in organizing the conference. In particular, we would like to thank the members of Program Committee of ICOMP’19, members of the congress Steering Committee, and members of the committees of federated congress tracks that have topics within the scope of ICOMP. Many individuals listed below, will be requested after the conference to provide their expertise and services for selecting papers for publication (extended versions) in journal special issues as well as for publication in a set of research books (to be prepared for publishers including: Springer, Elsevier, BMC journals, and others).    

Prof. Afrand Agah; Department of Computer Science, West Chester University of Pennsylvania, West Chester, PA, USA Prof. Emeritus Nizar Al-Holou (Congress Steering Committee); Professor and Chair, Electrical and Computer Engineering Department; Vice Chair, IEEE/SEM-Computer Chapter; University of Detroit Mercy, Detroit, Michigan, USA Prof. Hamid R. Arabnia (Congress Steering Committee); Graduate Program Director (PhD, MS, MAMS); The University of Georgia, USA; Editor-in-Chief, Journal of Supercomputing (Springer); Fellow, Center of Excellence in Terrorism, Resilience, Intelligence & Organized Crime Research (CENTRIC). Prof. Dr. Juan-Vicente Capella-Hernandez; Universitat Politecnica de Valencia (UPV), Department of Computer Engineering (DISCA), Valencia, Spain

      

         

 

  

 

Prof. Emeritus Kevin Daimi (Congress Steering Committee); Director, Computer Science and Software Engineering Programs, Department of Mathematics, Computer Science and Software Engineering, University of Detroit Mercy, Detroit, Michigan, USA Prof. Zhangisina Gulnur Davletzhanovna; Vice-rector of the Science, Central-Asian University, Kazakhstan, Almaty, Republic of Kazakhstan; Vice President of International Academy of Informatization, Kazskhstan, Almaty, Republic of Kazakhstan Prof. Leonidas Deligiannidis (Congress Steering Committee); Department of Computer Information Systems, Wentworth Institute of Technology, Boston, Massachusetts, USA; Visiting Professor, MIT, USA Prof. Mary Mehrnoosh Eshaghian-Wilner (Congress Steering Committee); Professor of Engineering Practice, University of Southern California, California, USA; Adjunct Professor, Electrical Engineering, University of California Los Angeles, Los Angeles (UCLA), California, USA Prof. Houcine Hassan; Department of Computer Engineering (Systems Data Processing and Computers), Universitat Politecnica de Valencia, Spain Prof. Tai-hoon Kim; School of Information and Computing Science, University of Tasmania, Australia Prof. Louie Lolong Lacatan; Chairperson, Computer Engineerig Department, College of Engineering, Adamson University, Manila, Philippines; Senior Member, International Association of Computer Science and Information Technology (IACSIT), Singapore; Member, International Association of Online Engineering (IAOE), Austria Prof. Dr. Guoming Lai; Computer Science and Technology, Sun Yat-Sen University, Guangzhou, P. R. China Dr. Andrew Marsh (Congress Steering Committee); CEO, HoIP Telecom Ltd (Healthcare over Internet Protocol), UK; Secretary General of World Academy of BioMedical Sciences and Technologies (WABT) a UNESCO NGO, The United Nations Dr. Ali Mostafaeipour; Industrial Engineering Department, Yazd University, Yazd, Iran Prof. Dr., Eng. Robert Ehimen Okonigene (Congress Steering Committee); Department of Electrical & Electronics Engineering, Faculty of Engineering and Technology, Ambrose Alli University, Nigeria Prof. James J. (Jong Hyuk) Park (Congress Steering Committee); Department of Computer Science and Engineering (DCSE), SeoulTech, Korea; President, FTRA, EiC, HCIS Springer, JoC, IJITCC; Head of DCSE, SeoulTech, Korea Dr. Xuewei Qi; Research Faculty & PI, Center for Environmental Research and Technology, University of California, Riverside, California, USA Dr. Akash Singh (Congress Steering Committee); IBM Corporation, Sacramento, California, USA; Chartered Scientist, Science Council, UK; Fellow, British Computer Society; Member, Senior IEEE, AACR, AAAS, and AAAI; IBM Corporation, USA Ashu M. G. Solo (Publicity), Fellow of British Computer Society, Principal/R&D Engineer, Maverick Technologies America Inc. Prof. Fernando G. Tinetti (Congress Steering Committee); School of Computer Science, Universidad Nacional de La Plata, La Plata, Argentina; also at Comision Investigaciones Cientificas de la Prov. de Bs. As., Argentina Prof. Hahanov Vladimir (Congress Steering Committee); Vice Rector, and Dean of the Computer Engineering Faculty, Kharkov National University of Radio Electronics, Ukraine and Professor of Design Automation Department, Computer Engineering Faculty, Kharkov; IEEE Computer Society Golden Core Member; National University of Radio Electronics, Ukraine Varun Vohra; Certified Information Security Manager (CISM); Certified Information Systems Auditor (CISA); Associate Director (IT Audit), Merck, New Jersey, USA Prof. Shiuh-Jeng Wang (Congress Steering Committee); Director of Information Cryptology and Construction Laboratory (ICCL) and Director of Chinese Cryptology and Information Security Association (CCISA); Department of Information Management, Central Police University, Taoyuan, Taiwan; Guest Ed., IEEE Journal on Selected Areas in Communications. Dr. Yunlong Wang; Advanced Analytics at QuintilesIMS, Pennsylvania, USA Prof. Layne T. Watson (Congress Steering Committee); Fellow of IEEE; Fellow of The National Institute of Aerospace; Professor of Computer Science, Mathematics, and Aerospace and Ocean Engineering, Virginia Polytechnic Institute & State University, Blacksburg, Virginia, USA Prof. Hyun Yoe; Director of Agrofood IT Research Center and Vice President of Korea Association of ICT Convergence in the Agriculture and Food Business (KAICAF); Director of Agriculture IT Convergence Support Center (AITCSC); Department of of Information and Communication Engineering, Sunchon National University, Suncheon, Republic of Korea (South Korea) Prof. Jane You (Congress Steering Committee); Associate Head, Department of Computing, The Hong Kong Polytechnic University, Kowloon, Hong Kong Dr. Farhana H. Zulkernine; Coordinator of the Cognitive Science Program, School of Computing, Queen's University, Kingston, ON, Canada

We would like to extend our appreciation to the referees, the members of the program committees of individual sessions, tracks, and workshops; their names do not appear in this document; they are listed on the web sites of individual tracks. As Sponsors-at-large, partners, and/or organizers each of the followings (separated by semicolons) provided help for at least one track of the Congress: Computer Science Research, Education, and Applications Press (CSREA); US Chapter of World Academy of Science; American Council on Science & Education & Federated Research Council (http://www.americancse.org/). In addition, a number of university faculty members and their staff (names appear on the cover of the set of proceedings), several publishers of computer science and computer engineering books and journals, chapters and/or task forces of computer science associations/organizations from 3 regions, and developers of high-performance machines and systems provided significant help in organizing the conference as well as providing some resources. We are grateful to them all. We express our gratitude to keynote, invited, and individual conference/tracks and tutorial speakers - the list of speakers appears on the conference web site. We would also like to thank the followings: UCMSS (Universal Conference Management Systems & Support, California, USA) for managing all aspects of the conference; Dr. Tim Field of APC for coordinating and managing the printing of the proceedings; and the staff of Luxor Hotel (Convention department) at Las Vegas for the professional service they provided. Last but not least, we would like to thank the Co-Editors of ICOMP’19: Prof. Hamid R. Arabnia, Prof. Leonidas Deligiannidis, and Prof. Fernando G. Tinetti. We present the proceedings of ICOMP’19.

Steering Committee, 2019 http://americancse.org/

Contents SESSION: SMART BUILDING, SMART CITY INITIATIVES, RELATED TOOLS AND METHODS Improving Smart Buildings by Integrating user Contexts Olaf Droegehorn, Henrique R. Sarmento

3

A Real-Time Wireless Ultra-wideband Indoor Positioning System with Fast Computation Algorithm Wen Piao Lin, Yu-Fang Hsu

10

An Integrated Holistic Success Model for Evaluating Smart City Initiatives Nurcan Alkis, Murat Tahir Caldag, Ebru Gokalp

16

SESSION: INTERNET COMPUTING, CLOUD COMPUTING, COMMUNICATION SYSTEMS, AND RELATED ISSUES Host-Independent Autonomous Decision of Operations of the Redundant Traffic Reduction Node Michi Ishikuro, Kenji Ichijo, Akiko Narita

25

Canarycurity: A Robust Security Surveillance System Built with Mobile and IoT Technologies 32 Brandon Horowitz, Chen-Hsiang Yu, Leonidas Deligiannidis Express Dialer Dwight Deugo

38

Detection and Monitoring of Tampering in Telecommunication and Utility Cable Patrick Mabadie, Marcel Ohanga Odhiambo

45

Energy Profiling of iOS Apps to Detect Causes of Energy Drain Gokila Dorai

52

MARLi - Molly Alternate Realities Language interactive: A New XML Markup Language for Defining Virtual and Augmented Reality Ronald P. Vullo, Christopher A. Egert, Andrew Phelps

59

An Experimental Study of SDN-OpenStack Cloud Environments Adam L. Grady, Ahyoung Lee

64

Internet of Drives: A Global Key-Value Store System using Kinetic Drives Xiang Cao, Cheng Li

68

Combating Online Defamation and Doxing in the United States Ashu M. G. Solo

75

PROCEEDINGS OF THE 2019 INTERNATIONAL CONFERENCE ON INTERNET COMPUTING & INTERNET OF THINGS

ICOMP’19 Editors Hamid R. Arabnia Leonidas Deligiannidis, Fernando G. Tinetti

Publication of the 2019 World Congress in Computer Science, Computer Engineering, & Applied Computing (CSCE’19) July 29 - August 01, 2019 | Las Vegas, Nevada, USA https://americancse.org/events/csce2019

Copyright © 2019 CSREA Press

SESSION: INTERNET OF THINGS + WEB-BASED APPLICATIONS + DATA SCIENCE Prototyping of a Personal Outlet-enabled Individualized Power Consumption Management System Hideki Kondo, Kenichi Dobashi, Kazumasa Takami

81

Fast Detection of Abnormal Data in IIoT with segmented Linear Regression Sejun Kim, Ihsan Ullah, Taeho Lee, Hee Yong Youn

88

Independent Algal Bloom Removal System Kenichi Arai, Ryota Nakashima, Tetsuo Imai, Toru Kobayashi

95

Safety Analysis System Using Smart Helmet Ahyoung Lee, JunYoung Moon, Se Dong Min, Nak-Jun Sung, Min Hong

102

A Model and Middleware for Composable IoT Services Ahmed Abdelmoamen Ahmed

108

An Exploratory Study on the use of Internet_of_Medical_Things (IoMT) In the Healthcare Industry and their Associated Cybersecurity Risks Rashad J. McFarland, Samuel Bo Olatunbosun

115

SESSION: LATE BREAKING PAPERS: COGNITIVE IOT COGITO: A Cognitive Dynamic System to Allow Buildings to Learn and Adapt Giandomenico Spezzano

125

Int'l Conf. Internet Computing and Internet of Things | ICOMP'19 |

SESSION SMART BUILDING, SMART CITY INITIATIVES, RELATED TOOLS AND METHODS Chair(s) TBA

ISBN: 1-60132-503-7, CSREA Press ©

1

2

Int'l Conf. Internet Computing and Internet of Things | ICOMP'19 |

ISBN: 1-60132-503-7, CSREA Press ©

Int'l Conf. Internet Computing and Internet of Things | ICOMP'19 |

Improving Smart Buildings by integrating User contexts Olaf Droegehorn1, Henrique R. Sarmento1 1

Department of Automation & Computer Science, Harz University of Applied Sciences, 38855 Wernigerode, Germany

Abstract — Within the Kyoto protocol and the Paris agreement, as a follow up, the world’s countries have agreed to limit global warming to a maximum of 2°C. Automation technologies are widely acclaimed to have the potential to reduce energy consumption and energy-related costs in buildings significantly. These technologies not only enable buildings to be automated and centrally controlled but also help to provide a healthier and more comfortable living or working environment. Most of the systems are typically limited by implementing fixed schedules reassembling user behavior or routines. This usually results in inaccurate assumptions or bad alignment of the automation system with its users. In this context, the current research work investigates the use of actual contexts of users for adapting smart buildings in such a way that either more energy can be saved or the user goals and intentions are better met. In this paper, a smartphone is used to deliver context artefacts of the related user in order to derive higher context conclusions. Based on these higher contexts the smart building is adapted accordingly. Here, as a specific use case, the heating system is regulated in a more sophisticated manner, leaving a strict schedule whenever the user context demands to do so. Results show that the introduction of user contexts in smart buildings can be up to 50% more efficient when compared to other scheduled or strict scenarios. Keywords: Building Automation, User Context, Heating Efficiency, User Intentions

I.

INTRODUCTION

The residential sector represents great part of the energy consumption in a country. In Europe, households represented 25% of the total energy consumption in 2014 [1]. Furthermore, the energy consumption in Europe has also increased by 29% from 1990 to 2014 [1]. In Europe, space heating contributed around 67% of the total energy consumption in residential buildings in 2012 [2]. As shown in Fig. 1, in Germany, space heating corresponded to 75% of the households’ energy consumption, appliances and lighting to 12%, water heating to 13% and cooking to 1%.

reducing the carbon footprint by 77% compared to the 2050 baseline [3]. Another initiative promoted by the European Commission (EC) in 2007 was the 2020 package, which has the following goals [4]: 20% reduction in greenhouse gas emissions when compared to 1990, establish 20% of EU energy based on renewables and improvement of energy efficiency by 20%. As the residential sector stands for a large part of the carbon footprint, it is a potential target for improvements [15]. The World Business Council for Sustainable Development (WBCSD) showed in 2009 that it is possible to cut down the energy use in buildings dramatically by changing the user behavior [5][17][18]. One approach for saving energy is by introducing BASs, which can reduce up to 38% of the carbon footprint in the German residential building scenario, and for space heating specifically, it can reduce up to 42% of the carbon footprint considering the same scenario [6]. In order to get these results with BASs, the user context is relevant because the usage of appliances, heater, etc. may vary from one user to another. In order to make the system more aware of the user context, smartphones can be introduced as the possibilities and availability of communication and sensors of mobile devices has been growing [7]. It is also assumed that the user carries a smartphone outside the house, so more information regarding the user context can be detected as the user’s intention to arrive home. Therefore, building automation sensors cannot provide the same information as smartphones, and if they could, another device would be responsible for that, and the user would have to invest with extra sensors in the system [16]. In this work, a smartphone was integrated to a BAS in order to explore its benefits by increasing heating efficiency in residential BAS. For this purpose, a report by Siemens [8] which is based on the standard EN15232 was followed in order to achieve a Class A product. Furthermore, a discussion involving, lighting and appliances control, and social greening is held by the end of the paper. Results show that the integration of smartphone in BASs are relevant. This paper is structured in five sections. Initially, related works are described, then the methodology followed to compare heating response between scenarios is presented. After that, the research method shows implementation details and other relevant scenarios considered to compare to the smartphone-based solution. Next, the results, which are based on one use case, are presented. Finally, discussion and conclusion are also presented.

Figure 1. Household’ energy consumption by end-use [2]

In order to lower the energy consumption the International Energy Agency (IEA), which has 28 countries as members (including Germany), has a goal of

II. LITERATURE REVIEW & RELATED WORK Related works are divided in three sections: BAS and User Context, Smartphones and User Context and the

ISBN: 1-60132-503-7, CSREA Press ©

3

4

Int'l Conf. Internet Computing and Internet of Things | ICOMP'19 |

Siemens Report. The last one shows details regarding the approach followed for implementation of the current work. It was also noticed a lack of papers related to Smartphones and BAS. A. BAS and User Context In Cano et al [9], the proposed system deals with energy efficiency, comfort services, environmental monitoring and security for building automation. The main purpose of the paper was to control Heating, Ventilation and Air Conditioning (HVAC) and lighting of an office. In order to achieve that, an automation platform called CityExplorer gathers information from sensors and actuators, monitors them by identifying anomalies, and takes actions by dealing with key efficiency parameters as saving energy or water based on the condition of the indoor environment. Additionally, for the comfort management, it was used models for predicting the comfort response of occupants in building. The system also auto adapts its operation through user interaction. Results showed that about 20% of the energy consumption could be saved. In Mehrabi, Fung, and Raahemifar [10], three techniques were used in the BAS: role-based flow charts, adaptive home automation system, and thermal modeling of the area for energy conservation. Role-based flow charts were developed for each area and device of the house. Adaptive home automation system proposes that the system adapts itself to the needs of the user. Thermal modeling of the area for energy conservation, considers modeling the response of the HVAC system, so that it is possible to find the pattern for the control signal by understanding the valve position and radiator response. In Nguyen & Aiello [11] a survey was performed focusing on registering the most important user activities in order to save energy and improve the user comfort. For that purpose, the survey explored HVAC, lighting and plug loads. It was identified that occupant’s presence can improve energy consumption from 10% to 40%. For those results, approaches explored real-time occupancy, user’s preference settings, occupancy prediction and identification of activities as: working, having a meeting, sleeping, etc. In [9] and [10] adaptation of the system and prediction of the user behavior was implemented, and in [11] user presence shows to be an important factor for energy saving. One assumption of this work is that through the use of smartphones more information regarding the user behavior will be available for the BAS and the user presence could be detected in order to control the heater response. Therefore, prediction and adaptation could not be necessary for simpler scenarios as residential building because more information regarding the user activity is available for the BAS. B. Smartphones and User Context The survey described by Hoseini-Tabatabaei, Gluhak, and Tafazolli [7], reviews papers which used smartphones in order to determine the user context which were acquired by opportunistic sensing. The paper described sensors, algorithms for context recognition, accuracies for activity recognition and goals of each studied work. User’s location recognition was performed by using GPS, Wi-Fi and Bluetooth in some works which have achieved 90% accuracy for that purpose.

In Mafrur et al [12], it was described a system which used smartphones in order to identify user behavior. Three mobile sensors were used: GPS, accelerometer and magnetometer, and the following contexts were identified: user outdoor and indoor position, user activities as walking, running, standing, sitting and walking. The algorithms for identifying activities were based on Support Vector Machine (SVM). The actuators of the system were lighting and playing media, and the accuracy for activity recognition was 92%. As shown in [7] and [12], GPS is relevant sensor for user presence detection, and other smartphone sensors can also be used to verify a diversity of user activities. C. Siemens Report A report by Siemens [8] called “Building Automation Impact on Energy Efficiency” describes an approach to measure the relative performance of a BAS compared to another. They also presented some brief results on some typical baseline systems’ performance. Their approach is based on a model driven methodology to save energy in building automation, and they applied the standard “EN 15232: Energy performance of buildings - Impact of Building Automation, Control and Building Management” [13] which specifies the following methods to evaluate BAS and building management: Ɣ A structured list of control, and functions which have impact in the energy performance of buildings; Ɣ Methods to define the minimum requirements for controlling and managing buildings of different complexities (the report give extra recommendations); Methods to assess the BAS functions (mentioned Ɣ in the first bullet) on building energy performance. It also provides the impact of these functions which can be used for calculations in the energy performance rating and indicators from other standards; Ɣ A simplified method to get estimations of the functions impact for typical buildings. The report contains functions for a diversity of controls: HVAC, hot water supply, cooling, lighting, blind and technical home and building management. By applying these functions, a system can be categorized in different classes regarding energy efficiency: D (lowest), B, C and A (highest). Fig. 2 illustrates a comparison between classes D, C, B and A heating and cooling controller’s response in an office in relation to the user’s occupancy, and it is possible to observe that Class D has no automation for setting temperature for heater and cooler. On the other hand, Class C has specific set points when the user is occupying the place, and Class B has a better response than Class C because both heating and cooling set points are closer to the user’s occupancy, therefore, less energy is consumed. Class A applies adaptive set points in the cooling or demand control air-flow, so it performs better than the other classes. Results from the report show that: Class A when compared to Class D can save 29% and 16% of thermal and electrical energy. Table 1 shows the functions for heating, lighting and technical BAS management which

ISBN: 1-60132-503-7, CSREA Press ©

Int'l Conf. Internet Computing and Internet of Things | ICOMP'19 |

5

should be followed in order to implement a system for determined class. Figure 2. Classes for cooling (C line) and heating (H line)

Table 2. Six types of errors regarding the ideal and detected user context. Where EHNO-OFF means Error Heating Normal Operation (ideal) - OFF (detected)

The total error can be calculated by summing the occurrence of all types of errors during the day, and the daily system efficiency can be calculated by the formula: operations based on user presence in a building office [8]

In formula 1, all errors are considered and divided by a day period: 1440. IV.

RESEARCH METHOD

In order to obtain a Class A product, the functions specified in the Table 1: 1.1, 1.5, 5.1, 5.2, 7.1 and 7.2 were followed. In addition to that, another function was added for controlling appliances: Ɣ

Table 1: Control functions, where 1.1 and 1.5 correspond to heating control, 5.1 and 5.2 to lighting control, 7.1 and 7.2 to technical BAS management [8]

III.

METHODOLOGY

The methodology relies on comparing the ideal heater response with the detected one from different scenarios so that errors and performance can be calculated. These scenarios are described in detail in the Research Method section. The heater is considered to operate in three modes: Normal Operation (NO - 21ºC), Lowered (LO - 17ºC) and OFF (5ºC). For the ideal Normal Operation the user is arriving home, waking up soon or at home and awake, and the heater in this situation should respond before the user awake or arrive home in order to pre-heat the house. The Lowered operation corresponds when the user is sleeping or preparing to sleep. Finally, OFF is detected when the user is outside home. Additionally, the three operations of the ideal situation when summed must equate the duration of a day in minutes. The operation modes, temperatures and contexts were adapted from Sangogboye [6] and Siemens [8] works. Table 2 describes the ideal and detected contexts for six types of errors, which can occur, and “EH” in the “Error type” column means “Error Heating”.

Appliances - Occupancy Control ż Class D, C, B - Manual on/off ż Class A - Turning on specific appliances when the user arrives home. Turning off appliances when the user is not home.

The project also aims to identify when the user is sleeping, and lower the temperature of the place. The heating response for the user sleeping should follow the same strategy as occupancy in Fig 2. Therefore, the heater can lower the temperature of the place one hour before the user sleeping state is identified. When the user is identified to be outside home the heater response should be immediate and turn off (different from Fig. 1) because in the worked scenario it is considered only one person and only one room to control. For awakening activity, the heater should respond one hour before in order to prepare the environment to maintain normal temperature for the user when he/she is awaken. Besides the current implementation, three scenarios were simulated based on the Siemens report classes [8]: Ɣ No BAS (Class D): there is no heater controller, and the user set the radiator knob at a certain level, and no further control is performed. Ɣ Fixed Schedules (Class B): the heater controller is based on fixed schedules set by the user. The user can program outside home and sleeping intervals. Ɣ Presence (Class A): the heater controller turns on and off based on the user’s arrival home (e.g. through motion sensor), and departure (e.g. through exit button

ISBN: 1-60132-503-7, CSREA Press ©

6

Int'l Conf. Internet Computing and Internet of Things | ICOMP'19 |

near to the door), respectively. However, there is no sleeping detection. For implementing the proposed system, Figure 3 shows the components, modules and sensors which were used in the current approach for the smartphone and server. Additionally, Table 3 shows the smartphone and server implementation details for developing the functions described in Table 1.

this purpose, it was assumed that when the user is going to sleep the smartphone is left under the pillow or flipped so the screen touches the mattress. If the phone is left on the bedside table upwards, and there is nothing touching its screen, the system would not work properly. When the user is waking up the smartphone is left in a position where nothing is covering the front screen, and if there is (e.g. smartphone is in the pocket), the position of the smartphone is not lying down. It was also assumed that the native alarm clock from the smartphone is used, so it is possible to anticipate when the user is going to awake. The next alarm clock is identified only when the user is going to sleep. Logs were registered and updated to the user’s view. If the system crashes there is also a log file which reports the last messages and errors.

Figure 3. The current implementation

For simulating the Fixed Schedules scenario, the user can configure intervals where he/she assumes to be sleeping and outside home. Charts are available to see results regarding the scenario comparison. All implementation was performed in an Android device (Galaxy S5) where the application ran in background, for this purpose a service was created. It is possible to configure MQTT connection and GPS (interval and distance filters), and the app also auto reconnects if the user changes from Wi-Fi to Mobile data and viceversa. Finally, the smartphone sends the following messages through MQTT to the server: GPS coordinates, sleeping, awakening, alarm clock, and scheduled intervals.

Table 3. Implementation of Siemens functions for Class A through Smartphone and Server

The following subsections detail the smartphone and server app. implementations. A. Smartphone App. Implementation Details The smartphone acted as a client which sends and receive messages through Message Queue Telemetry Transport (MQTT) [14] to and from the server. The mobile software was implemented by using the framework Apache Cordova with Ionic Software Development Kit (SDK) and AngularJS as the Model-View-Controller (MVC). The sensors used in the smartphone were: GPS, Light, Proximity and Game Rotation Vector (accelerometer and gyroscope). Some modules were also included in the approach as the MQTT, Charts, Background Execution, Notifications, Clock Alarm, Google Calendar and Logs. For presence detection, the GPS sensor was used and only activated when the user moves the smartphone so that energy in the device could be saved. For activity recognition sleeping/awaken states were detected through the game rotation vector, proximity and light sensor. For

B. Server App. Implementation Details The server integrates Raspberry PI with a radio receptor module . The Raspberry PI acted as the server in which Home Assistant , which is a home automation platform, was installed. Homematic devices were used in order to control heating, lighting and appliances. The communication between Home Assistant and Homematic devices was integrated through Homegear which was also installed in the Raspberry PI. Home Assistant uses the following modules: Google Calendar, History, User Speed and Distance, Google Travel Distance Matrix, Sun Position, Logs and MQTT. For presence detection, the user’s distance from home is calculated. The user is not home if the distance is higher than 100m, and in this case appliances, lighting and heating are shut down and sleeping/awakening states are not verified. If the user is identified at home heating is set to 21ºC, appliances are turned on and if the sun position is below horizon, lights are also turned on. The user direction is also identified. If the user direction is towards home the state arriving home is identified, and if the user is arriving in less than 60 minutes, the heating controller is set to 21º C. If there are no GPS updates from the user smartphone in 3 minutes during his/her arrival, the user is considered to be not home.

ISBN: 1-60132-503-7, CSREA Press ©

Int'l Conf. Internet Computing and Internet of Things | ICOMP'19 |

If the user is identified as sleeping, the heating controller is set to 17º C, and all lights are dimmed off and appliances turned off. The user is expected to awake when the next alarm clock is triggered, or when an awakening state message is received. The heater is set to 21ºC 60 minutes before the next alarm clock is triggered. If an awakening message is received, the heater is set to 21ºC immediately, and the routine scheduled 60 minutes before the next alarm clock is cancelled.

7

scenario and 23% more efficient when compared to the Presence scenario. Besides that, the total operation time for NO is 630 minutes, LO is 360 minutes and OFF is 450 in the ideal scenario.

If a party is identified, the heater is set to 17ºC as people’s body already produces heat. If the user is throwing a party, activities are not verified (GPS, sleeping, awaken). Logs are registered through files, for crashes verification, and in the front-end. A weekly history also shows the response of devices and components. User activities time are stored, so it is possible to identify user’s presence and sleeping intervals. By using these intervals, it was also calculated the ideal situation in which the heating should respond with 21ºC, 17ºC and off (5ºC). Finally, the server calculates and sends through MQTT efficiency and errors for all scenarios. C. Selected Use Case In order to simulate the Fixed Schedules intervals, the following context was adapted from Sangogboye [6] regarding the German BAS user requirement on a weekday: Ɣ The occupant is expected to be asleep between the interval 12:00am - 06:00am. Ɣ The occupant is expected to work from 9:00am, and arrive home at 6:00pm. Ɣ It is supposed that it takes 30 minutes for the user to arrive home from work. V.

Figure 4. Up: Ideal heating response vs. Smartphone heating response. Down: real response acquired by Home Assistant history collected data

RESEARCH RESULTS

For results comparing it is supposed that instead of following the Fixed Schedule described in the German Use Case subsection, the user started sleeping at 1:00am, and configured the alarm to trigger at 7:00am, so the ideal case the heating should lower the temperature to 17ºC at midnight and start heating normally (21ºC) at 6:00am. The user also decided to work earlier (8:00am), and arrive home earlier (4:30pm) on that day. In this case, the heater should turn off at 8:00 am, and operate normally (21ºC) at 3:30pm. As observed in Fig. 4 the smartphone approach would differ from the ideal heating response only by two errors: the lowered temperature (17ºC) one hour before sleeping and the normal operation (21ºC) 30 minutes prior to user’s arrival. The response collected by home assistant history was simulated in order to show changes in temperature, but this simulation was reproducible in the real application and scenario. Table 4 shows the response of the smartphone and other scenarios regarding all types of errors described in Table 2, and by calculating the total scenarios error and performance (by using Formula 1). It is possible to observe that by using the proposed comparing method and for this specific scenario, the smartphone approach can be 50% more efficient when compared to the No BAS

Table 4: Efficiency and error results between different scenarios

According to the results shown in Table 4 it is possible to observe that the No BAS scenario could not identify sleeping, neither outside home intervals, as the radiator valve is set to a fixed value. Therefore, this case study was associated with the errors EHOFF-NO and EHLO-NO, and they represent the total operation time of the ideal situation for the heating off and lowered operations respectively Fixed Schedules could have all types of errors. For example, the user believes he/she is going to sleep at 11:00pm every day, but for a specific day the user goes to an event or party with friends outside home, in this case the error EHOFF-LO is increased. The Presence scenario cannot identify the user sleeping activity neither when the user is arriving home,

ISBN: 1-60132-503-7, CSREA Press ©

8

Int'l Conf. Internet Computing and Internet of Things | ICOMP'19 |

so this scenario is mostly affected by EHLO-NO and EHNO-OFF. The smartphone approach is mostly affected by three types of errors EHON-OFF, EHLO-NO and EHOFF-NO. EHON-OFF occurs if it takes less than 60 minutes for the user arriving home, so the heater is set to 21ºC only when the arriving home state is identified, but in the ideal case, it should be set 60 minutes before the user is at home. EHLO-NO occurs because this approach cannot predict when the user is going to sleep, so the heater response does not lower down 60 minutes before the user sleeping. EHOFF-NO occurs when the user is going to his/her home direction, but actually goes to another place, which is in the same direction, and the user goes home only later. Therefore the user state would change from arriving home to not home, and mislead the system to turn on the heater when it was not supposed to. VI. DISCUSSION As observed in the results section the smartphone approach when compared to the fixed schedule reveals a unique opportunity for BASs when the users cannot estimate their behavior through schedules. This situation is even more relevant for weekends and vacation periods when the user has no regular routine at all. The efficiency for the smartphone approach when compared to fixed schedule would depend on the user’s strictness on following the intervals configured. It would be possible to implement this scenario only with a heater controller, as far as this device has schedule configuration. The Presence scenario relies not only on the same controllers like the smartphone approach, but uses two more controllers (exit button and motion sensor). It can be seen that this approach would not identify the user sleeping state, and neither his/her arriving home state, therefore, less comfort and efficiency is guaranteed. The No BAS scenario would only be interesting to choose if the user does not leave home frequently (e.g. the user work at home), but even in this condition the sleeping state would not be identified. The user could also set manually the heating valve when planning to sleep, but this would depend on the user’s strictness on following this behavior, and it would not be possible to guarantee a specific temperature without a controller either. Light and appliances were also controlled in the smartphone scenario. In the No BAS scenario, it would not be possible implement this function. In the Presence and Fixed Schedule scenarios (if the user integrates Raspberry PI and other actuators) it would be possible to implement it, and they would also have the similar issues as the heating controller discussion when compared to the Smartphone approach. For detecting parties, other scenarios could also identify them as far as they have a system which uses calendar, and is integrated in the home automation platform. The inclusion of this function for smartphones was implemented because of the relevance of Google Calendar in the smartphone devices. Social greening is promoted by efficiency and error results, which are shown in the user’s smartphone application so that it is possible to track the relevance of the scenarios according to the user behavior. After that, the

users can understand which solution is more relevant for their behaviors, and maybe change their current BAS approach. For future works, it would be interesting to apply the proposed system with more samples, and analyze the response through a long period in order to find and consolidate further benefits regarding this approach. VII. CONCLUSION This paper has shown that the integration of smartphones in Building Automation Systems are relevant, and through simulation of different scenarios it is possible to improve and offer better energy efficiency for the heating controller depending on the user’s behavior. The results analyzed by scenario comparison show that heating efficiency can be improved by 50% when compared to other scenarios. The smartphone approach also can be used in order to manage lights and appliances. Further analysis by using more samples is still required in order to consolidate findings. REFERENCES [1]

[2]

[3]

[4]

[5]

[6]

[7]

[8]

[9]

[10]

[11]

[12]

[13]

EuroStat. 2016. Statistics Explained. [ONLINE] Available at: http://ec.europa.eu/eurostat/statistics-explained/index.php/. [Accessed 28 March 2017]. Odyssee-Mure. 2015. Energy Efficiency Trends and Policies in the Household and Tertiary Sectors. [ONLINE] Available at: http://www.odyssee-mure.eu/publications/br/energy-efficiencytrends-policies-buildings.pdf. [Accessed 26 April 2017]. International Energy Agency. 2013. Transition to Sustainable Buildings - Strategies and Opportunities to 2050. [ONLINE] Available at: http://www.iea.org/publications/freepublications/publication/Buil ding2013_free.pdf. [Accessed 20 April 2017]. European Commission. 2015. Europe 2020 in a nutshell. [ONLINE] Available at: http://ec.europa.eu/europe2020/europe2020-in-a-nutshell/index_en.htm. [Accessed 20 April 2017]. World Business Council for Sustainable Development. 2009. Transforming the Market: Energy Efficiency in Buildings. [ONLINE] Available at: http://www.wbcsd.org/Projects/EnergyEfficiency-in-Buildings/Resources/Transforming-the-MarketEnergy-Efficiency-in-Buildings. [Accessed 20 April 2017]. Sangogboye F.C., Droegehorn O., Porras J. (2016) Analyzing the Payback Time of Investments in Building Automation. In: Dastbaz M., Gorse C. (eds) Sustainable Ecological Engineering Design. Springer, Cham, ISBN 978-3-319-32645-0, pp. 367-381, 2016 Hoseini-Tabatabaei, S.A., Gluhak, A. and Tafazolli, R., 2013. A survey on smartphone-based systems for opportunistic user context recognition. ACM Computing Surveys (CSUR), 45(3), p.27. Siemens Building Technologies. 2012. Building automation – impact on energy efficiency. [ONLINE] Available at: https://www.downloads.siemens.com/downloadcenter/Download.aspx?pos=download&fct=getasset&id1=A6V10 258635. [Accessed 20 April 2017]. Cano, M.V.M., Santa, J., Zamora, M.A. and Gómez-Skarmeta, A.F., 2013, November. Context-Aware Energy Efficiency in Smart Buildings. In UCAmI (pp. 1-8). Mehrabi, T., Fung, A.S. and Raahemifar, K., 2014, May. Optimization of home automation systems based on human motion and behaviour. In Electrical and Computer Engineering (CCECE), 2014 IEEE 27th Canadian Conference on (pp. 1-5). IEEE. Nguyen, T.A. and Aiello, M., 2013. Energy intelligent buildings based on user activity: A survey. Energy and buildings, 56, pp.244257. Mafrur, R., Khusumanegara, P., Bang, G.H., Lee, D.K., Nugraha, I.G.D. and Choi, D., 2015. Developing and Evaluating Mobile Sensing for Smart Home Control. International Journal of Smart Home, 9(3), pp.215-230. Standard, C.S.N., 2012. Energy Performance of Buildings—Impact of Building Automation, Controls and Building Management. European Committee for Standardization.

