Icef 2023 PDF
Uploaded by
tima.darbekIcef 2023 PDF
Uploaded by
tima.darbekICEF
2023
International Conference on
Electrical Facilities and
informational technologies 2023
August 22(TUE) - 25(FRI), 2023
Almaty University of Power Engineering and
Telecommunications, Kazakhstan
▶ Information
▶ Program at a Glance
▶ Contents
▶ Papers
KIEE Electrical
Organized by : Facilities Society
Co-Organized by :
Welcome Message
We warmly welcome you to International Conference on Electrical
Facilities and information technologies 2023
Dear Attendees,
On behalf of the ICEF2023 Organizing Committee, it is my pleasure to
welcome you to our signature event for this year, the 2nd International
Conference on Electrical Facilities and Information Technologies 2023
(ICEF2023) to be held in historic Almaty, Kazakhstan from August 22nd to 25th, 2023.
Under the theme “New Intelligence Technology: Past, Present and Future” this
hybrid conference aims to bring together scholars, practitioners, educators, and students
to exchange research ideas, outcomes, and experiences related to innovative electrical fa-
cilities and information technologies.
We have prepared an impressive array of keynote speakers, with ample time set aside
for colleagues to discuss amongst themselves in intimate settings on key issues, as well as
informal “icebreaker sessions” to build on our international students’ network for future
collaborative study. Furthermore, we have also included for you this year within our con-
ference benefits package, a one-day tutorial course on state-of-the-art Smart
Railway System and Logistics.
A very attractive and stimulating social program has also been prepared for you, which
will start with a Welcome Reception and Registration Party on Tuesday evening, August
22st. On Thursday evening, August 24th, the Conference Gala Dinner will take place at
Mountain Resort «Oi-Qaragai», the largest yurt in the world.
Many thanks to the members of KIEE Electrical Facilities Society, the ICEF2023
Organizing Committee members, and sponsoring organizations for providing their gen-
erous financial support. Particular thanks to Almaty University of Power Engineering and
Telecommunications, Kazakhstan for the local support and their interest and commitment
in hosting this event.
I wish the ICEF2023 Conference to be an unforgettable occasion for you and
look forward to meeting you in person and online in the days ahead.
Warmly,
Prof. Dr. SangYule Choi, Ph.D.
Conference Chair
Professor of Mechatronics
Induk University, Seoul, Korea
-2-
Committee
❚CONFERENCE HONORARY CHAIRMAN
Murat Syzdykov (Rector, AUPET, Republic of Kazakhstan)
Saltanat Amirgaliyeva (Rector, Academy of Logistcs and Transport, Kazakhstan)
Seung-Joe Yoon (President, Korea National University of Transportation, Korea)
❚CONFERENCE CHAIR
Sangyule Choi (Induk University, Korea)
❚ORGANIZING CHAIR
Jeong Won Kang (Korea National University of Transportation, Korea)
Minjae Lee (Gwangju Institute of Science and Technology, Korea)
Hyunchul Kim (Korea Railroad Research Institute, Korea)
❚PROGRAM CHAIR
Juhee Choi (Sangmyung university, Korea)
Jong-In Song (Gwangju Institute of Science and Technology, Korea)
Hag-Wone Kim (Korea National University of Transportation, Korea)
Ainur Begimbetova (AUPET, Republic of Kazakhstan)
❚TECHNICAL PROGRAM CHAIR
Moongu Jeon (Gwangju Institute of Science and Technology, Korea)
Feel-Soon Kang (Gyeongsang National University, Korea)
HONGJE RYOO (Chung Ang University, Korea)
Ishwar Sethi (Oakland University, USA)
Asifullah Khan (Pakistan Institute of Engineering and Applied Sciences (PIEAS), Pakistan)
Dragan Pamucar (University of Belgrade, Serbia)
Ashish Khare (University of Allahabad, Prayagraj, Uttar Pradesh, India)
Harish Garg (Thapar Institute of Engineering & Technology, Deemed University, India)
Sang Woong Lee (Gachon University, Korea)
US Tiwary (IIIT Allahabad, India)
Benlian Xu (Changshu Institute of Technology, China)
Alok Kumar (Singh Kushwaha, Guru Ghasidas University, India)
-3-
Manish Khare
(Dhirubhai Ambani Institute of Information and Communication Technology, India)
Jin Hyuk Park (Korea Railroad Research Institute, Korea)
Gunbok Lee (Korea Railroad Research Institute, Korea)
Yernar Amitov (AUPET, Republic of Kazakhstan)
Maxim Korobkov (AUPET, Republic of Kazakhstan)
Assel Mukasheva (AUPET, Republic of Kazakhstan)
Gani Balbayev (Academy of Logistcs and Transport, Kazakhstan)
Kuanysh Alipbayev (AUPET, Republic of Kazakhstan)
Joonho Ahn (Seoul National Univ. SEPRI, Korea)
Seungkwon Shin (Korea Railroad Research Institute, Korea)
❚INTERNATIONAL TECHNICAL AND SCIENTIFIC PROGRAM CHAIR
Zhanna Suimenbayeva (AUPET, Republic of Kazakhstan)
Sevinc Gulsecen (Istanbul University, Turkey)
Nguyen Thanh Binh (Ho Chi Minh City University of Technology, VNU-HCM, Viet Nam)
❚FINANCIAL CHAIR
Jongnam Kim (YSFS Co.,Ltd, Korea)
Yongbae Im (Korea Electrotechnology Research Institute, Korea)
Gulziya Rakhmetova (AUPET, Republic of Kazakhstan)
❚AWARD CHAIRS
Jeongju Kim (Hoseo University, Korea)
Zhubanysh Abdimuratov (AUPET, Republic of Kazakhstan)
❚PUBLICATION CHAIRS
Jeonghwan Gwak (Korea National University of Transportation, Korea)
Yongho Yoon (Gwangju University, Korea)
Hoseong Jeong (Korea Railroad Research Institute, Korea)
Vyacheslav Stoyak (AUPET, Republic of Kazakhstan)
Algazy Zhauyt (AUPET, Republic of Kazakhstan)
❚LOCAL CHAIR
Zhadyra Sagyndykova (AUPET, Republic of Kazakhstan)
-4-
Conference Topics
Conference Topics Paper Topics Include, But Are Not Limited To:*
❚Session 1 : Future Electric Facility Planning and Operations
• Power Distribution Network Modeling, Analysis and Control
• HVDC, MVDC & LVDC: Design, Control and Protection
• AC-DC Hybrid Grid Operation, Control and Protection
• IOT(Internet Of Thing)
• Electrical Safety : Design, Control and Protection
• Electric Machinery and Drives
• Energy Storage Systems, Renewable Energy
• Heat Power Engineering
❚Session 2 : Smart Electric Facility
• Smart Grids, Micro-grid, and Utility Applications
• Smart Energy Management for Home, Building and Factory
• Smart Materials and Devices, Media Facade and Large Format Display
• Illuminating Design and Light Installation
• Electrical Installation Design and Construction Technology
• Technical Specification for Electrical Installation
• Converge Technique for Electric Facility
• Electrical Power Utilities
❚Session 3 : Information Technology
• Software System & Engineering
• Artificial Intelligence
• Wireless Communications and Networking
• Embedded System, IoT/Edge Computing
• VR/AR, Metaverse
• Data Science, Big Data, Database, Data Mining
• 3D Printing, Digital Twin
• Cybersecurity
-5-
❚Session 4 : Intelligent Transportation Technology
• Railway System & Technology
• Intelligent Transportation System
• Future Smart Automobile Technology
• Road Operation Technology
❚Session 5 : Aerospace Sciences and Aviation Technology
• Aerospace Systems and Technologies
• Aerospace Guidance, Navigation, and Control
• Aerospace Cybersecurity
• Aerospace Avionics Systems
• Remote Sensing
• Unmanned Aerial Vehicles
-6-
Venue Information
❚Almaty University of Power Engineering and Telecommunications,
Kazakhstan
• Almaty University of Energy and Communications 126/1 Baitursynuly street, Almaty
050013, Kazakhstan
▪ MAP
▪ Address
Almaty University of Energy and Communications 126/1 Baitursynuly street, Almaty
050013, Kazakhstan
▪ CONTACT US
Contact numbers: (727)323-11-75 (int.6724)
E-mail: [email protected]
-7-
Program at a Glance
1st Day August 22 (TUE)
PROGRAM Place: AUPET
Time PROGRAM Room (D building. Assembly Hall)
13:00~ Registration
◾Keynote Session (1)
Infrastructure Solutions and Experience of Batys Transit JSC: Projects for the
14:00~14:40
Construction of Engineering Systems and Power Supply
Askar Iskakov (JSC "Batys Transit")
◾Keynote Session (2)
14:40~15:20 Digital strategies in times of uncertainty
Miras Kassymov (Axellect Kazakhstan and Central Asia LLP)
15:20~15:50 Coffee Break
PM ◾Keynote Session (3)
Human capital is the main value of energy companies
15:50~16:30
Talgat Temirkhanov (Chairman of the Association of Legal Entities "Kazakhstan
Electric Power Association", Kazakhstan)
◾Keynote Session (4)
Space Science and Technology: Challenges of Our Time
16:30~17:10
Marat Nurguzhin (Chairman of the Board JSC "National Center for Space
Research and Technology", Kazakhstan)
◾WORKSHOP (1)
17:10~19:00
SMART RAILWAY SYSTEM WORKSHOP
18:00~20:00 Welcome Reception (AUPET)
-8-
Program at a Glance
2nd Day August 23 (WED)
PROGRAM Place: AUPET
Time Room (D building. Assembly Hall)
08:30~ Registration
◾Keynote Session (5)
09:00~09:40 Electric Vehicle Parking Lots for the Sustainable Energy Supply of Smart Cities
Aydogan Ozdemir (Professor of Istanbul Technical University, Turkey)
◾Keynote Session (6)
Secure and Scalable IoT: An IoT Network Platform Based on Network Overlay and
09:40~10:20
MAC Security
Junwon Lee (Ph.D. of Samsung SDS, Korea)
AM ◾Keynote Session (7)
Advancing Anomaly Detection Systems: Application of Normalizing Flows in
10:20~11:00
Image and Time-Series Data
Jeonghwan Gwak (Professor of Korea National University of Transportation, Korea)
11:00~11:30 Coffee Break
◾Opening Ceremony
11:30~12:30 - Opening Speech
- Welcome Speech
12:30~14:00 Lunch (AUPET)
PROGRAM Place: AUPET (D building.Assembly Hall)
Room 1 Room 2 Room 3 Room 4 Room 5
Time
(A428) (A432) (A434) (A435) (A437)
14:00~15:40 Oral Session (1) Oral Session (2) Oral Session (3) Oral Session (4) Oral Session (5)
15:40~16:20 Coffee Break
PM 16:20~18:00 Oral Session (6) Oral Session (7) Oral Session (8) Oral Session (9) Oral Session (10)
16:00~17:00 ◾Kazakhstan- Korean student AUPET Tour [A building, room 213]
17:00~18:00 ◾Kazakhstan- Korean student workshop [A building, room 213]
-9-
Program at a Glance
3rd Day August 24 (THU)
PROGRAM Place: AUPET
Time Room (D building. Assembly Hall)
09:00~ Registration
◾Keynote Session (8)
09:00~09:40 Preventive Maintenance Scheduling & Its Actions
HARISH, GARG (Professor of Thapar Institute of Engineering & Technology, Patiala, India)
◾Keynote Session (9)
09:40~10:20 MaxViT-Unet: Multi-Axis Attention for Medical Image Segmentation
Asifullah Khan (Professor of PIEAS, Islamabad, Pakistan)
10:20~10:40 Coffee Break
AM ◾Keynote Session (10)
10:40~11:20 Introduction of Korea Railway System - Past, Present, and the Future
Hag-Wone Kim (Professor of Korea National University of Transportation, Korea)
◾Keynote Session (11)
An Intelligent Surveillance System for Human Behavior Analysis
11:20~12:00
Om Prakash (Professor of HNB Garhwal University, Srinagar Garhwal, Uttarakhand,
India)
12:00~12:40 ◾Poster Session 1 [Building A 4th Floor]
12:40~14:00 Lunch (AUPET)
PROGRAM Place: AUPET (D building.Assembly Hall)
Time Room 1 (A428) Room 2 (A432) Room 3 (A434) Room 4 (A435) Room 5 (A437)
Oral Session Oral Session Oral Session Oral Session
14:00~15:40
(11) (12) (13) (14)
PM
14:30~15:30 ◾Poster Session 2 [Building A 4th Floor]
18:30~20:30 Banquet
- 10 -
Program at a Glance
4th Day August 25 (FRI)
PROGRAM Place: AUPET
Time Room (D building. Assembly Hall)
◾Special Session (1): ◾Korea-Kazakhstan Industrial
09:00~10:00
Electric Facility Workshop Exchange Technical Forum (1)
◾Special Session (2): Intelligent ◾Korea-Kazakhstan Industrial
AM 10:00~11:00
Railway & Transportation Workshop Exchange Technical Forum (2)
◾Special Session (3): Intelligent ◾Korea-Kazakhstan Industrial
11:00~12:00
Science & Open Source Workshop Exchange Technical Forum (3)
PM 10:00~14:00 Industrial Exhibition
PROGRAM Place: ALT
Time Academy of Logisitics and Transport
AM 09:00~13:00 Kazakhstan-Korea Global Smart Railway Forum- II
- 11 -
Social Programs
Welcome Reception
August 22 (TUE) 18:00-20:00
Bakhshasaray event venue (Buy Ticket)
On the first day of the conference, please join to know each other!
Opening Ceremony
August 23 (WED) 11:30-12:30
Room (D building. Assembly Hall)
❚Opening Speech
- Prof. SangYule Choi (ICEF2023 Chairman, Korea)
❚Welcome Speech
- Rector Murat Syzdykov (AUPET, Republic of Kazakhstan)
- Rector Saltanat Amirgaliyeva (Academy of Logistic and Transport, Republic of
Kazakhstan)
- Consul General Naecheon Park (Consulate General of the Republic of Korea in
Almaty)
- Dr. Jun Hoon Park (Honorable President of Korea National University of
Transportation, Korea)
- Aidyn Ayimbetov (Kazakhstani astronaut)
❚Scholarship Award Ceremony
The ICEE 2023 Opening Ceremony is coming up on Thursday. You all are cordially in-
vited to the grand opening ceremony. We are looking forward to your gracious presence
on this occasion.