ISBN: 1-60132-503-7, CSREA Press ©

Int'l Conf. Internet Computing and Internet of Things | ICOMP'19 |

[14] O.Droegehorn, P.Trenz, B.Brausse, T.Schwan, C.Voscort, M.Wemmer, "Capability Based Communication for Green Buildings and Homes - a REST-like API within the conex.io Project -", Proceedings of the 51st Hawaiian International Conference on System Sciences, ISBN 978-0-9981331-1-9, p.5767-5776, URI: http://hdl.handle.net/10125/50612, 2018 [15] I. Schrader, O.Droegehorn, “Transforming Business Moments into Business Models”, Proc. of the 17th International Conference on eLearning, e-Business, Enterprise Information Systems, and eGovernment, Ed. Hamid R. Arabnia, Azita Bahrami, Leonidas Deligiannidis, Fernando G. Tinetti, ISBN 1-60132-474-X, 2018 CSREA Press, pp. 87-93, USA, July 2018 [16] O.Droegehorn, M.Pittumbur, J.Porras, “Front-End Development for Home Automation Systems using JavaScript Frameworks“, Proc. of the18th International Conference on Internet Computing

(IComp’17), ISBN 1-60132-461-8, Ed. Hamid R. Arabnia, Leonidas Deligiannidis, Fernando G. Tinetti, LasVegas, pp. 98104, Nevada, USA, July 2017 [17] I.Schrader, O.Droegehorn, “Digitization of companies understanding IT as an enabler -”, Proc. of the 16th International Conference on e-Learning, e-Business, Enterprise Information Systems, and e-Government, ISBN: 1-60132-454-5, Ed. Hamid R. Arabnia, Leonidas Deligiannidis, Fernando G. Tinetti, pp. 83-87, LasVegas, Nevada, USA, July 2017 [18] I. Schrader, O.Droegehorn, “Process-oriented IT-Management as management approach to face digitization”, Proc. of the 15th International Conference on e-Learning, e-Business, Enterprise Information Systems, and e-Government, LasVegas Nevada, USA, July 2016

ISBN: 1-60132-503-7, CSREA Press ©

9

10

Int'l Conf. Internet Computing and Internet of Things | ICOMP'19 |

A Real-Time Wireless Ultra-wideband Indoor Positioning System with Fast Computation Algorithm Wen Piao Lin1 and Yu-Fang Hsu2 Department of Electrical Engineering, Chang Gung University, Tao-Yuan, Taiwan, R.O.C. 2 Department of Electronic Engineering, Chienkuo Technology University, Changhua, Taiwan, R.O.C. 1 E-mail: [email protected], 2E-mail: [email protected] 1

Abstract - An experimental real-time wireless ultrawideband indoor positioning system consisting of three anchors and a tag is presented. The anchor and tag are composed of a microprocessor with an ultra-wideband wireless transceiver device. The scheme based on the timeof-arrival technique and fast triangle midpoint algorithm provides accurate localization in a limited indoor area. The average localization error of the proposed positioning system placed in the indoor space of 732cm×488cm×220cm is 12.87cm. The experimental results demonstrated that the proposed system has the characteristics of high precision localization and less computation time. Keywords: Indoor positioning; ultra-wideband (UWB); time-of-arrival (TOA); triangle midpoint algorithm (TMA). I.

INTRODUCTION

Many different positioning technologies are applied to the indoor environment for research, including Radio Frequency Identification Devices (RFID) [1][2], infrared sensors [3], Zigbee [4], Wireless Networks (Wi-Fi) [5], low Power Bluetooth [6] and so on. These different technologies are usually selected from the Received Signal Strength Indicator (RSSI) to achieve indoor positioning. Signals are relatively easy to be blocked by objects and cause poor penetration. They are easily affected by multiple path interferences, and the accuracy is mostly low. Therefore, it is necessary to set multiple reference points in the indoor environment or build the database in advance to improve the indoor positioning accuracy. Ultra-wideband (UWB) wireless transmission technology has excellent transmission quality in complex indoor environments and has the advantages of high transmission speed, excellent resistance to multi-path interference, low power, high penetration, and high time accuracy [7]-[9]. UWB transceiver utilizes a very short Radio Frequency (RF) pulse to achieve high bandwidth connections. It can execute an accurate measurement of time delay and distance difference [9][10]. Many algorithms for ŖŘŃ indoor positioning have been proposed such as calculating the Time-Of-Arrival (TOA) or Time-Difference-Of-Arrival (TDOA) schemes [12]. From the above, it is known that UWB is quite suitable for indoor positioning, and the positioning accuracy is high, even reaching a minimum error of 10 cm. This research mainly uses a microprocessor to control UWB device, and then uses TOA triangulation method,

combined with a fast positioning Triangle Midpoint Algorithm (TMA) to implement an accuracy indoor positioning system. First, a single tag based on a single base station measures its positioning accuracy and Environmental Parameter Calibration (EPC). After that, three base stations are placed in the three corners of the laboratory space, and the distances of the tags are respectively captured to test the arrival time method. Finally, the test results input into computer to evaluate and explore the positioning accuracy and computation time. This paper is organized as follows; the indoor wireless positioning system is presented in Section II. Then, the UWB positioning scheme is briefly explained in Section III. After that, the implementation of the positioning algorithm is described in Section IV. Then, experimental setup and results are discussed in Section V. Finally, Section VI concludes the paper. II.

INDOOR WIRELESS POSITIONING SYSTEM

Figure. 1. Block diagram of UWB Indoor wireless location system

Figure 1 is a block diagram of UWB indoor wireless positioning system. The green ellipse is the base station that is an anchor. In the middle of the figure, the gray inverted triangle is a tag to be located. The base station is placed in three corners of the experimental space and connected to the computer using Universal Asynchronous Receiver and Transmitter (UART) through a Transistor-Transistor LogicUniversal Serial Bus (TTL-USB) module. Then the measured data are written in JAVA programming language of TOA triangulation positioning scheme, and finally by the coordinate system to calculate a positioning point and its

ISBN: 1-60132-503-7, CSREA Press ©

Int'l Conf. Internet Computing and Internet of Things | ICOMP'19 |

error for real-time operation. The tag and anchors are integratedġ by the Arduino microprocessor with a UWB device (DW1000) as shown in the inset of Figure 1. The control program is programming in the microprocessor and communicates with the UWB device via the Serial Peripheral Interface Bus (SPIB). The base station and the tag are a wireless signal transmission using ultra-wideband pulses. The DW1000 [13] is a single chip radio transceiver Integrated Circuit (IC) compliant with the IEEE 802.15.42011 UWB standard. It facilitates real time location of assets into an accuracy of +/- 10 cm using either two-way ranging TOA measurements or one-way TDOA schemes. Moreover, DW1000 spans 6 radio frequency bands from 3.5 GHz to 6.5 GHz and also supports data rates of 110 kbps, 850 kbps and 6.8 Mbps. The transmitting or receiving signal for the DW1000 is used as a semi-directive antenna. The signal will be transmitted in an arc towards the antenna's facing side. This means that the back of the antenna has poor signal. It is a factor that affects the accuracy of the distance measurement on the anchor and needs to be taken into account. III.

B. Double-sided Two-way Ranging Double-Sided Two-Way Ranging (DS-TWR), is an extension of the basic SS-TWR in which two round trip time measurements are used and combined to give a TOA result which has a reduced error even for quite long response delays. Figure 3 shows the DS-TWR of multiple stations for single tag [13]. It can be seen in the graph that the tag sends a poll message which is received by three anchors in the infrastructure who reply in successive responses with packets RespA, RespB and RespC after which the tag sends the Final message received by all three anchors. This allows the tag to be located after sending only 2 messages and receiving 3. Anchor A and the tag can calculate the corresponding time T propA, and then multiplied by the speed of light c can be obtained the distance D A between the two devices:

(3)

UWB POSITIONING SCHEME

This section describes various methods of implementing UWB two-way ranging scheme between two nodes. In all of the schemes that follow one node acts as initiator, initiating a range measurement, while the other node acts as a responder listening and responding to the initiator, and calculating the range. A. Single-sided ŕŸŰ-way Ranging Single-Sided Two-Way Ranging (SS-TWR) involves a simple measurement of the round trip delay of a single message from one node to another and a response sent back to the original node. The operation of SS-TWR is as shown in Figure 2, where Tag initiates the exchange and Anchor A responds to complete the exchange and each device precisely timestamps the transmission and reception times of the message frames, and so can calculate time Tround A, and Treply A , by simple subtraction. The resultant TOA, TprorA, may be estimated by the equation: ġġġġġġġġġġġġġġġġġġġĩIJĪ and then multiplied by the speed of light c can be obtained the distance DA between the two devices: (2)

ġ

11

Figure. 2. Single station SS-TWR scheme

(4)

Figure. 3. Multiple station DS-TWR scheme

Similarly, between anchor B, anchor C and the tag can calculate the corresponding time TprorB and TprorC, and then multiplied by the speed of light c can be obtained the distance of DB and DC between the two devices. IV.

POSITIONING ALGORITHM

ġġġġġġThe concept of the Triangle Centroid Algorithm (TCA) was first proposed by Prof. Nirupama Bulusu of the University of Southern California [14]. The main reason is that if unknown nodes can receive signals from N anchor nodes, unknown nodes can consider as anchor nodes and the triangle centroid of the polygons formed by overlapping places [15]. However, the actual space has a high degree of complexity. In the real environment, the distance from the tag to the base station is slightly larger than the actual value. The circle drawn according to the TOA method not only intersects at one point but overlaps with the triangle area in the figure. Figure 4 shows the principle diagram of TCA where A, B, C are base stations, and T is the target tag. ġġġġġġThe three-circle intersections obtained from the TOA method in all cases can locate in the same triangle as the previous section, and the target can estimate by triangle

ISBN: 1-60132-503-7, CSREA Press ©

12

Int'l Conf. Internet Computing and Internet of Things | ICOMP'19 |

centroid algorithm. When F anchor leaves far away, the area of the three-circle interaction will not approximate as a triangle-shaped DEF. When TCA is used to estimate the target coordinates, the accuracy will decrease because of too narrow triangles. Therefore, we propose a TOA Triangle Midpoint Algorithm (TMA) to improve positioning accuracy.ġ ġġġġFrom Figure 4, we can speculate that the tag position will be very close to the arc DF and arc EF so that the tag will fall in the upper right corner of the yellow area near the arc DE. Here, we propose to use the point M of the intersection DE as the method of positioning coordinates to improve the error of the TCA.

to locate and calculate the intersection points D and E between the origin circle and the other two circles. Second, calculate the midpoint M between the two points D and E, and finally, use M as the positioning coordinate. After ending the algorithm, determine if the tag is offline. If the tag is disconnected, the base station returns to the standby state. If the tag continues to transmit, the bidirectional ranging is continued.

ġġġġġġġġġ ġġġġġġġġġġġġġġġġFigure. 5. Flow chart of TMA scheme

ġġġġġġġġġġġġġġFigure. 4. The Principle diagram of the TCA The coordinates of D(xd, yd) shown in the Figure 3 can be inferred from the triangle formula

(5)

Similarly, the coordinates of E(xe, ye) are derived. Finally, we find the coordinates of the midpoint M(xm, ym) of arc DE as coordinates of the target after positioning: , (6) The TMA software processing flow is shown in Figure 5. When the base station starts up, it will enter the standby state and determine whether there is a message in the tag. If no signal is received, it will continue to standby. Otherwise, the base station will synchronize with the tag. After bidirectional ranging performs after synchronization, the distance data is obtained. The TOA positioning calculation is performed using the JAVA programming language. The next step is to establish a coordinate system and place the three groups of base station positions as the center of the circle at the origin (0, 0), the X-axis point (X, 0) and the Y-axis point (0, Y). The distances measured by the three groups of base stations and tags plot as a radius, and then the TMA is used

V. EXPERIMENTAL SETUP AND RESULTS In this experiment, a 12×8 coordinate system was established on a lab room space of 732cm×488cm×220cm by a 61cm×61cm square grid on the ceiling. The base station placed in the three corners of the lab and connected it to the computer, and then measure the distance of the tag unilaterally as shown in Figure 1. The measured distance is estimated by the of the computer positioning algorithm. Finally, the positioning error is calculated by the coordinate system. A. One-to-One Ranging Test One-to-one obstacle-free distance test between devices was first performed. The measured actual distance starts from 50 cm, and then an experiment is conducted every 50 cm. The test results were recorded, and 30 test values were averaged as test values until the actual distance was up to 10 m. A total of 20 one-to-one accessibility measurements are performed. Figure 6 is the error calculated from the one-toone accessibility test value versus actual value. It can be found that the error is less than 6% at a distance which is less than 6 m. The converted error is approximately 20 to 30 cm. When the distance is more than 6 meters, the error starts to increase gradually. Since the one-to-one ranging distance between devices directly affects the positioning accuracy, we correct the measured values from 0.5m to 10m interval for every 0.5m.

ISBN: 1-60132-503-7, CSREA Press ©

Int'l Conf. Internet Computing and Internet of Things | ICOMP'19 |

The value is corrected as the demarcation point between the minimum value of the subsequent segment and the maximum value of the previous segment; for example, the average value of 30-data measured at distance of 6.5m is 6.889m, and the minimum value is 6.68m. The measured error is 0.389m. Next, the maximum value of 30-data measured at distance of 6m is 6.39m. Therefore, we can calculate that the demarcation point will be (6.68+6.39)/2 =6.54m. When the measured distance is greater than 6.54 m, the measured value must be deducted 0.389m. The error of the one-to-one distance test after correction is about 0.1m, and the accuracy is much improved.

13

B. TCA Localization Test

Figure. 8. TCA localization of tag at (7, 5) TABLE I. LOCALIZATION ERROR FOR VARIOUS ALGORITHMS

Coordinates

ġġ ġġġġġġġġġġġġġġġFigure. 6. Calibration of localization error

ġġ

The height of the tag on the Z axis will affect the accuracy of the distance measurement. When the distance between two nodes on the XY plane is 6m, the error is 0.082m under the height of 1m on the Z axis. And when the distance between two nodes is 1m, the error is 0.414m under the height of 1m on the Z axis. Therefore, in order to improve the accuracy of the three-dimension measurement, tag position should be as far as possible with three anchors in the same plane.

Yġ ġġġġġġ ġ Xġ ġġġġġFigure. 7. TCA localization of tag at (5, 3)

(1,2) (3,1) (3,6) (5,3) (7,3) (7,5) (10,7) (11,3) (11,5) (12,6) Average error (cm) Standard deviation

Error (cm) of triangle centroid algorithm

Error(cm) of triangle midpoint algorithm

23 3.31 24.3 1.92 19.95 63.68 188.45 80.49 145.55 202.96

26.19 10.85 8.048 16.99 13.59 8.57 14.48 10.03 6.93 13

Error (cm) of inner triangle centroid algorithm 20.22 10.84 7.88 14.58 12.11 8.54 14.48 9.98 6.93 12.92

75.36

12.87

11.85

72.91

5.36

3.76

We placed three base stations on the origin (0,ġ 0), Yaxis (0, 8), and X-axis (12,ġ 0); the tag is placed at different points in the coordinates. When a positioning coordinate of the tag is located at (5,ġ 3) shown in Figure 7, the calculated positioning coordinate after TCA execution is (4.97, 2.99). The error is converted to approximately 1.92 cm; however, when the positioning coordinate of the tag is located at (7,ġ 5) shown in Figure 8, the calculated positioning coordinate after TCA execution is (6.43, 4.12). The actual error is converted to approximately 63.68 cm. Finally, when the positioning coordinate of the tag is located at (11, 5), the calculated coordinate after TCA is (9.73, 2.98). The actual error is converted to about 145.55 cm. From test results when the tag

ISBN: 1-60132-503-7, CSREA Press ©

14

Int'l Conf. Internet Computing and Internet of Things | ICOMP'19 |

is located far away from the XY coordinates, the positioning using the TCA scheme will be invalid. The main reason is that the triangle completed by the TCA is too narrow, resulting in the measurement error is greater when the tag is far away from the XY coordinates. The detailed experimental results of 10 coordinate points by TCA scheme are shown in Table 1.

coordinate is (7.09, 5.1). The error is converted to about 8.57 cm. When the positioning coordinate of the tag is located at (11, 5) shown in Figure 10, the calculated positioning coordinate is (11.1, 5.04), and the error is converted to about 6.93 cm. The detailed experimental results of 10 coordinate points by TMA scheme are shown in Table 1.

C. TMA Localization Test To solve the problem when the tag is placed on the X and Y coordinates with larger values, the triangle is too narrow and long to decrease accuracy shown in Figure 8. Therefore, we propose a solution for the midpoint of the triangle. The intersection of the red circle and the blue circle and the intersection of the red circle and the green circle are selected, and the middle point is taken as the rear coordinate after positioning shown in Figure 9. This can avoid the large change of the blue circle and the green circle.

Figure. 10. TMA localization of tag at (11, 5)

Figure. 9. TMA localization of tag at (7, 5)

When a positioning coordinate of the tag is located at (7, 5) shown in Figure 9, the calculated positioning

D. Comparison of Algorithm To understand the performance of proposed TMA, we write a computer program to compare localization error forġ various algorithms. The Inner Triangle Centroid Algorithm (ITCA) [16] published by Nantong University of China. The comparison error is the distance between the coordinates of the tag calculated by each algorithm and the actual position of the target. A total of ten points in the ordinates take for ranging experiments. Finally, the average error and standard deviation figure out. It can be seen from Table 1 that the accuracy of the ITCA is the highest with an average error of approximately 11.848 cm.ġ The accuracy of

Figure. 11. Computation time for different algorithms

ISBN: 1-60132-503-7, CSREA Press ©

Int'l Conf. Internet Computing and Internet of Things | ICOMP'19 |

TMA is slightly lower than that of the ITCA. The average error is about 12.869 cm. Moreover, theġ TCA performs well in some coordinates at the positioning (3, 1) and positioning (5, 3). However, when the target is farther away from the origin, the TCA cannot accurately locate the target. The average error is 75.36 cm. ġġġġġġġŕhe computation time of the algorithm at the positioning (11, 5) shown in Fig. 11 that the algorithm of the proposed TMA is simple and requires only 13.654 Ps for positioning; and the algorithm of ITCA is comparatively complexity, therefore, takes 33.28 Ps to execute it once. It can be seen that the accuracy of proposed algorithm is very high. At the same time, the simple calculation process can be reduced the computation time. VI. CONCLUSION This study experimented and demonstrated a highprecision real-time wireless indoor positioning system based on time-of-arrival technique and fast triangle midpoint localization algorithm in a limited indoor space. We characterized existing ultra-wideband localization algorithm schemes and explored a high-accuracy algorithm method. According to experimental results that the accuracy of the one-to-one ranging between the anchor and a tag is almost within 10 cm. Moreover, the average error of the ultrawideband positioning systems is about 12.87 cm to use the triangle midpoint localization algorithm in a laboratory space of 732cm×488cm×220cm. Finally, we also compare the execution time of various positioning algorithms. The experimented results show that the proposed algorithm has a simple calculation process which can reduce the computation time and reach a real-time process. ġġ REFERENCES [1] X. Liu, M. Wen, G. Qin, and R. Liu, “LANDMARC with improved k-nearest algorithm for RFID location system”, IEEE International Conference on Computer and Communications, pp. 2569-2572, 2016. [2] S. Sahin, H. Ozcan, and K. Kucuk, “SmartTag: an indoor positioning system based on smart transmit power scheme using active tags”, IEEE Access, vol. 6, pp. 23500 -23510, Mar. 2018. [3] Y. Kobiyama, Q. Zhao, and K. Omomo “Privacy preserving infrared sensor array based indoor location awareness”, IEEEġ

International Conference on Systems, Man, and Cybernetics, pp. 001353-001358, 2016. [4] C. H. Chu, et al., “High-accuracy indoor personnel tracking system with a zigbee wireless sensor network”, Seventh International Conference on Mobile Ad-hoc and Sensor Networks, pp. 398-402, 2011. [5] X. Ge and Z. Qu, “Optimization WIFI indoor positioning KNN algorithm location-based fingerprint”, 7th IEEE International Conference on Software Engineering and Service Science, pp. 135-137, 2016. [6] M. E. Rida, F. Liu, Y. Jadi, A. Algawhari, and A. Askourih, “Indoor location position based on bluetooth signal strength”, 2nd International Conference on Information Science and Control Engineering, pp. 769-773, 2015. [7] S. Gezici, Z. Tian, and G. B. Giannakis, “Localization via ultrawideband radios: a look at positioning aspects for future sensor networks”, IEEE Signal Processing Magazine, vol. 22, no. 4, pp. 70-84, 2005. [8] D. B. Jourdan, D. Dardari, and M.Z. Win, “Position error bound for UWB localization in dense cluttered environments”, IEEE Transactions on Aerospace and Electronic Systems, vol. 44, no. 2, pp. 613-628, 2008. [9] D. Munoz, F. Bouchereau, C. vargas, and R. Enriquez-Caldera, “Position location techniques and applications”, Academic Press, Elsevier Inc., 2009. [10] S. Gezici, et al., “Localization via ultra-wideband radios: a look at positioning aspects for future sensor networks”, IEEE Signal Processing Magazine, vol. 22, no. 4, pp. 70–84, 2005. [11] C. Falsi, D. Dardari, L. Mucchi, and M. Win, “Time of arrival estimation for UWB localizers in realistic environments”, Journal on Applied Signal Processing, pp. 152–152, 2006. [12] M. M. Saad, C. J. Bleakley, M. Walsh, and T. Ye, “High accuracy location estimation for a mobile tag using one-way UWB signaling”, Ubiquitous Positioning, Indoor Navigation, and Location Based Service, pp.1-8, 2012. [13] DW1000 User Manual, DecaWave Ltd, Dublin, version 2.05, 2015. [14] N. Bulusu, J. Heidemann, and D. Estrin, “GPS-less Low cost outdoor localization for very small devices”, IEEE Personal Communications Magazine, vol.7, no.5, pp.28-34, 2000. [15] Z.-M. Wang and Y. Zheng, “The study of the weighted centroid localization algorithm based on RSSI”, International Conference on Wireless Communication and Sensor Network, pp. 276-279, Dec. 2014. [16] W. Pei, J. Ping, J. He, and H. Zhang, “Ultra-wideband indoor localization based on inner triangle centroid algorithm”, J. of Computer Applications, vol. 37, no. 1 pp. 289-293, 2017.

ISBN: 1-60132-503-7, CSREA Press ©

15

16

Int'l Conf. Internet Computing and Internet of Things | ICOMP'19 |

An Integrated Holistic Success Model for Evaluating Smart City Initiatives Nurcan Alkış1, Murat Tahir Çaldağ2, and Ebru Gökalp3 1, 2, 3 Technology and Knowledge Management, Başkent University, Ankara, Turkey Abstract - The smart city vision arises from the necessity of management, automation, optimization, and exploration all aspects of a city for improvement purposes. The needs of cities are growing and the citizens are forcing governments to undertake new “smarter” path to effective and efficient utilization of the resources. In order to provide a guidance to implement successful smart city projects, there is a need for a structured evaluation model for smart city initiatives. As a result of literature review, it is determined that the literature focused on evaluation of smart cities, however there is limited number of study investigating evaluation of smart city initiatives. Accordingly, in order to fulfill this gap, this study proposes an integrated holistic success model for evaluating smart city initiatives. Keywords: Smart City, Smart City Initiatives, Success Model, Success Factors.

1

Introduction

In the era of information, a revolutionary approach of sustainability in development, green environment and quality of life, the concept of "Smart City" is appearing to be the wave of the future urban planning. With the increased migration to urban areas, the management of transportation as well as environment and development of a sustainable economy becomes more complicated. The statistics show that more than 55% of the world population lives in cities and the percentage will rise to 68% by the year 2050 [1].With the interest smart city market is estimated to reach 3 trillion US$ by 2020 [2].This increasing in city population brings new kinds of problems, such as traffic congestions, waste management difficulties, pollution and health issues [3]– [7].Increase in population and the shift from rural to urban could add 2.5 billion to cities by 2050 [1]. As cities generate %80 of worlds greenhouse gas emissions and demands more than %75 energy production projects have been put in motion to achieve the expectations of 2020 European Strategy [8]. From the social and organizational viewpoint, another set of problems which are associated with high levels of interdependence, competing objectives and values, social and political complexity as well as multiple and diverse stakeholders are arising. To this end, the city problems become catastrophic [9], [10]. The needs of cities are growing and the citizens are forcing governments to undertake new “smarter” path to effective and efficient utilization of the resources. Smart city concept has many definitions there isn’t a description that is commonly accepted. The reason there isn’t

a consensus on smart city definition is it has been applied on two different domains, which one consists of buildings, natural resources, mobility, energy grids and the other one includes education, culture, policy innovations [11]. The difference between these domains is the decisive role ICT plays in the functions of the systems [12]. The division can be seen in literature as technical perspective and social perspective. Because of this divide there are different models and dimensions expressing smart city initiatives. As some studies research governance and people factors, others look into mobility and infrastructure. The current study focuses on all aspects of smart city initiatives and the related success factors to create a general success model by attempting to fill this gap. As a result of literature review, it is determined that the literature focused on evaluation of smart cities[13] , however there is limited number of study investigating evaluation of smart city initiatives[14], [15]. Accordingly, in order to fulfill this gap, this study aims to propose an integrated holistic success model for evaluating smart city initiatives. By exploring an extensive array of literature from various fields, an integrated success model is provided for smart city initiatives to give direction to future developments.

2

Conceptualizing a Smart City

Although there are numerous smart city definitions one common aspect which is ICT-based applications. Also there are various terms including digital city, eco city, intelligent city, livable city etc. which complicates a unified definition of smart city in the literature [16]. While some studies discuss it as a general case-study, others analyze specific parts, such as: Intelligent Transportation System (ITS), smart home, smart environment, smart health, smart living, smart tourism, smart food and other. There is a necessity of definition the term of Smart City in a more “general” sense. European Union’s definition of smart city is “a city seeking to address public issues via ICT-based solutions on the basis of multi-stakeholder, municipally partnership” [2]. An explanatory definition of smart city is “an integrated system in which human and social capital interact, using technologybased solutions which aims to efficiently achieve sustainable and resilient development and a high quality of life on the basis of a multi stakeholder, municipality-based partnership.”[8]. According to Caragliu et al. [3] smart cities can be conceptualized by six characteristic features including improvement of political and economic efficiency and development of culture and society, emphasis on business oriented urban development, a strong focus on aiming the

ISBN: 1-60132-503-7, CSREA Press ©

Int'l Conf. Internet Computing and Internet of Things | ICOMP'19 |

participation of different urban residents in public services, signifying the role of high-tech and creative industries in growth, reminding the importance the social and relational capital in city development, and taking in to consideration of social and environmental sustainability [3]. These definitions of smart cities conceptualize some critical factors of smart city initiatives. According to Chourabi et al. [15] success factors of smart city initiatives are given in eight dimensions as management and organization, technology, governance, policy, people and communities, economy, built infrastructure and natural environment. In their research Albino et al. [12] defined essential dimensions of smart city initiatives as smart economy, smart people, smart governance, smart mobility, smart environment and smart living. In this current study smart city initiatives critical success factors are conceptualized in eight dimensions including built infrastructure, economy, environment, governance, government, management and organization, people communities and technology. Most smart city initiatives in Europe are located in UK, Spain and Italy, although when assessed by highest percentages Italy, Austria, Denmark, Norway, Sweden, Estonia and Slovenia come on top. Smart environment and smart mobility are the most focuses projects in Europe [2]. Zaragoza traffic monitoring system is a successful example for smart city initiative. With Bitcarrier’s Citysolver solution and 150 sensors Zaragoza’s traffic information is managed in real time which is also published on a public website for citizen use [2]. Another example is Cologne’s Klima Strasse which is a building infrastructure project for reduction CO2 emissions in general [2]. Other initiatives examples can be given as Munich’s Smart Grid system, Gotenberg’s Celsius, Stockholm Royal Seaport etc.[2] Given the conceptual description of a smart city and some examples from different domains, it could be thought as monitoring and integrating conditions of all of its critical infrastructures, including transportation, health, waste management, education, energy management, water management, open government can better utilize its resources, plan its maintenance activities, and monitor security aspects as well as providing improved quality of life to the citizens. The smart city is a large organic system consisting of many subsystems and components. Dirks and Keeling [17] consider a smart city as the organic integration of systems. The interrelationship between a smart city’s core systems is taken into account to make the system of systems smarter. Systems cannot provide their operations in an isolated circumstance. Therefore, there should be increasingly effective combination of communication networks, ubiquitously embedded intelligence, sensors and tags, and software from the technical perspective, it also should have citizen centered approach to improve the quality of life of them. Our proposed model explained in the following section integrates all aspects of the smart city initiatives and success factors of each aspect.

3

17

Smart City Initiatives Success Factors

As a result of the literature review, it is observed that although there are studies for evaluating success of smart city initiatives from different aspects as economy or environment, there is limited study for defining success factors of smart city initiatives from a holistic viewpoint covering all aspects. Accordingly, we proposed a critical success factor model with eight dimensions for smart city initiatives including all aspect of smart city initiatives; built infrastructure, economy, environment, governance, government, management and organization, people communities and technology (given in Fig. 1.).

Built Infrastructure Technology

People and Communities

Economy

Smart City Initiatives Success Factors

Management and Organization

Environment

Governance Government

Fig. 1. Dimensions of the Proposed Holistic Success Model

3.1

Built Infrastructure

Infrastructure refers to cables and pipes networks and other supportive functions for services and facilities [18]. Built infrastructure for smart cities consist of basic infrastructure and ICT infrastructure for the use of intelligent and integrated communication systems [18]. We have identified 4 different factors (given in Table 1. with their references) from existing studies in the literature, which are categorized under built infrastructure dimension. For smart city initiatives IT infrastructure is considered a core necessity that it’s quality and implementation affect the services provided in a system. Smart information services are defined as the purpose of the initiatives to solve the city’s inhabitants needs [19].

3.2

Economy

Economy in the context of smart city initiatives is associated with overcoming economic challenges, establishing new businesses, reduce unemployment with creating new jobs, increase regional attractiveness, create competitive advantage and improve productivity [20]. In this study, economy and smart economy dimensions are considered as same. According to Kogan and Lee [18] level of economy which includes innovation, entrepreneurship, trademarks, labor market productivity and integration of international markets etc. is significant in planning smart city initiatives. One of the aims of smart economy is creating a multisectoral economy to safeguard cities against economic crisis [8]. In

ISBN: 1-60132-503-7, CSREA Press ©

18

Int'l Conf. Internet Computing and Internet of Things | ICOMP'19 |

this research, 14 different success factors are identified from economy and smart economy aspects and they are represented in Table 2.

Table 3 Environmental success factors Dimension

Table 1 Built infrastructure success factors Dimension

Built Infrastructure

Factors

References

Infrastructure / Technological infrastructure / IT Infrastructure Security and privacy

[15][18][19][14]

Operational cost

[15][14]

Smart information services

[19]

[15][14]

Environment

Table 2 Economy success factors Dimension

Economy

3.3

Factors

References

Social diversity as source of innovation Entrepreneurship & Innovation Productivity

[8]

Local and Global Conexion Economic image and trademarks Flexibility of labor market

[21]

International embeddedness Level of Economy

[21]

Factors

References

Energy saving

[8]

Shrinking cities

[8]

Holistic approach to environmental and energy issues Urban ecosystems under pressure Climate change effects

[8]

Attractivity of natural conditions /natural resources / natural environment Pollution

[21][22][15]

Environmental protection

[21]

Sustainability/environmen tal sustainability

[23][18]

[8] [8]

[21]

[21]

3.4

[21][1]

Smart governance and governance are taken as one dimension in this study and both factors from each aspect are grouped under governance dimension. Smart governance is referred as an important characteristic of smart city initiatives which is based on citizen participation [24] and public / private partnership [25]. Internal and external stakeholders are expected to engage in decision making and public services in the context of smart governance [12].With the emergence of ICTs governance of smart city initiatives have new ways to keep decision making processes more transparent and collaborative with e-government and other ICT-based applications. Cities which successfully integrated smart governance have different parties collaborating as a core characteristic [26]. In this research, the identified governance factors are given in Table 4.

[21] [21]

[18]

Environment

Environmental concerns raised more interest in smart city initiatives since one of the goals is to create self-sustaining cities and reduce environmental damage. Sustainability can be described as the means of economic and social development without interfering with the environment [23]. Environment and smart environment are considered the dimension in this study. Smart cities generally make its community more environmentally friendly by reducing the environmental impact of urban life with decreasing the carbon footprint [23]. Smart cities initiatives reduce energy consumption, pollution and CO2 emissions with projects to achieve ecological sustainability [8]. 9 different environmental factors are identified from literature, which are shown in Table 3.

3.5

Governance

Government

Smart government and government aspects of smart city initiatives are taken together. Smart government is described as an administration, that integrates information, communication and operational technologies, creates and sustains planning, management and operations across multiple domains, create policies and generates sustainable public value [18]. Government collaborates with communities to become more transparent, inform citizens of decision that affect their lives and make them participants in decision making processes [30].Smart governments goal is creating and sustaining operations and services with the mindset of being citizen-centric [30]. Success factors of smart government are given in Table 5.

ISBN: 1-60132-503-7, CSREA Press ©

Int'l Conf. Internet Computing and Internet of Things | ICOMP'19 |

Table 4 Governance success factors Dimension

Factors

Leadership and champion / Leadership environment Participation and partnership Communication

Governance

References

[27][15][23] [15][23][25][30] [15][23]

Data-exchange Service and application integration Accountability Transparency / Transparent governance Flexible governance

[15][23]

Shrinking cities

[8]

Territorial cohesion Citizen participation/Citizen engagement /Civic engagement Engaged Communities

[8]

Public and social services Public-people-Private Partnership

[21]

19

From the communities and groups perspective, respecting their needs and wishes are also crucial [18]. People and communities’ factors from the literature are given in Table 7. Table 6 Management and organization success factors Dimension

[15][23] [15][23] [15][23][21] [8]

[22][28][18]

Management and organization

[29]

[19]

Table 5 Government success factors Dimension

Government

3.6

Factors

Open Government / Open Data Political strategies and perspectives

References

[21] [21]

Management and Organization

Although management and organization are crucial factors for implementing and sustaining the smart city initiatives and projects it is the least researched dimension in the literature [15]. In this context manager’s attitudes and behavior as leading, choosing best practices and being responsible for staff training are significant [19]. Also resource management is crucial as one of the biggest challenges of smart city management is limited funding [20]. Factors related to management and organization are given in Table 6.

3.7

People Communities

Since people and communities are the core components of every city, they should be the focus of smart city initiatives and projects. Smart people is defined as electronic skills, working in ICT-based work, having access to necessary education and training, human resources and capacity management, within a society that values creativity and encourages innovation [2]. Improving the social cohesion, reducing the digital divide, developing a better quality of life, reducing unemployment, poverty and improving education and safety are some of the main objectives of this context [8].

Factors

References

Project size Manager’s attitudes and Behavior Users or organizational diversity Lack of alignment of organizational goals and project Multiple or conflicting goals Resistance to change

[15][14]

Digital asset management Physical asset management Performance management Customer/stakeholder focus Service enablement

[27]

Service delivery Resources Management / Sustainable resource management Lack employees with integration skills and culture

[27]

[15][14] [15][14] [15][14] [15][14] [15][14]

[27] [27] [27] [27]

[21] [15]

Table 7 People communities success factors Dimension

People communities

Factors

References

Social cohesion

[8]

Cyber Security

[8]

Inclusion

[21]

Education

[21][15][14]

Social & human capital

[22][18]

Social Infrastructure

[19]

Digital divide(s) Information and community gatekeepers Participation and partnership Communication

[15][14]

Quality of life

[15][14]

Social and ethnic plurality

[21]

Flexibility Cosmopolitanism/openmindedness Participation in public life

[21]

ISBN: 1-60132-503-7, CSREA Press ©

[15][14] [15][14] [15][14]

[21] [21]

20

Int'l Conf. Internet Computing and Internet of Things | ICOMP'19 |

3.8

[2]

Technology

Smart city definitions suggest that infrastructure is a core component and that technology is the enabler which makes the vision possible [30]. ICT-based applications and infrastructure are one of the key drivers of smart city initiatives [31]. Although ICT integration promote smart city initiatives it also causes digital divide and reduces social inclusion [25]. Some of the technological challenges smart cities endure are associated with IT skills and organizational factors [32][8]. Technology success factors are given in Table 8. Table 8 Technology success factors Dimension

Factors

IT Skills (IT training programs)

Availability of ICTinfrastructure ICT infrastructure deficit

Technology

4

[1]

[4] [5] [6]

[7]

References

[15] [21]

[8]

[8]

Technology Infrastructure

[21]

Mobility

[21][8]

Speed

[33]

Real-Time monitoring

[34]

Interoperability

[35][30]

[9]

[10]

[11]

Conclusion

This study focuses on success factors of smart city initiatives from all aspect of smart cities. By a comprehensive literature review presented above, we have identified eight dimension of smart cities and the success factors under each dimension. Identified eight dimensions are: Technology, Governance, People Communities, Economy, Environment, Government, Built Infrastructure, Management and Organization. These dimensions represent different aspects of smart city initiatives and the factors under each dimension give critical points to be considered for successful smart city initiatives and each factor provides a basis for evaluating success of smart city initiatives. Also the factor list can be used to compare and evaluate the success of different smart city initiatives in different domains. Furthermore, the factor list can be used to develop smart city strategies. As a future work, it is planned to identify the Key Performance Indicators (KPIs) for each success factor and conduct an explanatory case study for validating usefulness and adequacy of the proposed model.