- 12 -
Keynote session
August 10 (WED) ~ August 12 (FRI)
Conference Hall
◾Keynote Session (1)
14:00~ Infrastructure Solutions and Experience of Batys Transit JSC: Projects for the
14:40 Construction of Engineering Systems and Power Supply
Askar Iskakov (JSC "Batys Transit")
◾Keynote Session (2)
14:40~
Digital strategies in times of uncertainty
15:20
Miras Kassymov (Axellect Kazakhstan and Central Asia LLP)
August 22
◾Keynote Session (3)
(TUE)
15:50~ Human capital is the main value of energy companies
16:30 Talgat Temirkhanov (Chairman of the Association of Legal Entities "Kazakhstan
Electric Power Association", Kazakhstan)
◾Keynote Session (4)
16:30~ Space Science and Technology: Challenges of Our Time
17:10 Marat Nurguzhin (Chairman of the Board JSC "National Center for Space
Research and Technology", Kazakhstan)
◾Keynote Session (5)
09:00~
Electric Vehicle Parking Lots for the Sustainable Energy Supply of Smart Cities
09:40 Aydogan Ozdemir (Professor of Istanbul Technical University, Turkey)
◾Keynote Session (6)
09:40~ Secure and Scalable IoT: An IoT Network Platform Based on Network Overlay
August 23
10:20 and MAC Security
(WED)
Junwon Lee (Ph.D. of Samsung SDS, Korea)
◾Keynote Session (7)
10:20~ Advancing Anomaly Detection Systems: Application of Normalizing Flows in
11:00 Image and Time-Series Data
Jeonghwan Gwak (Professor of Korea National University of Transportation, Korea)
◾Keynote Session (8)
09:00~ Preventive Maintenance Scheduling & Its Actions
09:40 HARISH, GARG (Professor of Thapar Institute of Engineering & Technology, Patiala,
India)
◾Keynote Session (9)
09:40~
MaxViT-Unet: Multi-Axis Attention for Medical Image Segmentation
10:20
August 24 Asifullah Khan (Professor of PIEAS, Islamabad, Pakistan)
(THU) ◾Keynote Session (10)
10:40~
Introduction of Korea Railway System - Past, Present, and the Future
11:20 Hag-Wone Kim (Professor of Korea National University of Transportation, Korea)
◾Keynote Session (11)
11:20~ An Intelligent Surveillance System for Human Behavior Analysis
12:00 Om Prakash (Professor of HNB Garhwal University, Srinagar Garhwal,
Uttarakhand, India)
- 13 -
Special session
August 22 (TUE) ~ August 25 (FRI)
Room (D building. Assembly Hall), Room 5, A437, ALT
August 22 17:10~ D building. Assembly
◾SMART RAILWAY SYSTEM WORKSHOP
(TUE) 19:00 Hall
16:00~ D building. Assembly
◾Kazakhstan-Korean High Power Lab Workshop
August 23 17:00 Hall
(WED) 17:00~ D building. Assembly
◾Kazakhstan-Korean Student Workshop
18:00 Hall
August 24 16:00~ Room 5
◾Kazakhstan-Korea Global Smart Railway Forum- I
(THU) 18:00 A437
August 25 09:00~ Academy of Logisitics
◾Kazakhstan-Korea Global Smart Railway Forum- II
(FRI) 13:00 and Transport
Oral Session
August 23 (WED) ~ August 24 (THU)
Room 1 A428, Room 2 A432, Room 3 A434, Room 4 A435, Room 5 A437
14:00~ ◾Oral Session 1 Room 1
15:40 Session 3: Information Technology A428
14:00~ ◾Oral Session 2 Room 2
15:40 Session 1: Future Electric Facility Planning and Operations A432
14:00~ ◾Oral Session 3 Room 3
15:40 Session 3: Information Technology A434
August 23 14:00~ ◾Oral Session 4 Room 4
(WED) 15:40 Session 3: Information Technology A435
14:00~ ◾Oral Session 5 Room 5
15:40 Session 3: Information Technology A437
16:20~ ◾Oral Session 6 Room 1
18:00 Session 3: Information Technology A428
16:20~ ◾Oral Session 7 Room 2
18:00 Session 1: Future Electric Facility Planning and Operations A432
- 14 -
16:20~ ◾Oral Session 8 Room 3
18:00 Session 3: Information Technology A434
August 23 16:20~ ◾Oral Session 9 Room 4
(WED) 18:00 Session 3: Information Technology A435
16:20~ ◾Oral Session 10 Room 5
18:00 Session 3: Information Technology A437
14:00~ ◾Oral Session 11 Room 1
15:40 Session 4: Intelligent Transportation Technology A428
14:00~ ◾Oral Session 12 Room 2
August 24 15:40 Session 5: Aerospace Sciences and Aviation Technology A432
(THU) 14:00~ ◾Oral Session 13 Room 3
15:40 Session 1: Future Electric Facility Planning and Operations A434
14:00~ ◾Oral Session 14 Room 4
15:40 Session 5: Aerospace Sciences and Aviation Technology A435
Poster Session
August 24 (THU) 12:00 ~ 15:10
[Building A 4th Floor]
DATE Session Conference Topics Session
12:00~ Poster ◾Session 1: Future Electric Facility Planning and Operations
12:40 Session 1 ◾Session 3: Information Technology
August 24
(THU) ◾Session 2: Smart Electric Facility
14:30~ Poster
◾Session 4: Intelligent Transportation Technology
15:10 Session 2
◾Session 5: Aerospace Sciences and Aviation Technology ETC
- 15 -
Online & Offline Assistance
❚Registration Desk | Lobby
- August 22 (TUE) 13:00 – 18:00
- August 23 (WED) ~ August 25 (FRI) 09:00 – 11:00
- Lobby D building. Assembly Hall
❚Registration Fee Covers [On-Site]
[Kazakhstan]
- For AUPET 18 000 KZT
- For other Kazakhstan Universities 32 000 KZT
- For Students 14 000 KZT
[normal distribution]
- Regular USD 600
- Student USD 300
- [Early-Virtual]
- Regular USD 300
- Student USD 150
Keynote Sessions, Conference Kit, Lunches
❚Personal Envelope
1. Name Badge
2. Program Book
3. Souvenir
❚Lunch
- August 23 (WED) 12:30 – 14:00
- August 24 (THU) 12:40 – 14:00
❚Coffee Break | Lobby
- August 22 (TUE) 15:20~15:50
- August 23 (WED) 11:00~11:30 / 15:40~16:20
- August 24 (THU) 10:20~10:40 / 14:50~15:00
Fresh coffee, tea and refreshments will be served during the break times.
- 16 -
❚DOWNLOAD MATERIALS
You can download your materials at my page or virtual website
❚How do I access the online session?
- Only participants who have completed
- Registration can access the ICEE 2023
- Virtual website. Please use the official
- Website login information (ID/PW).
❚What are the technical requirements needed to participate in the ICEF 2023
online?
You need an Internet connection and web
browser. For the best viewing experience,
we recommend that you join the conference
using a supported internet browser, which
includes Google Chrome, Firefox and Safari.
ICEF 2023 VIRTUAL Platform Guide
❚How do I access the online session?
- Only participants who have completed registration can access the ICEF 2023
(www.icef2023.org) virtual website. Please use the official website Registration login
information (ID/PW).
- 17 -
Keynote, Oral Presentation (on-site/online)
ZOOM Access Link For Each Conference Room [GMT +06 / Almaty, Kazakhstan]
❚Opening Ceremony (on-site/online)
DATE Live Sessions Access Link Log-in ID
August 23 11:30~ https://us06web.zoom.us/j/8700834425?pwd=
◾ Opening Ceremony Log-in ID:870 083 4425
(WED) 12:30 WEtHK0llaWZmTWtmTnhPUlVqRmlGUT09
ACCESS CODE: 1
❚Keynote Presentation (on-site/online)
DATE Live Sessions Access Link Log-in ID
14:00~ https://us06web.zoom.us/j/8700834425?pwd=
◾ Keynote session (1) Log-in ID: 870 083 4425
14:40 WEtHK0llaWZmTWtmTnhPUlVqRmlGUT09
ACCESS CODE: 1
14:40~ https://us06web.zoom.us/j/8700834425?pwd=
◾ Keynote session (2) Log-in ID: 870 083 4425
15:20 WEtHK0llaWZmTWtmTnhPUlVqRmlGUT09
August 22 ACCESS CODE: 1
(TUE)
15:50~ https://us06web.zoom.us/j/8700834425?pwd=
◾ Keynote session (3) Log-in ID: 870 083 4425
16:30 WEtHK0llaWZmTWtmTnhPUlVqRmlGUT09
ACCESS CODE: 1
16:30~ https://us06web.zoom.us/j/8700834425?pwd=
◾ Keynote session (4) Log-in ID: 870 083 4425
17:10 WEtHK0llaWZmTWtmTnhPUlVqRmlGUT09
ACCESS CODE: 1
09:00~ https://us06web.zoom.us/j/8700834425?pwd=
◾ Keynote session (5) Log-in ID: 870 083 4425
09:40 WEtHK0llaWZmTWtmTnhPUlVqRmlGUT09
ACCESS CODE: 1
09:40~ https://us06web.zoom.us/j/8700834425?pwd=
◾ Keynote session (6) Log-in ID: 870 083 4425
10:20 WEtHK0llaWZmTWtmTnhPUlVqRmlGUT09
August 23 ACCESS CODE: 1
(WED)
10:20~ https://us06web.zoom.us/j/8700834425?pwd=
◾ Keynote session (7) Log-in ID: 870 083 4425
11:00 WEtHK0llaWZmTWtmTnhPUlVqRmlGUT09
ACCESS CODE: 1
11:30~ https://us06web.zoom.us/j/8700834425?pwd=
◾ Opening Ceremony Log-in ID: 870 083 4425
12:30 WEtHK0llaWZmTWtmTnhPUlVqRmlGUT09
ACCESS CODE: 1
- 18 -
DATE Live Sessions Access Link Log-in ID
09:00~ https://us06web.zoom.us/j/8700834425?pwd=
◾ Keynote session (8) Log-in ID: 870 083 4425
09:40 WEtHK0llaWZmTWtmTnhPUlVqRmlGUT09
ACCESS CODE: 1
09:40~ https://us06web.zoom.us/j/8700834425?pwd=
◾ Keynote session (9) Log-in ID: 870 083 4425
10:20 WEtHK0llaWZmTWtmTnhPUlVqRmlGUT09
August 24 ACCESS CODE: 1
(THU)
10:40~ https://us06web.zoom.us/j/8700834425?pwd=
◾ Keynote session (10) Log-in ID: 870 083 4425
11:20 WEtHK0llaWZmTWtmTnhPUlVqRmlGUT09
ACCESS CODE: 1
11:20~ https://us06web.zoom.us/j/8700834425?pwd=
◾ Keynote session (11) Log-in ID: 870 083 4425
12:00 WEtHK0llaWZmTWtmTnhPUlVqRmlGUT09
ACCESS CODE: 1
❚Oral Presentation (on-site/online)
DATE Live Sessions Access Link Log-in ID
14:00~ https://us06web.zoom.us/j/7386073435?pwd=
◾ Oral Session 1 Log-in ID: 738 607 3435
15:40 SHAzL1kwTmcyYm1PdkdzNVNLNUZ6dz09
ACCESS CODE: 1
14:00~ https://us06web.zoom.us/j/6118289105?pwd=
◾ Oral Session 2 Log-in ID: 611 828 9105
15:40 V0lmWFRRL0R6Tm56QTdRUHNoMWI3QT09
ACCESS CODE: 1
14:00~ https://us06web.zoom.us/j/9127203688?pwd=
◾ Oral Session 3 Log-in ID: 912 720 3688
15:40 dTVreXJLQXFzdVNrTUp0aVpZUzdJUT09
ACCESS CODE: 1
14:00~ https://us06web.zoom.us/j/2857636750?pwd=
◾ Oral Session 4 Log-in ID: 285 763 6750
15:40 SHAzL1kwTmcyYm1PdkdzNVNLNUZ6dz09
August 23 ACCESS CODE: 1
(WED)
14:00~ https://us06web.zoom.us/j/8700834425?pwd=
◾ Oral Session 5 Log-in ID: 870 083 4425
15:40 WEtHK0llaWZmTWtmTnhPUlVqRmlGUT09
ACCESS CODE: 1
16:20~ https://us06web.zoom.us/j/7386073435?pwd=
◾ Oral Session 6 Log-in ID: 738 607 3435
18:00 SHAzL1kwTmcyYm1PdkdzNVNLNUZ6dz09
ACCESS CODE: 1
16:20~ https://us06web.zoom.us/j/6118289105?pwd=
◾ Oral Session 7 Log-in ID: 611 828 9105
18:00 V0lmWFRRL0R6Tm56QTdRUHNoMWI3QT09
ACCESS CODE: 1
16:20~ https://us06web.zoom.us/j/9127203688?pwd=
◾ Oral Session 8 Log-in ID: 912 720 3688
18:00 dTVreXJLQXFzdVNrTUp0aVpZUzdJUT09
ACCESS CODE: 1
- 19 -
DATE Live Sessions Access Link Log-in ID
16:20~ https://us06web.zoom.us/j/2857636750?pwd=S
◾ Oral Session 9 Log-in ID: 285 763 6750
18:00 HAzL1kwTmcyYm1PdkdzNVNLNUZ6dz09
August 23 ACCESS CODE: 1
(WED)
16:20~ https://us06web.zoom.us/j/8700834425?pwd=
◾ Oral Session 10 Log-in ID: 870 083 4425
18:00 WEtHK0llaWZmTWtmTnhPUlVqRmlGUT09
ACCESS CODE: 1
14:00~ https://us06web.zoom.us/j/7305475584?pwd=d
◾ Oral Session 11 Log-in ID: 730 547 5584
15:40 TVreXJLQXFzdVNrTUp0aVpZUzdJUT09
ACCESS CODE: 1
14:00~ https://us06web.zoom.us/j/5104981166?pwd=b
◾ Oral Session 12 Log-in ID: 510 498 1166
15:40 3M4cFJxUHFnZnpuU3kyWW8vNzg0QT09
August 24 ACCESS CODE: 1
(THU)
14:00~ https://us06web.zoom.us/j/6118289105?pwd=V
◾ Oral Session 13 Log-in ID: 611 828 9105
15:40 0lmWFRRL0R6Tm56QTdRUHNoMWI3QT09
ACCESS CODE: 1
14:00~ https://us06web.zoom.us/j/5520682143?pwd=
◾ Oral Session 14 Log-in ID: 552 068 2143
15:40 WEtHK0llaWZmTWtmTnhPUlVqRmlGUT09
ACCESS CODE: 1
- 20 -
How The Session Will Go
Oral Session
• Presentation time: 15 minutes Presentation + 5 minutes Live Q&A
• Make sure your presentation materials do not have compatibility issues beforehand.
• Given the ZOOM livestreaming, please arrive in the session room at least 15 minutes
before the session.
• Summit your file to the staff of each presentation room before session starts.
• Presentations are to be done on-site
Poster Session
• Poster authors should print, bring and post the posters at the designated spot for
on-site availability. [A1 Size : 594*841mm]
• Please make sure to stand by your poster during your assigned presentation time.
• Q&As will be made live on-site and via virtual website with comments and email.
• All posters will be posted on both platforms (online/offline) simultaneously during the
conference dates.
- 21 -
Special Sessions
09:00~13:00 FRI, August 25, 2023
Room (Room 5, A437, ALT)
Kazakhstan-Korea Global Smart Railway Forum
Time Contents Note
Reception and Registration
Korea Railroad Company
Seoul Metro
Kazakhstan Temir Zholy (KTZ)
09:30~11: 30 Incheon Metro
Almaty Metro
Korea Railroad Research Institute (KRRI)
SR
Coffee Break or Campus Tour
Time Contents Speaker Note
- Opening speech - Prof. J.W. Kang
- Welcome speech - Rector Amirgaliyeva
- Congratulatory speech S.N.
- Lavrinenko Y.I.
Chairman of the Board of
Directors Kazakhstan Temir
Zholy (KTZ)
Introduction of ALT Dr. Gani Balbayev ALT
History of High Speed Trains in Korea Dr. Gil-Hyun Kang KNUT
History of Kazakhstan Railway Alik Assel Deputy Dean, ALT
Development and demonstration of a
transfer type automatic charging system Dr. Seung-Kwon Shin KRRI
for electric bus
11:30~13:00
Automatic Cleaning System for Rigid
Dr. Cheol-Min Park KRRI,
Conductor in the Underground Section
Professional staff is the key to the effective
development of the transport and logistics Prof. T. Chigambayev ALT
complex
Almagambetov Kanat Deputy Chairman of the
Introduction of Kazakhstan Temir Zholy
Esmukhanovich Board of JSC "NC "KTZ"
Andreev Pavel Head of automation and
Almaty metro
Veniaminovich communication service
Leading Company in Electric Railway
Mr. Sang Woong Lee Sejong Engineering Ltd
Construction: Sejong Engineering
Closing Speech ALT Rector
Group Photo and Lunch
- 22 -
Contents
Oral Session 1
Improved LQR control using fuzzy logic and Kalman filter controller of DC motor ··············· 39
Adil Maidanov1*, Sabyrzhan Atanov1, and Hüseyin Canbolat2
1
L. N. Gumilev Eurasian National University, Kazakhstan, 2Ankara Yildirim Beyazit University,
Kazakhstan
Digital Transformation of Document Support for Enterprise Quality Management System ·· 50
Shara Toibayeva and Irbulat Utepbergenov
Almaty University of Power Engineering and Telecommunications, Kazakhstan
On the issue of constructing and identifying the parameters of the piezometric pattern of
pumping wells in the underground leaching of metals ····················································· 55
Bakhyt Mukhanov1, Zhanar Omirbekova2, Yerbol Orakbayev1, and Gulmira Bazil1
1 2
Almaty University of Power Engineering and Telecommunications, Kazakhstan, Satbayev University
Almaty, Kazakhstan
Analysis and synthesis of single-bit adders for multi-bit adders with sequential transfers ···· 59
1 1 2
Sakhybay Tynymbayev , Assel Mukasheva , and Ibragimov Kuanyshbek
1 2
Almaty University of Power Engineering and Telecommunications Almaty, Kazakhstan, Al-Farabi
Kazakh National University, Kazakhstan
Oral Session 2
Analysis for Fault Characteristics of MMC in DC Distribution System ··································· 65
Jin-Sol Song1, Jong-Myoung Kim1, Seung-Whan Bae1, Arif Mehdi1, S Jarjees UI Hassan1,
1 1 1 2
Abraham. D. Arefaynie , Zeeshan Haider , Chul-Hwan Kim *, Myung-Hwan Min , and
2
Tae-Pung An
1 2
Sungkyunkwan University, Korea, ENTEC Electric & Electronic, Korea
Assessment of Locational and Total Hosting Capacity of Distribution Network with Multitype
DGUs by Iterative OPF Method ··························································································· 70
S Jarjees Ul Hassan, Arif Mehdi, Jin-Sol Song, Abraham D. Arefaynie, Zeeshan Haider, and
Chul-Hwan Kim*
Sungkyunkwan University, Korea
Optimal Parameter Design of Thyristor-based Multi-module SSCB in MVDC Network ········ 75
Jong-Myoung Kim, Jin-Sol Song, Seung-Whan Bae, and Chul-Hwan Kim*
Sungkyunkwan University, Korea
- 23 -
Investigation of Grounding Schemes in Hybrid (AC/DC) Medium Voltage Distribution
Networks ··························································································································· 81
Zeeshan Haider, Arif Mehdi, S Jarjees Ul Hassan, Jin-Sol Song, Abraham D. Arefaynie, and
Chul-Hwan Kim*
Sungkyunkwan University, Korea
Oral Session 3
Research and development of an information system for speech recognition based on neural
network algorithms. ·········································································································· 88
Maksat Kanashev1, Sergiy Gnatyuk2, Mirambayeva Nazym3, and Assel Mukasheva4
1
Almaty University of Power Engineering and Telecommunications, Kazakhstan, 2National Aviation
3
University National Aviation University, Ukraine, Almaty University of Energy and Communications
4
College Almaty, Kazakhstan, Almaty University of Power Engineering and Telecommunications,
Kazakhstan
Research and development of sign language recognition system using neural network
algorithm. ·························································································································· 95
Yegor Matveyas, Arailym Keneskanova, Dastan Kambarov, and Assel Mukasheva
Almaty University of Power Engineering and Telecommunications, Kazakhstan
Research and Development of a Chatbot Using Neural Network Technologies to Process
Requests ·························································································································· 101
Kyzyrov Daniyar Ersainuly and Sapakova Saya Zamanbekovna
International Information Technologies University, Kazakhstan
A Study on the Effective Multimodal Approach for Anomaly Detection using Gramian Angular
Difference Field from Heating, Ventilation, and Air Conditioning (HVAC) System of
Railways ·························································································································· 106
Beomjun Kim1, Younghoon Jeon1, Jeong Won Kang1, Chulsu Kim1, Oh Kuen Kwon2*, and
1
Jeonghwan Gwak *
1
Korea National University of Transportation, Korea, 2Semyung University, Korea
Advancing Railway Track Segmentation: Harnessing the Power of ResUNet and
Self-Supervised Learning Paradigms ············································································· 110
Changjoon Park, Jeong Won Kang, and Jeonghwan Gwak*
Korea National University of Transportation, Korea
- 24 -
Oral Session 4
Calculating FMEA variables of software failure risks based on journals and metrics ········ 114
Denis Valyayev1, Bolshibayeva Aigerim2, Dinargul Mukhammejanova1, and Assel Mukasheva1
1 2
Almaty University of Power Engineering and Telecommunications, Kazakhstan, International IT
University, Kazakhstan
Comparative analysis of Hadoop and Spark performance through different data processing
tasks ································································································································ 120
Yerulan Suynullayev1, Sergiy Gnatyuk2, Didar Yedilkhan3, Aisulu Serik1, and Assel Mukasheva1
1 2
Almaty University of Power Engineering and Telecommunications, Kazakhstan, National aviation
university, Ukraine, 3Astana IT University, Kazakhstan
The Use Of Machine Learning Methods In The Field Of Retail ············································· 127
Kuanyshbek N.S and Sapakova S.Z*
Al-Farabi Kazakh National University, Kazakhstan
A study on false defect management of PCB ICT inspection using failure rate Estimation ··· 128
Changmok Park* and Sangyule Choi
Induk University, Korea
Oral Session 5
VIDEO STREAM ANALYSIS ALGORITHMS BASED ON THE OPENCV LIBRARY USING THE
RASPBERRY P S PLATFORM ···························································································· 131
1 2 1 1
Adilzhan Kereyev , Bauyrzhan Shokanov *, Lyazzat Kaparova *, and Elubayeva Dina *
1 2
K.Zhubanov Aktobe Regional University, Kazakhstan, Nazarbayev Intellectual Schools, Kazakhstan
Internet of Things attack and security analysis ··································································· 136
Aigul Adamova1 and Tamara Zhukabayeva2
1 2
Astana IT University, Kazakhstan, L. N. Gumilev Eurasian National University, Kazakhstan
IoT watering robot in technologies of variable irrigation rationing in precision
agriculture ······················································································································· 142
Madina Mansurova1*, Baurzhan Belgybaev1, Yerlan Zaitin1, and Askhat Asset2
1
Al-Farabi Kazakh National University, Kazakhstan, 2Almaty University of Power Engineering and
Telecommunications, Kazakhstan
Cache Replacement Policy for Approximate Computing in Many Core Systems ················· 147
Juhee Choi
Sangmyung University, Korea
- 25 -
Oral Session 6
ANALYSIS OF CLIENT-SIDE AND SERVER-SIDE ATTACKS IN THE CONTEXT OF WEB
APPLICATIONS ················································································································· 150
Adlet Smailov* and Saukhanova Zhanat
L. N. Gumilev Eurasian National University, Kazakhstan
Security of containers and containerization environments ·················································· 153
Amir Abdreev* and Assel Nurusheva
L. N. Gumilev Eurasian National University, Kazakhstan
Research and development of two-factor authentication methods using neural
networks ·························································································································· 157
Temirkhan Darbek1, Timur Umarov2, Sabina Rakhmetulaeva3, Zhanna Suimenbayeva4, and
1
Assel Mukasheva
1
Almaty University of Power Engineering and Telecommunications, Kazakhstan, 2British
3 4
Management University, Kazakhstan, International IT University, Kazakhstan, Satpayev University,
5
Kazakhstan, Almaty University of Power Engineering and Telecommunications, Kazakhstan
Design and Implementation of an Electronic Voting System Using the IBM Blockchain
Platform ··························································································································· 163
Tolegen Aydynov*, Olzhas Sabiev, and Dina Satybaldina
L. N. Gumilev Eurasian National University, Kazakhstan
Study of Network Virtualization Security in Public Network ················································ 164
Junwon Lee1 and Juhee Choi2*
1
Samsung SDS, Korea, 2Sangmyung University, Korea
Developing An Attack Detection Algorithm Model and Compared with The K-Means
Algorithm ························································································································· 167
1,2 1 1 2 3
Uyangaa. Kh , Odonchimeg. L , Dashdorj. Ya *, Liji Wu , and Heemin Park
1 2
Mongolian University of Science and Technology, Mongolia, Tsinghua University, China,
3
Sangmyung University, Korea
Oral Session 7
Development of mobile application for copper wire molten mark classification ················· 173
Seong-jun Hong*, Dong-hyun Tae, and Jin-sik Lee
Korea Electrical Safety Corporation Electrical Safety Research Institute, Korea
Isolated control circuit design for heater power supply of high-power magnetron ············· 176
Hyun-Bin Jo and Hong-Je Ryoo
Chung-Ang University, Korea
- 26 -
Oral Session 8
Simulation of computer vision systems with artificial intelligence ····································· 180
Perizat Rakhmetova1*, Makabbat Zhamuratova1, Vinera Baiturganova1, Beibit Shingissov2, Laura
2 3
Tasbolatova , and Symbat Akparova
1 2 3
Satbayev University, Kazakhstan, Academy of Logistics and Transport, Kazakhstan, Ultra ALM,
Kazakhstan
Integration of Penalty Functions and Performance Criteria for Enhanced Optimization in
Optimal Control Problems using Artificial Immune System Algorithms ························· 185
1 2 1
Olga Shiryayeva *, Timur Samigulin , and Dmirtyi Panchenko
1 2
Satbayev University, IICT, Kazakhstan, Kazakh-British Technical University, Kazakhstan
A Comparative Study on One-Class Classification Based Autoencoder Models for Anomaly
Detection in Brake Operating Units ················································································· 190
Inki Kim, Jeong Won Kang, and Jeonghwan Gwak*
Korea National University of Transportation, Korea
Advancement in Rail Defect Detection Using YOLO-NAS: A Neural Architecture Search
Approach ·························································································································· 193
Namjung Kim, Junhwi Park, Jeong Won Kang, and Jeonghwan Gwak*
Korea National University of Transportation, Korea
Oral Session 9
A Neural Network Based Approach for Disease Prediction Using Symptom Data: Uncovering
Hidden Relationships and Improving Diagnostic Accuracy ·············································· 197
1 1 2 3
Zoljargal Jargalsaikhan , Zolzaya Dashdorj *, Stanislav Grigorev , Tae-Koo Kang ,
1 1
Munguntsatsral Dovdondash , and Erdenebaatar Altangerel
1 2
Mongolian University of Science and Technology, Mongolia, Irkutsk National Research Technical
3
University, Russia, Sangmyung university, Korea
System of Intelligent Control of Processes of Development and Harmonization of Standards in
the Field of Military and Space Activities on the Basis of Ontological Engineering ········· 202
Anar Utegenova1,2, Gulnaz Yermoldina2, Zhanna Suimenbayeva1,3*, Alisher Aden1,3, and
3,4
Vitaliy Naumenko
1 2
Almaty University of Power Engineering and Telecommunications, Kazakhstan, Institute of
3
Information and Computational Technologies, Kazakhstan, INT-SAT Alatau LLP, Kazakhstan,
4
International University of Information Technology, Kazakhstan
- 27 -
Oral Session 10
Algorithms of Methodology for Applying Augmented Reality in Teaching Physics ·············· 208
Beken Arymbekov
Al-Farabi Kazakh National University, Kazakhstan
Utilizing Plane Detection Technology based on Actual Physical Environment for Classroom
Experiments ···················································································································· 213
Yeji Heo, Woojeong Shim, and Juhee Choi*
Sangmyung university, Korea
Cost Comparison and Analysis of VR Tracker Manufacturing with Low-cost Sensor and High
value Sensor ···················································································································· 215
Beomjun Cho1, Chanhee Jang1, Juhee Choi1*, and Joonyoung Kim2*
1
Sangmyung university, Korea, 2IMEC, Belgium
Detection and Classification of the Volatile Organic Compounds from the Petroleum and
Petroleum-Derived Products Using an Electronic Nose and Machine Learning
Algorithms ······················································································································· 218
Olzhas Alseitov, Aliya Issainova, Dina Satybaldina*, and Assem Konyrkhanova
L. N. Gumilev Eurasian National University, Kazakhstan
LoRaWAN Optimization for monitoring of randomly fluctuated parameters ······················· 224
Kim D.K.1*, Turlikov A.M.2, and Markovskaya N.V2
1 2
Narxoz University, Kazakhstan, HSE University, Russia
Oral Session 11
Enhancing Public Bus Line Classification and User Route Estimation through Deep Learning:
A Comprehensive Analysis and Predictive Modeling Approach ······································· 227
Tsetsentsengel Munkhbayar1, Zolzaya Dashdorj1*, Tae-Koo Kang2, Tae-Koo Kang2,
2 2 1
Kyoung-Geun Cho , Taekgwon Nam , and Erdenebaatar Altangerel
1 2
Mongolian University of Science and Technology, Mongolia, Sangmyung University, Korea
Development of Fault Diagnosis Simulator of Pneumatic Door System for Electric Railway
Vehicles ···························································································································· 232
Gil Hyun Kang1, Young Min Lim2, In Su Chung1, and Chul Su Kim1*
1 2
Korea National University of Transportation, Korea, ROTECO, Korea
- 28 -
Oral Session 12
Mathematical model for determining the orientation of a star sensor ································ 235
Kuanysh Alipbayev and Kamila Saurova*
Almaty University of Energy and Communications named after G.Daukeyev, Kazakhstan
Target selection by a group of unmanned aerial vehicles based on a swarm approach ······ 242
Aigerim Mazakova1, Vadim Zhmud2, Sholpan Jomartova1, and Тalgat Mazakov1
1 2
Al-Farabi Kazakh National University, Almaty, Kazakhstan, Altae-Sayan Branch of the Federal
State Budgetary Institution of Science of the Geophysical Service of the RAS, Novosibirsk, Russia
DEVELOPMENT OF AN ADAPTIVE DRIVE FOR AEROSPACE ENGINEERING ························· 247
Ivanov K.1, Alipbaev K.1*, and Tulekenova D.2
1
Almaty University of Power Engineering and Telecommunications, Kazakhstan, 2Al-Farabi Kazakh
National University, Kazakhstan
Underwater robot propulsion system ·················································································· 251
Ivanov K., Alipbayev K.*, and Saurova K.