5

[3]

References

[12] [13] [14] [15] [16] [17] [18] [19] [20]

M. Brockerhoff and U. Nations, “World Urbanization Prospects: The 1996 Revision,” Population and Development Review, vol. 24, no. 4. UN Department of Economic and Social Affairs, p. 883, 2006.

[21]

M. Lancaster, “Summary for Policymakers,” Clim. Chang. 2013 - Phys. Sci. Basis, vol. 53, pp. 1–30, 1989. A. Caragliu, C. Del Bo, and P. Nijkamp, “Smart cities in Europe,” J. urban Technol., vol. 18, no. 2, pp. 65–82, 2011. J. Borja, “Counterpoint: Intelligent cities and innovative cities,” Univ. Oberta Catalunya Pap. E-Journal Knowl. Soc., vol. 5, 2007. J. MARCEAU, “Introduction: innovation in the city and innovative cities. Innovation: Management, Policy & Practice,” Innovation, vol. 49, no. 2–3, pp. 136–145, 2008. D. Toppeta, “The Smart City Vision: How Innovation and ICT Can Build Smart, ‘Livable’, Sustainable Cities. The Innovation Knowledge Foundation,” http//www.thinkinnovation.org/file/research/23/en/Top peta_Report_005_2010.pdf. [57], 2010. O. S. Kim, P. Meincke, O. Breinbjerg, and E. Jørgensen, “Method of moments solution of volume integral equations using higher-order hierarchical Legendre basis functions,” Radio Sci., vol. 39, no. 5, p. 17, 2004. A. Monzon, “Smart Cities Concept and Challenges: Bases for the Assessment of Smart City Projects. University of Madrid Transport Research Centre.,” in 2015 international conference on smart cities and green ICT systems (SMARTGREENS), 2015, pp. 1–11. E. P. Weber and A. M. Khademian, “Wicked problems, knowledge challenges, and collaborative capacity builders in network settings,” Public Adm. Rev., vol. 68, no. 2, pp. 334–349, 2008. S. S. Dawes, A. M. Cresswell, and T. A. Pardo, “From ‘Need to Know’ to ‘Need to Share’: Tangled Problems, Information Boundaries, and the Building of Public Sector Knowledge Networks,” Public Adm. Rev., vol. 69, no. 3, pp. 392–402, 2009. P. Neirotti, A. De Marco, A. C. Cagliano, G. Mangano, and F. Scorrano, “Current trends in Smart City initiatives: Some stylised facts,” Cities, vol. 38, pp. 25–36, 2014. V. Albino, U. Berardi, and R. M. Dangelico, “Smart cities: Definitions, dimensions, performance, and initiatives,” J. urban Technol., vol. 22, no. 1, pp. 3–21, 2015. R. Giffinger, C. Fertner, H. Kramar, and E. Meijers, “Cityranking of European medium-sized cities,” Cent. Reg. Sci. Vienna UT, pp. 1–12, 2007. S. D. B. Devaru, “Influence of Smart City on Entrepreneurial Activities.” H. Chourabi et al., “Understanding smart cities: An integrative framework,” in 2012 45th Hawaii international conference on system sciences, 2012, pp. 2289–2297. M. Eremia, L. Toma, and M. Sanduleac, “The Smart City Concept in the 21st Century,” Procedia Eng., vol. 181, pp. 12–19, 2017. S. Dirks and M. Keeling, “A vision of smarter cities, how cities lead the way into a prosperous an sustainable future,” New York IBM Glob. Serv., p. 18, 2009. N. Kogan and K. J. Lee, “Exploratory research on success factors and challenges of Smart City Projects,” Asia Pacific J. Inf. Syst., vol. 24, no. 2, pp. 141–189, 2014. G. Maccani, B. Donnellan, and M. Helfert, “A Comprehensive Framework for Smart Cities.,” in SMARTGREENS, 2013, pp. 53–63. S. Alawadhi et al., “Building understanding of smart city initiatives,” in International conference on electronic government, 2012, pp. 40–53. B. Benamrou, B. Mohamed, A. S. Bernoussi, and O. Mustapha, “Ranking models of smart cities,” in Colloquium in Information Science and Technology, CIST, 2017, pp.

ISBN: 1-60132-503-7, CSREA Press ©

Int'l Conf. Internet Computing and Internet of Things | ICOMP'19 |

[22] [23] [24] [25]

[26]

[27] [28] [29]

[30]

[31] [32] [33]

[34]

[35]

872–879. S. P. Caird and S. H. Hallett, “Towards evaluation design for smart city development,” J. Urban Des., pp. 1–22, 2018. S. Joshi, S. Saxena, and T. Godbole, “Developing smart cities: An integrated framework,” Procedia Comput. Sci., vol. 93, pp. 902–909, 2016. A. Mahizhnan, “Smart cities,” Cities, vol. 16, no. 1, pp. 13– 18, 1999. N. Odendaal, “Information and communication technology and local governance: understanding the difference between cities in developed and emerging economies,” Comput. Environ. Urban Syst., vol. 27, no. 6, pp. 585–607, 2003. J. R. Sanchez-Valencia, R. Longtin, M. D. Rossell, and P. Gröning, “Growth Assisted by Glancing Angle Deposition: A New Technique to Fabricate Highly Porous Anisotropic Thin Films,” ACS Appl. Mater. Interfaces, vol. 8, no. 13, pp. 8686–8693, 2016. T. C. Dearing and M. J. Cronin, Managing for Social Impact. [electronic resource] : Innovations in Responsible Enterprise. Springer, 2017. L. Pham, T. T. Mai, and B. Massey, “Key Factors for Effective Citizens Engagement in Smart City: The Case of Cork City.” A. Cavoukian and M. Chibba, “Cognitive cities, big data and citizen participation: The essentials of privacy and security,” in Towards Cognitive Cities, Springer, 2016, pp. 61–82. T. Nam and T. A. Pardo, “Conceptualizing smart city with dimensions of technology, people, and institutions,” in Proceedings of the 12th annual international digital government research conference: digital government innovation in challenging times, 2011, pp. 282–291. R. G. Hollands, “Will the real smart city please stand up?,” City, vol. 12, no. 3, pp. 303–320, 2008. Z. Ebrahim and Z. Irani, “E-government adoption: architecture and barriers,” Bus. Process Manag. J., vol. 11, no. 5, pp. 589–611, 2005. E. Asimakopoulou and N. Bessis, “Buildings and crowds: Forming smart cities for more effective disaster management,” in 2011 Fifth International Conference on Innovative Mobile and Internet Services in Ubiquitous Computing, 2011, pp. 229–234. A. Attwood, M. Merabti, P. Fergus, and O. Abuelmaatti, “SCCIR: Smart cities critical infrastructure response framework,” in 2011 Developments in E-systems Engineering, 2011, pp. 460–464. L. Anthopoulos and P. Fitsilis, “From online to ubiquitous cities: The technical transformation of virtual communities,” in International Conference on e-Democracy, 2009, pp. 360–372.

ISBN: 1-60132-503-7, CSREA Press ©

21

22

Int'l Conf. Internet Computing and Internet of Things | ICOMP'19 |

ISBN: 1-60132-503-7, CSREA Press ©

Int'l Conf. Internet Computing and Internet of Things | ICOMP'19 |

SESSION INTERNET COMPUTING, CLOUD COMPUTING, COMMUNICATION SYSTEMS, AND RELATED ISSUES Chair(s) TBA

ISBN: 1-60132-503-7, CSREA Press ©

23

24

Int'l Conf. Internet Computing and Internet of Things | ICOMP'19 |

ISBN: 1-60132-503-7, CSREA Press ©

Int'l Conf. Internet Computing and Internet of Things | ICOMP'19 |

25

ost-Independent Autonomous Decision of Operations of the Redundant Traffic Reduction Node Michi Ishikuro, Ken i Ichi o, and Akiko Narita Graduate school of Science and Technology, Hirosaki University, Hirosaki, Aomori, Japan

Keywords: traffic reduction, packet cache, TCP/IP, IPv6, autonomous decision.

Our research group has developed a network node with packet cache to reduce the redundancy in the traffic [1]-[8]. We call it the TR (traffic reduction) node. The TR node on the upstream side changes received data into small identifier and some information to rebuild them. The node on the downstream side reproduces the original data using the identifier and information. We can reduce redundant traffic between the TR nodes. Therefore, each node must execute appropriate operation including cache synchronization. Otherwise, the system cannot accomplish data transmission rather than reducing redundant traffic. A simple way is assigning the functions of the TR nodes statically. However, dynamic decision of TR node operations is desirable for flexible and effective utilization. In this paper, we present a method for host-independent autonomous decision of roles and operations of the TR node using IPv6 extension header.

1

2

Abstract - In computer networks, we often find the same data transferred through the same rout repeatedly. We have developed the TR (traffic reduction) node that reduces such redundancy of concurrent transmission in a TCP/IP network for efficient use of network resources. The TR nodes reduce redundancy in traffic cooperatively using packet cache. They must fulfill a proper role in their relationship. Otherwise, they disturb the transmission rather than yielded benefit. For flexible use and easy management, it is desirable for them to choose their operation dynamically according to conditions. In this paper, we proposed the method that enables the TR node to acquire information about alignment and to decide their operations host-independently and autonomously.

Introduction

In computer networks, we often find the same data transferred through the same rout repeatedly. Such redundancy in transmission should be reduced for productive employment of limited resources of the computer networks. If we transmit data with substantially multiple destinations, we can restrain generation of redundant traffic. Therefore, multicast is capable for efficient transmission. In general, IP multicast, by which packets are duplicated at branch points of transmission paths, is considered the most powerful technique to constrain redundancy of data in concurrent transmission. However, requirements for equipment on the transmission route and transport layer protocol restrict coverage of it. On the other hand, host computers duplicate data and forward them to the next destination in application level multicast. The strategy needs nothing special for routers instead of overheads of hosts and disadvantage in efficiency compared with IP multicast. Caching is another means for eliminating redundant traffic. Proxy server is generally used to shorten transmission paths for redundant data. However, it is difficult to reduce redundancy of concurrent transmission for delay. The server usually stores data as files on secondary storage. For concurrent transmission, unit of caching data must be small to utilize fast primary storage.

Related Works

Several research groups have studied packet caching for redundant traffic reduction [9]-[12]. Springer et al. proposed basic concept of the packet caching as shared cache and presented its potential. The purpose of their cache was application independent traffic elimination for bandwidthconstrained channel. Anand et al. explored the benefits of deploying packet-level redundant content elimination across the entire internet [10]. They assumed that all routers are capable of executing redundancy aware routing managed by centralized rout computation. In [11], authors designed decoding nodes to be resource saving so that not all of the routers need to store the same data. Yamamoto et al. [12] designed packet-caching system focusing on P2P traffic exchanged by BitTorrent traversing multiple domain networks. This restriction enables management of data structure in their cache and cache synchronization simple. We have constructed the TR node as transparent to host computers and covering internet similar to [10]. Additionally, we have strived to build the TR node system without central management node for flexible operation and management in practical convenience. Therefore, data management and data detection algorithm must be application independent. Avoiding overhead for detecting data in the cache, the TR node has used the leading byte of data as a hash value for fast searching, while Rabin fingerprint [13] was employed in [9], [10], and [11]. We found that overhead of applying Rabin fingerprint was not negligible for the TR node [5]. For

ISBN: 1-60132-503-7, CSREA Press ©

26

Int'l Conf. Internet Computing and Internet of Things | ICOMP'19 |

memory saving, we have studied cache utilization. We developed more efficient cache replacement algorithm by modifying FIFO (first in first out) [4]. We obtained more successful result for the TR node than using pure FIFO in [9], [10], and [11] and LRU (least recently used) in [12]. Furthermore, we utilized selective recording of data in cache based on difference of byte offset among redundant streams [6] and transmission rate [8]. For cache synchronization, Yoshida et al. proposed a method to confirm the existence of TR nodes at the beginning of TCP connection and to enable the TR nodes to decide their operation autonomously [4]. Ikeda et al. improved it [7]. However, the method has a disadvantage that detracts the transparency from the TR node.

3

Traffic reduction with the TR node

We have designed the TR node as a network router with high functionality to reduce redundant traffic of data transmitted concurrently. They are live broadcast, video conference, concurrent distribution of update files, and so on. Fig. 1 presents basic operation of the TR node. The TR node on upstream side change received data into code of data identifier and some information. The TR node on the downstream side reproduces the original data from the code. Size of the code is much smaller than that of the original data. Therefore, traffic is reduced on the network segment between the TR nodes. Ordinary routers or the other TR nodes may be situated between them. Detailed operations are explained in [1] and [3]. We implemented functions of the TR node on a program. It works on Linux operating system covering TCP/IP/Ethernet environment for easy validation in development. We have improved the TR node in processing speed and efficiency. There are two types of roles for the TR node essentially. We call one of them encoding node and the other decoding node. One encoding node and one or more decoding nodes are required for redundant traffic reduction on a transmission path for a TCP stream. Operations of the encoding node are monitoring streams, synchronizing cache, searching data in cache, retransmission detection, and encoding data. The node monitors streams and categorize them according to redundancy, sameness, and so on. Categories of streams are necessary for efficient utilization of the cache. When the encoding node receives a packet, the node searches the same data in its cache. The node encodes the data if it finds them, or records the data in its cache and directs cache synchronization to the TR node on the downstream side. For the second or latter time the encoding node receives the same data, the node can cut down size of data and send them directing to rebuild original data. The reason why the node detects retransmission is described later. Operations of the decoding node are synchronizing cache, forwarding data, and rebuilding original data. The decoding node at the most down streamside deletes instruction from the encoding node and rebuilds the original data if required. The other decoding nodes forward packets with the instruction. All the decoding

nodes record data of a packet with instruction for cache synchronization.

Redundant data YYYYY... YYYYY... YYYYY... cache ID data 0 XXXXX... 1 1 YYYYY... 1 YYYYY... Converted small data Original data for cache synchronization cache ID data 0 XXXXX... 1 YYYYY... YYYYY...

YYYYY...

1 YYYYY... Original data YYYYY...

Figure 1. Basic operation of the TR node. The TR nodes must accomplish appropriate operation to assure normal communications for application programs on the host computers. In this respect, the most important operation is rebuilding the original data. If the most downstream node forwards encoded data, encoded data arrives at the destination host. The host detects checksum error and discards it. After that, TCP retransmission timer on the sender expires and the TCP module retransmits the data. The encoding node detects the retransmission. Then the node forwards the retransmitted data without decoding. The data are delivered to the receiver host correctly. The TR node avoids more retransmissions caused by wrong operation in this manner. Still TCP module on the sender host assumes congestion for lack of acknowledgement and hinders transmission rate. That is, wrong operation of the TR node interferes with transmission. The simplest method to let the TR node operate appropriately is static assignment of the roles. However, the method lacks in flexibility. Furthermore, we will improve the TR node to suspend and restart its function for traffic reduction in response to status such as redundancy in traffic, load of relay nodes, and so on. Autonomous decision of the TR node operation is indispensable for independent suspending and restarting of respective TR nodes. We improved the TR node for autonomous decision in the

ISBN: 1-60132-503-7, CSREA Press ©

Int'l Conf. Internet Computing and Internet of Things | ICOMP'19 |

previous study for [4] and [7], and realized host-independent decision of its operation in this study.

4 4.1

Proposed TR node Basic operation for autonomous decision

For autonomous decision of the TR node, the node must exchange information about coworkers. In [4], host computers make space in the IPv4 header for information about TR node operations, and detect number of TR nodes on the transmission route when a TCP connection is established. For simple implementation of host-independent function in this study, we have chosen IPv6 for network layer protocol and utilized hop-by-hop option of extension header. The TR node adds the option field to IPv6 header on demand and removes it after decoding. Fig. 2 presents the formats for the option used. We assigned 5316 as option type of them. This annexure increases packet length. If the length exceeds MTU (maximum transmission unit) of the link layer, the TR node divides the packet and adds IPv6 fragment header to the fragments. The receiver unites them. This fragmentation increases traffic slightly. The encoding node monitors streams whether they are redundant or not to evade adding the option to the packet of non-redundant streams. Reduced size is much larger than incensed size by the option for redundant streams. Initially, all the TR nodes are neutral in terms of the types. When the neutral node receives the first packet of a stream, the node adds coworker detection option presented in Fig. 2 (a) setting 1 in TR hop field. The value of the field is number of TR nodes on the upstream side for the receiver node. Stream ID is an integer given by the node. If the node receives a packet with coworker acknowledgement option, the node becomes an encoding node. Alternatively, the neutral node that receives a packet with coworker detection option becomes a decoding node associating with the source node of the option. The node adds 1 to the value in the TR hop field and relays the option, and returns coworker acknowledgement option. All neutral nodes on the downstream side of the transmission path receive the coworker detection option and acknowledge. The packet with the coworker detection option arrives at the receiver host eventually. The host neglects the option. Thus, the TR nodes can know their types and obtain information of coworkers. The TR node system begins traffic reduction after that. The encoding node detects coworkers and registers configuration of coworkers for each redundant stream. Additionally, the node also sends packets with the option periodically. The system can find change of configuration. Moreover, when the decoding node suspends its function as a TR node, it can advertise the change of its status to the encoding node for prompt action using option of Fig. 2(b). For traffic reduction, the encoding node adds IPv6 extension header shown in Fig. 2 (c) to the packets of object streams. The node set number of coworker TR nodes as the value of the total hop filed. The initial value of past hop field is 1. Each decoding node increments past hop field by 1. If the instruction field indicates cache synchronization, coworker

27

TR nodes record TCP data field of the packet on their packet caches. If the instruction is to decode and the total hop value and incremented past hop value are equal, the node rebuilds the original data. 0 31 Next Header Hdr Ext Len option type Opt Data Len instruction TR hop stream ID reserved IPv6 address of the source node(for coworker detection only) (a) Coworker detection option (instruction f316) and coworker acknowledgement option (instruction f516). 0 Next Header instruction

Hdr Ext Len

option type reserved

31 Opt Data Len

(b) Suspend notification option (instruction f616). 0 Next Header instruction

Hdr Ext Len total hop

option type past hop

31 Opt Data Len reserved

(c) Traffic reduction option. Instruction codes are the same as in [3]. See [14] about Next Header, Hdr Ext Len, option type, Opt Data Len, and text about the other fields. Figure 2. Proposed IPv6 extension header.

4.2

Branch topology

The TR node must care for cache coherency especially for branch topology. The encoding node keeps information about streams to guarantee cache coherency. Each stream is identified with IP addresses and port numbers of source and destination specified in TCP and IP headers. The encoding node forms a stream group with such streams that are conveying the same contents. Redundant data are stored with a stream group identifier in the cache. The encoding node knows which coworker nodes relay each stream. If transmission paths of redundant streams of the same content spread out toward downstream side, the encoding TR node must send the same data multiply for cache synchronization. For example, when TR node A in Fig. 3 may relay a packet to a receiver host beyond TR node B. The packet does not go through TR node D. The encoding node lists coworker TR nodes for management of cache coherency. Before the encoding node sends encoded data, the node confirms that the coworkers have copy of the data using stream information. If the node finds incoherency with coworkers for the stream, the node sends a packet with timely synchronization instruction toward the most downstream coworker. All coworkers on midstream record the data. Data path for timely synchronization is not optimal in this study. It can be shorten with modification.

ISBN: 1-60132-503-7, CSREA Press ©

28

Int'l Conf. Internet Computing and Internet of Things | ICOMP'19 |

If several branches come together from upstream side, cache synchronization may fail without exclusive control, due to competition. For example of the competition, TR node F in Fig. 3 (b) may overwrite data that come through TR node E in its cache purging data that has been sent through TR node D. TR node E is unaware of the replacement so that the node may send an encoded packet assuming that TR node F can decode it properly. TR node F may execute procedure for decoding referring incoherent. We can solve this problem by hiding TR nodes on downstream side from the encoding node. As mentioned in the former subsection, when the neutral TR node receives the first packet with coworker detection option, the node recognizes the source of it as a peer node for encoding. After that, when the TR node receives packet with the option from the other TR node, the decoding node removes the option and does not acknowledge. We avoid competition by limiting number of associating encoding node as 1 for a deciding node. TR node A senders

TR node B

TR node C

senders

(a) Configuration of the experiment for decision of operations.

6 redundant streams

receivers receivers

(a) Branches spread out toward downstream side. TR node D (encoder) senders

5 redundant streams

(b) Configuration of the experiment for cache synchronization for branch topology.

3 redundant streams

TR node F (decoder) receivers

TR node E (encoder) (b) Branches from upstream side meet.

Figure 3. Branch topologies. Arrows indicate direction of transmission.

5

Evaluations and discussion

In this section, we present three validations of the proposed method with simple computer networks. We present the configurations of them in Fig. 4. Specifications of the hardware and software used are given in Table I. Line speed in the experiments were 100 Mbps. We set up a 1000-record packet cache in each TR node. Receiver machines emulated multiple receivers appropriately for number of redundant streams using processes. The application program on the servers continued to request transmission. Consequently, the server tried to utilize bandwidth fully. The program sent data in rotation of the streams piece by piece. The data are sequence of random numbers. We measured amount of data size received and sent during each second on the TR node. For comparison, we also performed measurements without coworker detection and related functions.

3 redundant streams

(c) Configuration of the experiment for exclusive control for branch topology. Figure 4. Configurations for measurements in this study. TABLE I.

HARDWARE AND SOFTWARE SPECIFICATION OF THE EXPERIMENTAL SYSTEM

Function Sender Sb1, Sc1

Specifications CPU Memory OS Intel Core i5-3470T (2.90GHz)/ Ubuntu 14.0.4_64 bits/ 4 GB

The other senders and receivers

Intel Pentium G645 (2.90 GHz)/ Ubuntu 14.0.4_32 bits/ 2 GB

TR nodes Tb1, Tc1

Intel Core i5-3570 (3.40GHz)/ Ubuntu 16.0.4_64 bits/ 8 GB

The other TR nodes

Intel Core i3-3220 (3.30 GHz)/ Ubuntu 14.0.4_64 bits/ 8 GB

ISBN: 1-60132-503-7, CSREA Press ©

Int'l Conf. Internet Computing and Internet of Things | ICOMP'19 |

received by Ta1

80 60 40

sent by Ta1

20 0

0

30

60

90 120 time[s]

150

180

(a) Results without coworker detection. amount of data per second [Mbits/s]

100

received by Ta1

80 60 40

sent by Ta1

20 0

0

30

60

90 120 time[s]

150

180

(b) Results with the proposed method. Figure 5. Experimental results for decision of operations. TR node Ta1 received approximately 100Mbytes and sent about 27Mbytes per second in both measurements with and without coworker detection over the first 50 seconds. These values are almost ideal in traffic reduction by the TR node for 5 redundant streams. During the next 50 seconds, throughput of transmission was very low without coworker detection. In this period, Ta1 without coworker detection encoded redundant data in spite of absence of the decoder TR node. Then packets with encoded data arrived at receiver Ra1. They were discarded for check sum error. Sender Sa1 could not receive acknowledgement so that restrained transmission rate assuming congestion. With coworker detection, amount of data received and sent by Ta1 were the same in this period. We can see that Ta1 suspended encoding. At the time point of 100 seconds, Ta2 became a decoding node. Proper

Cache synchroni ation in branch topology

We evaluated cache synchronization for the topology with a spread of branches toward downstream side using a configuration given in Fig. 4 (b). Server Sb1 sent data of 6 redundant TCP streams. Receivers Rb1 and Rb2 were destination of 3 streams, respectively. Fig 6 presents average of amount of transmitted data size per minute over 300 seconds. Throughputs were very low without coworker detection throughout this measurement. They were hindered by TCP congestion control and accommodation of delivery caused by characteristics of the application program. The program executed on server Sb1 requested transmission to receiver Rb1 first and Rb2 next in the rotation. When the packet sent to Rb2 arrived at TR node Tb1, the node found the same as ones received in its cache. Tb1 encoded and passed them to Tb3 even though cached data were incoherent. Tb3 converted them wrongly. Rb2 discarded packets for check sum error. Sb1 assumed congestion and diminished the transmission rate. Whereas, the application program on the server continually requested data transmission rotationally, and transmissions to Rb1 proceeded successfully. Before long, the system buffer for the transmission to Rb2 became full of unsent data in Sb1. Then the program suspended transmission request until vacancy was generated in the buffer following the order of transmission. Therefore, transmission rate to Rb1 was also hindered. By contrast, throughputs were kept if we utilize the coworker detection, since appropriate cache synchronization was accomplished. 100 80 60

sent by Tb3 received by Tb1

40

sent by Tb2 received by Tb3 received by Tb2

20 0

without coworker detection with coworker detection

amount of data per second [Mbits/s]

100

5.2

without coworker detection with coworker detection

We inspected transmitted data using the configuration given in Fig. 4 (a) for proper decision of roles and operations of the TR nodes. We changed the function of TR node Ta2 during the measurements. Ta2 worked as a TR node in the first 50 seconds, behaved as an ordinary router in the next second 50 seconds, and became a TR node again after that. Interval of coworker detection by Ta1 was 30 seconds. Server Sa1 established 5 TCP connections to transmit redundant data. Fig. 5 presents amount of data received and sent in each second by Ta1.

transmission was accomplished and throughput recovered in the last time region. Note that Ta1 with coworker detection restart encoding at the time point of 120 seconds at last. Time interval of coworker detection of Ta1 caused the delay of restarting encoding. We can shorten the interval time to decrease the delay.

without coworker detection with coworker detection

Decision of operations

average of amount of transmitted data size per second [Mbits/s]

5.1

29

Figure 6. Experimental results for cache synchronization in branch topology.

ISBN: 1-60132-503-7, CSREA Press ©

Int'l Conf. Internet Computing and Internet of Things | ICOMP'19 |

5.3 .

Exclusive control for branch topology

amount of data per second [Mbits/s]

We confirmed exclusive control avoiding competition of cache utilization in branch topology in which several branches come together from upstream side using a configuration given in Fig. 4 (c). Servers Tc1 and Tc2 sent different data sets using 3 redundant streams respectively. Transmission rates of the servers are limited by the bandwidth of the switching hub in this experiment. Topologically, amount of data size reconstructed and sent by Tc3 equals to that received by Tc1 and Tc2. Fig. 7 and Fig. 8 present the results without and with coworker detection. 100 received by Tc1

80 60

sent by Tc1

40

received by Tc2 sent by Tc2

20 0

0

100

time[s]

200

300

amount of data per second [Mbits/s]

(a) Received and sent data by Tc1 and Tc2. 100

node encoded and sent them since it is unaware of the replacement in Tc3. Tc3 convert the encoded data wrongly. Receiver Rc1 detected error of the packet referring TCP check sum and discarded it. Sc2 waited the acknowledgement for retransmission time out. Then transmission rate of Sc2 was limited by congestion window while Sc1 continued to deliver increasing congestion window size. Eventually, retransmission timer of Sc2 expired and the server assumed congestion on the transmission route to Rc2. Then the server suppressed transmission rate. Cached data from Sc2 on Tc3 were frequently overwritten in a short while because of much larger transmission rate of Sc1. Tc2 repeatedly constructed incorrect data and so that Sc2 failed to deliver. On the other hand, Tc3 also replaced cached data from Sc1 with those from Sc2. The purged data were often old for delivery of Sc1 so that the replacement caused transmission failure from Sc1 infrequently. Thus, The TR nodes disturbed transmission from Sc2 considerably. amount of data per second [Mbits/s]

30

100

80

60 40

60

sent by Tc1 0

100

time[s]

200

300

(a) Received and sent data by Tc1 and Tc2.

0

100

time[s]

200

300

(b) Sent data by Tc3. Figure 7. Experimental results for competition without coworker detection and exclusive control in branch topology. As shown in Fig. 7, both TR nodes Tc1 and Tc2 without coworker detection cut down size of received data. We find large difference between throughput of Tc1 and Tc2. Servers Sc1 and Sc2 share the bandwidth between the switching hub and Tc3 almost equally unless one of them refrains from transmission voluntarily. However, order of packets caused the disproportion. Server Sc1 and Sc2 began to deliver redundant data independently and almost simultaneously. In this measurement, Tc3 received the first packet for cache synchronization through Tc2. Nevertheless, Tc3 replaced the cached record with data of a packet through Tc1 soon. When Sc2 sent the same data as recorded in the cache of Tc2, the

amount of data per second [Mbits/s]

20 0

received by Tc1

20 0

40

received by Tc2 sent by Tc2

80

100 80 60 40 20 0

0

100

time[s]

200

300

(b) Sent data by Tc3. Figure 8. Experimental results for exclusive control with coworker detection in branch topology. With coworker detection, only TR node Tc1 cut down size of data as shown in Fig. 8. In this experiment, both TR nodes Tc1 and Tc2 tried to detect coworkers. Tc3 received

ISBN: 1-60132-503-7, CSREA Press ©

Int'l Conf. Internet Computing and Internet of Things | ICOMP'19 |

the first packet with coworker detection option from Tc1 first. Then Tc3 positioned Tc1 as an encoding node and acknowledged to only Tc1 as a coworker. Tc2 could not find coworkers so that the node did not send packets with options of the TR node operations. In this experiment, streams through Tc1 and Tc2 used bandwidth between Tc3 and Rc1 successfully but unequally. The streams through Tc2 employed about 60 % of it. Nonetheless, we can avoid fatal interference of the transmission with the proposed coworker detection and exclusive control.

6

Conclusion and Future works

In this paper, we proposed the method that enables the TR node to decide operations for redundant traffic reduction host-independently and autonomously. Before the TR nodes begin reducing redundant traffic, they construct a group for redundant traffic reduction. A candidate of an encoding node tries to find coworker nodes and other nodes advertise their existence to become decoding nodes. Using information about coworkers and instruction from the encoding node, the TR nodes know their roles and choose appropriate operations including cache synchronization. We employed IPv6 hop-byhop option of extension header for exchanging the information among the TR nodes. For future works, we will improve utilization of packet cache by organizing cache groups. In the present system, all TR nodes on the transmission path keep cache coherence with one another. However, if a TR node only forwards data for particular contents, cached data of the contents are unnecessary. We can utilize such a node as a decoding TR node for another encoding node potentially and achieve more effective traffic reduction.

7

References

[1] Yuki Otaka and Akiko Narita. Efficient Packet Cache Utilization of a Network Node for Traffic Reduction”: Proceedings of the 2013 International Conference on Internet Computing and Big Data (ICOMP '13), pp. 109-112, Jul. 2013. [2] Yuki Otaka and Akiko Narita, Efficient Assignment of Packet Cache Region for Traffic Reduction of Multiple Redundant Contents”; Proceedings of the 2014 International Conference on Internet Computing and Big Data (ICOMP '14), pp. 117-123, Jul. 2014. [3] Shunsuke Furuta, Michihiro Okamoto, Kenji Ichijo, and Akiko Narita, “Expansion of Region for Redundant Traffic Reduction with Packet Cache of Network Node, Proceedings of the 2015 International Conference on Internet Computing and Big Data (ICOMP '15)”, pp66 -72, Jul. 2015. [4] Tomohiro Yoshida, Yamato Ikeda, Kenji Ichijo, Akiko Narita, “Autonomous Decision of Function of the Redundant Traffic Reduction Node Using a New IP Option,”

Proceedings of the 2016 International Conference on Networking and Network Applications (NaNA 2016), pp. 233-238, Jul. 2016. [5] Kyosuke Doh, Kenji Ichijo and Akiko Narita. “Detection of partial agreement in byte streams for the Redundant Traffic Reduction Node”, Tohoku-Section Joint Convention of Institutes of Electrical and Information Engineers, 1H02, Aug. 2016. [6] Wataru Yokota, Kenji Ichijo, Akiko Narita, “Optimization of Packet Cache Utilization of Network Node for Redundant Traffic Reduction,” Proceedings of the 2017 International Conference on Internet Computing and IoT, pp.52-58, Jul. 2017. [7] Yamato Ikeda, Kenji Ichijo, Akiko Narita, “Autonomous decision of redundant traffic reduction with network nodes,” Proceedings of the 309th SICE Tohoku Chapter work shop, http://www.topic.ad.jp/sice/htdocs/papers /309/309-9.html, Jun. 2017. [8] Kyosuke Doh, Kenji Ichijo, Akiko Narita, “Cache Optimization in Reduction of Redundant Traffic with Variety in Delivery Progress and Transmission Rate”, Proceedings of the 2018 International Conference on Internet Computing and Big Data (ICOMP '18)”, pp17 -23, Jul. 2018. [9] N. Spring and D. Wetherall. “A protocol-independent technique for eliminating redundant network traffic”; Proceedings of the ACM SIGCOMM 2000 Conference on Applications, Technologies, Architectures, and Protocols for Computer Communication, pp. 87-95, Aug. 2000. [10] Ashok Anand, Archit Gupta, Aditya Akella. “Packet Caches on Routers: The Implications of Universal Redundant Traf c Elimination”; Proceedings of the ACM SIGCOMM 2008 conference on Data communication, pp. 219-230, Aug. 2008. [11] Ashok Anand, Vyas Sekar, and Aditya Akella. “SmartRE: an architecture for co-ordinated network-wide redundancy elimination”; SIGCOMM Computer Communication Review, 39 (4), pp. 87-98, Sep. 2009. [12] Shu Yamamoto, Akihiro Nakao. “P2P packet cache router for network-wide traffic redundancy elimination”; Proceedings of International Conference on Computing, Networking and Communications (ICNC), 2012, pp. 830-834, Jan. 2012. [13] M. Rabin. “Fingerprinting by random polynomials”; Harvard University Technical Report, TR-15-81, 1981. [14] S. Deering and R. Hinden, “Internet Protocol, Version 6 (IPv6) Specification,” RFC8200, Jul. 2017.

ISBN: 1-60132-503-7, CSREA Press ©

31

32

Int'l Conf. Internet Computing and Internet of Things | ICOMP'19 |

Canarycurity: A Robust Security Surveillance System Built with Mobile and IoT Technologies Brandon Horowitz, Chen-Hsiang Yu, Leonidas Deligiannidis Department of Computer Science and Networking Wentworth Institute of Technology, Boston, MA, USA {horowitzb | yuj6 | deligiannidisl}@wit.edu

Abstract — Due to the increase of property crimes, home surveillance and security systems have become more and more important. However, these systems are normally expensive and not customizable. Instead of relying on one-size-fits-all solutions, the research question is how to build a scalable system with low cost equipment, including mobile devices, Internet of Things (IoT) technologies, etc. In this paper, we introduce Canarycurity, which is a Raspberry Pi (RPI)-based home surveillance system that addresses the issue of home security with a consideration of redundancy and scalability. The system consists of several RPI devices, sensors, cameras, and an Android application. Compared to existing solutions on the market, early testing of the system shows that Canarycurity not only provides surveillance security to a residential area, but it is also easy to scale the system to meet different needs. Keywords — Surveillance system, redundancy, IoT

1. INTRODUCTION

According to Criminal Victimization released by the U.S. Bureau of Justice in 2018, there were 108.4 victimizations per 1000 households in the category of property crime [1]. Following is a common scenario happening for many people: a family has been saving up money for months to buy a new television for the holidays. They purchase the television through online companies, such as Amazon, and since the family is at work when the package is supposed to arrive, they opt to have it left at their doorstep [2,3]. Days go by, and the package that was supposed to arrive never did. This has been happening to their neighbors as well. The family finds out that a porch pirate has recently been going around the neighborhood, stealing packages from families when they’re not home. The family files a police report, but neither the shipping company nor the credit card company offer a refund. Since none of the

robbed houses had a surveillance system, the robber is never identified. The family briefly considers investing in a home surveillance system, but after considering their financial situation and the expensive monthly payments most home surveillance systems cost, they opt against it. In this research, we argue that it is possible to design and develop a surveillance security system for the public. This system needs to be cost efficient, easy to use, scalable and redundant for future needs. In this paper, we propose Canarycurity, an inexpensive Raspberry Pibased surveillance system. In the following sections, we will address related work, the proposed idea, system design, implementation, and results of the project.

2. RELATED WORK

A lot of research has worked on home security issues. For example, Total security [4], is a home security system similar to our system. It also uses Raspberry PIs to cut down the costs, MQTT to communicate between devices, and an Android application for end users to interact with the system. It does not address redundancy and load balancing topics, however. While most solutions incorporate Wi-Fi connectivity, others are based on Bluetooth technologies [5]. Solutions and techniques presented in [7] could be used with our system. The authors in [7] propose a fire alarm system which uses wireless sensor networks. It uses pressure, humidity, temperature, etc. sensors, and if any of those values exceeds a predefined threshold, it sends a message through the user’s android application telling them of the problem. Of course, IoT monitoring solution can include other sensors. For example, Mathur et. al. [6] proposed a system, based on RPIs, for remote monitoring sensors to record the lower limb health of amputees. Mainwaring et. al. [8] use a wireless sensor network to monitor realworld habitats. It uses a wireless sensor network with

ISBN: 1-60132-503-7, CSREA Press ©

Int'l Conf. Internet Computing and Internet of Things | ICOMP'19 |

barometric pressure, humidity, temperature, etc. sensors to help identify problems in the animal habitat. The data is then sent to a mobile phone. Ying-Wen et. al. [9] use an embedded surveillance system based on ultrasonic sensors. The system uses an ultrasonic transmission and an ultrasonic receiver. When an intruder walks between the emitter and receiver, the signal is blocked (like a trip-wire), and the system uses that blocked signal as a trigger to start recording. An extension to this solution is presented in [10]. This solution proposes embedded surveillance systems with ultra-low power alerts. The user installs pressure sensors and PIR (Pyroelectric Infrared) Sensors that are built around a microcontroller in their home’s windows and doors. In Patil et. al.’s work [11], when motion is detected by a PIR sensor, the user is notified of the intruder activity via email or SMS. The email contains an attachment image with the intruder’s image and a message with a warning. In [12], a camera-based Raspberry Pi-based surveillance system is presented. It uses the changes detected between frames to determine when to save video. An elegant solution built entirely on cameras is presented in [13] Computationally expensive video analysis is performed by utilizing CUDA’s parallel processing capabilities. Several algorithms are presented to detect motion, motion in sub-area of the field of view, and fire.

3. REDUNDANCY

The facts and examples stated in the introduction show both the necessity and the cost of mainstream home surveillance systems. As explained in the papers under the Related Work section above, a Raspberry Pi-based IoT security system is a cheap and effective solution. In this paper, we are offering a similar solution: a cheap, IoT-based home surveillance system with potential upgradeability (easily upgrade failed or older components), customization, and redundancy options. It’s main difference between our solution and other IoT security systems is its redundant architecture. In other IoT-based systems, the broker is a major failure point of the system. If the broker ceases function for any reason, all devices in that IoT system are no longer able to communicate. In Canarycurity, redundancy is offered through the ability to set up additional devices as backup brokers. When one broker goes down, the backup broker will detect the failed broker and assume its identity and functionality.

4. SYSTEM DESIGN

Canarycurity uses the Message Queuing Telemetry Transport (MQTT) protocol to communicate between devices. The brokers run on EMQ X, which is a distributed, massively scalable, highly extensible MQTT message broker. We chose EMQ X because it is easy to use, and it supports clusters (needed for redundancy) and can run on RPIs; redundancy is the duplication of critical components of a system that enable a system to be reliable even at times when critical components fail. In order to setup failover clusters, we used keepalived and haproxy. Keepalived is the software that allows redundancy and high availability using virtual IPs and master/backup connections. Haproxy is a load-balancer program that allows for traffic to be split between any number of brokers. The proof of concept system uses a Pyroelectric Infrared (PIR) motion sensor connected to a Raspberry Pi to detect when a person comes within its line of sight. Once the sensor detects a person, the RPI sends a signal to the attached camera to take a picture and then starts recording. The RPI then sends the recording to the brokers on the IoT network. The brokers then do two things: a) download and store the recording, b) send the picture to the devices that subscribe to the appropriate topics. Figure 1 shows the back-end communication between the master and backup brokers, RPIs, and the mobile application.

Figure 1. Diagram of Canarycurity’s functionality. The Android App publishes (Log Request) and Subscribes to (Log). The Pi with the camera and the PIR sensor publishes the video recordings. The Pi for Log and video storage publishes (Log File) and subscribes to (Log)

ISBN: 1-60132-503-7, CSREA Press ©

33

34

Int'l Conf. Internet Computing and Internet of Things | ICOMP'19 |

5. IMPLEMENTATION

Canarycurity consists of four main parts: a) the Brokers, b) the Sensors/Cameras, c) the Storage/Logs, and d) a mobile application. 5.1 Brokers When a device wants to subscribe or publish to a topic, it must first connect to a broker. Normally, when one tries to connect to a broker, they use the IP address of the machine that the broker is running on. However, since Canarycurity is designed around the ability to use several machines as brokers (for redundancy reasons), using the IP address of a broker that can potentially go down is not preferable. To resolve this, Canarycurity uses keepalived to enable all brokers to use a single shared virtual IP address. Keepalived monitors the status of all other brokers running on the network and can automatically make any machine take over the virtual IP address to receive new connections. The haproxy configuration file, shown in figure 2, controls several things. “bind *:80” tells the broker to listen for connections on port 80. “balance leastconn” tells the broker which server to forward the connection to. In the current implementation, leastconn is used to keep the connections on all brokers evenly distributed. “Server emqx_node_# ip:1883 check” is used to determine which brokers connections can be forwarded to and checks how many connections each broker has when the broker is up.

Figure 3: keepalived configuration file

Figure 2: haproxy configuration file The keepalived configuration file, shown in Figure 3, does several things. vrrp_script chkhaproxy makes sure haproxy is continuously running. State and priority are used to determine whether it should be a master node or a backup node. The virtual IP address determines which IP address is used on the machine when it is in state “master”.

Figure 4. Installation instructions for brokers running Raspbian OS

ISBN: 1-60132-503-7, CSREA Press ©

Int'l Conf. Internet Computing and Internet of Things | ICOMP'19 |

Figure 4 shows the commands to install the required software to implement Canarycurity. 5.2 Sensor/Camera The PIR sensor is attached to the Raspberry Pi via its GPIO pins. A python script uses the Sensor to detect body movement. When the sensor detects movement, the RPI sends a signal to the attached camera to start recording. The recording stops after a predefined time. Once the recording is finished, the RPI saves it as a H264 file with a filename that includes the date/time, and the Raspberry Pi that recorded it. Figure 5 shows the function that performs the recording.

35

If the message was from a sensor/camera Pi, it works in stages. When the header is sent, the storage Pi opens a new file with the contents of the header as the name of the new file. It then appends the name of that file to the beginning of the log file. Afterwards, the storage Pi continuously receives the 100,000-bit chunks of data and saves it to the new file until the end message is sent. 5.4 Android Application The Android application is a Java-based program that can display logs and request videos. The user first enters the virtual IP address of the brokers, which is specified in keepalived.conf file, and then presses the connect button. This publishes a message to the brokers, and in response the storage/log Pi sends a log of all recordings back to the Android device as shown in figure 7.

Figure 5. Function that performs the recording. After the RPI saves the video file, it publishes the name of the file as a header, breaks the recording up into 100,000-bit chunks, and publishes them piece by piece. Afterwards, it publishes an end message to tell the receiver that the file is finished sending it, as shown in figure 6.

Figure 6. File Sender

Figure 7. Android App Code Snippet The user can then look through the logs and enter the name of the video they want to download. Figure 8 shows a screenshot of the running application. Figure 9 is a picture of the equipment used in this project. Two networks were configured, one with redundant brokers and 5 RPIs with sensors, and the second with one broker and two sensors. We implemented and experimented with redundant brokers configuration and load balancing between the brokers.

5.3 Storage/Logs When the storage RPI (the RPI which collects and stores all the video recordings from all cameras) detects a message, it first checks whether it is a request for the log file, or a message from the sensor/camera Pi. If it’s a log request, the RPI converts the log file to a string and publishes it.

ISBN: 1-60132-503-7, CSREA Press ©

36

Int'l Conf. Internet Computing and Internet of Things | ICOMP'19 |

Figure 8. Screenshot of the Android App

and out of the sensor’s view, and the main broker going down unexpectedly. The mobile application of Canarycurity provides the ability to view videos and logs, provided the android phone is in the network. It is also possible to send a photo to the android along with a link to a live recording, instead of transmitting a video recording after the recording stopped. Although Canarycurity demonstrates a redundant surveillance system based on IoT technologies, there are still a few potential problems. Firstly, when one of the RPI cameras is in the middle of sending its recording to the broker, it cannot be interrupted. This means that if someone were to trigger the PIR sensor, then wait for it to start transmitting, the RPI would not record anything until the transmission is completed. The simple alternative to this is to cancel the transmission upon the sensor being triggered and start recording immediately, but that would lead to the previous recording not being fully transmitted. The second issue of this implementation is in the hardware itself. Since the RPIs fully record video before sending it, rather than streaming it to the broker, the hardware itself can potentially be tampered with before any video is sent out. For example, if someone were to unplug or break the RPI before it finished recording, that video clip would not be sent to the broker. Finally, there is the issue of the broker going down midtransmission. Whilst the backup brokers can detect when the main broker has gone down and replace it, replacing the broker is not instantaneous. Therefore, if a broker goes down mid-transmission, a recording may not be properly transmitted.

7. CONCLUSIONS AND FUTURE WORK

Figure 9. Ten raspberry pi zero Ws and two Wi-Fi routers used for the project.

6. RESULTS AND DISCUSSION

The final design of Canarycurity includes the following hardware: several Raspberry Pi zeros, some of the RPIs were equipped with a camera and PIR sensors, and an android phone running our mobile application. The system has been tested for a variety of different conditions, including people moving continuously around in front of the PIR sensor, people moving into

Although there are some improvements yet to be made and features yet to be implemented, this project has reached its original goal of providing redundancy to an IoT-based surveillance system. There are three major items that should be prioritized in future work: streaming, the android app, and packaging. Streaming is both a useful feature, and an option to alleviate some of the issues caused by transmission errors. If streaming were implemented, it would allow the video to be saved on other machines whilst the camera is still recording. Therefore, if the machine went down mid-recording, not all the video data would be

ISBN: 1-60132-503-7, CSREA Press ©

Int'l Conf. Internet Computing and Internet of Things | ICOMP'19 |

lost; we could end up at least with partial video recordings. The Android application also has some major features that could be useful. For example, the android phone could be able to access or receive messages from the IoT when it is not connected to the system’s Wi-Fi. One idea would be for the Android phone to receive a link to the recording and a photo from the broker when not connected to the local Wi-Fi. That way the user could choose to view the recording at their leisure or take immediate action away from home. Finally, the system should be simplified, packageable, and easily installable by none-technology savvies.

REFERENCES

[1] Rachel E. Morgan and Jennifer L. Truman. “Criminal Victimization, 2017”. U.S. Department of Justice, Office of Justice Programs Bureau of Justice Statistics, Dec. 2018 NCJ 25 https://www.bjs.gov/content/pub/pdf/cv17_sum.pd f (Retrieved Apr. 15, 2019) [2] Nick Wingfield. “’Porch Pirates’ Steal Holiday Packages as They Pile Up at Homes”. The New York Times, Dec. 19, 2017. https://www.nytimes.com/2017/12/19/technology/ packages-holiday-season-porch-pirates-strike.html (Retrieved Apr. 15, 2019) [3] Jeff Tavss. “Too tough to handle, 'porch pirate' struggles with TV box he stole”. Local10.com, Dec. 14, 2018. https://www.local10.com/news/tootough-to-handle-porch-pirate-struggles-with-tvbox-he-stole (Retrieved Apr. 15, 2019) [4] P. Han, G. Price, L. Deligiannidis, C. Yu. “A Security System Built with Mobile and IoT Technologies” The 5th Annual Conference on Computational Science & Computational Intelligence - Internet of Things & Internet of Everything (CSCI-ISOT 2018), Dec. 2018. [5] D. Sunehra and M. Veena. “Implementation of interactive home automation systems based on email and Bluetooth technologies” 2015 International Conference on Information Processing (ICIP), Pune, 2015 pp.458-463 [6] N. Mathur, G. Paul, J. Irvine, M. Abuhelala, A. Buis, and I. Glesk. “A Practical Design and Implementation of a Low-Cost Platform for Remote Monitoring of Lower Limb Health of

Amputees in the Developing World,” IEEE Access, vol. 4, pp. 7440-7451, 2016. [7] Muheden, Karwan, Ebubekir, Erdem, and Sercan, Vanin. “Design and implementation of the mobile fire alarm system using wireless sensor networks”, IEEE Int. Symposium on Computational Intelligence and Informatics, 2016, pp. 000243000246. [8] Alan Mainwaring, Joseph Polastre, Robert Szewczyk, David Culler, and John Anderson. “Wireless Sensor Networks for Habitat Monitoring.” WSNA’02, September 28, 2002, Atlanta, Georgia, USA. [9] Bai, Ying-Wen, Li-SihShen, and Zong-Han Li, “Design and implementation of an embedded home surveillance system by use of multiple ultrasonic sensors”, IEEE Transactions on Consumer Electronics 56, no. 1 (2010). [10] Bai, Ying-Wen, Zi-Li Xie, and Zong-Han Li, “Design and implementation of a home embedded surveillance system with ultra-low alert power”, IEEE Transactions on Consumer Electronics 57, no. 1 (2011). [11] Neha Patil, Shrikant Ambatkar, and Sandeep Kakde. “IoT Based Smart Surveillance Security System using Raspberry Pi”, International Conference on Communication and Signal Processing, IEEE, April 6-8, 2017. [12] Shakthi Murugan K.H, V.Jacintha, S.Agnes Shifani. “Security System using Raspberry PI” Third International Conference on Science Technology Engineering & Management (ICONSTEM), IEEE,2017. [13] Leonidas Deligiannidis, Hamid Arabnia. “Security Surveillance Applications Utilizing Parallel Video Processing Techniques in the Spatial Domain”. Emerging Trends in Image Processing, Computer Vision and Pattern Recognition. 1st Edition, in Emerging Trends in Computer Science and Applied Computing Series. Morgan Kaufman/Elsevier. Dec. 2014. pp117-130.

ISBN: 1-60132-503-7, CSREA Press ©

37

38

Int'l Conf. Internet Computing and Internet of Things | ICOMP'19 |

)#$%%
$




'"
/@=
.;@GGD
G>
GEHML=J
.;A=F;=
9JD=LGF
0FAN=JKALQ
*LL9O9
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

.D]XPDVD7$.$0, Graduate school of Engineering Soka University +DFKLRMLVKL7RN\R-DSDQ NBWDNDPL#WVRNDDFMS

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³SHUVRQDO RXWOHW´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

ISBN: 1-60132-503-7, CSREA Press ©

82

Int'l Conf. Internet Computing and Internet of Things | ICOMP'19 | ,,, ,1',9,'8$/,=('32:(5&2168037,210$1$*(0(17 6@ 6SHFLILFDOO\WKHRXWOHWGHWHFWVWKH,'ZLWKWKHKLJKHVW566, LQHDFKTVHFRQGSHULRGIURPWKHVWDUWWRWKHHQGRISRZHU FRQVXPSWLRQ7KHGHWHFWLRQFRXQWHURIWKLV,'LVLQFUHPHQWHG E\  $W WKH HQG RI SRZHU FRQVXPSWLRQ WKH ,' ZLWK WKH KLJKHVWGHWHFWLRQFRXQWLVVHOHFWHGDVWKHWDUJHW,'RYHUWKH REVHUYDWLRQSHULRG. 2) Identification of the Appliance used and the User: 7KH DSSOLDQFHXVHGLVLGHQWLILHGE\DSSO\LQJWKHPHWKRGGHVFULEHG LQ ,,,C.1) WR DSSOLDQFH WDJV 6SHFLILFDOO\ D,' VFDQ LV H[HFXWHG DQG WKH D,' WKDW KDV WKH KLJKHVW 566, DQG LV GHWHFWHGWKHPRVWRIWHQLVVHOHFWHGDVWKHD,'RIWKHDSSOLDQFH XVHG 7KH XVHU LV LGHQWLILHG DV IROORZV GHSHQGLQJ RQ WKH GHWHFWHGDSSOLDQFHDWWULEXWHVHH7DEOH,  a) The appliance attribute is Public: $ Public DSSOLDQFHLVXVHGE\RQHRUPRUHSHUVRQV7KHLQURRPWLPH RIHDFKUHVLGHQWLVPDQDJHGEDVHGRQK,'7KHSHUVRQVZLWK K,'V WKDW KDYH EHHQ UHFRUGHG DV EHLQJ LQ WKH URRP DUH LGHQWLILHG DV XVHUV 7KH DPRXQW RI SRZHU FRQVXPSWLRQ LV DOORWWHGWRHDFKXVHULQSURSRUWLRQWRKLVKHULQURRPWLPH b) The appliance attribute is PP1: $PP1DSSOLDQFHLV XVHGE\DVLQJOHSHUVRQZKRGRHVQRWOHDYHWKHRXWOHWZKLOH WKHDSSOLDQFHLVLQXVH7KHXVHULVLGHQWLILHGE\DSSO\LQJWKH PHWKRGGHVFULEHGLQ,,,C.1)WRKXPDQWDJV6SHFLILFDOO\LQ HDFKTVHFRQGSHULRGZKLOHWKHDSSOLDQFHLVLQXVHWKHK,' ZLWKWKHKLJKHVW566,LVGHWHFWHG7KHK,'ZLWKWKHKLJKHVW GHWHFWLRQ FRXQW DW WKH HQGRI WKH SHULRG LV LGHQWLILHG DV WKH K,' RI WKH XVHU 7KH SRZHU FRQVXPSWLRQ GXULQJ S S  T  VHFRQGVLVDOORWWHGWRWKHK,'WKDWLVIRXQGWRKDYHWKHKLJKHVW

ISBN: 1-60132-503-7, CSREA Press ©

Int'l Conf. Internet Computing and Internet of Things | ICOMP'19 | 566,LQHDFKTVHFRQGSHULRG$WHDFKDOORFDWLRQWKHDPRXQW RI SRZHU FRQVXPSWLRQ LV DGGHG WR WKH SRZHU FRQVXPSWLRQ FRXQWHU RI WKH UHOHYDQW K,' $W WKH HQG RI WKH XVH RI WKH DSSOLDQFHWKHSRZHUFRQVXPSWLRQDGGHGWRWKHXVHUZLWKDQ K,' RWKHU WKDQ WKH LGHQWLILHG WDUJHW K,' LV VXEWUDFWHG DQG DGGHGWRWKHXVHURIWKHWDUJHWK,'. c) The appliance attribute is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d) The appliance attribute is Personal: $ Personal DSSOLDQFHLVXVHGRQO\E\WKHSHUVRQZKRRZQVLW7KHUHIRUH WKH RZQHU RI WKH DSSOLDQFH UHJLVWHUHG LQ WKH VHUYHU LV LGHQWLILHGDVWKHSRZHUXVHU D. Functional Requirements for a Personal Outlet 7KH DXWKRUV HDUOLHU SURSRVHG WKH EDVLF IXQFWLRQV RI D SHUVRQDORXWOHW>@:HDGGHGD:L)LIXQFWLRQWRPDNHWKH RXWOHW DQ ,R7 GHYLFH $ SHUVRQDO RXWOHW UHFHLYHV ,'V IURP KXPDQWDJVDQGDSSOLDQFHWDJVLGHQWLILHVWKHSRZHUXVHUDW WKHHQGRIWKHXVHRIWKHDSSOLDQFHDQGVHQGVWKDWLQIRUPDWLRQ WR WKH VHUYHU $W WKH VWDUW RI WKH XVH RI DQ DSSOLDQFH WKH SHUVRQDORXWOHWDOVRGRZQORDGVWKHWDJPDQDJHPHQWWDEOHVRI WKHUHVLGHQWVDQGDSSOLDQFHVIURPWKHVHUYHU7KHIXQFWLRQDO EORFNVRIDSHUVRQDORXWOHWDUHVKRZQLQ)LJ 3HUVRQDO2XWOHW $SSOLDQFH

I I

,' %/( WDJV

,' ,'Q

I

0DQDJHPHQW VHUYHU

I I

I

I

I I

I

I

I

,'L  3RZHUFRQVXPSWLRQ ,QURRPWLPH



I 0HDVXUHV SRZHUFRQVXPSWLRQ LQFOXGLQJ GDWHDQGWLPH I 5HFHLYHV,' I ,GHQWLILHV,' I ,GHQWLILHV KXPDQ WDJ I ,QFUHPHQWV WKHLQURRPWLPHFRXQWHU I DQGI 0DNHVDVHOHFWLRQEDVHGRQ566, I (QWHUVWKHYDOXHRII LQWKHSRZHUFRQVXPSWLRQ FRXQWHURIWKHSRZHUXVHU I ,GHQWLILHVDSSOLDQFHWDJ I ,GHQWLILHVDSSOLDQFHDWWULEXWH I ,GHQWLILHV SRZHUXVHUEDVHGRQWKHDSSOLDQFHDWWULEXWH I (QWHUVWKHYDOXHRII LQWKHSRZHUFRQVXPSWLRQ FRXQWHURIWKHSRZHUXVHULGHQWLILHGLQI

)LJ )XQFWLRQDOEORFNVRIDSHUVRQDORXWOHW

E. Individualized Power Consumption Management Functions 7KHIXQFWLRQVUHTXLUHGWRPDJHXVHUVWDJ,'VDQGSRZHU FRQVXPSWLRQDUHGHVFULEHGEHORZ

83

• Registration and confirmation of a BLE tag7KH%/( WDJRZQHGE\DXVHULVUHJLVWHUHGE\HQWHULQJWKHWDJ QDPHWDJ,'DQGSDVVZRUG • Registration and confirmation of an appliance(DFK RIDOODSSOLDQFHVXVHGLQDVKDUHGKRXVHLVUHJLVWHUHGE\ HQWHULQJ DSSOLDQFH QDPH WDJ ,' DSSOLDQFH DWWULEXWH DQGSDVVZRUG 7KHUHODWLRQVDQGWUDQVLWLRQVEHWZHHQWKHVHIXQFWLRQVDUH VKRZQ LQ )LJ  ([FHSW IRU WKH WUDQVLWLRQ IURP WKH SRZHU FRQVXPSWLRQ GHWDLO IXQFWLRQ WR WKH PDLQ PHQX IXQFWLRQ WUDQVLWLRQVEHWZHHQIXQFWLRQVFDQEHELGLUHFWLRQDO7KHUHLVD IXQFWLRQWRGLVSOD\LQIRUPDWLRQDIWHUWKHXVHUKDVHQWHUHGLW 5HJLVWHU DUHVLGHQW

0DNHDQHZ UHJLVWUDWLRQ

&RQILUP UHJLVWUDWLRQ

/RJLQ

&RQILUPSRZHU FRQVXPSWLRQ

'LVSOD\SRZHU FRQVXPSWLRQ GHWDLOV

&RQILUP DSSOLDQFH UHJLVWUDWLRQ

0DLQPHQX

5HJLVWHU D%/(WDJ

5HJLVWHU DQDSSOLDQFH

&RQILUP%/( WDJUHJLVWUDWLRQ

)LJ 5HODWLRQVDQGWUDQVLWLRQVEHWZHHQIXQFWLRQV

F. Protocol used between a 3HUVRQDO Outlet and the Server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

• Registration and confirmation of users and login $ XVHULVUHJLVWHUHGLQWKHVHUYHUE\HQWHULQJWKHXVHU,' XVHU QDPH SDVVZRUG DQG FRQILUPDWLRQ SDVVZRUG $ XVHUORJVLQWKHV\VWHPXVLQJWKHXVHU,'DQGSDVVZRUG • Confirmation of power consumption 7KH XVHU FDQ GLVSOD\ WKH WRWDO SRZHU FRQVXPSWLRQ DVVRFLDWHG ZLWK KLVKHU %/( WDJ DV ZHOO DV GHWDLOHG LQIRUPDWLRQ ORFDWLRQRIXVHWKHVWDUWRIXVHWKHHQGRIXVHLQURRP WLPHDQGSRZHUFRQVXPSWLRQ

)LJ 2YHUYLHZRILQIRUPDWLRQH[FKDQJHGEHWZHHQDSHUVRQDORXWOHWDQG WKHPDQDJHPHQWVHUYHU

ISBN: 1-60132-503-7, CSREA Press ©

84

Int'l Conf. Internet Computing and Internet of Things | ICOMP'19 | 7$%/(,,

,1)250$7,21(/(0(176(;&+$1*('$1'7+( 5(/(9$1785/6

,QIRUPDWLRQ HOHPHQWQDPH 2XWOHWB,'

'HVFULSWLRQ

7DJB,' 6WDUW7LPH (QG7LPH 7RWDO $WWULEXWH 1XPBRIB7DJV 7DJB,'BLQB5RRP ,Q5RRPB7LPH

,'RIWKHSHUVRQDORXWOHW 3RZHUXVHULGHQWLILHGE\ WKHSHUVRQDORXWOHW 6WDUWWLPHRISRZHUXVDJH (QGWLPHRISRZHUXVDJH 7RWDOSRZHUFRQVXPSWLRQ IURPWKHVWDUWWRWKHHQGRI SRZHUXVDJH $WWULEXWHRIWKHDSSOLDQFH XVHG 1XPEHURIWDJVWKDWWKH SHUVRQDORXWOHWKDV GHWHFWHGIURPWKHVWDUWWR WKHHQGRISRZHUXVDJH 7DJ,'RIDSHUVRQLQWKH URRP 'XUDWLRQLQZKLFKD SHUVRQVWD\VLQWKHURRP

85/

85/

,QFOXGHG

,QFOXGHG

,QFOXGHG

ʊ

,QFOXGHG ,QFOXGHG

,QFOXGHG ʊ

,QFOXGHG

ʊ

ʊ

,QFOXGHG

ʊ

,QFOXGHG

ʊ

,QFOXGHG

ʊ

,QFOXGHG

5$72& 6\VWHPV 7KH VPDUWSKRQH FRPPXQLFDWHG ZLWK WKH VHUYHUYLD:L)L7KHWLPHZKHQDSHUVRQHQWHUHGWKHURRP DQGWKHWLPHZKHQKHVKHOHIWWKHURRPZHUHFDOFXODWHGIURP WKHKXPDQWDJLQIRUPDWLRQGHWHFWHGE\WKHSHUVRQDORXWOHW

)LJ 7KHFRQILJXUDWLRQRIWKHGHYHORSHGSHUVRQDORXWOHW

 

B. Individualized Power Consumption Management Server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

$FTXLVLWLRQRI PHDVXUHGSRZHU FRQVXPSWLRQ 3+3OLQHV

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

:HESDJHIRUPDQDJHPHQW +70/3+3OLQHV

3+30\64/ $SDFKH

)LJ ([DPSOHRIWKHVHTXHQFHRIVHQGLQJPHDVXUHGSRZHUFRQVXPSWLRQ

7R HYDOXDWH WKH SURSRVHG V\VWHP ZH GHYHORSHG DQ H[SHULPHQWDOV\VWHPWKDWFRQVLVWHGRIDSHUVRQDORXWOHWDQGWKH LQGLYLGXDOL]HGSRZHUFRQVXPSWLRQPDQDJHPHQWVHUYHU

$FTXLVLWLRQRILQ URRPLQIRUPDWLRQ 3+3OLQHV

&HQW26 

)LJ 6RIWZDUHFRQILJXUDWLRQRIWKHPDQDJHPHQWVHUYHU

9 (;3(5,0(176$1'(9$/8$7,216 8VLQJWKHH[SHULPHQWDOV\VWHPZHYHULILHGWKHRSHUDWLRQ RIWKHSURSRVHGV\VWHPDQGHYDOXDWHGWKHUDWHDWZKLFKWKHXVHU RUWKHXVHGDSSOLDQFHZDVLGHQWLILHGFRUUHFWO\LGHQWLILFDWLRQ UDWH  A. Usage Scenarios and Experiments to Verify the Operation :H YHULILHG WKH RSHUDWLRQ RI WKH V\VWHP ZLWK WZR XVHUV 8VHU  DQG 8VHU   DV IROORZV 7KH FKDQJHV LQ SRZHU FRQVXPSWLRQWKDWRFFXUUHGZKHQWKHWZRXVHUVSHUIRUPHGWKH IROORZLQJWKUHHVFHQDULRVLQVHTXHQFHZHUHGLVSOD\HGRQWKH HDFK8VHU¶VVPDUWSKRQHVFUHHQ • Usage scenario 1 (Fig.8(a)):KLOH8VHU7DJB1DPH  ,'  ZDV XVLQJ D PP2 WRDVWHU RYHQ 7DJB1DPH  ,' 8VHU7DJB1DPH ,' FDPHQHDU8VHU7KH SRZHUFRQVXPSWLRQGXULQJWKLVSHULRGZDVDOORWWHGWR 8VHU

ISBN: 1-60132-503-7, CSREA Press ©

Int'l Conf. Internet Computing and Internet of Things | ICOMP'19 | • Usage scenario 2 (Fig.8(b)):KLOH8VHUZDVXVLQJ DPP1GU\HU7DJB1DPH ,' 8VHUZDONHGDURXQG 8VHU  7KH SRZHU FRQVXPSWLRQ RI 8VHU  UHPDLQHG XQFKDQJHG DQG WKH SRZHU FRQVXPSWLRQ GXULQJ WKLV SHULRGZDVDOORWWHGWR8VHU

85

FRQGXFWHG XQGHU WKH FRQGLWLRQV GHVFULEHG LQ 9B.1) DQG 9B.2)&RPPRQH[SHULPHQWFRQGLWLRQVDUHVKRZQLQ7DEOH ,,,

• Usage scenario 3 (Fig.8(c))8VHUVDWLQ WKHOLYLQJ URRPDQGZDVFKDUJLQJKLVVPDUWSKRQH7DJB1DPH  ,'  8VHU  VDW LQ IURQW RI 8VHU  7KH SRZHU FRQVXPSWLRQGXULQJWKLVSHULRGZDVDOORWWHGWR8VHU



D 8VDJHVFHQDULRDQGDVFUHHQVKRZLQJWKHPHDVXUHPHQWUHVXOW 

 )LJ /RFDWLRQVRIWKHDSSOLDQFHDQGUHVLGHQWVWKDWVLPXODWHPRYHPHQWVRI UHVLGHQWVLQDVKDUHGVSDFH 7$%/(,,,



E 8VDJHVFHQDULRDQGDVFUHHQVKRZLQJWKHPHDVXUHPHQWUHVXOW



 F 8VDJHVFHQDULRDQGDVFUHHQVKRZLQJWKHPHDVXUHPHQWUHVXOW )LJ 8VDJHVFHQDULRVDQGPHDVXUHPHQWUHVXOWV

B. Rate at which the Used Appliance or the User is Identified Correctly (Identification Rate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

&20021(;3(5,0(17&21',7,216

,WHP ([SHULPHQWWLPH 1XPEHURIH[SHULPHQWV ,'VFDQF\FOHT  &\FOHDWZKLFKWKHVHUYHUUHFHLYHVSRZHU FRQVXPSWLRQGDWDIURPWKHZDWWFKHFNHUS 

9DOXH VHFRQGV  VHFRQGV VHFRQG

1) Conditions for Experimenting used Appliance Identification: :H FKHFNHG ZKHWKHU WKH DSSOLDQFH WDJ DWWDFKHGWRDQDSSOLDQFHVHQGVWKH,'WRWKHSHUVRQDORXWOHW DQG ZKHWKHU WKH SHUVRQDO RXWOHW FDQ FRUUHFWO\ LGHQWLI\ WKH XVHGDSSOLDQFH%\FKDQJLQJWKHORFDWLRQRIWKHDSSOLDQFHWDJ HYHU\  VHFRQGV ZH FUHDWHG  SDWWHUQV WKDW VLPXODWHG WKH PRYHPHQWVRIWKHUHVLGHQWV7ZRDSSOLDQFHVZHUHXVHGLQWKH H[SHULPHQWVDPP2DSSOLDQFHDQGDPersonalDSSOLDQFH7KH DVVXPHGVFHQDULRZDVDVIROORZV:KLOH8VHUZDVXVLQJD PP2DSSOLDQFH8VHUZKRFDUULHGDPersonalDSSOLDQFHDQG ZDVQRWWKHXVHURIWKHPP2DSSOLDQFHPRYHGDURXQGLQWKH URRP DQG OHIW RU HQWHUHG WKH URRP 'HWDLOHG H[SHULPHQW FRQGLWLRQVDUHVKRZQLQ7DEOH,9. 2) Conditions for Experimenting the User Identification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

ISBN: 1-60132-503-7, CSREA Press ©

86

Int'l Conf. Internet Computing and Internet of Things | ICOMP'19 | PD\RFFXULQUHDOOLIH$QH[DPSOHRIDQH[SHULPHQWSDWWHUQ IRUHDFKDWWULEXWHQDPHLVVKRZQLQ)LJ)RUH[DPSOHLQ DWWULEXWH ³Public´ GXULQJ WKH SHULRG IURP WKH VWDUW RI WKH H[SHULPHQWXSWRVHFRQGVODWHU8VHUZDVORFDWHGQHDU DQG 8VHU  ZDV ORFDWHG QHDU LQ )LJ  )URP  VHFRQGV ODWHU XS WR  VHFRQGV ODWHU 8VHU  UHPDLQHG DW WKH VDPH ORFDWLRQEXW8VHUPRYHGWRGLVWDQWDQG8VHUHQWHUHGWKH URRPDQGPRYHGWRQHDU)URPVHFRQGVODWHUXSWR VHFRQGVODWHU8VHUZDVVWLOODWWKHVDPHORFDWLRQEXW8VHU PRYHGEDFNWRQHDUDQG8VHUSURPSWO\OHIWWKHURRP:H XVHGGLIIHUHQWWKUHVKROGVIRUWKH33&RXQWHUWRVHHZKLFK RQHZDVWKHPRVWXVHIXO'HWDLOVRIWKHH[SHULPHQWFRQGLWLRQV DUHVKRZQLQ7DEOH9 7$%/(,9