Almaty University of Power Engineering and Telecommunications, Kazakhstan
Research and design of an efficient adaptive drive with balancing friction coupling ·········· 255
Konstantin Ivanov, Aidos Sultan*, Kuanysh Alipbayev, and Komekbayev Arman
Almaty University of Power Engineering and Telecommunications, Kazakhstan
Oral Session 13
Method of predicting the service life of a linear adjustment transformer based on the analysis
of operational reliability data ··························································································· 263
I.T. Utepbergenov1*, A.A. Tazhibayev1,3, Ye.U. Zhumagaliyev2, D.A. Yesengaliyev2, and
3
S.A. Yermetov
1 2
Almaty University of Power Engineering and Telecommunications, Kazakhstan, ZhubanovAktobe
Regional University, Kazakhstan, 3TrencoR&D LLP, Kazakhstan
Sensorless operation of six-phase PMSM based on high-frequency signal injection with torque
ripple reduction method ·································································································· 269
Seong-hoon Kim, Kwan-yuhl Cho*, Hag-wone Kim, and Sang-min Lee
Korea National University of Transportation, Korea
Study on the Operation Algorithm of TVPP Considering Distribution Line Constraints in Jeju
Power System ·················································································································· 271
Ji-Won Lee1, Dong-Il Cho2, Seong-Jun Park1, Tae-Yang Nam1, Byung-Ki Kim3, Jae-Sun Huh4,
Nam-Gil Paik4, Jae-Chul Kim1, and Won-Sik Moon1*
1
Soongsil University, Korea, 2Korea University, Korea, 3Korea Institute of Energy Reasearch, Korea,
4
KEPCO, Korea
- 29 -
Study of the direct and inverse problem of the thermal conductivity of a rod ····················· 274
Sholpan Jomartova, Тalgat Mazakov, Nurdaulet Isimov, Aigerim Mazakova, and
Kalamkas Begaliyeva
Al-Farabi Kazakh National University, Kazakhstan
Switching Loss Analysis by Interleaving Method of 3-LV Boost Converter with Coupled
Inductor ··························································································································· 279
Hae In Kim, Hag Wone Kim*, Kwan Yuhl Cho, and Jeong Won Kang
Korea National University of Transportation, Korea
Oral Session 14
Finding trends in the development of narrow beam inter-satellite communication antennas for
Cubesat satellites using patent analysis ········································································· 283
Zh.Suimenbayeva1,2, B.Aitchanov1, K.Alipbayev1*, A.Komekbayev1, A. Aden1, and
1
A.Mukasheva
1 2
Almaty University of Power Engineering and Telecommunications, Kazakhstan, Satpayev
University, Kazakhstan
A method for improving the accuracy of navigation receivers using GPS as an
example. ·························································································································· 288
Khussainov Ansar and Ibraim Merey*
Almaty University of Power Engineering and Telecommunications, Kazakhstan
Poster Session 1
A Study on the Partial Discharge Characteristics of VTs and CTs by Applying Repeated
Impulses ·························································································································· 300
Ji Man Park, Ja Yoon Kang, Young Chae Mun, and Ga Ram Han*
Korea Electrical Safety Corporation Research Institute, Korea
An Application of Terahertz Wave for Detecting Insulation Defects in 22.9kV XLPE Power
Cable ································································································································ 303
Ji-Man Park, Young-Chae Mun, and Ja-Yoon Kang*
Korea Electrical Safety Corporation Research Institute, Korea
A Study on the On/off-line Framework for Asset Management of 154 kV Electrical
Facilities ·························································································································· 305
Young Kyu Mo*, Min Hwang, Pil-sung Woo, Joon-Ho Son, and Youngseok Kim
Korea Electrical Safety Corporation Research Institute, Korea
- 30 -
Design of Real-time Monitoring/Diagnosis Algorithm for Electrical Facilities ···················· 307
Hoon Jung, Young-Taek Lim, Jong-Chan Lee, and Joonho Ahn*
Seoul National University, Korea
Leakage Current Analysis of PCS considering the Ground System ····································· 311
Dongmin Kim and Hanju Cha*
Chungnam National University, Korea
Estimation of the probability of disconnection of urban electric networks in emergency
situations based on the use of GIS maps ········································································· 313
Y.Umbetkulov*, K.Bakenov, T.Аgimov, M.Khazimov, and L.Uteshkalieva
Almaty University of Power Engineering and Telecommunication named G.Daukeev, Kazakhstan
A Study on Common Mode Effects According to Battery Rack Configurations in Energy Storage
Systems ··························································································································· 317
Yong-Eun Choi, Hyo-Seok Oh, Byung-Min Moon, Chin-Young Chang, and Jae-Moon Kim*
Korea National University of Transportation, Korea
Development of 48V PbC Hybrid Battery for Electric Forklift Using High Quality Activated
Carbon ····························································································································· 319
Hoon Seo1, In Seok Seo1, Hyon-ok Lim1, and Kyung-Wan Koo2*
1
DongYoung Ind. Co. Ltd., KOREA, 2HoSeo University, KOREA
Experimental studies of a model of a dusted flow air heater for the development of a new
hot-water boiler ··············································································································· 322
Otynchiyeva, M. T.1, Bakhtyar B.T.2*, Korobkov M.S.3, and Tursunbayeva G.U.2
1 2
Narxoz University, Kazakhstan, S.Seifullin Kazakh Agro Technical Research University,
Kazakhstan, 3Almaty University of Power Engineering and Telecommunications, Kazakhstan
Comparison of VARC DC Interrupting Module Topologies for Enhancing Superconducting VARC
DCCB ································································································································ 335
Daeseong Jang and Hyosang Choi*
Chosun University, Korea
Resonance Frequency Analysis for LC Resonant DC Circuit Breaker Design ······················ 340
Jihye Kim and Hyosang Choi*
Chosun University, Korea
Failure Analysis of Arc and Leakage Generation Signal in DC Line ····································· 343
SoYeon Han, SuChul Bang, SungIn Jeong, and YongHo Yoon*
Gwangju University, Korea
Detection of Leakage Current Occurrence Location of LVDC Lines through Level-Shift Signal
Processing ······················································································································· 346
Su-Chul Bang, So-Yeon Han, Sung-In Jeong, and Yong-Ho Yoon*
Gwangju University, Korea
- 31 -
Using IoT technologies to prevent fires ··············································································· 349
Akyltay Burgegulov1, Gulzat Ziyatbekova1,3*, Тalgat Mazakov1,3*, Aigerim Sametova1,2*, and
1
Magzhan Aliaskar *
1 2
Al-Farabi Kazakh National University, Kazakhstan, Almaty University of Power Engineering and
3
Telecommunications, Kazakhstan, Institute of Information and Computational Technologies,
Kazakhstan
Implementation of Real-time Water Quality Monitoring for Energy Consumption Optimization
Utilizing IoT Sensors in a Smart Fish Farm ····································································· 356
Hunseo Lee1, Eunsu Jung1, Juhyeon Youk1, June-young Park2, Hyun-il Kang1, and Young Park1*
1 2
Hanbat National University, Korea, CNC Aid Co., Ltd., Korea
Implementation of IoT-based smart elevator remote regeneration control monitoring system
platform ··························································································································· 358
Hyeong Ju Kwon, Hag-wone Kim, and Sang-min Lee*
Korea National University of Transportation, Korea
An ontological model for automating the process of preparing electronic courses in the field of
information and communication technologies. ······························································· 360
N.ZH. SABITOVA1, B.Sh. RAZAKHOVA1, and S.O. GNATYUK2
1
L. N. Gumilyov Eurasian National University, Kazakhstan, 2National Aviational University, Ukraina
A Study of Deep Learning Model for Color Measuring Process in Industrial Facilities ······· 362
1 2 1 1
Kyung Suk Kim , Sanghun Lee , Ki-Sub Kim *, and Jeong Won Kang *
1 2
Korea National University of Transportation, Korea, Central Research Institute, KCC, Korea
Research and development of an automated installation based on modern technology to
determine the degree of ambient air pollution ································································ 363
Madina Yessimkhanovna Mansurova1, Samal Beksultanovna Abdreshova1, Sandugash
2 1
Kudaibergenovna Orazalieva , and Gulmira Nurlankyzy Abdreshova
1 2
Al-Farabi Kazakh National University, Kazakhstan, Almaty University of Power Engineering and
Telecommunications, Kazakhstan
Data mining of leading indicators ························································································· 365
Laura Alimbayeva1 and Kuralay Belgibayeva2
1 2
Author One’s Affiliation: Al-Farabi Kazakh National University, Kazakhstan, Author Two’s
Affiliation: Kazakh University of Technology and Business, Kazakhstan
Data-Driven Enhanced Travel Planning by Utilizing Real-Time User Data ·························· 372
Sooah Park and Juhee Choi*
Sangmyung University, Korea
A Study on Reuse Policy for Worn-out Cells in Non-Volatile Memories ······························ 374
Juhee Choi and Seonuck Paek*
Sangmyung University, Korea
- 32 -
Development of an analytical computation system and its application for automation of
equations derivation in robotics systems ········································································ 376
Magzhan Aliaskar1, Sholpan Jomartova1*, Aigerim Mazakova1*, Talgat Mazakov1*, and
1,2
Gulzat Ziyatbekova *
1 2
Al-Farabi Kazakh National University, Kazakhstan, Institute of Information and Computational
Technologies, Kazakhstan
MATHEMATICAL MODELING AND DEVELOPMENT OF AUTOMATED FIRE MONITORING
SYSTEM ···························································································································· 381
A.A. Sametova1,2*, T.Zh. Mazakov1,3, G.Z. Ziyatbekova1,3, and A.D. Burgegulov1
1 2
Al-Farabi Kazakh National University, Kazakhstan, Almaty University of Power Engineering and
3
Telecommunications, Kazakhstan, Institute of Information and Computational Technologies,
Kazakhstan
Phased Array Ultrasonic Inspection Model for High-speed Train Wheels ··························· 385
Jae Hwan Cha, Cheol Su Kim*, Jeong Hyung Kang*, and Kyung Suk Kim*
Korea Railroad Corporation, Korea
ICT Convergence Roof Installation Solar Location Tracking Solar Power Generation
System ····························································································································· 389
DeokGun Woo1*, HyoJai Lee1, and ChoonWon Seo2
1 2
Korea-polytechnics, Korea, Neo Solartec Inc., Korea
Study on the collapsing of artificial retaining wall facility using acoustic emission ············ 391
Kwang Bok Kim1, Bong Ki Kim1, Woon Kyung Kim1, Tae Hun Kim1, Sung Hun Park1,
Saurabh Kumar1, Jun Gil Kang1, and Kyung-Wan Koo2*
1
Integrity Diagnostic Korea Co. Ltd., Republic of Korea, 2Hoseo University, Korea
Technique for the use of matching circuits in the transmit-receive path of the uav radio control
channel in conditions of icing of the antenna device ······················································· 393
Nurbolat Ondyrbayev1*, Nursultan Nurzhauov1, Anna Sutiko2, Pavel Boikachev2,
1 1
Katipa Chezhimbayeva , and Asel Yerzhan
1 2
Almaty University of Power Engineering and Telecommunications, Kazakhstan, Military Academy
of Belarus, Belarus
Designing a special mobility (hot air balloon) communication interface ····························· 397
Seung-Hwan Byun1*, Kyung-Wan Koo2, Mun-Sung Kim1, and Jong-Gyu Kang3
1 2 3
Daejeon University, Korea, Hoseo University, Korea, MetaCom Ltd., Korea
VLAN Network Architecture for Trains to Trains Communication ······································· 400
Yongjung Kim1, Marina Asankulova2, and Kyuhyoung Choi1*
1 2
Seoul National University of Science and Technology, Korea, LLC Tendik Audit, Kyrgyzstan
- 33 -
Poster Session 2
A rectangular gate structure with attachable sanitary lights and human monitoring
camera ····························································································································· 405
1 1 2 3 3
Woo-Young Kim , Hyung-Kon Lee , Young-Min Kim , Seung-Youn Yang , Jin-Tae Kim ,
4 5 6 6 7
Kyung-Wan, Koo , Min-Soo Kim , Tae-Ho Kim , Seong-Kweon Kim , In-Choel Jeong ,
8 8
Chang-Jun Ahn , and Jae-Sang Cha *
1 2 3 4
Boseung Electric Co., Ltd., Korea, Jooyoung F&G, Korea, Fivetek. Co.ltd., Korea, Dongseoul
University, Korea, 5Hoseo University, Korea, 6Seoul National University of Science and Technology,
7 8
Korea, SungKongHoe University, Korea, Chiba University, Korea & Japan
A development of IoT switching and control box application used in various power source
choice and sanitary light control functions ······································································ 407
1 1 2 3 4
Woo-Young Kim , Hyung-Kon Lee , Gyewon Choi , Minsoo Kim , Tae-Ho Kim ,
4 5 6 7 7
Seong-Kweon Kim , Kyung-Wan Koo , In-Cheol Jeong , Chang-Jun Ahn , and Jae-sang Cha *
1
Boseung Electric Co., Ltd., Korea, 2Sungkyunkwan University, Korea, 3Dongseoul University, Korea,
4 5
Seoul National University of Science and Technology, Korea, Hoseo University, Korea,
6 7
SungKongHoe University, Korea, Chiba University, Korea & Japan
On the Electrical Safety Analysis of Multi-chute Classification System for Joint Courier
Logistics Base Infrastructure based STPA ······································································ 409
Sunwoo Hwang1, Joouk Kim2, and Youngmin Kim1*
1 2
Ajou University, Korea, Korea Railroad Research Institute, Korea
The Remote Driving Control System Based on Telepresence for Mini-tram ······················· 411
Jung-Ju Kim*, Gi-Won Kim, Jun-Seok Lee, Jae-Hun Choi, Hyung-Woo Jin, and Chan-Jong Lee
Hoseo University, Korea
On the Derivation of Demand Time for Response to Variable Function Focus Camera
Situation ··························································································································· 415
Min Joong Kim, Myung Sung Kim, and Young Min Kim*
Ajou University, Korea
Machine-Learning based Condition Diagnosis Model of Heating, Ventilation and Air
Conditioning System for Train Cars ················································································· 417
Juhee Choi1, Kangwoo Nam2, and Chul-Su Kim2*
1 2
Sangmyung University, Korea, Korea National University of Transportation, Korea
A Study on Route Selection Methods of Autonomous Vehicles for Public Mobility ·············· 419
Jung-Ju Kim*, Gi-Won Kim, Chan-Jong Lee, Jun-Seok Lee, Jae-Hun Choi, and Hyung-Woo Jin
Hoseo University, Korea
A Model for Implementing Network-Connected Systems to Train Control Systems ··········· 422
Gyuoyoung YEON1, Jaeho LEE2, and Jongwoo LEE1*
1 2
Korea National University of Transportation, Korea, Korea Railroad Research Institute, Korea
- 34 -
A Study on Transient Relief Methods through Characteristic Analysis of Electric Railway
Vehicle Spark Gap ············································································································ 427
Jongrok Choi, Chinyoung Chang, Jaemoon Kim, Jaeyoung Kim, and Hojun Kwon
Korea Railroad Research Institute, Korea
Development of Trolley for Detection Installation Errors of Catenary System ···················· 429
Chulmin PARK1*, Dongkue Kim1, and Sangjin SON2
1
Korea Railroad Research Institute, Korea, 2Myeongsung Rollingstock & Plant Co. Ltd, Korea
Research on the development and application of lightweight catenary poles for emergency
recovery ··························································································································· 431
Ho Ryung Cho, Jong Nam Kim, Hong Kuk You, Young Jae Na, and Jae Moon Kim*
Korea National University of Transportation, Korea
A Study on the Characteristics of Power System Protection System According to Regenerative
Braking of Urban Railway Vehicles ·················································································· 436
Jong Hak Park1,2, Jeong Hyung Kang1,3, and Seung Ho Song1,3, Sung Cheol Han1,2, and
1
Jeong Won Kang *
1 2 3
Korea National University of Transportation, Korea, Seoul METRO, Korea, KORAIL, Korea
A Study on Stabilization of Code Frequency Control Between Wayside and Onboard Signal
System of AF Non-insulated Track Circuit ······································································· 437
Dong Hoon Park, Ho Hyun Han, Seon Kyo Kim, and Jeong Won Kang*
Korea National University of Transportation, Korea
A Study on the Integrated Operation of VHF and Complex Communication Facilities ········· 438
Seon Kyo Kim, Dong Hoon Park, Ho Hyun Han, and Jeong Won Kang*
Korea National University of Transportation, Korea
A Study on CBM Application of Power Capacitor for SIV in Railway Vehicles ······················· 439
Hyo-Seok Oh, Yong-Eun Choi, Byung-Min Moon, Seung-Pyo Jeon, and Jae-Moon Kim*
Korea National University of Transportation, Korea
A Study on SoH of Lithium Polymer Battery Applied to Railway Vehicles ···························· 441
Hyo-Seok Oh, Yong-Eun Choi, Byung-Min Moon, Seung-Pyo Jeon, Chin-Young Chang, and
Jae-Moon Kim*
Korea National University of Transportation, Korea
Pantograph Charging System for Electric Bus ····································································· 443
Seungkwon Shin1*, Kiwon Lee1, Hyungchul KIM1, Rakgyo Jeong1, and Sangryul Choi2
1 2
Korea Railroad Research Institute, Korea, Induck University, Korea
Preparation Analysis of the Reduction Effect the Increase Harmonic Resonance in the AC
Electric Railway Feeding System ····················································································· 445
Guk Hyun Son, Tae Hoon Kim, and Jae Moon Kim*
Korea National University of Transportation, Korea
- 35 -
A Study on the Train Position Detection for the Integrity of Freight Train Based on ETCS Level 3
With GPS and 6-Axis Sensors ·························································································· 448
Seong Yong Jeon1, Nurlan Berdigozhin2, and Kyu Hyoung Choi1*
1 2
Seoul National University of Science and Technology, Korea, Hua Xia Company, Kazakhstan
An Analysis of the Changes of Regenerative Energy Utilization Facilities by the Tap Switching of
Urban Railway Transformers ··························································································· 455
Chi-Myeong Yun1, Hosung Jung2*, Hanmin Lee2, and Hyungchul Kim2
1
University of Science and Technology, Korea, 2Korea Railroad Research Institute, Korea
A Network Architecture for A Substation Automation Based On IEC 61850 ························· 457
Kyungsoon Park1, Bekzhan Rysmendeev2, and Jongwoo LEE1*
1 2
Seoul National University of Science and Technology, Korea, National Company Kyrgiyz Temir
Jolu, Kygyzstan
A Study on the LCC Analysis for High-Speed Train On-Board Signaling Systems ··············· 461
Jungjae LEE1, Meruyert Zhumakhanova2, and Jongwoo LEE1*
1 2
Seoul National University of Science and Technology, Korea, JSC National Company Kazakhstan
Temir, Kazakhstan
MCB Interrupting and Sequential Energizing of EMU in DC and AC Transition Section ······· 466
Dalseo PARK, Hyosang KIM, and Jaeseok SHIM*
Seoul National University of Science and Technology, Korea
Specification Analysis of Reducer-less Motor for Railway Vehicle Door ······························ 469
Byung-Min Moon, Hyo-Seok Oh, Yong-Eun Choi, Seung-Pyo Jeon, In-Soo Chung, and
Jae-Moon Kim*
Korea National University of Transportation, Korea
Analysis of Platform Screen Door Using Reliability Block Diagram Analysis Tool ·············· 471
Jong Nam Kim, Tae Gil Ha, Guk-Hyun Son, and Jeong Won Kang*
Korea National University of Transportation, Korea
A Study on High-Power IGBT Gate Drivers for Railway Vehicle ··········································· 472
Seung-Pyo Jeon, Hyo-Seok Oh, Yong-Eun Choi, Byung-Min Moon, Jeong-Won Kang, and
Jae-Moon Kim*
Korea National University of Transportation, Korea
Full-Stack Simulator for Non-Volatile Memories-based Computing Elements for Railway
Safety ······························································································································· 474
Zheng Wang1, Seong Tae Jhang2, and Juhee Choi3*
1 2 3
Linyi University, China, Suwon University, Korea, Sangmyung University, Korea
A Study on the Optimal Traffic Signal Timing Transition Algorithms for Diverse Demand
Fluctuations on Corridors ································································································ 476
Yong-Bin, Cho1, Jin-Tae, Kim1*, and Thomas JW Kim2
1
Korea National University of Transportation, Korea, 2UI networks, Korea
- 36 -
Investigation of the output parameters of a definable adaptive gear variator. ···················· 480
Konstantin Samsonovich Ivanov1 and Gulmira Nurlankyzy Abdreshova2
1 2
AUES, Almaty, Kazakhstan, Al-Farabi Kazakh National University, Almaty, Kazakhstan
DEVELOPMENT OF AN INTEGRATED MULTIFUNCTIONAL COMPLEX FOR LIGHTING THE SPACE
SITUATION ······················································································································· 481
Anargul Boranbayeva and Dana Tulekenova
AIPET, AUES, Kazakhstan
Safety of aircraft operation in modern conditions ································································ 489
Gulnara Nauryzbayeva, Guldana Aldzhambekova, and A Tuktibayev
Almaty University Power Engineering and Telecommunications after G. Daukeyev, Kazakhstan
A Molecular Dynamics Simulation Study on Superlubric Carbon-based Nanodevice ········· 494
Tae Hoon Kim, Jeong Won Kang*, Haram Kim, and Ki-Sub Kim*
Korea National University of Transportation, Korea
Extracted Cellulose with Deep Eutectic Solvent and Ionic Liquid: Changes in Structural and
Thermal Characteristics ·································································································· 495
Yeung Chan Kim, Ho Hyun Han, Jeong Won Kang*, and Ki-Sub Kim*
Korea National University of Transportation, Korea
Implementation of carbon neutrality through the RE100 verification system in the agricultural
sector ······························································································································· 496
Minseong Kang1, Sangho Park1, Kihyun Kim2, and Kyungwan Koo2
1 2
Korea Testing Laboratory, Korea, Hoseo University, Korea
Study on the collapsing of artificial retaining wall facility using acoustic emission. ··········· 500
1 1 1 1 1
Kwang Bok Kim , Bong Ki Kim , Woon Kyung Kim , Tae Hun Kim , Sung Hun Park ,
1 1 2 2
Saurabh Kumar , Jun Gil Kang , Sang-Tae Han , and Kyung-Wan Koo
1 2
Integrity Diagnostic Korea Co. Ltd., Korea, HoSeo University, Korea
Estimation of AIS-based ship GHGs emission using PAQmanⓒ System : focusing on
Pyeongtaek･Dangiin Ports ······························································································· 503
Yongchan Lee1, Heekwan Lee1, CheolGyo Yim2, and KyungWan Koo3
1
Incheon National University, Korea, 2JIN Electrical Engineering Co.,Ltd., Korea, 3Hoseo
University, Korea
Development of a Waste Plastic Energy Resource System by Integrating Pyrolysis and
High-Temperature Reduction Processes ········································································· 506
Song Jea Jun1 and Kyung Wan KOO2
1
HOWON CO., LTD PRESIDENT. Korea, 2Hoseo University, Korea
- 37 -
August 22(TUE) - 25(FRI), 2023
Oral Session 1
14:00~15:40 Wed, August 23, 2023
Room : A428
Chair: Assel Mukasheva (AUPET, Kazakhstan)
- 38 -
Improved LQR control using fuzzy logic and
Kalman filter controller of DC motor
Adil Maidanov 1*, Sabyrzhan Atanov 2 and Hüseyin Canbolat 3
1
Ph.D. student, L. N. Gumilyov Eurasian National University, Astana, Kazakhstan,
Email: [email protected]
2
Professor, Dr., L. N. Gumilyov Eurasian National University, Astana, Kazakhstan,
Email: [email protected]
3
Professor, Dr., Ankara Yildirim Beyazit University, Ankara, Turkey,
Email: [email protected]
Abstract— Direct current (DC) motors are widely utilized in Practically, applying improved LQR control using fuzzy
various industries due to their simplicity, reliability, and ease of logic and Kalman filter controller can result in enhanced
control. Achieving precise and efficient control of DC motors performance, reduced energy consumption, and increased
has been the subject of extensive research, leading to the system reliability. However, it is essential to note that
development of various control techniques, including linear implementing these control systems may necessitate additional
quadratic regulator (LQR) control. However, more than LQR hardware and software, which can impact the overall cost of the
control is required to handle the non-linearities and system. Therefore, careful consideration should be given to the
uncertainties often encountered in DC motor control specific requirements and constraints of the target application.
applications. This has prompted the exploration of enhanced
control strategies by combining LQR with other techniques, In conclusion, this study demonstrates the efficacy of
such as fuzzy logic and Kalman filtering. integrating LQR control with fuzzy logic and Kalman filter
controller for DC motor control. Combining these techniques
This study focuses on the integration of LQR control with improves control performance and robustness, enabling precise
fuzzy logic and Kalman filter controller to improve the and efficient management of DC motors in various industrial
performance of DC motor control systems. The objective is to applications. The findings contribute to the advancement of
evaluate the effectiveness of this combined approach in terms of control strategies for DC motors, providing valuable insights for
control performance, robustness, and applicability in different researchers, engineers, and practitioners in the field. Future
industrial settings. research could explore further enhancements and optimizations
of the integrated approach, considering additional factors such
The proposed control systems are implemented and as system dynamics and external disturbances to refine DC
simulated on the Simulink platform to accomplish this. A motor systems' control performance.
comparative analysis is conducted to assess the performance of
the integrated LQR with fuzzy logic and Kalman filter Keywords — DC motors, control techniques, linear quadratic
controller against using each technique separately. The results regulator (LQR), fuzzy logic control, Kalman filter control,
provide insights into the advantages and limitations of the performance evaluation, control performance, robustness, non-
combined approach and its potential for enhancing the control linearities, uncertainties, noise, disturbances
of DC motors.
I. INTRODUCTION
The findings demonstrate that the integration of LQR with DC motors are widely used in various applications such as
fuzzy logic and Kalman filtering leads to improved control robotics, automation, aerospace, and automotive industries
performance and robustness compared to individual
due to their simplicity, reliability, and ease of control. In the
techniques. Fuzzy logic is well-suited for handling non-
field of robotics, DC motors are used to power the movement
linearities and uncertainties, allowing for better adaptation to
varying operating conditions. In addition, the Kalman filter
of robotic arms, grippers, and other components [1]. In the
estimator provides accurate state estimation, even in the automation industry, DC motors are used in conveyor belts,
presence of noise and disturbances. By leveraging this estimated pumps, and other machinery [2]. In the aerospace industry,
state information, the LQR controller can generate control DC motors are used in aircraft systems such as flight control
inputs that optimize the behavior of the DC motor system. surfaces, landing gear, and navigation systems. Similarly, in
the automotive industry, DC motors are used in power
Combining LQR, fuzzy logic, and Kalman filter control windows, windshield wipers, and other subsystems. In order
techniques exhibits several benefits for DC motor control to achieve precise and efficient control of DC motors, various
systems. The integrated approach enhances the precision and control techniques have been developed over the years [3].
stability of control, making it particularly suitable for
applications that require precise control of DC motors. One popular approach is the linear quadratic regulator
Industries such as robotics, automotive, aerospace, and (LQR) control [4], which is a feedback control technique that
manufacturing stand to benefit from implementing these minimizes a cost function while optimizing the system's
advanced control systems. performance. LQR control has been applied to various DC
motor systems, allowing for stable and efficient operation.
However, LQR control assumes that the system is linear and
- 39 -
time-invariant [5], which is not always the case in practice. Types of DC Motors
Additionally, LQR control can be sensitive to measurement
noise, which can affect the stability and performance of the DC motors can be categorized into two types: brushed DC
system [6]. [15] motors and brushless DC motors [16]. Brushed DC
motors are commonly used in low-cost applications where
To overcome these limitations, other control techniques speed control is not critical. Brushless DC motors are used in
have been proposed. One such approach is the use of fuzzy high-performance applications where high-speed and precise
logic control [7], which allows for the management of control are required.
uncertainties and nonlinearities in the system [8]. Fuzzy logic
is a mathematical method that uses linguistic variables to Principles of Operation
describe the behavior of a system [9], making it particularly DC motors operate based on the interaction between
useful for DC motor control where nonlinearities are common magnetic fields. The interaction between the magnetic field of
[10]. the stator and the magnetic field of the rotor generates torque,
Another approach is the use of a Kalman filter estimator, which causes the rotor to rotate. The speed and direction of
which accurately estimates the system states and reduces the rotation can be controlled by varying the voltage or current
impact of measurement noise on the control system. Kalman supplied to the motor.
filter estimators are commonly used in DC motor control Control Techniques
systems to improve stability and performance.