&21',7,216)25(;3(5,0(17,1*$33/,$1&(7$* ,'(17,),&$7,21

,WHP 1XPEHURI H[SHULPHQW SDWWHUQV 1XPEHURI %/(WDJV



9DOXH 



+XPDQWDJV



$SSOLDQFH WDJV



33 3HUVRQDO

33&RXQWHU VHFRQGV 

8VHU 8VHU 7RDVWHURYHQ 6PDUWSKRQH FKDUJHU



(7 $WWULEXWHQDPH

 VHFRQGV  QHDU0  QHDU1  

3XEOLF

 VHFRQGV

 VHFRQGV

QHDU0  GLVWDQW1  QHDU2

QHDU0 QHDU1, OHDYH2

 QHDU0, GLVWDQW1

33

QHDU0,  QHDU1

33

QHDU0  QHDU1

 QHDU0, GLVWDQW1

3HUVRQDO

 QHDU0 QHDU1

GLVWDQW0,  1 QHDUHVW

QRWH (7(ODSVHGWLPHIURPWKHVWDUWRIWKHH[SHULPHQW

 QHDU0,  OHDYH1    QHDU0,  OHDYH1    GLVWDQW0,  QHDUHVW1  

C. Experiment Results 1) Used Appliance Identification: 6LQFH DQ\ DSSOLDQFH WKDWZDVEHLQJXVHGZDVFRQQHFWHGWRWKHSHUVRQDORXWOHWZLWK DSRZHUFRUGDQGWKXVGLGQRWPRYHXVHGDSSOLDQFHVZHUH DOZD\VLGHQWLILHGFRUUHFWO\. 2) User Identification: 7KHH[SHULPHQWUHVXOWVDUHVKRZQ LQ 7DEOH 9, 7KH SRZHU FRQVXPHG E\ WKH Public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ersonal DSSOLDQFH WKHUH LV QR QHHG WR GHWHFW KXPDQWDJ,'VWRDOORWSRZHUFRQVXPSWLRQEHFDXVHWKHRZQHU RIWKHPersonalDSSOLDQFHUHJLVWHUHGLQWKHVHUYHULVWKHSRZHU XVHU7KXVWKHXVHULGHQWLILFDWLRQUDWHLVHTXDOWRWKHDSSOLDQFH WDJ,'LGHQWLILFDWLRQUDWH6LQFHWKH,'LGHQWLILFDWLRQUDWHZDV  LQ WKH H[SHULPHQW RI DSSOLDQFH WDJ LGHQWLILFDWLRQ WKH XVHU LGHQWLILFDWLRQ UDWH IRU WKH Personal DSSOLDQFH ZDV DOVR  7$%/(9,

$SSOLDQFHDWWULEXWHDQG FRQGLWLRQV 3XEOLF 33

)LJ$QH[DPSOHRIDQH[SHULPHQWSDWWHUQ7KHORFDWLRQRIHDFKXVHUDW GLIIHUHQWHODSVHGWLPHIURPWKHVWDUWRIWKHH[SHULPHQW 7$%/(9

&21',7,216)25(;3(5,0(17,1*7+(+80$17$* ,'(17,),&$7,21

,WHP 1XPEHURI H[SHULPHQWSDWWHUQV



1XPEHURI%/(WDJV





3RZHUXVHU 7KUHVKROGIRU 33&RXQWHU VHFRQGV 

3XEOLF 33 33 3HUVRQDO

9DOXH

    8VHU +XPDQWDJ  8VHU 8VHU 'LVSOD\ 'U\HU $SSOLDQFH  WDJ 7RDVWHURYHQ 6PDUWSKRQHFKDUJHU 8VHU DQG

86(5,'(17,),&$7,215$7(

33



7KUHVKROGIRU 33&RXQWHU VHFRQGV  3HUVRQDO

   

,GHQWLILFDWLRQUDWH         

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

ISBN: 1-60132-503-7, CSREA Press ©

Int'l Conf. Internet Computing and Internet of Things | ICOMP'19 |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¶VPHPEHUVZKRZRUNHGZLWK\RX RQ\RXUHYDOXDWLRQH[SHULPHQW 5()(5(1&(6 >@ ³1DWLRQDOVWUDWHJ\JUHHQSROLF\RXWOLQHIURPWKHIHWDOPRYHPHQWWR WKHJURZWKRIWKHJUHHQHQHUJ\UHYROXWLRQ´ KWWSZZZFDVJRMSMSVHLVDNXQSXSROLF\SGIVKLU\R SGIDFFHVVHGRQ  >@ 77DNDKDVKL',VKLXUDDQG0.XUR\DQDJL ³1HZYDOXH SURSRVLWLRQLQUHQWDOKRXVLQJSRVVLELOLW\RIVKDUHKRXVH´0LWVXELVKL 8)-5HVHDUFK &RQVXOWLQJ KWWSZZZPXUFMSWKLQNWDQNUFUHSRUWFRQVXOWLQJBUHSRUWFUB SGIDFFHVVHGRQ 

>@ 'RPLQLTXH ' *XLQDUG DQG 9ODG 0 7ULID ³%XLOGLQJ WKH :HE RI 7KLQJV´ 0DQQLQJ 3XEOLFDWLRQV -XQH  ,6%1  SDJH >@ +.RQGRDQG.7DNDPL³,QGLYLGXDOL]HG(OHFWULF3RZHU0DQDJHPHQW 6\VWHP 8VLQJ %/( 7DJ DW D 6KDUHG +RXVH´ LQ SURFHHGLQJV RI 7KH QG ,QWHUQDWLRQDO &RQIHUHQFH RQ ,QIRUPDWLRQ 1HWZRUNLQJ ,&2,1  -DQXDU\ >@ &+/LHQ-&/LQDQG@ 7LDJR0)HUQiQGH]&DUDPpV³$Q,QWHOOLJHQW3RZHU2XWOHW6\VWHP IRU WKH 6PDUW +RPH RI WKH ,QWHUQHW RI 7KLQJV´ +LQGDZL 3XEOLVKLQJ &RUSRUDWLRQ ,QWHUQDWLRQDO -RXUQDO RI 'LVWULEXWHG 6HQVRU 1HWZRUNV 9ROXPH$UWLFOH,'SDJHV >@ %HONLQ :H0R 6ZLWFK -XQH  KWWSZZZEHONLQFRP DFFHVVHG RQ  >@ &+ /LHQ @ “%OXHWRRWKZDWWFKHFNHU5(;%7:$77&+>5$72&@´ KWWSZZZUDWRFV\VWHPVFRPSURGXFWVVXESDJHEWZDWWFKKWPO DFFHVVHGRQ  >@ “0\%HDFRQŠ3UR[LPLW\DUHDVSHFLDOW\SH0%$W'5｜$SOL[ &RUSRUDWLRQ” KWWSEXVLQHVVDSOL[FRMSSURGXFWP\EHDFRQPEDWGUDFFHVVHG RQ 



ISBN: 1-60132-503-7, CSREA Press ©

87

88

Int'l Conf. Internet Computing and Internet of Things | ICOMP'19 |

Fast Detection of Abnormal Data in IIoT with Segmented Linear Regression Sejun Kim1, Ihsan Ullah1, Taeho Lee1, and Hee Yong Youn1 1 Department of Electronic, Electrical and Computer Engineering, Sungkyunkwan University, Suwon, Korea

Abstract - Industrial IoT (IIoT) is considered as an important component of manufacturing system nowadays. By collecting the sensed data from the facilities with IIoT, the operation condition is properly analyzed and handled. Here abnormal data requires to be quickly detected for the safety and productivity of entire system. The existing threshold-based approach is not suitable for IIoT since it cannot detect any dormant error or abnormal behavior under the threshold. In this paper a novel approach for the detection of abnormal data is proposed which is based on segmented linear regression with predicted interval and priority-based scheduling. Computer simulation reveals that the proposed scheme is superior to the existing scheme employing threshold, typical linear regression, or FCFS policy in the speed of detecting abnormal data. Keywords: IIoT, Machine learning, Packet scheduling, Emergency detection, Segmented linear regression

1

Introduction

Internet of Things (IoT) has been recognized as a crucial technology in the information and communication field for several years. In IoT environment, each object is connected using wired or wireless network so that the data can be effectively collected and analyzed to serve various end users. Numerous implementations based on IoT have been developed for real world problems. One of them is for applying to industrial manufacturing system, which is called industrial IoT (IIoT). In IIoT, sensors are placed the in the production facilities to monitor the status of the equipment and surroundings. Then the sensed data are analyzed locally on the facility, and then globally on the centralized control system. As a result, the manufacturing facilities can be effectively monitored and controlled. Since numerous sensors are deployed in the IIoT environment, effective ingestion of the data obtained from the sensors is a challenging problem. There have been various studies on monitoring the status of the nodes and processing the packets in IIoT. Especially, early identification of the loss of the sensor nodes which are in charge of critical facilities is of utmost importance since their loss can cause catastrophic damage to the entire premise. Therefore, it is imperative to quickly gather the data from the node of abnormal behavior and analyze them in the gateway (GW) before malfunctioning.

Two processes are involved in handling such node; detection and manipulation. Threshold is usually employed to detect any emergency or abnormal condition of a sensor node or target object. The main shortcoming of this approach is that it cannot detect any abnormal data pattern under the threshold. Detecting such dormant abnormal data needs to use a dynamic approach such as predicting future data based on support vector machine (SVM) [1]. SVM has been considered as an efficient technique for the classification of data. However, a large volume of data need to be accumulated at the GW for a while to achieve a good quality classification, which is a big overhead to the resource limited GW. More importantly, slow operation is not acceptable to IIoT in which the connected nodes need to be handled in real-time according to the degree of the emergency. The GW sets the scheduling priority of the nodes, and a scheduling method based on the priority was proposed in [5] for local optimization. Here an effective scheme for detecting abnormal data in real-time in IIoT environment is required to be developed. In addition, the loss of the packets of other regular nodes due to increased processing rate of emergency node needs to be taken into account. In this paper a novel approach for detecting emergency node connected to the GW is proposed, which generate abnormal data. It is achieved by employing the segmented linear regression technique with predicted interval. This can effectively solve the disadvantage of threshold-based approach which detects only the data exceeding the threshold. Segmented linear regression is an efficient technique for the regression of data between the breakpoints by detecting the inflection of regression, which is often used to determine the pattern of non-linear data. Since the data generated in IIoT environment is periodic, segmented linear regression is effective to analyze the regression of periodic data. The method controlling the priority of the packets processed in the GW is also proposed based on the multi-queue structure for quickly handling the emergency node. By increasing the priority of packet processing, the emergency data can be processed with high priority. The queueing theory is used to estimate the average waiting time, and the priority raise is limited not to overly delay other nodes. Computer simulation reveals that the proposed scheme can detect emergency node much faster than the existing schemes for various operational conditions. In addition, the emergency node is processed much faster than other regular nodes.

ISBN: 1-60132-503-7, CSREA Press ©

Int'l Conf. Internet Computing and Internet of Things | ICOMP'19 |

The rest of the paper is organized as follows. Section 2 discusses the work related to the proposed scheme. In Section 3, the proposed scheme is presented, and then its performance is evaluated in Section 4. The conclusion and future work are discussed in Section 5.

2 2.1

Related Work Detection of Abnormality

With the growth of IoT, a plenty of researches have been conducted, aiming to monitor and handle any emergency situation in the system. Especially, highly reliable technique is needed for the detection of abnormal situation to minimize the damage in the IIoT environment. Figure 1 shows the structure of IIoT. Most existing schemes employ threshold to detect any malfunctioning of the target object. The threshold-based approach has an advantage of simple implementation. However, any abnormal data within the threshold cannot be detected which may eventually cause an emergency situation. In addition, considering the diversity of IIoT, adjusting the threshold for each different type of facilities will be quite cumbersome. Therefore, there have been various researches on accurate and adaptive detection of abnormalities in IIoT environment.

89

proposed the operation criterion for compound relay protection which is based on the law of current and classification algorithm for detecting abnormal data. They classify abnormal data by detecting discontinuous wave point among continuous wave of current data. The existing methods are for detecting abnormal data from huge amount of sensed data. However, they are not effective to be implemented on resource limited IIoT GW requiring real time operation on small amount of data. Therefore, in this paper, an efficient classification approach is proposed for identifying abnormal data in IIoT environment.

2.2

Priority-based Scheduling

After detecting an emergency node, the data from the node has to be processed with high priority. The purpose to detect emergency node is to handle urgent situation. For this, the central manager, whether it is human or computer, requires enough data to correctly analyze the situation. Therefore, the data from emergency node has to be rapidly sent to the manager and accumulated. In the GW operating with the FCFS policy, the data from different sources are not able to be processed according to its priority. To effectively handle urgent data, various priority-based scheduling schemes have been proposed. Figure 2 shows the operational flow of priority-based scheduling. Reference [5] proposed an emergency scheduling method based on priority and local optimization for smart grid. The proposed scheme exchanges the location information of the nodes to reduce the number of hops and distance between the sink and source node. Then, based on the emergency information, the destination node decides the sequence of packet scheduling to minimize packet loss, waiting time, and latency.

Figure 1. The structure of IIoT system. For example, a method predicting and controlling the stability of a power system based on the voltage and velocity of generator was proposed in [1]. This is done by analyzing the existing data and predicting the future data using SVM. It predicts the time approaching to instability by 95% of accuracy. In [2], they focused on the classification of noise data with a new approach of Support Vector Data Description (SVDD) for singular classification. This approach adopts Ɛinsensitive function with distance-based local density and negative sample to reduce the noise of data which in turn increases the accuracy of detecting abnormal data. A method using 3-parameter Weibull distribution for the Harmonic current data was proposed to classify abnormal data by properly setting the threshold value [3]. Reference [4]

Figure 2. The flow of priority-based scheduling. In [6] event-aware backpressure (EABS) scheduling was proposed for solving the network congestion problem. It implements queue backlog to select the node for next hop of shortest path with backpressure to send the emergency packet.

ISBN: 1-60132-503-7, CSREA Press ©

90

Int'l Conf. Internet Computing and Internet of Things | ICOMP'19 |

The Two-Ford mechanism [7] achieves QoS during the process of new load by arranging the loads based on the characteristics of the first packet and delay requirement. Reference [8] considers the latency occurring in the aggregation of data to reduce the power consumption of M2M devices. The M/G/1 queueing model is utilized to analyze the priority for the data aggregation in the M2M gateway. As a result, the system latency and power efficiency were improved based on the priority. Various scheduling schemes employing multi-queue structure have also been proposed. Reference [9] proposed a scheme with dynamic job selection based on burst time for cloud computing. In [10] the Round robin scheduling based on time-sliding was combined with the priority-based scheduling to reduce the waiting of the tasks. The existing schemes focus on load balancing or efficient utilization of resources, with less concern on efficient handling of emergency situation. In this paper the multi-queue structure is adopted along with priority for proper treatment of emergency node.

3 3.1

where PB is the breakpoint at which the regression result is inflected. If the independent variable x is lower than PB, Y is determined by the first linear function. Otherwise, the other function determines Y. The breakpoint signifies the reference point which divides the data set. It is determined as Eq. (4).

PB

The Proposed Scheme Detection of Abnormality

Linear regression is an efficient technique used to analyze the trend of the target data, which is done by minimizing the error between the data and a linear function. Due to simple calculation, linear regression is suitable to be implemented in IIoT GW. However, the data generated in IIoT environment do not often display linearity, and therefore regression for inflected data is required. Regardless of the data type in IIoT environment such as temperature, vibration, and pressure, the emergency node generates abnormal data during its operation period. Note that the typical IIoT sensors operate with a fixed cycle in monitoring the environment. Therefore, it is important to compare the pattern of the data collected in each cycle and the regression of the data has to be applied to them. Segmented linear regression is an efficient technique for the regression of data between the breakpoints by detecting the inflection of regression, which is often used to determine the pattern of non-linear data. Since the data generated in IIoT environment is periodic, segmented linear regression is effective to analyze the regression of periodic data. In the proposed scheme, predicted interval is used to get the range of normal data based on the result of regression. The models of the proposed scheme are as follows. Linear regression is a model predicting a dependent variable, y, from an independent variable, x, by analyzing the linearity of the data. The training process models the relationship between x and y for the given data set {yi, xi1 ,đ, xip}. The model is as follows.

yi

regression. Also, Xi is a transposed form of Xi, and X7iβ is inner product of X7i and β. As a result, the linear regression model is formed as a linear function, Y = Ax + B. With segmented linear regression data are divided into several segments along x-axis, and linear regression is applied to each segment. Assume that the data set is divided into two segments. With linear regression, there would be only one segment of a line. However, two linear functions are obtained with segmented linear regression as follows. ­i 1, x d PB (3) Y Ai x  Ki ® ¯i 2, x ! PB

E1 xi1 

 E p xip  H i

i 1, 1

,n

(1)

xip  m

X TiE  m

i 1,

,n

(2)

In the equations above, Ei is a coefficient of independent variable and p is the number of parameters assumed by linear

min xX

¦( y  Y ) r

2

(4)

where y is the data value and Yr is predicted regression result for the given x. (y−Yr)2 is called loss function. Therefore, the breakpoint, PB, is an x value which minimizes the loss function. After the segmented linear regression is applied, the predicted interval of each segment is obtained to get the boundaries of normal data. The predicted interval is obtained as follows. lP X P uP J P (l  X  u ) P (   ) (5) V V V lP uP Z P( ) (6) V V where V is standard deviation of the given data set and P is mean value. l and u are the minimum and maximum value, respectively. It can be derived as follows. lP z V (7) uP z

V

Here z is z-score value. Then, l and u can be expressed as follows. l P  zG (8)

u

P  zG

(9)

When the sensed data are started to be sent from the nodes, the GW decides the period of the data. Then it searches the breakpoints where the data rapidly changes. With them, linear regression and predicted interval of the data between the break points are obtained. The predicted intervals are treated as boundaries of normal data. After this setup, the GW gets the sensed data from the nodes and their data are compared with the normal boundary of the data cycle. The overall procedure of the detection of emergency node is shown below.

ISBN: 1-60132-503-7, CSREA Press ©

Int'l Conf. Internet Computing and Internet of Things | ICOMP'19 |

Procedure 1. Detection of emergency node Input: Node connected to GW, Ni Controller resource capacity, M Input data, D

Pi (10) T Here Pi represents the number of packets to process for Node-i. Then the service rate of GW for it becomes Pi since it is the processing rate. The number of packets of Node-i, Ni, at steady state is as below.

Pi

Begin 1: Xs, Ys ˥Segmented linear regression model 2: Yhigh ˥Predicted interval of Xs, Ys 3: 4: for input data d (d  D) do 5: if(d_value > Yhigh) then 6: Emergency_count ˥ Emergency_count + 1 7: 8: if(Emergency_count > Emergency_threshold) then 9: Ni ˥ Emergency node 10: 11: End if 12: End for End

The GW is able to identify abnormal data by just comparing the value of the data and boundary. Figure 3 shows the process of abnormal data detection.

Figure 3. The process of detecting abnormal data.

3.2

91

Priority Control with Multiple Queues

To implement priority for each node, multi-queue structure is employed. The packet from each node is handled by separate waiting queue of the GW. Figure 4 shows the structure of the proposed multi-queue scheduling for the GW.

f

Ni

In normal state the packet processing sequence is decided according to the number of packets in each queue. The processing rate of Node-i, PI, in a period of T is as follows.

(1 

k 0

Oi f § Oi · ) k¨ ¸ Pi k 0 © Pi ¹

¦

k

(11)

where Sk is k-th steady state probability of Node-i, while the state is defined as the number of packets. UL is the service utilization which is the average number of packets to be processed. The average wait time, TQ, is the time for a packet to wait to be processed. TQ

1 n

n

¦P i 1

Ni

(12)

i

The emergency node is given higher priority as follows. Note that each sensor node generates and sends packets to a queue associated with it in the GW. The GW allocates the processing time of each queue in proportion to the input rate. Assume that three nodes send the packets to the GW (hence three queues), and the input rate of Queue-1,2, and 3 are 5, 3, and 2 packets/sec, respectively. Then the GW processes the packets of the queues in circular fashion, from Queue-1 to Queue-3, in regular operation condition, with 50%, 30%, and 20% of processing time for the three queues, respectively. Assume that the node sending packets to Queue1 is detected as an emergency node when a packet of Queue-1 has just been processed. Then another packet of Queue-1 is processed instead of processing that of Queue-2. Then the average waiting time of the nodes is calculated. If it is still lower than the threshold, the packet of Queue-1 is processed repeatedly until the moment when the average waiting time becomes larger than the threshold. Then the packets from other queues are processed in the original circular sequence. This approach allows to handle the packets of emergency node more rapidly than the other nodes, while avoiding excessive delay of other nodes.

4 4.1

Figure 4. The structure of multi queue scheduling.

¦

kS ik

Performance Evaluation Simulation Setup

The experiments were conducted on a PC consisting of 8GB memory and Intel i5-7500 CPU based on Windows OS, and the scheme was implemented using Python language. In order to evaluate the effectiveness of the proposed scheme, it is compared with the emergency node detection scheme using linear regression, threshold, or FCFS policy in terms of processing time in different operational conditions.

4.2

Simulation Results

In order to evaluate the performance of the schemes, a network consisting of six nodes is built. The data generated

ISBN: 1-60132-503-7, CSREA Press ©

92

Int'l Conf. Internet Computing and Internet of Things | ICOMP'19 |

from the nodes and the results of predicted interval of the proposed segmented linear regression and linear regression are shown in Figure 5.

Figure 5. The result of regression with the data. It is assumed that each node senses a specific data of the manufacturing facilities. When the facility is on the rest, the sensed value is 30 as shown in the figure. When it operates, the value rapidly grows to about 80 and decreases to the default value of 30. Notice form the figure that the emergency node gets malfunctioning from the third period, and becomes totally abnormal from the fourth period. The predicted interval of each regression is shown in Figure 6. Since it is modeled for the data of each period, the model is separately applied to each period. As the figure reveals, the upper boundary of the predicted interval of linear regression follows the trend of the data to a certain degree. However, the difference from the actual data is about 10 in average. On the other hand, the predicted interval of the proposed segmented linear regression match the inflection point. Therefore, the change in the trend can be accurately infered. As a result, the model of the proposed scheme well fits the data, and thus it is possible to accurately detect the abnormal data. Figure 6 compares the accumulated proccessing times of the emergency packet with different schemes. In this simulation the average rate of each node is 70 packets/sec and the processing rate of the GW is 500 packets/sec. The threshold for the delay in controlling the priority of emergency node is 0.5 sec, and the confidence level of the predicted interval is 95%. In the figure the processing time linearly increases with the growth of the number of processed packets. Then it bends at the point when the GW detects the emergency node. After the detection, the priority of emergency node gets higher than the other nodes. As a result, the processing rate of emergency packets gets higher and the gradient becomes lower. For the FCFS case, the priority is not changed, and therefore the slope does not change. Observe from Figure 7 that the proposed scheme using segemented linear regression detects the emergency node much earlier than the scheme with linear regression, at about 160th packet compared to 190th packet. It is because the proposed scheme accurately models the upper bound of the data. Therefore, if an emergency node gets malfunctioning and exceeds normal data, the proposed

scheme immediately detects it. However, the scheme with linear regression or threshold requires more time to hit the boundary of normal range.

Figure 6. The comparison of processing times of emergency packet. (Confidence level = 95%) Figure 7 is the case 90% confidence level. When the confidence level is decreased, the boundary of the data which affects the regression model gets narrow. It means that the upper bound of the predicted interval becomes closer to the data. Therefore, detecting abnormal data becomes more sesitive than with higher confidence level. As the figure shows, the proposed scheme detects the emergency node at about 110th packet. It is also faster than the other schemes. In addition, unlike with Figure 6, the difference in the detection time between the linear regression and threshold technique gets bigger. This is because the predicted interval of linear regression becomes narrower as the confidence level decreases while that with threshold does not change.

Figure 7. The comparison of processing times of emergency packet. (Confidence level = 90%) Figure 8 shows the comparion of processing times for 300 packets generated from an emergency node with different input rates. As the input rate increases, the processing time decreases because the processing rate is much higher than input rate. As shown in the figure, the time becomes flat after 80 packets/sec since the capacity of GW meets the transmission rate from all the nodes. Meanwhile, the processing time of the proposed scheme is always smaller than the scheme with linear regression and threshold. It means that

ISBN: 1-60132-503-7, CSREA Press ©

Int'l Conf. Internet Computing and Internet of Things | ICOMP'19 |

if there exists surplus capacity in the GW, the proposed scheme can serve more emergency data.

Figure 8. The comparison of processing times with varying input rate. Figure 9 shows the comparion of the processing times with different service rates. Similar to input rate, the processing time gradually decreases as the service rate increases. Notice that the proposed scheme of segmented linear regression consistently performs better than the other schemes.

5

Table 1 presents the processing ratio of emergency node. The number of emergency packets over entire packtes is measured with 70 packets/sec of input rate and 500 packets/sec of processing rate. The FCFS scheme processes emergency packets equally as other packets. Since there are six nodes, the packets from each node are processed equally likely at about 16%. However, with the proposed scheme, the GW handles about 2% more pakects from the emergency node than other nodes. Table 1. The comparison of processing ratios. Time(s)

Processing ratio (%)

5

Proposed 18.707

FCFS 16.138

50

18.518

16.047

500

18.866

16.284

Conclusion

In this paper we have proposed a scheme based on the predicted interval of segmented linear regression to quickly detect emergency node in IIoT environment. The proposed scheme compares the sensed data with the upper bound data in the predicted interval to quickly detect abnormal data. Also, a priority control scheme has been proposed to handle the emergency node preferentially. The control scheme is based on multi-queue structure, and queueing theory is adopted not to overly degrade the normal nodes. Computer simulation reveals that the proposed scheme is superior to the existing threshold-based, linear regression, and FCFS scheme in the speed of detecting abnormal data. Also, the data from emergency node can be processed at the GW with higher rate than the others. Future research includes the study on the optimal setting of the threshold and waiting time since they are important parameters for the detection and management of emergency node. A formal modeling and experiment will be conducted to systematically decide the values showing the best performance in various IIoT environments. Also, the proposed scheme will be enhanced to support different pattern of operation of the nodes, data distributions, etc.

6

Figure 9. The comparison of processing times with varying service rate.

93

Acknowledgement

This work was partly supported by Institute for Information & communications Technology Promotion(IITP) grant funded by the Korea government(MSIT) (No. 2016-000133, Research on Edge computing via collective intelligence of hyperconnection IoT nodes), Korea, under the National Program for Excellence in SW supervised by the IITP (Institute for Information & communications Technology Promotion) (2015-0-00914), Basic Science Research Program through the National Research Foundation of Korea (NRF) funded by the Ministry of Education, Science and Technology(2016R1A6A3A11931385, Research of key technologies based on software defined wireless sensor network for real-time public safety service, 2017R1A2B2009095, Research on SDN-based WSN Supporting Real-time Stream Data Processing and Multiconnectivity), the second Brain Korea 21 PLUS project.

7

References

[1] F. R. Gomez et al., “Support Vector Machine-Based Algorithm for Post-Fault Transient Stability Status Prediction Using Synchronized Measurements”, IEEE Transactions on Power Systems, Vol.26, Issue.3, pp.1474-1483, Nov. 2010. [2] G. Chen et al., “Robust support vector data description for outlier detection with noise or uncertain data”, KnowledgeBased Systems, Vol.90, pp.129-137, Dec. 2015. [3] T. L. Zang et al., “A detection method for abnormal harmonic current monitoring data using three-parameter Weibull distribution”, IEEE 11th Conference on Industrial

ISBN: 1-60132-503-7, CSREA Press ©

94

Int'l Conf. Internet Computing and Internet of Things | ICOMP'19 |

Electronics and Applications(ICIEA), pp.2366-2371, Jun. 2016. [4] X. Zhang et al., ”A Compound Relay Protection Operation Criterion Based on Kirchhoff’s Current Law and Abnormal Data Detecting Algorithm”, Energy and Power Engineering, Vol.9, No.4B, pp.88-94, Apr. 2017. [5] T. Qiu et al., “A Local-Optimization Emergency Scheduling Scheme With Self-Recovery for a Smart Grid”, IEEE Transactions on Industrial Informatics, vol.13, Issue.6, pp.3195-3205, Dec. 2017. [6] T. Qiu et al., “EABS: An Event-Aware Backpressure Scheduling Scheme for Emergency Internet of Things”, IEEE Transactions on Mobile Computing, Vol.17, No.1, pp.72-84, Jan. 2018. [7] P. Kaur and P. Singh, “Priority based Scheduling Algorithm with Fast Task Completion Rate in Cloud”, Advances in Computer Science and Information Technology (ACSIT), Vol.2, No.10, pp.17-20, 2015. [8] S. A. AlQahtani, “Analysis and modelling of power consumption-aware priority-based scheduling for M2M data aggregation over long-term-evolution networks”, The Institution of Engineering and Technology (IET), Vol.11, Issue.2, pp.177–184, 2017. [9] A. V. Karthick, E. Ramaraj and R. G. Subramanian, “An Efficient Multi Queue Job Scheduling for Cloud Computing”, 2014 World Congress on Computing and Communication Technologies (WCCCT), pp.164-166, 2014. [10] U. Rafi, M. A. Zia, A. Razzaq, S. Ali and M. A. Saleem, “Multi-Queue Priority Based Algorithm for CPU Process Scheduling”, International Conference on Management Science and Engineering Management (ICMSEM 2017), pp.47-62, 2017. [11] G. Patti et al., “A Priority-Aware Multichannel Adaptive Framework for the IEEE 802.15.4e-LLDN”, IEEE Transaction on industrial electronics, Vol. 63, pp.6360-6370, 2015. [12] S. R. Sarangi et al., “Energy Efficient Scheduling in IoT Networks”, 33rd Annual ACM Symposium on Applied Computing, pp.733-740, 2018. [13] C. Kan et al., “Parallel computing and network analytics for fast Industrial Internet-of-Things (IIoT) machine information processing and condition monitoring”, Journal of Manufacturing Systems, Vol.46, pp.292-293, 2018 [14] J. Singh and D. Gupta, “An Smarter Multi Queue Job Scheduling Policy for Cloud Computing”, International Journal of Applied Engineering Research, Vol.12, No.9, pp.1929-1934, 2017. [15] L. Zhou, H. Guo, “Anomaly Detection Methods for IIoT Networks”, IEEE International Conference on Service Operations and Logistics, and Informatics (SOLI), 2018.

ISBN: 1-60132-503-7, CSREA Press ©

Int'l Conf. Internet Computing and Internet of Things | ICOMP'19 |

95

Independent Algal Bloom Removal System Kenichi Arai, Ryota Nakashima, Tetsuo Imai, and Toru Kobayashi Graduate School of Engineering, Nagasaki University, Nagasaki, Japan [email protected]

Abstract - In this study, we propose an independent algal bloom removal system that can remove algal bloom independently through automatic detection of where they are occurring by coordinating a drone, an automated algal bloom removal robot, and a cloud server. Through realization of this system, automatic detection of where algal bloom is occurring and independent removal of algal bloom are enabled. As this system requires automatic detection of where algal bloom is occurring, it is extremely important for the evaluation of this system to assess the performance of the algal bloom determination system which determines if there are algal bloom or not. Therefore, we conducted an evaluation experiment to assess its determination performance. In this paper, we show the result of this evaluation. Keywords: Internet of Things, Multi robot operation, Algal bloom removal robot, Convolutional Neural Network

1

Introduction

Due to the impact of global warming, increase of algal bloom, which causes deterioration of environment of lakes and marshes, is becoming a problem. An algal bloom is an overgrowth of cyanobacteria (such as Microcystis), which is phytoplankton, in lakes and marshes eutrophicated by nitrogen or phosphorus that turns the surface of water into powdery green state. A large growth of algal bloom is so green that it is compared to green paint, and it ruins the view of lakes and marshes. Furthermore, the algal bloom that was blown by wind to the shore can rot and emit strong and bad smell, and when it glows on a lake or marsh used as a water sauce, it becomes an obstacle to the filtering process that removes trashes from the water, causing the water treatment process to become inefficient. Indeed, the damage caused by toxin in algal bloom is occurring around the world, and health damages and restriction on domestic water caused by its impact on water sources are widely reported. Thus, measures against algal bloom have become an important international issue [1]. Several algal bloom removal devices have been developed. However, these devices require workers to go to where an algal bloom is occurring, carry, install, and retrieve them on a boat. Moreover, the workers must work on the boat and therefore it is dangerous for them. Furthermore, the location of where an algal bloom is occurring changes by time, which requires the device to be moved each time the location

Fig. 1: Overview of Independent Algal Bloom Removal System changes. And as these devices are large, human resource required for moving is also a problem. In this study, we propose an independent algal bloom removal system that can remove algal bloom independently through automatic detection of where they are occurring by coordinating a drone, an automated algal bloom removal robot, and a cloud server. The characteristic aspect of this system is the fact that it can detect where algal bloom is occurring automatically by using a tablet PC to coordinate a drone, an automated algal bloom removal robot, and a cloud server in order to independently remove algal bloom (Fig.1). Through realization of this system, automatic detection of where algal bloom is occurring and independent removal of algal bloom are enabled. Therefore, the problems of troublesome usage of removal devices, dangerous work process, and limited operation area will be solved. As this system requires automatic detection of where algal bloom is occurring, it is extremely important for the evaluation of this system to assess the performance of the algal bloom determination system which determines if there are algal bloom or not. Therefore, we conducted an evaluation experiment to assess its determination performance. In this paper, we show the result of this evaluation. The structure of the remainder of this paper is as follows. In Section 2, we show the difference between related works and this study. In Section 3, we show the details of this system. In Section 4, we show the results of evaluation experiment

ISBN: 1-60132-503-7, CSREA Press ©

96

Int'l Conf. Internet Computing and Internet of Things | ICOMP'19 |

Fig. 2: System Configuration of Independent Algal Bloom Removal System using this system to assess the performance of the algal bloom determination system. In Section 5, we present our conclusion.

2

Related Works

In Japan, other algal bloom removal devices such as MIZUMORI [2] or Algae Hunter [3] have been developed in the past. However, these devices (hereafter referred to as existing devices) have several problems. Firstly, the operation of existing devices is troublesome. They require the workers to go to where an algal bloom is occurring, carry, install, and retrieve them on a boat. Thus, usage of existing devices requires large amount of work. Secondly, operation of existing devices is dangerous for the workers. As the workers must work on a boat, there is a possibility that it may capsizal during the work. Moreover, as algal bloom sometime contain toxin, if a worker fall into the lake or marsh and accidentally drink its water, it may cause health damage. Thirdly, the existing devices have limited operation area. The operation area of the existing devices is limited to the area around the device. However, the location where algal bloom is occurring changes by time. Thus, the device has to be moved according to this change. And as these existing devices are large, removal of algal bloom using them requires significant amount of human resource to move them.

With this system, automatic detection of where algal bloom is occurring and independent removal of algal bloom are enabled by coordinating a drone, an automated algal bloom removal robot, and a cloud server. Therefore, the problems of the existing devices, namely the troublesome usage of removal devices, the dangerous work process, and the limited operation area, will be solved.

3

Independent Algal Bloom Removal System

3.1

System Requirements

Following is the requirements for the realization of this system. x Requirement 1: It can automatically detect where algal bloom is occurring by using the drone. x Requirement 2: The automated algal bloom removal robot can automatically move to where algal bloom is occurring. x Requirement 3: The automated algal bloom removal robot can automatically return. x Requirement 4: It is possible to remotely monitor the automated algal bloom removal robot.

ISBN: 1-60132-503-7, CSREA Press ©

Int'l Conf. Internet Computing and Internet of Things | ICOMP'19 |

97

Fig. 3: Drone Application This study developed a “drone application”, an “algal bloom determination system” and a “navigation system for automated algal bloom removal robot” in order to satisfy these requirements. Note that the drone application was produced as an application for Android OS (DJI Mobile SDK [4]) .

3.2

Basic System Configuration

This system consists of a drone, an automated algal bloom removal robot, a cloud server, and a tablet PC that coordinates them. The drone application, the algal bloom determination system, and the navigation system for automated algal bloom removal robot we developed are installed on the tablet PC, the cloud server, and the automated algal bloom removal robot, respectively (Fig.2). Note that the automated algal bloom removal robot is equipped with a Raspberry Pi, and the navigation system is installed on this Raspberry Pi. This system photographs the occurrence situation of algal bloom from the sky by setting the photographing area for the drone on the tablet PC, which enables the drone to fly automatically in the set photographing area. The resulting aerial photographs are sent to the cloud server together with the positional information of the locations of the photographs via the tablet PC. At the cloud server, image recognition is conducted using Convolutional Neural Network (CNN) [5] in order to identify the locations where algal bloom is occurring. The identified occurrence locations of algal bloom are sent to the automated algal bloom removal robot via the PC tablet, and the removal of algal bloom is conducted independently based on the information of the occurrence locations of algal bloom. Following the completion of algal bloom removal, the

automated algal bloom removal robot returns to the home position automatically.