Linear Quadratic Regulator (LQR) is a popular control
The goal of this report is to evaluate and compare the technique used in DC motor control. LQR is a feedback
performance of LQR, fuzzy logic, and Kalman filter control control technique that minimizes a cost function while
techniques for DC motor control. The proposed control optimizing the system's performance. LQR control has been
systems will be implemented and simulated on the Simulink applied to various DC motor systems [17], allowing for stable
platform, and their performance will be compared to and efficient operation. However, LQR control assumes that
determine the most effective technique for DC motor control. the system is linear and time-invariant, which is not always
the case in practice. Additionally, LQR control can be
II. LITERATURE REVIEW AND PROBLEM STATEMENT sensitive to measurement noise, which can affect the stability
Despite the widespread use of DC motors in various and performance of the system [18].
applications, achieving precise and efficient control of these
The block diagram (Fig.2) of the LQR control system is
motors can be challenging due to nonlinearity and
composed of three main parts: the plant, the reference input,
measurement noise. While linear quadratic regulator (LQR)
and the LQR controller. The plant represents the system to be
control is a popular control technique, it has limitations when
controlled, in this case, a DC motor, and is modeled by a
applied to nonlinear and time-varying systems. As such,
transfer function. The reference input is the desired output or
alternative control techniques such as fuzzy logic and Kalman
setpoint, which is fed into the controller. The LQR controller
filter estimators have been proposed. However, there is a need
takes the difference between the reference input and the output
to compare the performance of these techniques for DC motor
of the plant and calculates the control signal to be applied to
control and determine the most effective approach. This
the plant. The control signal is also fed back to the controller
research aims to address this need by evaluating and
as feedback, which is used to adjust the controller's
comparing the performance of LQR, fuzzy logic, and Kalman
calculations.
filter control techniques for DC motor control and identifying
the most effective approach for achieving precise and efficient
control.
DC motors (Fig.1) are widely used in various industries
due to their simplicity, reliability, and ease of control. The
performance of DC motors [12] can be improved by using
advanced control techniques such as LQR, fuzzy logic, and
Kalman filter. This chapter provides an overview of the
literature on DC motor control, including the types of DC
motors, the principles of operation, and the methods of
control. Furthermore, this chapter reviews the previous Figure 2. LQR block diagram
research conducted on LQR, fuzzy logic [13], and Kalman
filter control techniques for DC motor control [14]. Fuzzy Logic Control
Fuzzy logic is a mathematical method that uses linguistic
variables to describe the behavior of a system [19]. Fuzzy
logic allows for the management of uncertainties and
nonlinearities in the system, making it particularly useful for
DC motor control where nonlinearities are common [20, 21].
Fuzzy logic controllers can operate effectively even when the
mathematical model of the system is unknown or uncertain
[22].
Key Components of FLC
Fuzzy logic control consists of five main components, as
Figure 1. DC motor illustrated in Fig.3 presented in [3]. The components are part
- 40 -
of a fuzzy logic system, which is a type of control system that • To determine which of the three proposed control
uses fuzzy set theory to model and control complex systems techniques is the most suitable for DC motor control
with uncertain and imprecise information. The components of based on their performance and stability in simulated
a fuzzy logic system include a fuzzifier, a rule base, an environments.
inference engine, a defuzzifier, and an output interface. These
components work together to convert input data into crisp The results of this study will provide valuable insights into the
strengths and limitations of each control technique and inform
output data, allowing the system to make decisions and take
future research on the development of advanced control
actions based on uncertain and imprecise information.
techniques for DC motor applications. The findings of this
study may also have practical implications for various
industries that use DC motors, such as robotics, automation,
aerospace, and automotive industries, by providing guidance
for the design and implementation of control systems that can
improve the performance and stability of DC motor systems.
The scope of the project
The scope of this project is to evaluate and compare the
performance of three different control techniques, namely
LQR, fuzzy logic, and Kalman filter [11], for DC motor
Figure 3. Fuzzy Logic control schematic control. The focus will be on the simulation and
implementation of these control systems using the Simulink
platform. The project aims to determine the most effective
Kalman Filter Estimator technique for achieving precise and efficient control of DC
Kalman filter estimator is a control technique that motors. The results of this study can be used to improve the
accurately estimates the system states and reduces the impact performance of DC motor control systems in various
of measurement noise on the control system [23]. Kalman applications such as robotics, automation, aerospace, and
filter estimators are commonly used in DC motor control automotive industries. However, it is important to note that
systems to improve stability and performance [24]. Kalman this project does not involve the physical implementation of
filter estimators can accurately estimate the state of the system the control systems on real DC motors. Therefore the findings
even when the measurements are noisy [25]. should be interpreted in the context of simulation-based
studies.
IV. MATERIALS AND METHODS
The simulation setup consists of a DC motor model and a
control system model. The DC motor model is based on
mathematical equations that describe the behavior of a DC
motor. The control system model is based on the LQR, fuzzy
logic, or Kalman filter control techniques.
The DC motor model includes the following parameters:
the resistance of the armature, the inductance of the armature,
Figure 4. Kalman Filter Block Diagram the back-emf constant, the moment of inertia of the rotor, the
viscous damping coefficient, and the mechanical load torque.
Previous Research on DC Motor Control Techniques The control system model includes the following parameters:
the gain matrix for LQR control, the membership functions
Previous research has been conducted on the performance
and rules for fuzzy logic control, and the Kalman filter gain
evaluation and comparison of LQR, fuzzy logic [26], and
matrix for Kalman filter control.
Kalman filter control techniques for DC motor control.
Several studies have shown that fuzzy logic and Kalman filter The simulation is performed using Simulink, which is a
controllers outperform LQR controllers in terms of stability software tool for modeling, simulating, and analyzing
and performance in certain scenarios. However, the results are dynamic systems. Simulink allows for the simulation of
highly dependent on the specific application and system under complex systems using block diagrams, which represent the
consideration. components of the system and their interactions. The Simulink
model for the DC motor control system includes blocks for the
III. THE AIM AND OBJECTIVES OF THE STUDY DC motor model, the controller model, and the inputs and
The aim of this study is to evaluate and compare the outputs of the system.
performance of three different control techniques, namely
Mathematical Models of the DC Motor
linear quadratic regulator (LQR), fuzzy logic, and Kalman
filter, for DC motor control and determine the most suitable The armature voltage equation is given by:
approach for achieving precise and efficient control of DC
motors. To achieve this aim, the following objectives are The armature voltage equation (1) describes the voltage
accomplished: applied to the armature of the DC motor. The equation states
that the armature voltage (𝑉! ) is equal to the sum of three
• To compare the effectiveness of three different control terms: the product of the armature resistance (𝑅! ) and the
techniques, namely LQR, fuzzy logic, and Kalman armature current (𝐼! ), the product of the armature inductance
filter, in terms of their ability to control a DC motor. (La) and the time derivative of the armature current, and the
- 41 -
back electromotive force (EMF) of the motor ( 𝐸! ). The 𝐾$ 𝐼! (𝑠) = 𝐽𝜔(𝑠) + 𝐵𝜔(𝑠) (8)
equation captures the dynamics of the armature voltage and its
dependence on the armature current and the back EMF of the
motor. If current is obtained from (8) and substituted in (9), we
𝑑𝐼! ($) (1) will have,
𝑉! 𝑡 = 𝑅! 𝐼! (𝑡) + 𝐿! . + 𝐸! (𝑡)
𝑑𝑡
1 (9)
𝑉! (𝑠) = 𝜔(𝑠) [𝐿 𝐽𝑠 # + 𝑅! 𝐽 + 𝐿! 𝐵(𝑠) + 𝐾$ . 𝐾" ]
𝐾" !
Equation for back emf of motor will be:
Equation (9) is obtained by solving equation (8) for the
The back EMF equation (2) relates the back EMF of the armature current and substituting it into equation (7). The
DC motor (𝐸& ) to its angular velocity (ω). The equation states resulting equation relates the Laplace transform of the
that the back EMF is proportional to the angular velocity with armature voltage to the Laplace transform of the angular
a constant of proportionality (𝐾' ). The equation captures the velocity.
physical relationship between the back EMF and the angular
velocity of the motor. Then the relationship between rotor shaft speed and
applied armature voltage is represented by the transfer
function (10):
𝐸& (𝑡) = 𝐾' 𝜔(𝑡) (2)
𝜔(𝑠) 𝐾𝑡 (10)
=
𝑉𝑎 (𝑠) (𝐽𝐿𝑎 𝑠2 + $𝐽𝑅𝑎 + 𝐵𝐿𝑎 %𝑠 + (𝐾𝑡 . 𝐾𝑏 + 𝐵𝑅𝑎 )
The torque balance equations (3)-(4) relate the torque
developed by the motor (𝑇( ) to its armature current (𝐼! ) and
angular velocity (ω). Equation 3 states that the torque is
proportional to the armature current with a constant of Finally, equation (10) is obtained by solving equation (9)
proportionality (𝐾$ ). Equation 4 states that the torque is equal for the transfer function relating the angular velocity and the
applied armature voltage. The transfer function captures the
to the product of the motor's moment of inertia (𝐽) and the time
relationship between the rotor shaft speed and the applied
derivative of the angular velocity plus a damping term (𝐵)
armature voltage in the frequency domain.
times the angular velocity. These equations capture the
physical relationship between the torque developed by the Consider the following values for the physical parameters
motor, the armature current, and the angular velocity. [27, 28]
The transfer function is as (11):
𝑇( (𝑡) = 𝐾$ 𝐼! (𝑡) (3)
𝜔(𝑠) 2 (11)
=
𝑑𝜔(𝑡) (4) 𝑉! (𝑠) 𝑠 ) + 12𝑠 + 20.02
𝑇( (𝑡) = 𝐽 + 𝐵𝜔(𝑡)
𝑑𝑡
The mathematical models used for the design and
Let us combine the upper equations together, implementation of the LQR, fuzzy logic, and Kalman filter
controllers are based on the state-space representation of the
Equation (5) is a reformulation of the armature voltage system. The state-space representation is a mathematical
equation that incorporates the back EMF term. Equation (6) is framework for describing the behavior of a system in terms of
a reformulation of the torque balance equation that relates the its states and inputs [29].
torque to the angular velocity and the armature current.
The state-space representation of the DC motor control
system is given by the following equations (12)-(15):
𝑑𝐼! ($) (5)
𝑉! 𝑡 = 𝑅! 𝐼! (𝑡) + 𝐿! . + 𝐾' 𝜔(𝑡)
𝑑𝑡 𝑥̇(𝑡) = 𝐴𝑥(𝑡) + 𝐵𝑢(𝑡) 𝑦(𝑡) = 𝐶𝑥(𝑡) (12)
𝑑𝜔(𝑡) (6)
𝐾$ 𝐼! (𝑡) = 𝐽 + 𝐵𝜔(𝑡) The system under consideration can be represented in
𝑑𝑡 state-space form, where x(t) denotes the state vector, u(t)
represents the input vector, and y(t) stands for the output
vector. The system's behavior can be described by a set of
Equations (7) and (8) are obtained by taking the Laplace
differential equations in which A is the state matrix, B is the
transform of equations 5 and 6, respectively. These equations
input matrix, and C is the output matrix. Specifically, the state
are in the frequency domain and relate the Laplace transform
vector, x(t), is comprised of the angular velocity and the
of the armature voltage and armature current to the Laplace
current of the DC motor, while the input vector, u(t),
transform of the angular velocity and torque, respectively.
represents the voltage applied to the DC motor. Finally, the
𝑉! (𝑠) = 𝑅! 𝐼! (𝑠) + 𝐿! . 𝐼! (𝑠) + 𝐾' 𝜔(𝑠) (7) output vector, y(t), denotes the angular velocity of the DC
motor, as reported in [30].
- 42 -
The state-space representation of the LQR control system The steady-state Kalman filter estimates the state variable
is given by the following equations: 𝑥[𝑛] using two main equations, Equations 18 and 19.
Equation 18 is the prediction equation, which estimates the
𝑥̇(𝑡) = (𝐴 − 𝐵𝐾)𝑥(𝑡) 𝑢(𝑡) = −𝐾𝑥(𝑡) (13)
state variable 𝑥[𝑛 + 1] based on the past measurements of the
system up to 𝑦[𝑛 − 1], and the input 𝑢[𝑛] and Gaussian noise
𝑤[𝑛] that affect the system dynamics. This estimate is updated
, where K is the gain matrix for LQR control. The gain over time as new measurements become available.
matrix is calculated using the Riccati equation, which is a
mathematical formula that solves for the optimal gain matrix Equation 19 is the update equation, which incorporates the
based on the system's parameters and a cost function. latest measurement 𝑦[𝑛] into the state estimate 𝑥[𝑛|𝑛 − 1] to
obtain a new estimate of the state variable 𝑥[𝑛|𝑛]. The
The state-space representation of the fuzzy logic control Optimal Innovation gains Mx and My are chosen to minimize
system [31, 32] is given by the following equations: the steady-state covariance of the estimation error and remove
noise from the estimate.
𝑢(𝑡) = 𝑓(𝑥(𝑡)) (14)
𝑥/[𝑛 + 1|𝑛] = 𝐴𝑥/[𝑛|𝑛 − 1] + 𝐵𝑢[𝑛] + 𝐺𝑤[𝑛] (18)
, where f(x(t)) is the fuzzy inference system [35], which
maps the input state vector x(t) to the output control signal The developed Kalman filter architecture will be obtained
u(t). The fuzzy inference system consists of membership
as:
functions[40], which describe the degree to which a given
input value belongs to a particular linguistic variable, and
rules [41], which describe the relationship between the input
𝑥 ̂[𝑛|𝑛] = 𝑥 ̂[𝑛|𝑛 − 1]𝑀' (𝑦[𝑛] − 𝐶𝑥 ̂[𝑛│𝑛 − 1] (19)
and output variables [42].
− 𝐷𝑢[𝑛])
The state-space representation of the Kalman filter control
system is given by the following equations [25, 28]:
Equation 20 calculates the estimated output of the system,
𝑦[𝑛|𝑛], based on the estimated state variable 𝑥[𝑛|𝑛 − 1] and
𝑥̂(𝑡) = 𝐴𝑥̂(𝑡 − 1) + 𝐵𝑢(𝑡) + 𝐾(𝑦(𝑡) − 𝐶𝑥̂(𝑡 − 1)) 𝑢(𝑡) (15) the input 𝑢[𝑛]. The filter can be implemented recursively, and
= −𝐾𝑥̂(𝑡)
the estimated state variable and output are updated with each
new measurement of the system output.