3.3

Prototype System

3.3.1 Tablet PC The drone application is installed on the tablet PC (HTC Nexus 9, Android 7.1.1 [6]). With this system, coordination between the processes of the drone, the automated algal bloom removal robot, and the cloud server is possible through only the operation of the drone application installed on the tablet PC (Fig.3) . The drone application consists of two functions, namely “data collection function” and “data transmission and reception function”. Following is the details of these two functions. x Data Collection Function When the map in this drone application (Google Map) is tapped, the photographing area (300m×300m for the prototype), for which the point that was tapped is the center, is set automatically. Note that, Google Maps API [7] is used to display the Google Map. When the photographing area is set, the positional information of the photographing location, as well as the route information, is calculated automatically. When this route information is sent to the drone, the drone flies automatically according to the route information and photographs the occurrence situation of alga blooms from the sky. The aerial photographs the drone took are saved on the SD card in the body of the drone. Note that the flight altitude and the speed of the drone can be specified when sending the route information to it. Moreover, as the drone (Phantom4 PRO [8]) is equipped with GPS, the positional information

ISBN: 1-60132-503-7, CSREA Press ©

98

Int'l Conf. Internet Computing and Internet of Things | ICOMP'19 |

Fig. 4: Diagram for Divided Image Location Coordinate Calculation (latitude/ longitude) of the photographing location is recorded as EXIF information on the aerial photographs taken by the drone. x Data Transmission and Reception Function When the “Auto” button on the drone application is tapped, the aerial photographs saved on the SD card in the body of the drone are sent to the tablet PC. After they are sent to the tablet PC, the aerial photographs saved on the SD card are deleted. Based on the aerial photographs taken by the drone, a CSV file for determining the occurrence of algal bloom is created. The file names of the aerial photographs and the positional information (latitude/ longitude) of their photographing locations are saved on this CSV file, with one line corresponding to one location. Following this, the aerial photographs and the CSV file are sent to the cloud server. After they are sent to the cloud server, the aerial photographs and the CSV file saved on the tablet PC are deleted. After sending the aerial photographs and the CSV file to the cloud sever, the drone application starts the algal bloom determination system of the cloud server. Please see Section 3.3.2 for the detail of this algal bloom determination system. The drone application regularly checks the determination result output directory until the algal bloom determination system of the cloud server identifies where algal bloom are occurring, in other words until there is an output of the algal bloom determination result. When the determination result is outputted to this directory, the occurrence locations of algal bloom are sent to the tablet PC from the cloud server. Note that after sending the occurrence locations of algal bloom to the automated algal bloom removal robot, the drone application starts the navigation system for the automated algal bloom removal robot. Please see Section 3.3.3 for the detail of this navigation system for the automated algal bloom removal robot. 3.3.2 Cloud Server The cloud server is equipped with the algal bloom determination system (Sakura no VPS + CentOS + Python) . “Sakura no VPS” is a commercial cloud service in Japan. The algal bloom determination system uses CNN for image

recognition in order to determine the locations algal blooms are occurring. Following is the detail of the algal bloom determination system. First, the algal bloom determination system starts image division program. The size of one divided image is equivalent of the size of the mesh for setting the movable area of the automated algal bloom removal robot using its navigation system (5m×5m for the prototype system). Next, the location coordinate in relation to each areal photograph taken by the drone is calculated using the CVS file for algal bloom determination. The calculation result is saved on the text file. Following is the detail of the calculation method of location coordinate. As the photographing area changes according to the angle of view in relation to the water surface, this system places the camera to the bottom of the drone (perpendicular to the water surface). Thus, the location coordinate of each areal photograph corresponds with the coordinate of the central point of the areal photograph. As shown in Fig.4, the location coordinate of each divided image is calculated in relation to the central coordinate of the areal photograph. Next, the algal bloom determination system determines if there are algal blooms or not on each divided image using CNN. CNN is one of neural networks that calculate through multiple layering of convolution layers or pooling layers [5]. Please see Section 4 for the detail of the determination method. The algal bloom determination result is saved on the text file. Note that on this text file, the location coordinates of the divided images that were judged to be “With algal bloom” are saved. Finally, the algal bloom determination result, in other words the locations where algal blooms are occurring, is sent from the cloud server to the tablet PC. After sending it to the tablet PC, the text file of the algal bloom determination result on the cloud server is deleted. 3.3.3 Automated Algal Bloom Removal Robot The automated algal bloom removal robot is equipped with a single-board computer called Raspberry Pi (Raspberry Pi 2 model B), and the navigation system (PHP + Python + Apache + HTML + JavaScript + MySQL) for the automated algal bloom removal robot is installed on this Raspberry Pi. Fig.5 shows the user interface of this navigation system. The navigation system is constructed as a Web application by installing Web server on the Raspberry Pi. Moreover, as the Raspberry Pi is connected to a camera module, a GPS module, a battery, an object sensor, and a 3G module, it can remotely monitor the automated algal bloom removal robot by using the information from these attachments. Following is the detail of this navigation system. This navigation system first takes the photographing area of the drone, divides this area into a mesh, sets the mesh area as the movable area, and displays this movable area on the map of the navigation system (Google Map). When displaying this movable area, it also displays the occurrence locations of algal blooms using the determination result of the algal bloom determination system. Moreover, it obtains the current location of the automated algal bloom removal robot and set

ISBN: 1-60132-503-7, CSREA Press ©

Int'l Conf. Internet Computing and Internet of Things | ICOMP'19 |

99

Fig. 5: User Interface of Navigation System for Automated Algal Bloom Removal Robot that location as the home position. Next, it sets the cleaning area based on the algal blooms occurrence locations information, and it also sets no-entry areas if necessary. Following this, it automatically generates the movement route of the automated algal bloom removal robot based on the information of the home position, cleaning area and no-entry area. A search algorithm A* [9] is used for the generation of the route. The automated algal bloom removal robot moves and removes algal blooms following this movement route, and returns to the home position upon completion. Note that the camera view of the automated algal bloom removal robot, its remaining battery and positional information can be checked on this navigation system. On the other hand, this navigation system also has the manual mode. Thus, it is possible to switch to the manual mode when there is an emergency and control the automated algal bloom removal robot manually. Thus, the system user can conduct every operation of the automated algal bloom removal robot by using this navigation system on a tablet PC.

4

Algal Bloom Determination Method and Its Evaluation Experiment

This section discusses the algal determination method of the algal determination system, its evaluation experiment and the evaluation result.

4.1

Algal Bloom Determination Method using CNN

The CNN model for this system is based on AlexNet [10], which is a representative CNN model for categorizing images. The algal bloom determination is classified into the following two classes: “With algal bloom” and “Without algal bloom”.

Fig. 6: View of Algal Bloom Determination using CNN Fig.6 shows the view of algal bloom determination using CNN. Note that precision, recall and F-value are used as the evaluation indexes of the algal determination method.

4.2

Data Set and Evaluation Experiment

The RGB images (areal photographs) used for the evaluation experiment were taken by actually flying the drone over the kururi dam located at togitsu in Nagasaki. The flight altitude of the drone was uniformly set at 20m from the water surface, and the camera was pointed directly below (perpendicular to the water surface) to take photographs. The data set used for the evaluation experiment is thus: 108 images were created by dividing the RGB images photographed by the drone into the mesh of 3×4 (vertical × horizontal) (With algal bloom: 39, Without algal bloom: 69). These images were increased to 2336 images through data extension. (With algal bloom: 1168, Without algal bloom: 1168). During this data extension, noise processing based on inversion, rotation and Gaussian distribution was conducted using OpenCV [11]. Among the extended 2336 pictures, 70% was assigned as the training data and 20% was assigned as the examination data to conduct the learning in order to create a classification model for determining whether there are algal

ISBN: 1-60132-503-7, CSREA Press ©

100

Int'l Conf. Internet Computing and Internet of Things | ICOMP'19 |

Fig. 7: Evaluation Experiment Flow blooms or not. The remaining 10% is used as the test data for conducting the evaluation experiment using the created classification model. Fig.7 shows the evaluation experiment flow.

4.3

Evaluation Result

The input images to CNN were resized to 224×224, which is also the case of the model of AlexNet. The batch size was 256. The learning rate for the optimization is unified at 0.001. Following is the result of the evaluation experiment. First, Fig.8 shows the transition of the accuracy of algal bloom determination by CNN. Next, Table 1 shows the result of performance evaluation of algal bloom determination by CNN. For the output layer of AlexNet, which is frequently used for multi class classifications, softmax function is used. However, for binary classifications, sigmoid function is commonly used. Thus, a CNN model that changed the output layer of AlexNet to sigmoid function was used for this experiment. Note that the threshold for the class probability at the output layer for determining whether there is an algal bloom or not is 0.5. This means that when the probability of “With algal bloom” is 0.5 or more, it is determined as “With algal bloom”, and when it is less than 0.5, it is determined as “Without algal bloom”. Moreover, the epoch number is evaluated at 100, which is where the accuracy of the learning converged, and its result is the evaluation of the performance of the model. As a result, a relatively high accuracy, with precision at 0.78, recall at 0.75 and F-value at 0.77, was obtained. It is inferred that the fact that CNN could capture the characteristic shapes of the target (characteristics such as many curved lines

Fig. 8: Transition of Epoch Number and Accuracy of CNN Table 1: Result of Performance Evaluation of Algal Bloom Determination by CNN Precision 0.78

Recall 0.75

F-value 0.77

of algal blooms) well led to the relatively high F-value being obtained. However, it is not yet at the level of practical application. Thus in future, further improvement in accuracy, for instance through using an optimal CNN model or a parameter, is required.

ISBN: 1-60132-503-7, CSREA Press ©

Int'l Conf. Internet Computing and Internet of Things | ICOMP'19 |

5

Conclusion

In this paper, we proposed an independent algal bloom removal system that can remove algal bloom independently through automatic detection of where they are occurring by coordinating a drone, an automated algal bloom removal robot, and a cloud server. Further, as this system requires automatic detection of where algal bloom is occurring, it is extremely important for the evaluation of this system to assess the performance of the algal bloom determination system which determines if there are algal bloom or not. Therefore, we conducted an evaluation experiment to assess its determination performance. As a result, a relatively high accuracy, with precision at 0.78, recall at 0.75 and F-value at 0.77, was obtained. The future challenges include a proposal of algal bloom determination system using near infrared images. Near infrared ray is a short infrared ray with wavelength at approximately 0.72 to 2.5 micrometers. Chlorophyll contained in plants absorbs visible light in the process of photosynthesis. However, it reflects near infrared ray. Therefore, the wavelength region of near infrared ray has strong reflection from plants but hardly has any reflection from water surface, which enables clear distinction between plants and water surface. Proposal of an algal bloom determination system using near infrared images in order to determine the occurrence areas of algal blooms and improve the algal bloom determination performance is the challenge for the future.

Acknowledgment This study is partly supported by the JSPS KAKENHI 18K11267.

References [1]

Algal bloom – Wikipedia. https://en.wikipedia.org/wiki/Algal_bloom, (accessed 2019-05-23). [2] Kumamoto University, “MIZUMORI.” https://www.kumamotou.ac.jp/daigakujouhou/kouhou/pressrelease/2013_file/release130812.pd f, (accessed 2019-05-23, in Japanese). [3] Ebismarine Corporation, “Algae Hunter.” http://ebismarine.com/algae.html, (accessed 2019-05-23, in Japanese). [4] Mobile SDK - DJI Developer. https://developer.dji.com/mobile-sdk/, (accessed 2019-05-23). [5] Y.LeCun, L.Bottou, Y.Bengio, and P.Haffner, “Gradient-based learning applied to document recognition,” Proceedings of the IEEE (Volume: 86 , Issue: 11), pages 2278-2324, 1998. [6] HTC Nexus 9. https://en.wikipedia.org/wiki/Nexus_9, (accessed 201905-23). [7] Google Maps API. https://cloud.google.com/maps-platform/, (accessed 2019-05-23). [8] DJI Phantom4 PRO. https://www.dji.com/phantom-4-pro, (accessed 2019-05-23). [9] P.E.Hart, N.J.Nilsson, and B.Raphael, “A Formal Basis for the Heuristic Determination of Minimum Cost Paths,” IEEE Transactions on Systems Science and Cybernetics, Volume 4, Issue 2, Pages 100-107, 1968. [10] A.Krizhevsky, I.Sutskever, and G. E.Hinton, “ImageNet Classification with Deep Convolutional Neural Networks,” Proceedings of the 25th International Conference on Neural Information Processing Systems (NIPS’12), Volume 1, Pages 1097-1105, 2012. [11] OpenCV. https://opencv.org, (accessed 2019-05-23).

ISBN: 1-60132-503-7, CSREA Press ©

101

102

Int'l Conf. Internet Computing and Internet of Things | ICOMP'19 |

Safety Analysis System using Smart Helmet Ahyoung Lee1 , JunYoung Moon2 , Se Dong Min3 , Nak-Jun Sung2 , and Min Hong4 1 Department of Computer Science, University of South Dakota, Vermillion, SD, USA 2 Department of Computer Science, Soonchunhyang University, Asan, Republic of Korea 3 Department of Medical IT Engineering, Soonchunhyang University, Asan, Republic of Korea 4 Department of Computer Software Engineering, Soonchunhyang University, Asan, Republic of Korea

Abstract— In this paper, we introduce a safety analysis system based on our previous research [1]. The proposed method uses a smart helmet to collect data through experimental tests on construction worksites, and we design a safety analysis algorithm that can be applied to the safety management of actual construction worksites based on the data. Smart helmet is designed and implemented to detect dangerous situations and prevent accidents on the construction worksites by using the following three sensors. Acceleration sensor for a sudden falling analysis of construction worker, carbon monoxide sensor for fire or gas leak analysis of construction sites, and ultrasonic sensor for objects falling analysis. As shown by our experimental test results, we expect that the proposed method can be used for prevention of accidents at the construction worksites and for quick response in case of an accident. Keywords: IoT, Smart Helmet, Safety Management, Bluetooth Communication

1. Introduction In the past, the safety management of construction work sites was only limited to the installation of safety structures to prevent accidents. However, in this way, the prevention of actual construction worksite accidents and the risk of construction worksite were not significantly reduced. In this research, the safety management system of construction worksite using some sensors was designed to detect and prevent the actual accidents which can be occurred on actual construction worksites [2]. Although the principle of safety management is simple, it is very difficult to apply properly in actual construction worksites. According to the Bureau of Labor Statistics (BLS) deadly catastrophic industrial disaster census in 2015 [3], the construction industry had 9,234 worker deaths and the fatal injury rate was around 10% in the construction worksites on USA during 10 years (2006 ∼ 2015) as shown in Fig.1. It can be seen that safety management on the construction worksite is actually not easy. For these reasons, a safety management system is needed as increasing demand for a safe, efficient, and convenient. However, most industrial companies, except large corporate businesses, do not have such well-organized safety management systems [4].

Fig. 1: Construction Worker Fatalities (2015)[3].

The common safety management system for construction worksites has been designed using some sensors. In a preliminary research [1], we analyzed the location and work performance of workers using beacons and smart helmets. In this paper, we analyze and verify the dangerous situations of the worker falling, fire, gas leakage, and object falling to the worker by accumulating the risk factor data that can occur in the construction work using the smart helmet. Therefore, it is expected to contribute to the prevention of safety accidents and quick response by allowing the accurate measurement of system and implementing the efficient safety management procedures. The rest of this paper is organized to present information in four different sections: related work 2, system model 3, performance evaluation 4, and conclusion 5.

2. Related Works Safety related accidents continue to occur at various construction worksites around the world. Most accidents at construction worksites are physical accidents such as crashes, chemical exposure, and natural accidents like gas leaks and fire with many of them life-threatening. In recent years, increasing attention has been given to preventing and coping with safety accidents. Thus, it is necessary to improve the efficiency of safety equipment in response to disaster situations. Some research in safety monitoring and implementation have been active, especially in the area of user safety analysis and sensor applications [5]. In this paper,

ISBN: 1-60132-503-7, CSREA Press ©

Int'l Conf. Internet Computing and Internet of Things | ICOMP'19 |

103

Fig. 2: Smart Helmet Configuration. we used the sensors in smart helmets to prevent and respond quickly to frequent or dangerous accidents. A construction worker fall is the most common example of accidents at construction worksites. This fall accident can be prevented or intervened with sensors where an algorithm uses an accelerometer to measure acceleration [6], [7], [8]. In addition to worker falls, collisions with falling construction related objects are also some of the frequently occurring safety accidents at many construction worksites. The distance and velocity between these objects can usually be measured and studied through ultrasonic sensors [9]. It is also common to find situations such as fire or gas leaks using carbon monoxide [10], [11]. These safety accidents are also very dangerous causing severe injury or even death.

3. Smart Helmet 3.1 Device Information The proposed smart helmet is implemented based on three sensors: acceleration sensor, ultrasonic sensor, and carbon monoxide sensor, and also based on an Arduino MCU (Micro Controller Unit) with a Bluetooth module. The MCU refers to a computer that has a microprocessor and an input/output module on one chip and performs predetermined functions. It uses less power than a normal PC, thus it is used to perform simple calculations or operations because of its low performance. Generally, it is used to automatically control product or device, such that Arduino, Raspberry and Beagle are MCU devices. The configuration of smart helmet used in this research is shown in Table 1 and in Fig. 2. The smart helmet is connected to the smartphone application by using the Bluetooth

Table 1: Smart Helmet Sensors MCU

Sensor JarduinoUNO BTmini

Accelerometer

MMA8452

CO Sensor

MQ-7

Ultrasonic Sensor

HC-SR04

Functions MCU&BT 3-axis accelerometer measurement Carbon monoxide measurement Distance measurement

Pin Number RX:4 TX:7 SCL:A5 SDA:A4 A0 Trig:10 Echo:9

module attached to the Arduino MCU. The Arduino MCU performs the process of collecting data of each sensor into one packet, and then the smartphone application receives the packet, and divides and stores the data according to the sensor type. Fig. 3 shows the movement of worker based on data received from the accelerometer sensor. The packets which include acceleration, ultrasonic and carbon monoxide related data sent from the Arduino MCU to the application are shown in Fig. 4. Fig. 5 presents a flow chart of the smart helmet data receiving application used in this research until it is connected to the smart helmet to receive and store the data. In this paper, we designed the system as shown in Fig. 6 to collect and analyze the data sensed by the smart helmet. It sends sensed data to the smartphone which is linked with Bluetooth communication from smart helmet. The smartphone stores the received data in the local DB and the server DB. The server sorts the data stored in the DB in the order of sensing time and sends it back to the smartphone. Finally, the application visualizes the data

ISBN: 1-60132-503-7, CSREA Press ©

104

Int'l Conf. Internet Computing and Internet of Things | ICOMP'19 |

Fig. 3: Snapshot of Smart Helmet Application.

Fig. 5: Smart Helmet Application Flow Chart.

Fig. 4: Smart Helmet Data Format.

collected from smartphone as a graph. Data analysis is done through the last graph and organized data.

3.2 Experimental Test Procedure The experimental test was divided into three cases: worker fall, gas leakage and fire, and falling objects. Regardless of the subject, gas leak, fire and object falling on the head which can be analyzed were carried out without recruiting the subjects.

Fig. 6: The Analysis System Flow Design.

3.2.1 Worker Fall Case A total of 15 subjects were measured in the worker fall test. Seven men in their twenties and eight women in their twenties participated in the experimental test. In the experiment, the accelerometer sensor data were measured through the walking and tumbling behavior while wearing the smart helmet, and the data were collected twice.

3.2.2 Gas Leak and Fire Case Gas leaks and fire tests were carried out by placing a smart helmet in a specific location and using carbon monoxide artificially to collect carbon monoxide data measured through the sensor.

ISBN: 1-60132-503-7, CSREA Press ©

Int'l Conf. Internet Computing and Internet of Things | ICOMP'19 |

Fig. 7: Snapshot of Implemented Smart Helmet.

105

Fig. 8: Walking Situation Data.

3.2.3 Object Falling on Head Case Finally, the object dropping onto the head was carried out by using ultrasonic sensor, dropping a little heavy object starting with light paper and collecting data about it.

4. Smart Helmet Data Analysis The data used for the analysis were collected through smart helmets and smartphone applications. Fig. 7 shows our implemented smart helmet.

4.1 Accelerometer Sensor Data Analysis Accelerometer data were used to analyze worker falls. The data were compared, and also the feature points were compared when general walking and falls occurred. Fall situations can come in many situations, such as falling forward, falling back, falling sideways, and so on. In this paper, we analyzed the common data of falling events in the front and back, and based on this, various fall situations are analyzed. Data analysis was measured data trends by averaging the data collected for each axis and using percentages as the main value for each axis. Fig. 8 is a graph of data displayed in the walking state. The X-axis of the graph represents the progress of the action over time, and the Y -axis represents the amount of data change about the mean. In the walking state, the X-axis and Z-axis values from acceleration sensor were showed to a small degree of shaking. In the case of the Y -axis value, the large degree of shaking was measured in which the data was maintained at a minimum of −4% from the maximum of 230%. Also, all three axes showed patterned motion. On the other hand, in Fig. 9, a large change was measured and data showing a large change in a short time without being patterned was measured. This figure is a graph showing the data for falls falling forward. In the forward fall situation, all three axes showed a large change, and the variation of the X-axis was very large. In the case of the X-axis with the largest variation of the maximum and minimum values, the maximum change from the maximum of about 471% to the minimum of 470% was more than twice as large as

Fig. 9: Forward Falling Situation Data.

that of the walking state. Also, the amount of data change during the displayed short time of the graph was measured to a large extent. Fig. 10 is a graph of the data of the backward fall situation. Similar to the forward fall situation, the data change was larger in the three axes than in the walking situation, and the data change over a short period of time indicated in the graph was measured to a large extent. When the angle of the head is changed in the fall situation, it is confirmed that the measured value of each axis of the accelerometer sensor is changed greatly. When falling forward, the value of the X-axis is measured to be high because the head is tilted without turning the head forward. However, when falling backward, the angle of head changes when tilted because it tends to see the back unconsciously. Therefore, the height of Z-axis was measured differently from the frontward fall. As a common feature of both cases, it is confirmed that the overall data change is large within a short time regardless of the axis data change according to the angle during measurement. Therefore, if a large amount of change appears in a short time regardless of the axis, it can be detected as falling situation. As a result, it was confirmed that the data change of two times or more was measured in a shorter time than the case of walking in all three axes in the forward and

ISBN: 1-60132-503-7, CSREA Press ©

106

Int'l Conf. Internet Computing and Internet of Things | ICOMP'19 |

Fig. 10: Backward Falling Situation Data. Fig. 12: Ultrasonic Sensor Data of light Object.

Fig. 11: CO Sensor Data. Fig. 13: Ultrasonic Sensor Data of heavy Object.

backward falling situation.

4.2 CO Sensor Data Analysis The CO sensor is a sensor that measures the ambient CO concentration. The experimental tests for CO sensor case were performed by placing the smart helmet in a specific place and artificially introducing carbon monoxide. The fire situation was experimented by burning papers to generate carbon monoxide, thus Fig. 11 shows the average of CO sensor data value. In a typical situation, the CO sensor data value is measured on average between 170 and 200. As a result, it was found that the sensor value increases proportionally as the ambient CO concentration increases slightly. The value of sensor rose to a maximum of about 500, which was measured in a situation where the smoke was full.

4.3 Ultrasonic Sensor Data Analysis An ultrasonic sensor can measure the distance between helmet and any other objects. Experimental tests were carried out assuming a situation where the object falls to the top of head. The objects were processed using different weights to distinguish whether the object falling on the head is a heavy object falling quickly or a light object falling slowly. The falling motion was measured for the same time under the same conditions. Fig. 12 is a graph of the experimental

test data with light weight objects. If the weight is not heavy, the relatively smooth curved shape of data is measured on average. Fig. 13 shows the data graph when the experimental test is carried out with relatively heavy objects. For the safety of developed smart helmet and the experimenter, in case of heavy object, the object was tied up with a string, so it was slowed down when the data value was less than 50, unlike the experimental test with light object. It can be seen that the amount of change in data value over a short period of time is larger than that of a light weight object. As the weight of object increases, the falling rate also increases. Therefore, the heavier object has to have the larger amount of data change in a short time. It can be expected that a very dangerous situation may occur when colliding with an object having a large data value change in a short time.

5. Conclusion We conducted data analysis for safety of workers and quick response to prevent the safety accidents using smart helmet which can be applied in a construction worksite safety management system. Through the smart helmet, the

ISBN: 1-60132-503-7, CSREA Press ©

Int'l Conf. Internet Computing and Internet of Things | ICOMP'19 |

data of acceleration sensor, ultrasonic sensor, and carbon monoxide sensor were collected, and the data were analyzed to detect the specific situations. The worker falling can be alerted using the change of acceleration sensor data, and the collision with the falling dangerous objects can be detected when the larger amount of data values is changed in a short time with the ultrasonic sensor. The fire or gas leakage can be readily predicted when the value of carbon monoxide sensor is increased dramatically. As the future works, we will standardize the data by collecting and analyzing various data through more complex experimental tests. In addition, it is expected that it will contribute to more accurate safety management and quicker safety accident introduction by applying it into safety accident management system.

References [1] K.-B. Seo, D. min, S.-H. Lee, and M. Hong, “Design and implementation of construction site safety management system using smart helmet and ble beacons,” will be published at the Journal of Interne Computing and Services (JICS), 2019. [2] H.-S. Lee, K.-P. Lee, M.-S. Park, H.-S. Kim, and S.-B. Lee, “A construction safety management system based on building information modeling and real-time locating system,” Korean Journal of Construction Engineering and Management, vol. 10, no. 6, pp. 135–145, 2009. [3] BLS, “Nationa census of fatal occupational injuries in 2015,” in USDL-16-2304, 2016. [4] J. M. Wilson and E. S. ¸ Koehn, “Safety management: problems encountered and recommended solutions,” Journal of construction engineering and management, vol. 126, no. 1, pp. 77–79, 2000. [5] A. Carbonari, A. Giretti, and B. Naticchia, “A proactive system for real-time safety management in construction sites,” Automation in Construction, vol. 20, no. 6, pp. 686–698, 2011. [6] J. Chen, K. Kwong, D. Chang, J. Luk, and R. Bajcsy, “Wearable sensors for reliable fall detection,” in 2005 IEEE Engineering in Medicine and Biology 27th Annual Conference. IEEE, 2006, pp. 3551–3554. [7] A. Bourke, J. OŠbrien, and G. Lyons, “Evaluation of a thresholdbased tri-axial accelerometer fall detection algorithm,” Gait & posture, vol. 26, no. 2, pp. 194–199, 2007. [8] J. Mantyjarvi, J. Himberg, and T. Seppanen, “Recognizing human motion with multiple acceleration sensors,” in 2001 IEEE International Conference on Systems, Man and Cybernetics. e-Systems and e-Man for Cybernetics in Cyberspace (Cat. No. 01CH37236), vol. 2. IEEE, 2001, pp. 747–752. [9] A. Carullo and M. Parvis, “An ultrasonic sensor for distance measurement in automotive applications,” IEEE Sensors journal, vol. 1, no. 2, p. 143, 2001. [10] S.-J. Chen, D. C. Hovde, K. A. Peterson, and A. W. Marshall, “Fire detection using smoke and gas sensors,” Fire Safety Journal, vol. 42, no. 8, pp. 507–515, 2007. [11] R. Pohle, E. Simon, R. Schneider, M. Fleischer, R. Sollacher, H. Gao, K. Müller, P. Jauch, M. Loepfe, H.-P. Frerichs, et al., “Fire detection with low power fet gas sensors,” Sensors and Actuators B: Chemical, vol. 120, no. 2, pp. 669–672, 2007.

ISBN: 1-60132-503-7, CSREA Press ©

107

108

Int'l Conf. Internet Computing and Internet of Things | ICOMP'19 |

A Model and Middleware for Composable IoT Services Ahmed Abdelmoamen Ahmed Department of Computer Science, Prairie View A&M University Prairie View, TX, USA [email protected]

Abstract—Internet of Things (IoT) can be considered the rising star in the sky of Distributed Computing. With IoT sensors becoming increasingly ubiquitous, there is tremendous potential for IoT services which can take advantage of the data collected by these sensors. An IoT service may involve a large number of end-devices (e.g., sensors and actuators) connected to gateways, which act as the aggregation points for a group of sensors and actuators to coordinate the connectivity of these devices to each other and to an external network. In this paper, we present a model for representing IoT services, allowing them to be carefully studied. The model provides a set of coordination mechanisms underlying such services and informing design and implementation decision about them. We have implemented these coordination mechanisms in the form of a distributed middleware over which such class of IoT services could be implemented relatively easily. The simplest form of a service is defined by having one data-contributor sending a one-time service message to one client; more complex services can be specified by composing simpler services. This paper presents the syntax and operational semantics of the model, and describes the design and implementation of the middleware. Keywords-IoT, Services, Model, Composition, Middleware, Actors.

I. I NTRODUCTION The growing ubiquity of IoT end-devices opens up an opportunity to offer innovative services based on both what the millions of sensors are sensing, as well as what actuators are actively contributing. Consider an IoT service for Precision Agriculture (PA) [1] which measures the spatial variability of soil properties in agricultural fields, monitors farm conditions, and plans irrigation and harvesting. Such services would tackle production costs and operational challenges for both small- and large-scale farmers. Consider other services, such as one of a livestock-monitoring application which utilizes a set of sensors mounted in a livestock farm to collect data regarding the location, well-being, and health of cattle. This information would help farmers in identifying animals that are diseased, so that it can be separated from the herd, thereby preventing the spread of diseases. These services rely on the state of the context in which sensing devices are located such

as geographical location, proximity, temperature, wind speed and direction, solar radiation, humidity, etc. [2]. Increasingly, sensed data could also inform decisions to activate actuators to carry out tasks automatically. A growing number of smart farm technologies offer good examples of such capability. We broadly refer to these as Distributed IoT Services. Although there is a growing body of work devoted to supporting such services (see [3]–[5]), the barriers to offering IoT services continue to be prohibitive because of the lack of precise understanding, specification, and analysis of such services. Furthermore, the focus has been on narrow application areas or specific concerns, making it difficult to utilize them for a broader class of services. Besides, such services often require complex communication and coordination mechanisms, which are not adequately supported by existing ones. This paper presents a formal model, DisIoTS (Distributed IoT Services), for representing distributed IoT services. The model allows key principles of these services to be rigorously studied. DisIoTS also identifies core mechanisms underlying such services, which inform design and implementation decision about them if they are implemented over a middleware. We first present the syntax and operational semantics of DisIoTS. Second, we show how services can be built by composing simpler services, and give examples to demonstrate the concept of service composition. We also present the design and implementation of a distributed middleware which can be used to simplify the burden of initiating and managing IoT services. The middleware is organized with two parts executing on the IoT end-devices, as well as highperformance servers. II. S ERVICES A service is something that receives input contributions from some contributing source, processes them, and creates output contributions for some client. We call these contributions as service feeds. In a system of services, we treat every client and contributor as a service: they are called the client service and the contributing service of the particular service. In other

ISBN: 1-60132-503-7, CSREA Press ©

Int'l Conf. Internet Computing and Internet of Things | ICOMP'19 |

…

…

……

…...…

…...…

service’s boundary message flow

Fig. 1.

…

…

…

We define services compositionally. In other words, simpler services can be composed to create more complex services by repeatedly applying a set of composition rules. A service s is written as: [[I : α : O]]s

……

III. S ERVICE C OMPOSITION

…

In a system of services, the required communication between services is carried out by sending and receiving asynchronous messages. Messages can be of one of two types: control messages or data messages. Control messages (also called inter-service messages) are communicated between services for administrative purposes; while the data messages are used to send service feeds from contributing services to client services. Each service has a dedicated agent called the service coordinator, which is responsible for handling control messages between its service and other services in the system. When a control message is sent to a service, the message is received by its coordinator agent. The rest of agents implement the service logic and process the received data messages. Figure 1 illustrates the interaction between services. A rectangle represents a service’s boundary. Each service has a set of ports using which it could communicate with other services in the system. The figure also illustrates the components of a service. Ovals are agents serving the service, white circles are input ports, black circles are output ports, and the lines with arrows represent message flows; the service encapsulates a set of agents implementing its logic. These agents are invisible to other services. To interact with a service, both contributing and client services have to be connected to the service’s ports: to send messages to the service, contributing services are connected to its input ports; to receive messages from the service, client services are connected to its output ports.

…

A. Communication between Services

…

words, the client service could simply be an end user receiving a feed from some service, but not necessarily producing a derivative service for another service; a contributing service, similarly, could be just a sensor without any service contributing to it. We model a service by a set of ports and a set of agents. Ports can be one of two types: input or output. A service receives input feeds from contributing services through its input ports and sends output feeds to client services through its output ports. Agents implement the service’s logic and convert the feeds arriving from contributing services into the feeds required by client services. We represent both ports and agents as active objects. Consequently, a service can be defined as a set of active objects.

109

output port

input port agent

Communication between Services

where s is the service’s unique name, I is a set of input ports, α is a set of agents of the service, and O is a set of output ports. To be precise in our presentation of services, we assume that the computations involved in a service are carried out by actors [6]. Actors are active objects, which use asynchronous messages to communicate. In particular, we use actors to model ports and agents. An actor with name a ∈ α and behavior b is represented by (b)a . Recipients of messages are identified by unique addresses (i.e., actor names). An actor can only communicate with actors whose names it possesses. In DisIoTS, input ports are receptionists of data. Therefore, contributing services must know the name of input ports of client services in order to send both control and data messages – through their output ports – to them. The primitive service is a one-time service, which has one input port, one agent, and one output port. The primitive service has one contributing service sending it a one-time service feed, ζ, through its input port, i (white circle), which it sends to a client service through its output port, o (black circle), can be written as: [[{i} : {a} : {o}]]s where i is an input port, o is an output port, and a is an agent which has a forwarder behavior. For improving readability, from here on, we will refer to {a}, {i} and {o} as simply a, i and o, respectively. The intent from the curly braces of sets in our case is obvious A. Composition Rules There are three service composition rules: union, output merge, and serial composition. 1) Union Composition: The union composition rule is used to compose a number of services by combining their ports and agents, as well as their service feeds. The purpose of this composition is to create a new service which utilizes the definitions of a number of existing services, to widen the scope of coverage provided by them over time, space, or both. For instance, this composition would be useful to create a new service which takes advantage of some existing services covering different

ISBN: 1-60132-503-7, CSREA Press ©

110

Int'l Conf. Internet Computing and Internet of Things | ICOMP'19 |

geographical locations in a city, observing various events of interest in the same location, etc. The union composition first creates a new service with empty sets of input ports, agents and output ports. Both ports and agents of the services being composed, which retain the old behaviors, are cloned and added to their corresponding sets ports and agents of the new (composed) service. Figure 2 shows an example of composing n > 1 services over space using the union composition rule. Each contributing service sj , which has a set of input Ij , a set of agents αj and a set of output ports Oj , produces a set of feeds Zj during the execution of the service, for 1 ≤ j ≤ n, where n is number the contributing service to be composed.

ȡ݊ Fig. 2.

ሼ ଵ ǡ ǥ ǡ ௡ ሽ

ሼଵ ǡ ǥ ǡ ௡ ሽ

ሼߙଵ ǡ ǥ ǡ ߙ୬ ሽ

…

…

ߙ

݊

ȡଵ

…

..…

…

݊

ߙͳ

ͳ …

…

ͳ

ሼȡଵ ǡ ǥ ǡ ȡ௡ ሽ

Union Composition

merging their output ports, the composition receives as parameters the set of output ports to be merged and one behavior. The composition first creates a new service with empty sets of input ports, agents and output ports. Then, the input ports of the services being composed, which retain the old behaviors, are cloned and added to the input ports of the composed service. The composition then creates a new output port, which becomes the output port of the composed service, and creates a new agent actor with the provided behavior targeting the newly created output port. Finally, the composition transforms the output ports to be merged into agents which have the newly created agent as their target. Figure 3 shows an example of composing n > 1 services using the output merge composition rule. Each contributing service sj , which has a set of input Ij , set of agents αj and set of output ports Oj , produces a set of feeds Zj during the execution of the service, for 1 ≤ j ≤ n, where n is number the contributing service to be composed. αc is the set of agents of the composed service. Zc is a set of feeds produced by the composed service. oc is the newly created output port of the composed service. This composition can be described as follow: n n   |nj=1 [[Ij : αj : Oj ]]sj ⇒ [[ Ij : αj ∪ {(b)ac } ∪ S o ,b

We use the composition operator symbol | to represent one of the composition rules defined in this section. The union composition of n > 1 services can be described as follow:  n n n    |nj=1 [[Ij : αj : Oj ]]sj ⇒ [[ Ij : αj : Oj ]]sc j=1

j=1

j=1

S :(

n 

j=1

j=1

o

Oj − S ) ∪ {oc }]]sc

j=1

where n > 1, S o is a set of output ports to be composed, b is the provided behavior, sc is the name of the composed service, oc is the newly created output port, ac is the name of the newly created agent actor with a behavior b targeting oc , and Dom(αj )’s are disjoined for 1 ≤ j ≤ n. Each output port in S o , which retains the same behavior, is added to the set of agents of sc , and they all have their target to the newly created agent ac .