, where x̂(t) is the estimated state vector, K is the Kalman
filter gain matrix, and y(t) is the measured output vector. 𝑥̂𝑦:[𝑛|𝑛] = 𝐶𝑥:[𝑛|𝑛 − 1] + 𝐷𝑢[𝑛] + 𝑀% (𝑦[𝑛] − 𝐶𝑥 ̂[𝑛│𝑛 − 1] (20)
Kalman Filter − 𝐷𝑢[𝑛])
The given equations describe a steady-state Kalman filter,
which is a mathematical algorithm that can be used to estimate LQR Controller
the state of a dynamic system from noisy measurements. The
filter is used to calculate the genuine plant output, 𝑦𝑡[𝑛], by The Linear Quadratic Regulator (LQR) methodology is a
estimating the state variable 𝑥[𝑛] of the system in real-time popular control design approach that is used to design
using noisy measurements 𝑦[𝑛] and input 𝑢[𝑛]. The objective feedback controllers for linear systems. The goal of the LQR
is to obtain an accurate estimate of the state variable 𝑥[𝑛] that methodology is to minimize a given cost function J, which is
minimizes the error between the estimated and true values of a function of the system states and control inputs. The cost
the state variable. function 𝐽 is defined in equation (21), where 𝑄 is the
weighting matrix of the system states and 𝑅 is the weighting
The true plant update equation (Equation 16) describes the matrix of the control inputs.
evolution of the system state 𝑥[𝑛] over time, where A, B, and
G are matrices that represent the dynamics of the system, u[n]
is the input to the system, and 𝑤[𝑛] is Gaussian noise that +
(21)
affects the system dynamics. 𝐽 = J ( 𝑥 $ 𝑄 𝑥 + 𝑢* 𝑅 𝑢) 𝑑𝑡
,
The LQR controller is given by. where 𝑢 is the control
𝑥[𝑛 + 1] = 𝐴𝑥[𝑛] + 𝐵𝑢[𝑛] + 𝐺𝑤[𝑛] (16) input and 𝑥 is the system state (22):
Equation 17 represents the measurement of the system 𝑢 = − 𝐾𝑥 (22)
output, 𝑦[𝑛], which is affected by both the state variable 𝑥[𝑛]
and the input 𝑢[𝑛], as well as measurement noise 𝑣[𝑛] and
Gaussian noise 𝑤[𝑛]. To design an LQR controller, the first step is to select
appropriate values for the weighting matrices 𝑄 and 𝑅. The
choice of 𝑄 and 𝑅 determines the relative importance of the
𝑦[𝑛] = 𝐶𝑥[𝑛] + 𝐷𝑢[𝑛] + 𝐻𝑤[𝑛] + 𝑣[𝑛] (17) system states and control inputs in the cost function 𝐽. The
value of 𝑅 should be chosen to weigh the control inputs more
- 43 -
than the system states, while the value of 𝑄 should be chosen represents a combination of input variables. The output
to weigh the system states more than the control inputs. variable is represented in the cell corresponding to the row and
column.
Once the weighting matrices 𝑄 and 𝑅 have been selected,
the next step is to compute the feedback gain matrix 𝐾 using TABLE II. RULES OF FUZZY LOGIC CONTROLLER
the solution of the discrete algebraic Riccati equation. The
feedback gain matrix 𝐾 is given by equation (23), where P* is NB NM NS Z PS PM PB
NB NB NB NB NB NM NS Z
the positive semidefinite matrix that is the unique solution to
NM NB NB NB NM NS Z PS
the Riccati equation (24). NS NB NB NM NS Z PS PM
Then the feedback 𝐾 can be computed and the closed loop Z NB NM NS Z PS PM PB
PS NM NS Z PS PM PB PB
system responses can be found by simulation.
PM NS Z PS PM PB PB PB
The gain matrix 𝐾 which solves the LQR problem is PB Z PS PM PB PB PB PB
For example, consider the row where the input variable is
𝐾 = 𝑅-. 𝐵* 𝑃∗ (23) Negative Small (NS). The columns in this row specify the
output for different combinations of the input variables. The
output for NS-NS combination is Negative Big (NB), for NS-
Where 𝑃∗ is unique, positive semidefinite solution to the NM combination is Negative Big (NB), for NS-Z combination
Riccati equation. is Negative Medium (NM), for NS-PS combination is Zero
(Z), and so on.
Similarly, each row in the table corresponds to a different
𝐴* 𝑃 + 𝑃𝐴 – 𝑃𝐵𝑅-. 𝐵* 𝑃 + 𝑄 = 0 (24)
input variable, and each column corresponds to a different
combination of input variables. The rules in the table are
designed based on the expert knowledge of the control system
The Riccati equation (24) is a matrix equation that must be and the desired control behavior. The Fuzzy Logic Controller
solved to compute the value of 𝑃∗ . This equation relates the uses these rules to determine the appropriate control action
value of 𝑃∗ to the system matrices A, B, Q, and R. The matrix based on the input variables received from the system.
𝑃∗ determines the cost function J, and its value is crucial for
computing the feedback gain matrix K. Input and Outputs
Once the feedback gain matrix 𝐾 has been computed using The design of the fuzzy logic controllers includes two
equation (23), the control input 𝑢 is computed using equation input variables, namely the error and the change of error, and
(22). The control input 𝑢 = −𝐾𝑥 is the optimal control input one output variable (Fig.5).
that minimizes the cost function 𝐽.
In summary, the LQR methodology is a control design
approach that is used to design feedback controllers for linear
systems. The controller is designed by selecting appropriate
values for the weighting matrices 𝑄 and 𝑅 , computing the
positive semidefinite matrix 𝑃∗ from the solution of the Riccati
equation, and then computing the feedback gain matrix 𝐾
using equation (23). Finally, the control input u is computed
using equation (22), and it is the optimal control input that
minimizes the cost function 𝐽. Figure 5. Membership function of inputs of Fuzzy Logic system
Table 1 shows the linguistic variables used in the fuzzy
logic system for DC motor control. These variables include In the Mamdani-based fuzzy inference system, linear
Negative Big (NB), Negative Medium (NM), Negative Small membership functions are employed for both the inputs and
(NS), Zero (Z), Positive Small (PS), Positive Medium (PM), outputs. The fuzzy logic controller utilizes triangular
and Positive Big (PB). membership functions for the input variables, which are the
error (e) and the rate of change of error (∆e), while the output
TABLE I. LINGUISTIC VARIABLES FOR FUZZY SYSTEM variable is the controller output (∆e), which can be seen in the
Fig.6.
NB NM NS Z PB PM PS
Big Medium Small Zero Big Medium Small
Negative Negative Negative positive Positive Positive
The rules in Table 2 represent the mapping between the
input linguistic variables and the output linguistic variables
used in the Fuzzy Logic Controller for DC motor control. The
rules specify the control actions that the controller should take
based on the input variables. Each row and column of the table Figure 6. Membership function of Output of fuzzy system
- 44 -
Each variable is divided into 7 membership functions, and error covariance output is a measure of the uncertainty in the
the ranges for the error and output are [-120, 120] and [-180, estimated state.
180], respectively. Table 3 provides the rule base for the DC
motor, and Fig. 10 provides a more detailed view of the fuzzy The Kalman filter block in Simulink allows for the
specification of the noise covariance matrices, which are
system rules, allowing for a deeper understanding of how the
important parameters in determining the filter's performance.
system operates.
In particular, the sensor noise covariance and the input plant
V. EXPERIMENTAL RESULTS AND SIMULATION noise covariance matrices play an important role in
determining the amount of noise rejection achieved by the
DC Motor Control using Kalman Filter filter.
The Simulink block model for DC motor speed control by The sensor noise covariance matrix represents the
Kalman filter is presented in Fig.7. The system output measurement noise, which is the noise that is present in the
response for DC motor speed is measured in rad/sec, and the measured output of the system. The value of the sensor noise
gains RPM to rad/sec are used. The obtained results for the covariance lies in the variance range of 0 to 1. A value of 1
speed output response are then converted back to RPM speed corresponds to maximum sensor noise rejection, meaning that
response. The system model is defined in the Transfer the filter is giving more weight to the predicted state rather
function block. than the measurements. In other words, the filter trusts the
To account for sensor noise, a sensor noise covariance is model more than the measurements. On the other hand, a
introduced with a value between 0 and 1, which represents the value of 0 corresponds to no sensor noise rejection, meaning
variance range. The maximum sensor noise rejection effect is that the filter is giving equal weight to the measurements and
achieved when the value is set to 1. However, the impact of the predicted state. In this case, the filter is trusting the
noise rejection can be observed in the input plant noise measurements more than the model.
covariance Q, where larger values result in a minimal noise
rejection effect.
The model in the Simulink environment is utilized for
controlling the speed of the DC motor using a Kalman filter.
The state space domain is used to specify the system model.
By employing the gains RPM to rad/sec, the system output
response is translated to rad/sec, and the results are then
converted back to RPM speed response.
In conclusion, the Simulink model for DC motor speed
control by Kalman filter offers a robust and accurate approach
to controlling the motor's speed. By introducing the sensor
noise covariance and input plant noise covariance, the impact
of noise on the system response can be minimized. The model
can be further optimized by adjusting the parameters of each
block to suit the specific application.
Figure 8. Kalman Filter block on Simulink
The graph (Fig.9) shows two lines: a blue line representing
the reference speed generated by the signal builder and a red
line representing the actual speed of the DC motor. The
motor's speed is controlled using the Kalman filter. The graph
illustrates that the actual speed closely follows the reference
speed, indicating that the Kalman filter has been successful in
controlling the DC motor speed.
The results indicate that the Kalman filter is effective in
controlling the speed of the DC motor. The low overshoot of
2.542% indicates that the Kalman filter is successful in
Figure 7. Simulink model of Kalman filter Controller for DC Motor accurately controlling the motor speed without introducing
significant deviation from the desired setpoint. This suggests
that the Kalman filter has been successful in accurately
Kalman filter Simulink block estimating the state of the DC motor and generating control
inputs that minimize the error between the reference speed and
The Kalman filter block in Simulink (Fig.8) is based on the actual speed.
the Kalman filter algorithm. The block has two main inputs:
the measurement input and the control input. The Furthermore, the rise time of the Kalman filter in
measurement input is the signal that is measured from the controlling the DC motor speed of 26.829 seconds is relatively
system, while the control input is the signal that is applied to short, especially if we consider that the reference speed itself
the system. The block also has two main outputs: the estimated has a rise time of 27 seconds. This indicates that the Kalman
state output and the estimation error covariance output.The filter has been successful in closely tracking the reference
estimated state output provides the best estimate of the current speed with a relatively short delay. Therefore, the Kalman
state of the system based on the input data. The estimation
- 45 -
filter is well-suited for applications that require accurate observe the performance of the LQR controller in tracking the
control of the DC motor speed with a short delay. reference speed and maintaining the desired speed of the
motor.
However, it is important to note that the rise time
requirement may vary depending on the specific application. Overall, the Simulink model of the DC motor speed
For applications that require a faster response time, the rise control with LQR controller is an effective tool for studying
time of the Kalman filter may need to be further optimized. the implementation of LQR control in a practical application.
Additionally, the performance of the Kalman filter may be It allows the user to observe the performance of the controller
affected by external disturbances or uncertainties in the in real-time and provides a visual representation of the system
system. Therefore, it is crucial to evaluate the controller's response.
performance in the context of the specific application
requirements and optimize its parameters accordingly.
In summary, based on these results, the Kalman filter is
effective in controlling the DC motor speed and is well-suited
for applications that require accurate control with a relatively
short delay. However, its performance should be evaluated in
the context of specific application requirements and optimized
accordingly.
Figure 10. Simulink model of LQR Controller for DC Motor
The performance of the LQR (Linear Quadratic Regulator)
Controller in controlling the speed of a DC motor is evident
in the graph. The graph (Fig.11) consists of two lines: the
black line represents the reference speed generated by the
signal builder, and the red line represents the actual speed of
the DC motor after being controlled by the LQR Controller.
Figure 9. Speed Response of DC motor with Kalman Filter Controller
DC motor control using LQR controller
The Simulink model of the DC motor speed control with
LQR controller is designed to demonstrate the implementation
of LQR control in a practical application. The model (Fig.10)
consists of four main parts: the reference speed signal, the DC
motor, the LQR controller, and the scope.
Figure 11. Speed Response of DC motor with LQR Controller
The reference speed signal is generated using the Signal
Builder block, which allows the user to define a custom signal
profile for the desired DC motor speed. This reference signal Based on the results, the LQR (Linear Quadratic Regulator)
is then fed into the LQR controller. controller is shown to have an overshoot of 4.237% and a rise
time of 26.873 seconds.
The DC motor model is implemented using the Transfer
function block in Simulink. The block provides a realistic DC The overshoot percentage indicates the degree of deviation
motor model, including the motor dynamics, of the controlled variable (DC motor speed) from its steady-
electromechanical behaviour, and back emf. The input to the state value before stabilizing. In this case, the overshoot
DC motor block is the control signal from the LQR controller, percentage of 4.237% suggests that the LQR controller has
and the output is the motor's actual speed. allowed some degree of overshoot before stabilizing the DC
motor speed. However, it is worth noting that an overshoot
The LQR controller is implemented using the LQR percentage of 4.237% is relatively small and may not cause
Control block from the Control System Toolbox in Simulink. significant problems in many applications.
The block allows the user to define the state-space model of
the system, the weighting matrices for the state and control The rise time indicates the time it takes for the controlled
inputs, and the desired closed-loop poles. The output of the variable to reach a steady-state value after the controller is
LQR controller is the control signal that is fed into the DC engaged. The rise time of 26.873 seconds suggests that the
motor block. LQR controller takes a relatively long time to stabilize the DC
motor speed. In many applications, a long rise time may be
Finally, the scope block is used to plot the actual speed of undesirable, particularly in systems that require fast response
the DC motor as a function of time. This allows the user to times.
- 46 -
Overall, the results suggest that the LQR controller is rise time of 27 seconds itself. The graph shows that the actual
effective in controlling the DC motor speed, albeit with some speed of the DC motor closely follows the reference speed,
limitations. The small overshoot percentage indicates that the indicating that the fuzzy logic controller has been successful
controller can maintain the motor speed close to its desired in controlling the DC motor speed.
value, while the relatively long rise time suggests that the
controller's response time may not be fast enough in some However, compared to the other two controllers, LQR and
applications. Therefore, it is essential to evaluate the Kalman filter, the fuzzy logic controller has a higher percent
controller's performance in the context of the specific overshoot and a longer rise time. This suggests that the fuzzy
application requirements and optimize its parameters logic controller may not be the most effective controller in
accordingly. this scenario. While it has shown success in controlling the
DC motor speed, it may not be the optimal choice for
DC Motor control using Fuzzy Logic achieving the desired speed response with minimal overshoot
and fast rise time.
To implement fuzzy logic control in a DC motor speed
control system, a Simulink model can be used. The model Overall, the effectiveness of the fuzzy logic controller for
consists of a signal builder block to generate a reference speed DC motor speed control depends on the specific application
signal and a fuzzy logic block to control the DC motor model. and the desired performance criteria. While it may not be the
The DC motor model can be represented by a transfer most effective choice in this scenario, it may still be a viable
function, which relates the input voltage to the output speed. option in other situations where different performance criteria
or constraints are present.
The fuzzy logic block consists of three main parts: the
fuzzification module, the rule base, and the defuzzification
module. The fuzzification module takes the input variables
and maps them to fuzzy sets, which represent the degree of
membership of each variable in each set. The rule base
contains a set of fuzzy logic rules, which map the input
variables to the output variable. The defuzzification module
takes the output of the rule base, which is a set of fuzzy sets,
and maps them to a crisp output value. This value is then used
as the input to the motor controller.
The results of the DC motor speed control using fuzzy logic
can be visualized using a scope block in Simulink. The scope
block displays the output speed of the motor in response to
the reference speed signal generated by the signal builder
block. The performance of the fuzzy logic controller can be
evaluated by comparing the actual performance of the motor
to the desired performance and adjusting the fuzzy rules and Figure 13. Speed Response of DC motor with Fuzzy Logic Controller
input variables as needed to improve the performance.in
Fig.12 the Simulink model of DC motor speed control using
fuzzy logic can be seen. VI. CONCLUSION
Based on the results obtained from the three controllers,
namely Kalman filter, LQR, and Fuzzy Logic, it can be
concluded that the Kalman filter controller outperforms the
other two controllers in terms of overshoot and rise time. The
Kalman filter controller resulted in the least overshoot
(2.542%) and a comparable rise time (26.829 seconds) to the
other two controllers, making it more efficient in controlling
the DC motor speed.