…

ߙ

݊

ȡͳ ȡ݊

Fig. 3.

ሼ ଵ ǡ ǥ ǡ ௡ ሽ …

..…

…

݊

ߙͳ

ͳ …

ͳ

…

where n > 1, sc is the name of the composed service, and the domains of all αj (Dom(αj )) are disjoined for 1 ≤ j ≤ n. 2) Output Merge Composition: The output merge composition rule is used to compose a number of services by merging their output ports. The purpose of this composition is to create a new service which enables a single client to receive service updates collected by a number of contributing services aggregated in some order. That said, the input feeds can be received by the composed service and processed as required by the service’s logic. Alternatively, they can be somehow aggregated. In its simplest form, the service simply forwards out each feed received in its original form. In more interesting forms, it can process received feeds in permitted ways, both to create aggregate feeds to be forwarded, and to make decisions about whether and when to forward aggregates [7]. It turns out that the output merge composition rule requires a set of parameters to be provided at the composition time. To compose a number of services by

o

ߙ…

‘ܿ

ȡܿ

Output Merge Composition

3) Serial Composition: The serial composition rule is used to create a new service by composing a number of services serially. The composition connects some of the

ISBN: 1-60132-503-7, CSREA Press ©

Int'l Conf. Internet Computing and Internet of Things | ICOMP'19 |

output ports of one service to some of the input ports of another service. The purpose of this composition is to create a new service which enables its client to consume feeds from a pipeline of services, where the feed flow is from the first service to the second service, and so on. In its simplest form, each service in this pipeline simply forwards out each feed received in its original form. Alternatively, it can process the received feeds before forwarding it to the next service. The composed service can be created by composing a number of services in serial if a port-map, M, is provided. Each entry in M has the form (osk , isk+1 , bk ), where osk is the name of an output port of the k th service being composed, isk+1 is the name of an input port of the (k+1)th service being composed, and bk is a binding behavior for a newly created agent actor which connects between osk and isk+1 , for 1 ≤ k < m, where m is the number of entries in M. Figure 4 shows an example of composing n > 1 services using the serial composition rule. Each contributing service sj , which has a set of input Ij , a set of agents αj and a set of output ports Oj , produces a set of feeds Zj during the execution of the service, for 1 ≤ j ≤ n, where n is number the contributing service to be composed. Ic , Oc and αc are the sets of input and output ports, and agents of the composed service, respectively. Zc is a set of feeds produced by the composed service. This composition can be described as follow: ։ n m   |nj=1 [[Ij : αj : Oj ]]sj ⇒ [[( Ij ) − ( {ik }) : M n 

αj ∪

j=1

m 

{(b)ak } : (

n 

j=1 m 

Oj ) − (

j=1

k=1

k=1

{ok })]]sc

k=1

where n > 1, and Dom(αj )’s are disjoined for 1 ≤ j ≤ n; M is a provided port-map, with its k th entry has the form of (osk , isk+1 , bk ), which is a triple identifying an output port osk of the k th service, which connects to an input port isk+1 of the next service, and a behavior bk of a newly created agent actor ak which implement that binding behavior, for 1 ≤ k < m, where m is the number of entries in M; and sc is the name of the composed service. ܿ

ȡ݊

ߙ…

ܿ …

݊

…

…

Fig. 4.

ߙ

…

ȡͳ

݊

…

ߙͳ

ͳ …

…

ͳ

ȡܿ

Serial Composition

IV. O PERATIONAL S EMANTICS This section presents the operational semantics for DisIoTS. We define the state of a service as follows:

111

Definition 1: Service State A service is denoted as s and is written as: [[I : α : O]]s where s is the service’s unique name, I is a set of input ports of the service, α is a set of agents of the service, and O is a set of output ports of the service. Because agents are modeled as actors, α can be considered as an actor map which maps a finite set of actor names to their behaviors. A coordinator actor of a service s, written as as ∈ α, receives all messages sent to s. Notice that there is a one-to-one mapping between s and as ; i.e., for each service s, there is exactly one coordinator actor. Definition 2: DisIoTS Configuration The instantaneous snapshot of a system of services is called a DisIoTS configuration. A DisIoTS configuration represents the state of a finite set of long-lived services, a finite set of contracts between services which define how services are connected to each other, and a finite set of control messages between services. Here, we are only interested in modeling the administrative interactions between services, so we do no show data messages communicated between services. A contract is negotiated between two services when one of them wants to consume service feeds produced by the other. A DisIoTS configuration can be represented by a 3-tuple: S | C | M S is a set of services. C is the set of contracts between the services, where each contract c ∈ C has the form (s1 , s2 , map), where s1 is the name of the first service, s2 is the name of the second service and map is a name table which says which output ports of the first service are connected to which input ports of the second service. The connections are represented using (o ; i) pairs where o ∈ O is an output port of s1 and i ∈ I is an input port of s2 . The contracts involving a service s ∈ S can be written as co(s) ⊂ C. M is a finite set of control (inter-service) messages in the system which are communicated between services, and are handled by coordinator actors of the communicating services. DisIoTS assumes that there is a special service, called the directory service, which has up-to-date information about the capabilities of all contributing services in the system. For example, the service capability of a sensor service is determined by the maximum sampling rate of that sensor. The directory service can be implemented as a federated hierarchy of directory services which can be distinguished by different metrics such as geographical locations, type of services, etc. Each contributing service in the system must be registered with the directory service in order to participate in serving clients.

ISBN: 1-60132-503-7, CSREA Press ©

112

Int'l Conf. Internet Computing and Internet of Things | ICOMP'19 |

A. Service Request One way to represent the requirements of a service is to represent them as sets of timed data feeds. For each of these feeds, the service needs to secure appropriate sensor data feeds and carry out the required aggregations and customizations for different clients. This representation of a service request gives the ability to a client to represent their service requirements without being too rigid. If the client is too rigid in defining their request, the service request is likely to be rejected if the directory service is too busy at these points of time. A service request is represented by ρ and is defined as follows: Definition 3: Service Request A service request ρ is a set of sets of timed service data feeds. Each set ρi ∈ ρ has a sufficient number of service feeds for serving that request. A service request is represented formally as follows: ρ = {ρ1 , . . . , ρn }, ρi = ζ1 , . . . , ζm where n ≥ 0 is the number of sets in ρ, m ≥ 0 is the number of service feeds in ρi , and ζ represents a single timed service feed. Only one set ρselected out of ρ needs to be served. B. Transition Rules We use transition rules to describe the progress of a system of services. Next, we present the transition rules of DisIoTS. 1) Sending a service request: In DisIoTS, a service is initiated by a client service which sends a service request ρ to the directory service with the intent of creating a new service. The client’s requirements are expressed in that request ρ. The same service request can also be used to subscribe to an existing service. The following transition shows how a client service sends a service request to the directory service: → [[I : [R[send(sd , (ρ, s))]]as , α : O]]s , S | C | M − [[I : [R[nil]]as , α : O]]s , S | C | sd ⇐ (ρ, s) , M where s is a client service, sd is the directory service, ρ is a service request, as is s’s coordinator actor, and send(sd , (ρ, s)) sends message (ρ, s) containing the received ρ and the client’s name to the directory service sd . This leads to the creation of message sd ⇐ (ρ, s) on the right hand side, and actor as continues execution. For convenience, we define a function search which is used by the directory service to determine the opportunity to serve a new service request ρ by selecting one existing service matching the requirements of ρ, or a set of contributing services which could collectively contribute to serving ρ. The search function takes as parameters a service request ρ and the name of the

client service s requesting ρ, and returns one of three pairs: (1) ({sm }, c), a pair of the name of an existing service sm which matches ρ’s requirements and a new contract c created between sm and s; (2) (Sρ , ∅), a pair of a set of potential contributing services Sρ which could collectively contribute to serving ρ based on their capabilities and an empty set ∅ indicating that no contract is created at this point; or (3) (∅, ∅) to indicate that there is no way of serving ρ. The search function is defined as follows: search(ρ, s) = ({sm }, c) | (Sρ , ∅) | (∅, ∅) 2) Receiving a service request: On receiving a service request ρ, the directory service uses the search function to determine the opportunity to serve ρ. This transition is written as: [[I : [R[ready(search)]]asd , α : O]]sd , S | C | sd ⇐ (ρ, s) , M − → [[I : [search(ρ, s)]asd , α : O]]sd , S | C | M where sd is the directory service, asd is sd ’s coordinator actor, s is the name of the client service which sent ρ, sd ⇐ (ρ, s) is a service request message sent to sd , and ρ is the service request. As a result of delivery of this message to actor asd , asd uses the search function to first searches for an existing service matching the requirements of ρ, or a set of contributing services which could collectively contribute to serving ρ. 3) Subscribing to an existing service: If a matching service is found, then the directory service tells the client service about this service, and a new contract is signed between the client service and the found service. If search(ρ, s) evaluates to ({sm }, c), then the transition rule for subscribing to an existing service is as follows: λ →X ({sm }, c) ⇒ search(ρ, s) − → [[I : [R[search(ρ, s)]]asd , α : O]]sd , S | C | M − [[I : [R[({sm }, c)]]asd , α : O]]sd , S | C ′ | s ⇐ sm , M where X = {asd } is the context in which search(ρ, s) is reduced to ({sm }, c); sd is the directory service; asd is sd ’s coordinator actor; ρ is the service request; s is the name of the client service which sent ρ; sm is the name of an existing service sm which matches ρ’s requirements; c is a new contract created between sm and s, which has the form (sm , s, (osm ; is )) where osm ∈ Osm is an output port of sm and is ∈ Is is an input port of s; and C ′ = C ∪ {c}. This transition happens only if and only if search(ρ, s) is evaluated to ({sm }, c). This leads to the creation of message s ⇐ sm containing sm which is sent to s, and the creation of a new contract c created between sm and s on the right hand side.

ISBN: 1-60132-503-7, CSREA Press ©

Int'l Conf. Internet Computing and Internet of Things | ICOMP'19 |

4) Creating a new service: Consider a function create, which when provided as parameters a service request ρ, a set of contributing services, Sρ , which could collectively contribute to serving ρ, and the name of the client service s requesting ρ, creates a new service which satisfies ρ’s requirements; the create function returns the name of the newly created service s′ , a set of contracts Cρ between s′ and Sρ , and a contract c between s′ and s. The create function is defined as follows: create(ρ, Sρ , s) = ({s′ }, Cρ , {c})

113

constructs implemented in a middleware to support the programmability of IoT services. We prototyped the distributed middleware as an actor system. The distributed run-time system for the proposed middleware is organized with parts executing on the IoT end-devices at the sensing side, and on remote servers at the service platform side. In the rest of this section, we discuss these two parts separately.

If there exist contributing services whose feeds are sufficient for generating ρ, then the service request is accepted. This operation is shown in the following: λ

search(ρ, s) − →X (Sρ , ∅) ⇒ → [[I : [R[search(ρ, s)]]asd , α : O]]sd , S | C | M − [[I : [R[(Sρ , ∅)]]asd , α : O]]sd , S | C | M where X = {asd } is the context in which search(ρ, s) is reduced to (Sρ , ∅), sd is the directory service, asd is sd ’s coordinator actor, ρ is the service request, and s is the name of the client service which sent ρ. This transition happens only if and only if search(ρ, s) is evaluated to (Sρ , ∅). The directory service then uses the create function to create a new service utilizing the contributing services. The transition rule for creating a new service is as follows: λ

Fig. 5.

The GUI for Creating an IoT Services

′

→X ({s }, Cρ , c) ⇒ create(ρ, Sρ , s) − → [[I : [R[create(ρ, Sρ , s)]]asd , α : O]]sd , S | C | M − ′ ′ ′ [[I : [R[({s }, Cρ , {c})]]asd , α : O]]sd , S | C | s ⇐ s′ , M {s′ } , Cρ and {c} are fresh where X = {asd } is the context in which create(ρ, Sρ , s) is reduced to ({s′ }, Cρ , {c}); sd is the directory service; asd is sd ’s coordinator actor; s is the client service which sent the service request ρ; s′ is the newly created service producing ρ from Cρ ’s contributions; Cρ is a new set of contracts created between s′ and Sρ ; c is a new contract created between s′ and s, which has the form (s′ , s, (os′ ; is )) where os′ ∈ Os′ is an output port of s′ and is ∈ Is is an input port of s; C ′ = C ∪ Cρ ∪ {c}; and S ′ = S ∪ {s′ }. This transition happens only if and only if search(ρ, s) is evaluated to (Sρ , ∅). This leads to the creation of message s ⇐ s′ containing the name of the new service s′ , the creation of new sets of contracts Cρ and {c}, and the creation of a new service s′ on the right hand side. V. A M IDDLEWARE FOR D ISTRIBUTED I OT S ERVICES The design of the middleware builds on the domainspecific mechanisms for DisIoTS described in the previous sections. Specifically, we built a set of programming

A. Service Platform Side The prototype implementation of the service platform is built using Actor Architecture (AA) [8], a Java library and runtime system for distributed actor systems. In this platform, a service is initiated by a client which sends a service request to the platform with the intent of creating a new service. The client’s requirements are expressed in that request. The same service request can also be used to subscribe to an existing service. On receiving a service request, the platform determines the opportunity to serve that request by selecting one existing service matching the requirements of the request, or a set of contributing services which could collectively contribute to serving the request. If a matching service is found, then the platform tells the client about this service, and the client subscribes to the found service. If none of the existing services match the target service’s requirements expressed in the request, the platform tries to compose a number of contributing services to create a new service based on the definition of the request. Otherwise, the request is rejected. A service request is represented as a set of timed data feeds, which are called service modes. Using this platform, programmers can specify the modes in which

ISBN: 1-60132-503-7, CSREA Press ©

114

Int'l Conf. Internet Computing and Internet of Things | ICOMP'19 |

actors

……

IoT Service Layer Runtime Environment OS/RTOS HAL Fig. 6.

Software stack for IoT devices

an IoT service can be, the sensing needs of each mode, and the sensed events which trigger mode transition. The platform then monitors for mode transition events, and dynamically adjusts the sensing frequencies to match the current mode’s requirements. Separating the mode change logic from a service’s functional logic leads to more modular code. A graphical interface (shown in Figure 5) is developed to allow programmers to program reusable multimodal sensing requirements for services. New modes can be specified explicitly or learned. Particularly, we have developed a way for new modes to be learned by recognizing mode signatures from examples of sensor data sampled in the mode. B. IoT End-Device Side At the sensing side, we developed an actor-based runtime environment to support the deployment and execution of distributed IoT services primarily. The runtime environment enables programmers to write a single actor program which will automatically become distributed to run on a large number of devices at the edge of the network. Actors will be the building blocks for a service which can be deployed on the top the runtime environment. The actors are connected in a dataflow to form the application. This simplifies actor migration between runtimes and matching of actor requirements with runtime capabilities. If the runtime does not meet the requirements posed by a currently deployed actor, then the actor will be automatically migrated to a runtime that can satisfy the requirements. Figure 6 shows the software stack of an IoT device in the proposed system which includes the following layers: (i) Hardware Abstraction Layer (HAL). HAL is a software layer that enables access to the hardware components of IoT devices such as RAM, ROM, serial interfaces, etc. (ii) OS/RTOS Layer: IoT devices have either embedded or Real-Time Operating Systems (RTOS) that are particularly suited for small constrained devices, and that can provide IoT-specific capabilities. (iii) Runtime Environment Layer: The actorbased runtime environment runs over the OS layer. It hosts several actors which can interact with each

other both locally and over the network. The runtime provides a uniform computing environment for all actors, regardless of hardware or operating system differences. (iv) IoT Service Layer: IoT services are deployed on top of the runtime environment. Services are defined as a set of functions which are represented as actors, where each actor can provide data generation (e.g., sensors), processing and data consumption (e.g., actuators). Actor code execution is triggered by certain events such as receiving a new message from another actor locally or remotely, or detecting a change in the state of a sensor. VI. C ONCLUSIONS In this paper, we presented DisIoTS, a formal model for representing distributed IoT services. DisIoTS defines a service from two different angles: compositionally and directly. From the composition angle, we presented a set of composition rules for constructing complex services by composing simpler ones. From the direct creation angle, we defined a service as a set of actors, which are ports and agents. A service receives input feeds from contributing services via its input ports and sends output feeds to client services via its output ports. The received feed messages are handled by the service’s agents, which sends the processed feeds to output ports. We also presented the operational semantics for DisIoTS, which are defined by a transition relation on DisIoTS configuration. Finally, this paper described a middleware that can be used to easily create new IoT services. For future work, we plan to experimentally evaluate the performance and scalability of the middleware. R EFERENCES [1] R. C. T. Bendre, M. R. and V. R. Thool, “Big data in precision agriculture: Weather forecasting for future farming,” in Proceedings of the IEEE International Conference on Next Generation Computing Technologies, ser. NGCT, 2015, pp. 744–750. [2] A. Abdelmoamen and N. Jamali, “An actor-based middleware for crowd-sourced services,” EAI Transactions on Mobile Communications and Applications (Accepted), no. vol TBD, pp. 1–15, 2017. [3] K. Incki, I. Ari, and H. Sozer, “Runtime verification of iot systems using complex event processing,” in Proceedings of the IEEE International Conference on Networking, Sensing and Control, ser. ICNSC ’17, Calabria, Italy, 2017, pp. 625–630. [4] S. Nastic, S. Sehic, M. Vogler, H. L. Truong, and S. Dustdar, “Patricia – a novel programming model for iot applications on cloud platforms,” in Proceedings of the IEEE International Conference on Service-Oriented Computing and Applications, ser. SOCA ’13, Hawaii, USA, 2013, pp. 53–60. [5] S. Mora, F. Gianni, and M. Divitini, “Rapiot toolkit: Rapid prototyping of collaborative internet of things applications,” in Proceedings of the IEEE International Conference on Collaboration Technologies and Systems, ser. CTS ’16, Orlando, USA, 2016, pp. 438–445. [6] G. A. Agha, I. A. Mason, S. F. Smith, and C. L. Talcott, “A foundation for actor computation,” The Journal of Functional Programming, vol. 7, no. 01, pp. 1–72, 1997. [7] A. Abdelmoamen and N. Jamali, “Coordinating crowd-sourced services,” in Proceedings of MS, Alaska, USA, 2014, pp. 92–99. [8] M. Jang, A. Momen, and G. Agha, Efficient Agent Communication in Multi-agent Systems, 2005, pp. 236–253.

ISBN: 1-60132-503-7, CSREA Press ©

Int'l Conf. Internet Computing and Internet of Things | ICOMP'19 |

115

An Exploratory Study on the use of Internet_of_Medical_Things (IoMT) In the Healthcare Industry and their Associated Cybersecurity Risks Rashad J. McFarland Dept. of Computer Science Norfolk State University 700 Park Avenue Norfolk, VA. 23504, USA

Abstract - The Internet of Things (IoT) has the potential to positively impact the Healthcare industry. The use of this constantly evolving technology has greatly improved patient care, sharing of information between providers and medical institutions to ensure patients are receiving consistent and advanced treatment across the world. With IoT devices comes the endless dialogue about security of patient privacy with network infrastructure, authentication and properly configured end devices. Along with secure devices providers, employees at medical facilities and users must remain vigilant as they prescribe and use the evolving technologies. There are great benefits and potential in the use of IoTs in the Healthcare industry. This research paper analyzed vulnerabilities, risks, and threats that comes with the use of Internet based medical Technology devices (IoMT). In addition to the problems identified, this work provided suggestions on mitigation techniques that can be used to prevent emerging security threats.

Keywords: IoT, IoMT, Cybersecurity, Healthcare Technology, Cybercrimes, Portable Devices

1.

Introduction

1.1 Act

Health Insurance Portability and Accountability

The Internet of Things (IoT) is transforming the daily lives of people all around the world. “The Internet of Things (IoT) is the global network of connected embedded systems. What distinguishes the IoT is that each node is connected to the Internet and is uniquely addressable” [24]. IoT devices include smart televisions, smart cars, smart refrigerators, and even smart locks to secure homes, all of which can all be connected to the internet. Most of these devices can be controlled remotely using devices such as smartphones to confirm actions to the user via text messages, emails, or phone calls. The Internet of Medical Things (IoMT) is an assembly of medical devices and applications that interface with healthcare information technology (IT) frameworks through computer systems. IoMT utilizes Wi-Fi as its base to assist medical devices with communicating with the machine. Instances of IoMT applications are logical dashboard linking remote monitoring of health, eHealth applications, mixture siphons, telemedicine and so on. These applications enable patients and healthcare providers to send medical information through their cell phones, other smart devices and internet networks.

Samuel BO Olatunbosun Dept. of Computer Science Norfolk State University 700 Park Avenue Norfolk, VA. 23504, USA When discussing healthcare, the Health Insurance Portability and Accountability Act of 1996 (HIPAA) must be addressed. “The Security and Privacy rules of HIPAA require covered entities to maintain appropriate administrative, technical, and physical safeguards for protecting Electronic Patient Health Information (e-PHI)” [2]. E-PHI is described as a patient physical or mental condition, financial cost of medical treatment, specifics concerning patient treatment, and the name, social security number, address and birthday of the patient [2]. All transmission of medical data must be transmitted in a secure manner. a. Traditional Medical Data Breaches As per an investigation by Symantec Corporation on Healthcare Cyber Security in 2015, Cyberattacks were the top reason for healthcare data breaches in 2015. Studies have shown that providers have changed their perspectives and raised awareness on healthcare related cybersecurity threats. For example, phishing scams and ransomware, are the expanded risk of patient safety and in healthcare delivery. In February 2015, according to the NPR report, which showed that an organized crime network had high demands to provide and sell healthcare data on the black market. b. Problem Motivation and Importance i) Problem: The healthcare sector experiences continuous cybersecurity threats. These issues can start from malware that negotiates the system’s integrity and disrupts the security and treatment of patients. IoMT devices will create greater problems due to technical deficiencies, and poor practices by patients and providers. ii) Motivation: Cyber criminals are continually searching for new approaches to gain access to sensitive data. One of the primary inspirations of cyber criminals is to make money and the healthcare industry must act aggressively to mitigate this problem. iii) Importance: Cybersecurity is most likely more critical for patients in the healthcare sector. Criminals can utilize information which is leaked to be auctioned in the black market, which, then, is later utilized for data crimes, c. Research Goals and Methodology This project examined the potential benefits of IoMT devices to the healthcare industry, the providers, and patients. The goal was to introduce IoMT, and to identify benefits that will help and threats that will hinder the healthcare sector in protecting sensitive information from cyberattacks. Literature reviews were used to understand the issues and challenges relevant to the IoMT concept in the healthcare sector. This included examining several popular IoMT devices in use today and the security risks they may pose to patient information.

ISBN: 1-60132-503-7, CSREA Press ©

116

2.

Int'l Conf. Internet Computing and Internet of Things | ICOMP'19 |

Literature Review

This section summarized the opinions and findings of others with regards to impacts and use of IoMT devices in the healthcare industry most of which are based on observations and authentic research. In a research article titled, “IoT Can Significantly Improve Healthcare in the Context of Smart City”, the long term view for IoMT is described as “Smart City” Healthcare. The aim of this concept was to increase quality, coverage of healthcare while making it more affordable [5]. “The Internet of Things (IoT) powered by Digital-5 forces (i.e. Mobility, Big Data, Cloud, Social and Robotics & Artificial Intelligence (AI)), is transforming these challenges into opportunities. As per Frost & Sullivan (2016), Smart City healthcare is anticipated to create business opportunities with a market value of $220 Billion by 2020” [5]. 2.1 Embracing IoMT Devices The IoT is an ecosystem community of different physical items, which are given with unique identifiers with software, electronics, network and sensors availability, to gather and exchange important information without human intercession. Various technologies including IoT are remote sensor systems, micro-electromechanical systems, cloud computing, future internet and semantic technology. The process of enhancing human health which includes prevention and treatment of diseases, accidents, injury, mental and physical impairments. It helps in typical mental, physical health and prosperity of individuals around the globe. This incorporates essential consideration, optional consideration and all work done by the health professional to enhance the tertiary consideration of general society. 2.2 Technical Challenges of IoMT Devices With regards to developing IoMT devices, the equipment configuration is a critical part for the achievement of the intended use and data security of the device. The device also must meet other standards such operating capacity and energy efficiency. With the expanding demand of associated devices, connected frameworks need to work with different devices and adjust to various system designs. Expanded technology developments and implementing new applications, embedded system designers confront numerous issues with adaptability and interoperability of IoMT devices. Healthcare industry security standards and practices are a technical vulnerability. As more IoMT devices are connected to the internet, the number of vulnerabilities will increase. A study conducted by the Barkley Research Group predicted each person would have seven IoT devices by 2020, fifty billion internet connected devices worldwide by 2020, and medical device revenue reaching $97 billion by 2024 [25]. The principle concepts for any security control evolve around the CIA triad; confidentiality, integrity, and availability [24]. As the CIA triad pertains to devices, the initial configuration, audit logs, quality improvement, update and patch management practiced by each developer or manufacturer can weaken the security posture of the IoMT infrastructure. It only takes one weak link in the security chain to cause a data breach or an interruption in service. Healthcare information is one of the top objectives of cyber hackers, and it has been estimated that the expense of information breaks a year ago

was $ 6.2 billion in the Healthcare Industry. The devices connected in healthcare offices and to patients, the aggregate number of IoMT devices is increasing, making new threat vectors for cybercriminals and new challenges of securing healthcare. Due to their technical knowledge ranging from novice to expert, hackers are described as the following in a 2018 Special Edition Time Magazine dedicated to Cybersecurity [26]: 1) State Sponsored - government sponsored hackers employed to breach foreign governments computer systems to obtain state secrets and intelligence information. 2) Cyber-criminal - uses keystrokes and phishing scams such as the “Nigerian Prince” to take advantage of online shoppers, health-care facilities, airports to steal data and money. 3) Vulnerability Broker - finds software exploits, such as zero-day exploits, and sells them to governments and high paying clients. Considers hacking as a legitimate business. 4) Hacktivist - targets governments, corporations, “bullies” in an attempt to fight injustice and right perceived wrongs against humanity. Anonymous is one of the most popular hacktivist. 5) Script Kiddies - considered attention seeking hackers targeting unprotected email accounts and websites. Known for defacing unprotected websites and sending emails containing attachments with viruses that replicate themselves and delete data. 2.3 IoT Medical Data Breaches Information security and protection are critical segments of routine healthcare practice since research facilities/labs are legally required to securely transmit and store electronic patient information. With expanding the availability of information frameworks, constant protection of healthcare information can be a challenging task. The Health Insurance Portability and Accountability Act (HIPAA) regulates the protection and security of medical information. Government and state regulations as well as HIPAA laws can be expensive as healthcare providers have to operate within these direct guidelines. As mentioned in section two, $8 million dollars in settlements were paid out in various data breaches that impacted more than 55, 650 patients [22]. HIPAA violations are extremely costly for healthcare offenders because lawsuits are brought against them. Depending upon the number of patients impacted by the data breach, the payout can be in the millions. Data breaches have destroyed the reputation of many healthcare providers and drove some out of business. For FileFax, one of the companies penalized, the data breach put the company out of business. FileFax was a medical records management company that left 2,150 patient medical records at a shredding and recycling facility, causing the breach [22]. The company paid $100,000 out of the receivership estate to the HHS Office for Civil Rights to settle potential HIPAA violations [22]. Due to data breaches in 2012 and 2013, the University of Texas, MD Anderson Cancer Center located in Houston was required to pay $4.3 million in civil penalties. This data breach exposed the health information of 33,500 people [22]. The largest settlement for a data breach was $16 million paid out by Anthem [23]. The cyber-attacks took place between December 2014 and January 2015. The compromised data included names, Social Security numbers, addresses, and dates of birth [22]. In January 2015,

ISBN: 1-60132-503-7, CSREA Press ©

Int'l Conf. Internet Computing and Internet of Things | ICOMP'19 |

Anthem discovered the attacks and was subsequently found negligent in having the necessary infrastructure in place to ward off a cyberattack that compromised the medical data of 79 million patients [23]. In a separate class-action lawsuit in August 2018, Anthem surpassed its largest settlement of $16 million by paying out $115 million to 19.1 million patients that were potentially impacted by the 2014-2015 data breach [23].

3. Assessment of specific IoMT Devices and Their Security Implications

As the evolution of IoMT devices continues, the capabilities of each device will greatly improve. Currently, there are three classes of IoM’s that provide constant monitoring and tracking [5]: Class A) tracking humans (e.g. patients, caregivers, and family members). Class B) tracking things (e.g. medical devices, supplies, and specimens). Class C) tracking humans and things (e.g. combination of class A and class B). 3.1 Description of Popular IoT Medical Devices The IoT Medical (IoMT) branches off of the general IoT concept the same as other industries. An article published in 2017 by Becker’s Health IT & CIO Report, described IoMT as consisting of five areas [4]. The five areas are: 1). On the body - includes wearable devices that can monitor vital signs (eg heart arrhythmias and blood pressure, and alert doctors and nurses to adverse changes in real time); implantable sensors (to monitor a patient's blood glucose levels), and ingestible sensors (to monitor medication adherence). 2). In the home includes activity monitors, virtual assistants, medication adherence tools connected to delivery systems (e.g. inhalers for chronic obstructive pulmonary disease, location tracking (e.g. in dementia), and falls notification mats. 3). Within the community - includes personal emergency response systems, automated kiosks (for blood pressure checks, vision testing). 4). In the clinic - includes portable diagnostic devices, leading to more accurate diagnoses at the point of care. 5). In the hospital facility - includes smart hospitals, where IoMT components are supporting core functions of a hospital realtime location services, patient and personnel flow tools” [17]. These areas provide a concentration point and direction for the implementation and growth of IoMT. 3.2 Wearable Devices and Ingestible Sensors Wearable sensors are primarily worn around the wrist and chest. The sensors worn around the chest are worn by athletes during practice or games. This gives the health industry something solid to measure as athletes condition themselves. Wrist sensors are used to treat a multitude of health issues. The use of these sensors are popular in the treatment of Alzheimer’s patients. Alzheimer patient wrist sensors are GPS enabled, which adds in tracking a patient the happens to wonder off [4]. If these devices are lost or stolen, the stored health information can become ‘at-risk’ for exploitation. It is highly recommended at a minimum, that all wearable devices are password protected and if available, a remote wiping capability that will erase all sensitive information when activated. Ingestible sensors are the newest of all sensors for the healthcare use. In this case, the patient wears an external patch on their body which is responsible for delivering information back to the provider. These wearable

117

and ingestible sensors provide a foundation for the continued development of IoMT. “Emerging technologies create new opportunities, and the robust monitoring of persons or things, alike, in indoor and outdoor environments, becomes of interest to many scientific and industrial applications, where one of the most important is the healthcare domain” [6]. 3.3 Pacemaker WhiteScole a research firm of security, assesses security on heart devices and home observing devices of four noteworthy makers of the healthcare field. Inside the pacemaker device, they found 8,000 shortcomings. The main reason behind such vulnerability in pacemakers and comparable devices is mostly because of the way that various merchants purchase 3rd Party for their software or equipment. Not many times, these parts have a vulnerability which is undetermined and unpublished [2]. In 2016, six versions of pacemakers developed by the Abbott healthcare firm were recalled due to hi-jacking vulnerability [19]. The vulnerable pacemakers impacted 465,000 patients and was resolved with a firmware push to each device [19]. This is an unthinkable life or death scenario that highlights the most extreme motives of cyber-criminals. 3.4 Implanted Defibrillators Apart from the pacemaker, the implanted defibrillator is additionally known for safety vulnerabilities. Used to screen the electrical action of the heart, they are essential to understand the unsafe rhythm and to give stuns. They can be checked through a radio transmitter. In the event that a digital criminal can hack into radio broadcast through a protocol of communication, for example, it involves time before it can deal with the device, where it tends to be restarted. It can be a disaster if a digital criminal is able to reset the clock of the defibrillator and keeping the device from responding to cardiovascular/arrhythmic activities [3]. 3.5 Patient Health Portals The patient portal is an online site which is connected to EHR, which focuses on access of the patient to health information. This resource provides patients quick access to lab results, provider notes, health history, and immunizations. Most portals also include appointment scheduling, direct secure messaging, prescription refill requests, and bill payment [4]. The vulnerabilities that face other websites and portals are the same for healthcare portals. These vulnerabilities include poor configuration, encryption, update and patch management, and complacency.

4.

Research Findings

4.1 IoMT Devices in Use Wearable devices had a promising introduction to the healthcare industry. For example, Parkinson’s disease is currently being treated with a product called Lee Silverman Voice Treatment (LSVT) [1]. Parkinson’s disease causes speech disorders (Dysarthria) that impact the lips, tongue, throat and vocal cords of the patients. The LSVT system provides treatment to patients by using speech exercises that include saying letters and reading sentences [1]. The patient wears a smartwatch type device called EchoWear to communicate back and forth with their providers for their treatment. Unfortunately, there has been two deaths during

ISBN: 1-60132-503-7, CSREA Press ©

118

Int'l Conf. Internet Computing and Internet of Things | ICOMP'19 |

football practice this year, one high school player and the other at the collegiate level from the University of Maryland. Jordan McNair, the 19-year-old college student died from complications related to extreme exhaustion and heatstroke [19]. Deaths like these have prompted sports at all levels to monitor hydration and exhaustion of athletes. This particular Class A sensor may not prevent death but again, it can provide valuable information for future research or athletic rule changes. With the use and information provided by the wearable and ingestible sensors, the healthcare provider can initiate the call to dispatch first responders to the patient to treat them onsite or transport them to a medical facility if required. “IoT is a breakthrough for ubiquitous health management and monitoring as it is uncompromised by distance or location. Embedded with wearable, implantable and even ambient (environmental) sensors, IoT technology enables the gathering and simultaneous analysis of multiple vital parameters. Equipped with communication and GPS capabilities, IoT facilitates storage of patient information in a secure cloud-based database which can be accessed via computer or Smartphone, by both patient and healthcare professionals” [15]. Patients suffering from Alzheimer’s diseases have been known to disappear at times and are found dead or alive at random places. This wearable sensor enables first responders and healthcare providers to quickly find the patient safe and secure. This type of devices was considered and used in cases involving suicidal patients and alcohol treatment and recovery patients [4]. The sensor provides information reflecting increases in heart rate, sweating and heavy breathing that may occur in suicidal patients. For alcoholism, the same concept is used tracking the mobility of the patient if they begin to wobble, lose their balance or their actions in general become unstable [4]. Along with wearables, ingestible sensors have shown promise for the treatment of mental health diseases and alcoholism. “In 2015, the U.S. Food & Drug Administration approved the first digital pill for general human consumption [16]”. The pill is filled with ‘Ability’, a popular drug that has proven successful for treating schizophrenia, bipolar disorder and other mental health diseases [4]. Once the pill is swallowed, it tracks when the pill hits the stomach, pushes that information to the patch and out to the healthcare provider. This is huge for the mental health area as a lot of focus has been placed recently on the proper treatment of military personnel with Post Traumatic Stress Disorder and others that suffer from anxiety or depression. “The ability of devices to gather data on their own removes the limitations of human-entered data— automatically obtaining the data doctors need, at the time and in the way they need it. The automation reduces the risk of error. Fewer errors can mean increased efficiency, lower costs and improvements in quality in just about any industry” [9]. Wearable devices monitor the vital signs and other symptoms of patients for any abnormal readings. In this situation, healthcare providers communicated with the patients to adjust the current dosage of medication to control the blood pressure or body temperature to decrease the chance of an emergency situation. For providers, this gives them the ability to treat patients earlier than later which can be the difference between

life and death. In a 2018 article written by Kayla Matthews, six examples of IoMT use cases were provided [26]: 1). Reducing emergency room visits and wait times. 2) Remote health and monitoring. 3) Ensuring the availability and accessibility of critical hardware. 4) Tracking staff, patients and equipment. 5) Enhance drug management. 6) Addressing chronic disease. 4.2 Healthcare Providers at Work EHR’s cuts back the amount of time spent by providers and their assistance providing information patients now have the ability to obtain themselves. This provides valuable time for healthcare providers to give to other patients with more critical needs. Healthcare providers can also share patient information to ensure the provider has access to the notes documented by the last provider prior to rendering any treatment to the patient. For example, if a patient gets ill on a work trip or family vacation, an EHR will provide critical information such as allergies, chronic problems and current medications to aid the provider in effectively treating the patient until they return home to their primary healthcare provider. The assisting provider can also leave detailed documentation of treatment rendered for the primary healthcare provider to review and continue the appropriate level of care. IoMT augments the capabilities of health care personnel, facilities and technologies within hospitals as well as remotely, by alleviating the need for constant physical oversight while maintaining supervision of patients [15]. The focus of healthcare providers is invested on caring and providing the best medical treatment to their patient. Healthcare providers include but are not limited to, “physicians, dentists, psychiatrist, physiotherapists, pharmacists, nursing practitioners, nursing auxiliaries, midwives, paramedical practitioners, etc. Health professionals are generally organized as self-employed professionals (except nurses and midwives)” [12]. 4.3 Potential Cybersecurity Related Risks The medical field has changed for better, technology enables individuals to live more and live more advantageous lives with fewer health complexities. Shockingly, as medical devices turn out to be further developed with the IoT being introduced, the security concerns emerge. It was not the attack of WannaCry ransomware recently that demonstrated the vulnerability of medical frameworks. At the point when Britain's medical center was influenced by WannaCry, it demonstrates what number of perilous clinics around the globe are there to trust IoT without remembering the ramifications of security. a. Seven Barriers to IoMT Devices Expected to be worth $158.1 billion dollars by 2022, here are seven barriers inhibiting IoMT adoption throughout the healthcare industry [20]: 1) Outdated funding, business and operating models that might not be equipped to address emerging technological solutions. 2) To work effectively, IoMT solutions must be interoperable with one another, along with the various players — such as providers and payers — that are using the technologies. 3) As more medical devices become connected to one another and other systems, there's an increased risk for

ISBN: 1-60132-503-7, CSREA Press ©

Int'l Conf. Internet Computing and Internet of Things | ICOMP'19 |

a data breaches or cyberattacks. 4) The regulatory landscape is still adjusting to emerging technologies, which IoMT vendors and developers must keep their eyes on. 5) A shortage of relevant digital talent among IT staff could affect an organization's ability to deploy IoMT tools. 6) Patients may not be comfortable with regular sharing of their health information, which means healthcare organizations and vendors must learn how to maintain trust in the digital age. 7) To fully realize the value of IoMT tools, healthcare organizations must scale them across their entire system to drive patient outcomes. b. Governance and Disparities Across America Government and state regulations can also be expensive as providers have to operate within these direct guidelines. Cost can be associated with meeting and maintaining technical standards. It can also be tied to implementation and operational execution of healthcare services. Proper planning, budgeting and forecasting can save providers more money in the long run. Violations of HIPAA and other regulations are extremely costly. For example, data breaches due to improper infrastructure or a flaw in policies and procedures can ruin the trust and confidence for providers. This loss of trust and confidence can be more damaging than the monetary lose suffered in the process. So far in 2018, over $8 million dollars in settlements have been paid out in various data breaches that impacted more than 55, 650 patients [22]. The cost of healthcare services is a major issue for providers just as it is for patients. The difference can be seen by cost and quality from state to state. Healthcare cost limited the available treatment in rural areas basic medical needs such as periodic check-ups, flu shots, birth control and school immunization for children. The access to specialized treatments for chronic illness was a burden for patients that required treatment only available large metropolitan areas. Patients were required to travel to other states to receive the specialized needed to potentially save their lives. Providers were not able to provide specialized treatment services in rural or low income areas because of operating cost, payroll difficulties and other expenses. b. Technical Threats Technical threats are an ongoing battle in the IoMT environment. As technology continues to evolve, the challenges have become more complicated and sophisticated. Network reliability and resource availability are “must haves” for operational success in any industry. For example, in wired network environments, troubleshooting loss of connectivity and other network concerns were fairly straight forward. Wired network environment troubleshooting can typically start with checking the power or physical coaxial, CaT5 or fiber optic cable. This approach may not work in a wireless network environment as the variables change to include electromagnetic or infrared waves. Wireless network environments are closing the gap on its counterpart but wired network environments still provide a higher rate of reliability and quality of service. Configuration management was a key element to keeping IoMT networks and devices protected from various attacks. Improperly configured devices created holes in the network infrastructure that allowed hackers of all

119

kinds access to sensitive medical data. Experienced hackers typically pursued avenues that provided access to escalated privileges used by system administrators. The network security posture of networks was compromised by basic vulnerabilities such as the use of default or weak passwords, poor patch management, inconsistent audit log review, and in cloud environments. There are many security features developed and proven to keep networks and information secure. While technologic security measures became more prevalent and more expensive, the expectation of these security measures became unrealistic. At times, excitement and the potential of certain developments caused premature implementation of devices that were not ready for the task at hand. New advances and devices typically devalued the presence and capabilities of basic security process and procedures in the minds of upper management and investors. The development of new IoMT devices also created new vulnerabilities that the industry was not prepared to take on. c. Human Threats The rapid and evolving pace of technology easily surpassed the knowledge of its providers, support staff, and users. In the IoMT environment, the typical providers and support staff may not be trained in information technology enough to drop a new software tool or device in their lap and they feel comfortable using it effectively. IoMT professionals are required to attend a basic level of mandatory security awareness training, typically annually or after a major data breach/cybersecurity incident via e-learning or attending a brief. For the user, there is a gap in their security awareness training. Where do they receive their security awareness training to learn safe, practical use of their wearable devices or medical portal access to download their medical records? If asked a question or asked for directions, most people are inclined to be as helpful as possible. This genuine willingness to be helpful can be viewed as a vulnerability in regards to IoMT and healthcare information. Mandatory security awareness training reviews social engineering, phishing, malware, explains the requirement for complex passwords, and covers the importance of standard physical security protocols. Complacency was another cause for data breaches. Human complacency became a tendency once responsibilities and required actions were considered “routine” by an individual. With exception to basic users, the insider threat can be potentially considered the most dangerous human threat. Insider threats are categorized as malicious insider threats and innocent or accidental insider threats. “The malicious insider is motivated by a number of factors, but most frequently by disgruntlement and/or financial gain” [24]. The malicious insider could range from an employee that was recently passed over for promotion or recognition, or recently reprimanded for poor job performance. Accidental insider threats fall into the helpful, kind human with sincere intentions category and the on the other side, simply became used to their job responsibilities and let their guard down. This lapse in judgement can cause grave danger in an IoMT environment. e. Knowledge Gaps in the Healthcare Industry Here are some of the interesting research findings that appeared as common themes during this study. Currently,

ISBN: 1-60132-503-7, CSREA Press ©

120

Int'l Conf. Internet Computing and Internet of Things | ICOMP'19 |

healthcare professionals have little to no knowledge of IoT, let alone IoMT. When asked, “how do you think IoMT will help healthcare providers and patients?”, all three healthcare professional (from three different medical concentrations) were absolutely clueless. The question essentially was broken down to a cellphone which was identified as an IoT device, then described how this cellphone could start their car, turn on the lights in their home, disarm their alarm system, and unlock the front door to their home. At this point, they began to understand the concept. At the end of the conversation, all three healthcare professionals had no valuable input due to their lack of knowledge of the IoT or IoMT concepts. All three professionals worked in medical concentrations that were discussed as showing promise and potential from IoMT device implementation. As IoMT developments continue to evolve, is there an injustice being done to current healthcare providers by not including them in this new medical movement? Or will their years of knowledge and experience in the healthcare field be phased out due to this potential oversight? The responses from potential IoMT users or patients were less surprising. As with the providers, the IoT and IoMT summary was given to the potential users. The surprise came when asked about healthcare data breaches. There was no major concern about loss of personal data due to the information exposed by Home Depot, Target, Equifax and other data breaches. The overall consensus was that their personal data was already in the hands of the “bad guys” and it was just a matter of time before it was used. This group were victims of identity theft or knew someone that been through the ordeal. The idea of healthcare data breaches held significantly less weight than financial breaches.

5. Recommendations

The following recommendations below are provided to support further research into improving the over results and security postures of IoMT devices and tools. 5.1 Standards and Interoperability Strategies, benchmarks and rules can be created, reported and actualized to support healthcare information security. Numerous measures about health security are distributed by global associations. These reports can be utilized to give security and classification. Every individual and organization involved in the process must have sufficient information about methods, rules and standards identified with information security and protection. They can be surveyed and updated routinely and can change per the requirements of the area of the health service. Healthcare industry security standards, government and state regulations, and HIPAA laws can create vulnerabilities for IoMT devices. Although these measures are in place to add structure to the healthcare industry, there are wide differences in legislation across the country. Currently, California, Texas, and Massachusetts are the only states that have comprehensive written state data security laws and 48 states have data breach notification laws [28]. Manufacturers typically have different practices to make their IoMT devices stand out from other IoMT devices that are designed to produce the same results for the healthcare industry. This practice can create interoperability and security issues for IoMT devices. Along the same lines, IoMT

manufacturers are limited to the expertise of their IT personnel and in the end, this creates limitations within their IoMT devices. These interoperability gaps need to be closed to help improve cybersecurity. Industry standardization can work towards some of the interoperability concerns. Standard written governance for all 50 states that include direct HIPAA requirements for IoMT devices and tools. This governance should define who, how, and when these HIPAA requirements will be inspected for compliance, and carry strict penalties for non-compliance. 5.2 Log Management All IoMT healthcare devices, IoMT healthcare applications and system segment be logged must be linked with the central log management framework. The logs are observed, broken down and assessed to prevent undesirable incidents of health frameworks. In the same way, there must be the management of a central log or examining of security information and event manages to guarantee security. So as to interfere in undesirable incidents, unwanted incidents must be reported for to the security group rapidly. Security Information or Central Log Management and Event Management must have solid authorization and authentication to screen review logs. Logs should be reviewed daily to notice any anomalies that can identify the frequency of attempted cyberattacks to disrupt it or start recovery efforts if required. 5.3 Training As more IoMT devices are connected to the internet, the number of vulnerabilities will greatly increase. IoMT devices connected unsecured networks are possibly at risk for exposure as some unsecured networks provide legitimate WiFi services others are in place to take advantage of users that connect to them with laptops or IoMT devices. Training must be set up to enhance the information security awareness of medical specialists. These trainings give insights concerning fundamental security and IoMT human services applications and demand the protection of health information. These training can be given routinely to the representatives. Security awareness training is typically carried out in two methods in medical facilities. The first method is an annual online training that is completed by each employee providing them basic information and scenarios to think through. The second method is face to face training given on an annual basis and each employee is required to attend. Research confirmed that annual online security awareness training is conducted by government healthcare facilities with little to no mention of IoMT devices. The annual requirement for both security awareness training methods can be considered ineffective. At the rate that malicious attacks and data breaches occur, quarterly security awareness training requirements of either type should be considered the minimum or each time that a malicious attack or data breaches is successfully carried out. The online security training is often the same training with the same scenarios given for 3 to 4 years at a time, with no mention of IoMT devices. The case was the same for the face to face security awareness training. Ultimately, if the healthcare employee is not engaged and receptive to the training, it can be considered ineffective. IoMT device users should be aware of malicious attacks designed to divulge

ISBN: 1-60132-503-7, CSREA Press ©

Int'l Conf. Internet Computing and Internet of Things | ICOMP'19 |

sensitive information. Phishing and spear phishing attacks are typically conducted via email and other messaging platforms to gain names, addresses, telephone numbers, and social security numbers for malicious use. As discussed in section four, there is increase in phishing attacks by attackers impersonating someone that knows their target. This method establishes a rapport faster with target causing them to let their guard down and divulge information. Cyber awareness training can be improved by implementing role playing scenarios that require critical thinking and action. 5.4 Good Cybersecurity Hygiene Connectivity of any healthcare network infrastructure requires a high level of reliability for healthcare information, services, and devices. The slightest disruption in connectivity can lead to unexpected outages, data loss, and an interruption in patient care. And for patients on vacation or business trip, connecting to Wi-Fi may be their only option to stay connected with their healthcare provider via the IoMT device. The use of IoMT devices should be accompanied with the options to connect to secure healthcare provider or home internet network service providers. Restrict the device from working on any other network resource outside of those specifically agreed upon and known to be secure. Keeping software up-to-date, patch management, performing periodic backups, and installing anti-virus software are critical to the security of IoMT devices. This is a constant process since vulnerabilities and threats are constantly emerging. All of this is accomplished by having a standard cybersecurity plan in place. This cybersecurity plan is developed with a defense in layers’ approach with no optional steps. Everything outlined within it is a requirement. Finally, implement periodic vulnerability and penetration testing to validate the overall effectiveness of the security posture to successfully mitigate cybersecurity risks.

6.

Conclusion

IoMT devices increasingly are essential for healthcare systems. IoMT tools gather huge amounts of sensitive, personal information for incorporation into these frameworks. In this way, these frameworks are vulnerable to numerous information security threats and require protection. In this paper, we have discussed various types of IoMT devices, and also types of security risk and vulnerability to those IoMT devices. IoMT devices can help to control chronic illness, improve overall healthcare cost, availability, and quality, while potentially extending human life. Keys to success can include interoperability, people over profits mentality, and greater investment of manufacturer resources in security during the development and implementation of IoMT devices. Expectation management of IoMT device capabilities, learning requirements for healthcare providers and patients need to be clearly defined. Government and state laws should be harmonized to both complement HIPPA regulations and to eliminate interstate interoperability gaps in as potential loopholes for data breaches. The potential benefits of IoMT far outweigh the potential dangers and fallouts of cyber threats.

7. References 1

A. Shahani, "The Black Market For Stolen Health Care Data : All Tech Considered : NPR," NPR, 2015.

2

3 4

5

6 7 8

9 10

11 12

13

14 15

16

17

121

IJSMI, "Overview of recent advances in Health care technology and its impact on health care delivery," International Journal of Statistics and Medical Informatics, pp. 1-6, 2018. D. J. Escher, "Types of Pacemakers and their Complications," Circulation, vol. 47, no. 5, pp. 1119-1131, 1973. G. H. Bardy, B. . Hofer, G. . Johnson, P. J. Kudenchuk, J. E. Poole, G. L. Dolack, M. J. Gleva, R. . Mitchell and D. . Kelso, "Implantable transvenous cardioverter-defibrillators.," Circulation, vol. 87, no. 4, pp. 1152-1168, 1993. K. . Terry, "Patient Portals: Beyond Meaningful Use," , . [Online]. Available: http://www.physicianspractice.com/technology/content/article /1462168/1890621. [Accessed 26 12 2018]. D. Wood, N. Apthorpe and N. Feamster, "Cleartext Data Transmissions in Consumer IoT Medical Devices," no. 1-6, 2018. A. P. Bernstein, "Anthem Hacking Points to Security Vulnerability of Health Care Industry," The New York Times, 2015. P. S. Mathew, A. S. Pillai and V. . Palade, "Applications of IoT in Healthcare," , 2018. [Online]. Available: https://link.springer.com/chapter/10.1007/978-3-319-706887_11. [Accessed 25 12 2018]. K. Chauhan, "Top 6 Hardware Design Challenges of the Embedded Internet of Things (IoT)," D Zone (IoT Zone), 2018. I. C. Cucoranu, A. V. Parwani, A. J. West, G. . RomeroLauro, K. . Nauman, A. B. Carter, U. J. Balis, M. J. Tuthill and L. . Pantanowitz, "Privacy and security of patient data in the pathology laboratory," Journal of Pathology Informatics, vol. 4, no. 1, pp. 4-4, 2013. Neil Weinberg, "Securing IoT in Healthcare is Critical," CSO, 2018. M. Haghi, K. Thurow, D. I. Habil and R. Stoll, "Wearable Devices in Medical Internet of Things: Scientific Research and Commercially Available Devices," US National Library of Medicine National Institutes of Health, pp. 4-15, 2017. T. . Tung and T. . Matsuyama, "Human Motion Tracking in Video: A Practical Approach," , 2010. [Online]. Available: http://vision.kuee.kyotou.ac.jp/japanese/happyou/pdf/tung_2009_eb_6.pdf. [Accessed 26 12 2018]. P. Mehrotra, "Biosensors and their applications – A review," Journal of Oral Biology and Craniofacial Research, 2016. H.-J. . Kim, H. . Chang, J.-J. . Suh and T. . Shon, "A Study on Device Security in IoT Convergence," , 2016. [Online]. Available: http://ieeexplore.ieee.org/document/7503989. [Accessed 27 12 2018]. K. . Daniluk and E. . Niewiadomska-Szynkiewicz, "Energyefficient security in implantable medical devices," , 2012. [Online]. Available: http://yadda.icm.edu.pl/yadda/element/bwmeta1.element.ieee000006354433. [Accessed 27 12 2018]. R. . Chow, "IoT Privacy: Can We Regain Control?," , 2015. [Online]. Available: https://doi.acm.org/10.1145/2756601.2756623. [Accessed 27 12 2018].

ISBN: 1-60132-503-7, CSREA Press ©

122

Int'l Conf. Internet Computing and Internet of Things | ICOMP'19 |

ISBN: 1-60132-503-7, CSREA Press ©

Int'l Conf. Internet Computing and Internet of Things | ICOMP'19 |

SESSION LATE BREAKING PAPERS: COGNITIVE IOT Chair(s) TBA

ISBN: 1-60132-503-7, CSREA Press ©

123

124

Int'l Conf. Internet Computing and Internet of Things | ICOMP'19 |

ISBN: 1-60132-503-7, CSREA Press ©

Int'l Conf. Internet Computing and Internet of Things | ICOMP'19 |

125

COGITO: A Cognitive Dynamic System to Allow Buildings to Learn and Adapt G. Spezzano CNR – National Council Research of Italy Institute for High Performance Computing and Networking (ICAR) Via P. Bucci 8/9c, Rende (CS), Italy Abstract— This paper presents the COGITO project focused on the integration of the Internet of Things (IoT) with dynamic cognitive systems (SDC) with the aim of improving the management of public and residential buildings with cognitive and self-developed capabilities. New custom services will be developed that will take into account building occupants, reduce costs through automation and optimization of facilities, and improve end-user satisfaction. The project will develop a Cognitive IoT (CIoT) architecture that will integrate thousands of sensors in buildings and will be able to learn about building behaviors and intuitively assist users in detecting and mitigating undesired events. Cognitive buildings will be able to analyze the use of space, monitor the comfort of the occupants and generate intelligent building management systems. To this end, a new generation of smart devices will be designed, namely Cognitive Objects, in order to increase the reliability, flexibility, and graininess of the data provided. Keywords: Cognitive IoT, Building Analytics, Fog/Mist Computing, Reinforcement Learning.

1. Introduction Modern building owners are looking for ways to improve efficiency, increase tenant comfort, and maintain the financial well being of your building. Buildings are becoming more complex, with more interconnected devices and the massive amounts of data these devices generate. Building Management Systems (BMS) do a great job at providing metering dashboards and providing alerts when problems happen, but they may not be giving you the whole story. Nowadays, buildings need of many services like light and ventilation, heating or cooling, and often access control and surveillance systems to ensure the safety of their occupants. These needs create business opportunities for service providers in order to provide systems that meet current user demands, but that cannot evolve and intelligently adapt to future needs. These static solutions are often referred to as building automation systems (BAS). Such systems spread rapidly, but their high cost has limited their use, especially in small and medium-sized buildings. The introduction of new technologies such as the Internet of Things (IoT), enables the so-called Smart Building Automation Systems

(SBAS) in the commercial and public buildings, including shops, factories, and corporate offices, schools, hospitals, and private residences. SBAS solutions are devoted to save energy, earn productivity, reduce costs, and change market dynamics for building management. The SBAS typically is a reactive process where actions can be triggered by stimuli without user intervention. These accomplished tasks are automatic, but not cognitive, i.e., they are not yet able to: (i) take into account user behavior, (ii) allow the building to learn and adapt to changes so as to improve tasks and optimize the offered services [1]. Cognitive buildings systems (CBS) [2] differ from the traditional ones due to the ability concerning build systems that are able to learn, reason and cooperate, and to take decisions in real-time. The COGITO project was conceived to go beyond the concept of SBAS, overcoming the traditional model of building management based on "a posteriori" analysis and actions, or with the analysis of historical data linked for example to the management of resources such as the energy use, water consumption, space occupancy rate, and then take action. COGITO is novel because it uses the new fog/mist computing paradigm to create a cognitive embedded IoT system [5]. In this approach, cognitive objects [8] (sensors, actuators, controllers, cameras, etc.) can communicate with each other and with the Cloud. Moreover, they have processors on board to perform locally cognitive algorithms according to an analytics approach on the edge. All of this permits to locally analyze data, extract information that can be translated immediately into actions or recommendations, or can be passed to a higher conceptual level that offers a computational layer where a decision is made by evaluating global information. With the use of cognitive objects, it is no longer necessary to use rule-based algorithms, but all the complexity of the management of the information is transferred to cognitive algorithms that will process information according to a data-driven approach. In such a case, users are only engaged in defining the characteristics of the raw data that they think may have relevant information, then, the functionalities and the actions to be taken are automatically learned. Automated learning systems, based on a data-driven approach, are excellent for analyzing large data like those produced by a large number of sensors. The use of the

ISBN: 1-60132-503-7, CSREA Press ©

126

Int'l Conf. Internet Computing and Internet of Things | ICOMP'19 |

cognitive IoT introduces a revolution in building automation systems, where decentralized architecture allows data analysis to be made on the edges of the network. This approach provides real-time intelligence and responsiveness in control operations. At the same time, it is possible to reduce bandwidth utilization as heavy communications are avoided. Current computing technology actually provides low-cost and high energy-save embedded processors that can be used for data analysis within the sensor itself. With this approach, data sent to the network can simply be the final result of a carried out analysis, which will have a smaller volume and will allow shorter response times. The expected results of the project will help in developing innovative market-based cognitive services that can ensure building management through a unifying platform by applying data-driven technology to the building automation ecosystem. Proposed solutions will drive towards the future of building automation systems. A key feature is the possibility of implicitly taking into account occupants needs and preferences through a "cognitive object model" which is able also to exploit the advantages related to the use of IoT devices equipped with computational capabilities allowing to support powerful and effective on-the-fly analysis. Newmarket services will make the buildings able to analyze space utilization, improving occupants comfort, and automatically portray intelligent management of the whole system. Cognitive Objects are complex objects that will benefit from innovative hardware technologies and advanced cognitive software algorithms. New embedded devices and/or dedicated processors (for example, Intel Movidius, NVIDIA Jetson Nano, Raspberry PI 3) will be evaluated to support the execution of cognitive algorithms. The use of such devices will allow the exploitation of a "local" computational power that will permit the processing of data close to data sources in order to create a low-latency fog/mist computing environment. The buildings, by incorporating cognitive technologies, will increase the value of the offered services by making them more effective, cheaper, safer, faster and more distinctive. The new 5G communication technology which will be tested for the realization of a demonstrator in the city of Matera, will bring significant performance improvements to support cognitive services in buildings.

2. Project contribution and activities In the introduced context, the aim of the project is the development of personalized cognitive services with "selfdeveloped" capabilities to make buildings adaptive and autonomous. In order to develop such services, we will proceed by: •

designing and developing self-contained cognitive objects with distributed control which rely on a multiagent based approach. In particular, we consider "situated agents", which are typically used to model and

control autonomous physical objects that are located in a real environment; • defining and implementing a control architecture based on machine (deep) learning and/or bio-inspired algorithms that will be used to control the defined cognitive objects. By using the most recent technologies on machine learning and artificial intelligence, a cognitive object will be able, for instance, to provide dynamically responses to user’s questions and to automatically create input-output relationships on gathered/observed data; • making the cognitive objects self-adaptive through the use of methodologies belonging to the field of autonomous systems which will be combined with various machine learning techniques and evolutionary algorithms; • proposing an open "cognitive framework" [3], which constitutes an intelligent supervisory and control infrastructure for all the facilities and objects located in a building. The framework will be capable of enhancing the environment so as to improve the quality of life of its residents by offering cognitive services related to safety, comfort, energy efficiency and assistance to elderly people. Within the COGITO project, the following issues will be considered: • Predictive maintenance. Data transmitted from connected assets such as boilers, pumps, chillers, and elevators, is analyzed and enriched to identify anomalies such as equipment operating outside of normal parameters. Predictive analytics for Remaining Useful Life (RUL) evaluates the length of time a machine is likely to operate before it requires repair or replacement. By taking RUL into account, engineers can schedule maintenance, optimize operating efficiency, and avoid unplanned downtime. For this reason, estimating RUL is a top priority in predictive maintenance programs. • Comfort and wellness. Through the deployment of IoT sensors in the building that can monitor temperature, humidity, people flow and CO2 concentration, it is possible to detect the presence and concentration of carbon dioxide in the air allows us to assess its quality and set up proper ventilation (or check its operation). By registering the daily occupant habits, the energetic waste and the incorrect behavior of the occupants will be detected, identifying the alternative actions to attain a better and virtuous behavior. Real-time monitoring and visualization, enriched with analytics to determine the next best action and trade-off analysis, triggers an intervention, either human or machine. • Smart Lighting Systems. Light sensors maximize the contribution of natural light within buildings by dimming artificial lighting when it is sunny and increasing artificial lighting when natural light levels fall. Au-

ISBN: 1-60132-503-7, CSREA Press ©

Int'l Conf. Internet Computing and Internet of Things | ICOMP'19 |

•

tomatic blinds use these sensors to curb overheating or excess light and determine when to deactivate for maximizing light and free warming. Algorithms able to take decisions without the human intervention will be developed, deciding to control shading devices in the function of weather conditions, solar irradiation, natural lighting level in indoor environment and the economical expense consequent to the artificial lighting plant, the last activated to contrast the dark conditions due to the operation of shading devices. This control system will allow us to reduce energy consumption and improve internal comfort. Safety and security efficiency. There is a clear growing trend in the need to secure access control, in order to authenticate authorised persons, detect any irregular intrusion and alert an emergency centre if needed, but also in a monitoring service warning if something abnormal happens such as a fire. We focus on the problem of evacuation of smart buildings in case of emergencies.

3. Cognitive objects and MAS Existing projects towards Smart Buildings use mostly a multi-agent system (MAS) methodology [3] In multiagent systems, agents can additionally communicate and coordinate with each other as well as with their environment. An agent is anything that can be viewed as perceiving its environment through sensors and acting upon that environment through effectors. The agents are autonomous software entities representing virtual things (devices). MAS provides a valuable framework for intelligent control systems to learn building and occupancy trends. In our cognitive framework, the agent also contains an intelligent and reasoning unit. The responsibility of this particular unit is to implement a plan on how to use these capabilities in order to achieve the given goals. Existing systems use mostly a pattern mining technique to determine what can be the next action. An example of this is the MavHome project by Cook et al. [6], a home that acts as a rational agent that aims to maximize the comfort and productivity of its inhabitants and minimize operating costs. In order to achieve these goals, the house must be able to predict, reason about, and adapt to its inhabitants. COGITO uses a machine learning approach to make decisions based on cognitive controllers. We use an event processing tool to handle the events from the home automation devices, prediction algorithms to predict the next action and reinforcement learning to decide which actions are suitable to be automated. Reinforcement Learning (RL) algorithms [7] are machine learning algorithms for decision making under uncertainty in sequential decision problems. The problems solved by RL are modeled among others through a Markov Decision Process (MDP). MDP is defined as a 4-tuple (S, A, R, T). It

127

defines a set of states S, a set of actions A, a reward function R, and a state-transition function T. In RL, an agent is immersed in an unknown environment. The agent is then asked to learn how to behave optimally (taking optimal actions) via a trial-and-error process. Figure 1 illustrates the high-level conceptual structure of a RL system.

Fig. 1: Agent-environment interaction in reinforcement learning. The learning process is as follows: • the agent is asked to select an action in a given environment state; • the selected action is executed and the environment rewards the agent for taking this action using a scalar value obtained via the reward function; • the environment performs a state transition using the state- transition function leading to a new environment state and a new learning step. The goal of the agent is to maximize the reward it gets from the environment by learning which action leads to the optimal reward. The policy followed by the RL agent that drives the selection of the next action is nothing more than a function that selects an action in a given environment state. Reinforcement Learning aims to imitate the learning process of humans. Through the reinforcement learning method, an agent can sense the environment through several sensor inputs. The agent uses these raw inputs to generalize the experience of the system for confronting new and unknown situations. The combination of reinforcement learning and deep neural networks - known as Deep Reinforcement Learning - has resolved several limitations of reinforcement learning including the limitation in the diversity of application domains, the need for manual engineering features, and their poor scalability for high-dimensional state-space domains. Cognitive buildings are artificially intelligent systems that need to adapt themselves based on user actions and surroundings. These systems need to carefully analyze the user needs and the conditions of the surroundings in order to predict future actions and also minimizes user interaction. The integration of cognitive analytics, sensors, and existing building systems can significantly improve occupant experience. Cognitive objects are managed by cognitive

ISBN: 1-60132-503-7, CSREA Press ©

128

Int'l Conf. Internet Computing and Internet of Things | ICOMP'19 |

agents that make decisions based on controllers driven by a reinforcement learning algorithm. Our cognitive agent must execute three key activities in sequence namely: (i) collect data from the thing; (ii) make decisions; and (iii) take actions. The task of data collection focuses on processing information coming from devices, such as reading data from input sensors. The collected data are used to set the inputs of the agents’ controller. Then, the controller processes a decision to be taken by the agent. Cognitive agents act based on the controller output. An action (effector activity) can be to interact with other agents, to send messages, or to set actuator devices, thus making changes to the environment.

terminal state as no more appliances are left to be scheduled. Each appliance has a specific kW unit requirement which is an indicative of how fast it uses up the supplied electrical energy. Reward counts the power energy consumption.

Fig. 2: MDP graph structure with two appliances.

4. Use case Cognitive IoT enables home appliances to be smart. The appliance connected to the Internet can be associated with a cognition behavior that allows the device to learn user requirements, activity patterns, and daily schedule. The autolearning makes home appliances smart, whether its a television, washing machine, dryer, coffee machine, refrigerator or security system. In this direction, commercial companies are updating their devices to make them smarter. Solving problems of high energy consuming by appliances of the smart building during peak hours is very complicated [4]. The optimization problem can be posed as a reinforcement learning task if it satisfies the Markov property. The Markov decision process, better known as MDP, is an approach in reinforcement learning to take decisions in the environment. The MDP tries to capture a world in the form of a grid or a graph by dividing it into states, actions, models/transition models, and rewards. The solution to an MDP is called a policy and the objective is to find the optimal policy for that MDP task. Thus, any reinforcement learning task composed of a set of states, actions, and rewards that follows the Markov property would be considered an MDP. Binary decision variables are required to plan the operation of household appliances. The binary variable indicates whether specific energy is being processed or not. In figure 2, we propose a graph that represents the environment dynamics when a request is placed to activate one appliance at a time. The structure of this graph enables us to define the state-to-state transition and reward functions for all stateaction pairs of the MDP. Starting from the initial state, the possible actions available are to activate one of the appliances as S = {0, 1} or S = {1, 0}, or activate both the appliances as S = {1, 1}, or delay the activation of both the appliances as S = {0, 0} by the maximum allowable delay time β as specified by the user at the time of request. If at time t, both the appliances are scheduled to turn on, then the terminal state is reached. If only one of the appliances is yet to be scheduled, then the possible actions available are to activate the appliance as At+1 = 1, or to delay the activation of the appliance as At = 0 by the maximum allowable delay time β n -1. At+1 = 1 is the

5. The Consortium The partners of the project are: CNR-ICAR (DIITET department), University of Calabria (DIMEG and DIMES departments), University of Reggio Calabria (DIIES department), University of Basilicata (DICEM department), ENEA, TIM s.p.a., ATERP Cosenza, CUEIM, DIGIMAT, SCAI-LAB, DS-TECH s.r.l., SMARTS s.r.l..

Acknowledgment This work has been partially supported by COGITO project funded by the Italian Government, (CUP:H56C180001000051), and by GLAMOUR (Green Learning and Adaptive Multi-interface iOt enabled devices throUgh social inteRactions), funded by POR Calabria FESR-FSE, Italy (CUP:J28C17000250006)

References [1] J. Ploennigs, A. Ba and M. Barry, "Materializing the Promises of Cognitive IoT: How Cognitive Buildings Are Shaping the Way," in IEEE Internet of Things Journal, vol. 5, no. 4, pp. 2367-2374, Aug. 2018. doi: 10.1109/JIOT.2017.2755376 [2] F. Cicirelli, A. Guerrieri, A. Mercuri, G. Spezzano, A. Vinci: ITEMa: A methodological approach for cognitive edge computing IoT ecosystems. Future Generation Comp. Syst. 92: 189-197, 2019. [3] F. Cicirelli, A. Guerrieri, A. Mercuri, G. Spezzano, A. Vinci: Cognitive smart environment: an approach based on concept hierarchies and sensor data fusion. IJSPM 13(5): 506-519, 2018. [4] Chaouch, Haithem & Ben Hadj Slama, J. (2015). Modeling and simulation of appliances scheduling in the smart home for managing energy. 2014 International Conference on Electrical Sciences and Technologies in Maghreb, CISTEM 2014. 10.1109/CISTEM.2014.7077028. [5] A. Afzal et al. "The cognitive Internet of Things: A unified perspective" Mobile Netw. Appl. vol. 20 no. 1 pp. 72-85, 2015. [6] D. J. Cook et al., "MavHome: an agent-based smart home," Proceedings of the First IEEE International Conference on Pervasive Computing and Communications, 2003. (PerCom 2003)., Fort Worth, TX, pp. 521-524, 2003. doi: 10.1109/PERCOM.2003.1192783 [7] Kai Arulkumaran, Marc Peter Deisenroth, Miles Brundage, Anil Anthony Bharath: Deep Reinforcement Learning: A Brief Survey. IEEE Signal Process. Mag. 34(6): 26-38, 2017. [8] Maryam Ashoori, Rachel K. E. Bellamy, Justin D. Weisz: The Impending Ubiquity of Cognitive Objects. AAAI Workshop: Symbiotic Cognitive Systems, 2016.

ISBN: 1-60132-503-7, CSREA Press ©

Author Index Ahmed, Ahmed Abdelmoamen - 108 Alkis, Nurcan - 16 Arai, Kenichi - 95 Caldag, Murat Tahir - 16 Cao, Xiang - 68 Deligiannidis, Leonidas - 32 Deugo, Dwight - 38 Dobashi, Kenichi - 81 Dorai, Gokila - 52 Droegehorn, Olaf - 3 Egert, Christopher A. - 59 Gokalp, Ebru - 16 Grady, Adam L. - 64 Hong, Min - 102 Horowitz, Brandon - 32 Hsu, Yu-Fang - 10 Ichijo, Kenji - 25 Imai, Tetsuo - 95 Ishikuro, Michi - 25 Kim, Sejun - 88 Kobayashi, Toru - 95 Kondo, Hideki - 81 Lee, Ahyoung - 64 , 102 Lee, Taeho - 88 Li, Cheng - 68 Lin, Wen Piao - 10 Mabadie, Patrick - 45 McFarland, Rashad J. - 115 Min, Se Dong - 102 Moon, JunYoung - 102 Nakashima, Ryota - 95 Narita, Akiko - 25 Odhiambo, Marcel Ohanga - 45 Olatunbosun, Samuel Bo - 115 Phelps, Andrew - 59 Sarmento, Henrique R. - 3 Solo, Ashu M. G. - 75 Spezzano, Giandomenico - 125 Sung, Nak-Jun - 102 Takami, Kazumasa - 81 Ullah, Ihsan - 88 Vullo, Ronald P. - 59 Youn, Hee Yong - 88 Yu, Chen-Hsiang - 32