The LQR controller also performed well with an overshoot
of 4.237% and a rise time of 26.873 seconds, but it was
outperformed by the Kalman filter controller. The Fuzzy
Logic controller had the highest overshoot of 6.034% and a
rise time of 26.954 seconds, making it the least effective
controller of the three tested (Table 3).
Figure 12. Fuzzy Logic controller for DC motor
TABLE III. PERFORMANCE COMPARISON OF KALMAN, LQR, AND
FUZZY LOGIC CONTROLLER
The graph (Fig.13) illustrates the fuzzy logic controller in
controlling the speed of a DC motor. The reference speed Controllers Overshoot (%) Rise Time (sec)
generated by the signal builder is shown as the black line,
while the red line represents the actual speed of the DC motor Kalman 2.542 26.829
after being controlled by the fuzzy logic controller.
LQR 4.237 26.873
The rise time of the fuzzy logic controller for DC motor
speed control is 26.954 seconds, while the percent overshoot Fuzzy Logic 6.034 26.954
is 6.034%. The reference speed used in this experiment has a
- 47 -
Therefore, the main goal of this study, which was to network-based control, which may have better performance
evaluate and compare the performance of three different under certain conditions.
control techniques for DC motor control, has been achieved.
The study has shown that the Kalman filter controller is the In conclusion, this study provides valuable insights into
the performance of three different controllers for DC motor
most suitable approach for achieving precise and efficient
speed control. The results suggest that the Kalman filter
control of DC motors.
controller is the most effective controller for this application,
The findings of this study provide valuable insights into but further research is needed to validate the results under
the strengths and limitations of each control technique, and different operating conditions and to compare the proposed
they may also have practical implications for various method with other control strategies. Overall, the study
industries that use DC motors, such as robotics, automation, contributes to the advancement of the field of control
aerospace, and automotive industries, by providing guidance engineering and provides useful information for practitioners
for the design and implementation of control systems that can and researchers alike.
improve the performance and stability of DC motor systems.
REFERENCES
In conclusion, the Kalman filter controller is
[1] M. Ruderman, J. Krettek, F. Hoffmann, T. Bertram, "Optimal State
recommended for controlling the speed of DC motors, but the Space Control of DC Motor", Proceedings of the 17th World Congress,
choice of controller may also depend on other factors such as The International Federation of Automatic Control, Seoul, Korea, July
the complexity of the system, computational resources, and 6-11, 2008. pp. 5796- 5801.
ease of implementation. [2] Katsuhiko Ogata: Modern Control Engineering; Prentice Hall
International, Inc. Fourth Edition 2002. [8] Rashid M.H., ‘Power
VII. DISCUSSION Electronics’, Prentice Hall, 1998.
[3] Aditya Pratap Singh, “Speed Control of DC Motor Design and E
In this study, we aimed to evaluate and compare the Controller Based on Matlab” Innovative Systems.
performance of three different controllers for DC motor speed [4] Nur Iznie Afrah Binti Mohdisa “DC motor controller using Linear
control, namely LQR, fuzzy logic, and Kalman filter. Our Quadratic Regulator (LQR) Algorithm implementation on PIC”,
results showed that the Kalman filter controller outperformed Bachelor of Electrical Engineering (Hons.) (Electronics), Universiti
the other two controllers in terms of overshoot and rise time, Malaysia Pahang, Nov 2008.
making it the most effective controller for DC motor speed [5] Che Ku Mohd Faizul “Linear Quadratic Regulator (LQR) speed control
control in our experiments. for DC motor using MC68HC11, Bachelor of Electrical Engineering
(Hons.) (Electronics), Universiti Malaysia Pahang, Nov 2008.
One possible explanation for the superior performance of [6] Ali M. Yousef and M. K. El-Sherbiny “Improvement of Synchronizing
the Kalman filter controller is that it uses a model-based and Damping Torque Co-efficients Based LQR Power Systemv
approach to estimate the state of the system and then adjust Stabilizer,” Proceedings of IEEE International Conference on
Electrical, Electronic and Computer Engineering, ICEEC 2004, pp.
the control action accordingly. This approach allows the 753-758.
controller to compensate for disturbances and model
[7] H.X.Li and S.K.Tso, "Quantitative design and analysis of Fuzzy
uncertainties more effectively than the other two controllers, Proportional- Integral- Derivative Control- a Step
which rely on a pre-defined control law and do not explicitly TowardsAutotuning", International journal of system science, Vol.31,
account for uncertainties in the system. No.5, 2000, pp.545-553.
[8] Zadeh, L. A., Fuzzy Sets. Information and Control, 8, 338-353, 1965.
The proposed method and the results obtained in
[9] Assilian, S. and Mamdani, E.H., An Experiment in Linguistic
comparison with existing methods are significant in several Synthesis with a Fuzzy Logic Controller. International Journal of Man-
ways. First, the study provides a comprehensive comparison Machine Studies, 7(1), 1-13, 1974.
of three popular control techniques for DC motor speed [10] an Jantzen, ‘Tuning of Fuzzy PID
control, which can help researchers and engineers to select an Controllers’,www.iau.dtu.dk/~jj/pubs/fpid.pdf
appropriate control strategy for their specific applications. [11] Mohinder S.Grewal ,Angus P.Andrews,”KALMAN FILTERING
Second, the study demonstrates the effectiveness of the Theory and practice using Matlab”.Fourth edition,Wiely, 2015
Kalman filter controller for DC motor speed control, which [12] Shoba E.S.R, T.V.Narmadha, ‘A Fuzzy PD+I controller for speed
can be used as a benchmark for future research in this area. control of DC Shunt Motor’, Proceedings of NCICCA 2005, Tamil
Nadu, India pg 366- 370, 2005.
However, there are several limitations to this study that [13] Yasser Ali Almatheel, Ahmed Abdelrahman “Speed Control of DC
need to be acknowledged. First, the experiments were Motor Using Fuzzy Logic Controller” IEEE,( International Conference
conducted under idealized conditions, and the performance of on Communication, Control, Computing and Electronics Engineering
(ICCCCEE), Khartoum, Sudan),2017.
the controllers may vary under different operating conditions
[14] Katsuhiko Ogata: System Dynamics; Prentice Hall International, Inc.
or in the presence of disturbances. Second, the study only Third Edition 1998.
evaluated the performance of the controllers based on two
[15] Y. S. Lai, et al., "Efficiency comparison of PWM-controlled and PAM-
metrics, overshoot and rise time, and other performance controlled sensorless BLDCM drives for refrigerator applications," in
criteria, such as steady-state error and settling time, were not Conference Record - IAS Annual Meeting (IEEE Industry Applications
considered. Society), 2007, pp. 268-273.
[16] S. Jianwen, et al., "A novel microcontroller-based sensorless brushless
In addition to the limitations, there are also some DC (BLDC) motor drive for automotive fuel pumps,", IEEE
disadvantages of this study that need to be addressed in future Transactions on Industry Applications, vol. 39, pp. 1734- 1740, 2003.
research. One of the main limitations is the use of a single DC [17] Roozbeh Molavi, and Davood A. Khaburi, "Optimal Control Strategies
motor model, which may not be representative of all types of for Speed Control of Permanent-Magnet Synchronous Motor Drives",
DC motors. Future studies could consider using different World Academy of Science, Engineering and Technology 44, 2008,
motor models for validating the effectiveness of the proposed pp.428-432.
method. Another limitation is the lack of comparison with [18] Ruba M.K. Al-Mulla Hummadi,(2012), Simulation Optimal Speed
Control For A Dc Motor Using Linear Quadratic Regulator (LQR),
other control strategies, such as adaptive control or neural Journal Of Engineering, Volume 18 March 2012, pp 340-348
- 48 -
[19] Ramesh Chandra Chourasia, Mukesh Kumar, “Speed Control of [31] H. L. Tan, N. A. Rahim and W. P. Hew, A Dynamic Input Membership
S.E.D.C. Motor by Using Pi and Fuzzy Logic Controller” International Scheme for a Fuzzy logic DC Motor Controller, The IEEE
Journal of Soft Computing and Engineering (IJSCE),Volume-3, Issue- International Conference on Fuzzy Systems, St. louis, Missouri, USA ,
2, May 2013. pp. 426-429, 2003
[20] [20] R. Radzi (2010). Dc Motor Speed Control Using Fuzzy Logic [32] N. Mort, M. F. Abbod and D. A. Linkens, Comparative Study of Fuzzy
Controller. University Technology Malaysia. Retrieved from DC Servo Motors and Stepper Motors for Mechatronic System, IEEE,
https://www.researchgate.net/publication/322913964_DC_motor_spe UK, 1995 60
ed_control_using_fuzzy_logic_controller [33] Robert Bicker, Zhongxu Hu, Kevin Burn,’A Self Tuning Robotic Force
[21] H. L. Tan, N. A. Rahim and W. P. Hew, “A dynamic input membership Controller’, Proceedings , Springer Verlag, 2002.
scheme for a fuzzy logic DC motor controller,” in 12th IEEE [34] Kapinjay Ucharia, Himmat Singh, “International Journal of Scientific
International Conference on Fuzzy Systems (Institute of Electrical and Engineering and Technology” Volume 3 Issue No.3, pp : 235 – 239 1
Electronics Engineers, Piscataway, NJ, 2003) pp. 426-429. March 2014.
[22] R.M. Aguilar, V. Muñoz and Y. Callero, Control Application Using [35] Shashi Bhushan Kumar, Mohammed Hasmat Ali, Anshu Sinha,
Fuzzy Logic: Design of a Fuzzy Temperature Controller, University of “Design and Simulation of Speed Control of DC Motor by Artificial
La Laguna, 2012. URL: http://www.intechopen.com/ Neural Network Technique international Journal of Scientific and
[23] Tetsuya Iwasaki and Teruo Kataoka, ”Application of an Extended Research Publications, Volume 4, Issue 7, July 2014.
Kalman Filter to Parameter identification of an induction motor”,1989 [36] Md Akram Ahmad, Pankaj Rai, “Speed control of a DC motor using
IEEE. Industry Applications Controllers” Automation, Control and Intelligent Systems,2014; 2(6-
[24] Aleksandar Pajkanovic and BrankoDokic , “Wheelchair control by 1): 1-9.
head motion”, Serbian journal of electrical Engineering, vol. 10, No. 1, [37] Control Tutorials for MATLAB and Simulink - Motor Speed Digital
135–151, 2013. Controller Design, 2009.
[25] Mohinder S.Grewal ,Angus P.Andrews,”KALMAN FILTERING [38] “A study on the brush noise reduction of a dc motor using multi-body
Theory and practice using Matlab”.Fourth edition,Wiely, 2015 dynamics”, Tae-Won Park, Il-Ho Jung and Jong-Hwi Seo, XXI
[26] J. Jantzen, Tutorial on Fuzzy Logic, Technical University of Denmark: ICTAM, 15-21 August 2004, Warsaw, Poland.
Department of Automation, Technical report number 98-E 868, 19 Aug [39] K.Ogata, “Modern Control Engineering”, Prentice-Hall, 1970
1998, URL: http://www.iau.dtu.dk/
[40] A. Shi, M. Yan, J. Li, W. Xu, Y. Shi, The Research of FUZZY PID
[27] K.Venkateswarlu and Ch. Chengaiah “Comparative study on DC motor Control Application in DC Motor of Automatic Doors, Jurnal of
speed control using various controllers” Volume 1 , Issue 6 / Dec 2013 Institute of Electrical and Electronics Engineers, pp. 1354-1358, 2011
[28] AlokRanjan Singh and V.K. Giri “Design and Analysis of DC Motor [41] S. Zhang, X. Wang, Study of Fuzzy-PID Control in MATLAB for
Speed Control by GA Based Tuning of Fuzzy Logic Controller” Twophaase Hybrid Stepping, Proceedings of the 2nd International
International Journal of Engineering Research & Technology (IJERT), Conference On Systems Engineering and Modelling (ICSEM-13), pp.
Vol. 1 Issue 5, July – 2012 1011-1014, 2013
[29] J.P. Anderson, State-space modeling, system identification and control [42] M. Muruganandam and M. Madheswaran, Modeling and Simulation of
of a 4th order rotational mechanical system . Master Thesis. Naval Modified Fuzzy Logic Controller for Various types of DC Motor
Postgraduate School, California, 2009 Drive, international conference on control, automation, communication
[30] M. Ruderman, J. Krettek, F. Hoffmann, T. Bertram, Optimal State and energy conservation, 4th-6th June 2009 pp. 1-6
Space Control of DC Motor, Proceedings of the 17th World Congress
The International Federation of Automatic Control Seoul, Korea, 2008
.
- 49 -
Digital Transformation of Document Support
for Enterprise Quality Management System
Shara Toibayeva1*, Irbulat Utepbergenov2
1
G.Daukeev Almaty University of Power Engineering and Telecommunications, Almaty, Kazahstan
2
G.Daukeev Almaty University of Power Engineering and Telecommunications, Almaty, Kazahstan
Email: [email protected], [email protected]
Abstract—This article deals with quality, which is one of the production, control, work with personnel, etc. is characterized
important strategic tools in business. Assessing the effectiveness by the concept of quality. For example, the quality of a product
of quality management systems of enterprises is of great reflects the quality of its design, development, manufacturing,
importance, given the specificity of quality indicators, the multi- and sales processes. Thus, quality management affects the
level nature of the system, and the need to select the optimal entire product life cycle: from research in marketing, research
number of performance indicators and assess the state of the and development, design stages, and up to the production of
system, which is associated with the formation of rational products, sales and maintenance. In this regard, the current
decisions in the management of quality management systems.
steps in the development of quality improvement techniques
The method and model of automated enterprise quality
cover both the problems of improving the quality of products
management and intelligent automated system of quality
management of enterprise integrated with ruling MICS
or services, and the quality of management itself [2].
subsystems (Management Information and Control System) are The main task for effective management are the stages of
offered allowing to automate QMS implementation and support formation, implementation and use of an automated system of
processes and increasing the validity, efficiency and quality management. The control system should provide
effectiveness of management decisions by automated a number access to the documentation of the enterprise, quality output
of functions of decision makers and personnel. The article also of the requested information to take management actions to
considers the development of the Model and algorithms of
solve the problems of the enterprise relevant at a certain
digital transformation of documentary support of quality
moment.
management system to identify contradictions and
inconsistencies in the documentation of quality management At product quality management, the objects of
system (QMS) of the economy of Kazakhstan. This solution will management are processes, and managing decisions are
eliminate the problems of processing large volumes of normative formed by comparing the indicators of the actual state of the
documents of the enterprise while maintaining in an automated managed process with its planned ones. To successfully
QMS. The solution is based on reality model and accompanied perform its functions, an enterprise needs to manage
by the development of a formal language using approaches significant correlated activities as a system of processes.
similar to the creation of well-known declarative programming
languages. This project was supported by a grant from the According to the results of the analysis of literary sources
Ministry of Education and Science of the Republic of for the last 20 years, we can say that in the world, in particular
Kazakhstan (Zhas Galym project No. AR 13268939 Research in Kazakhstan, the scientific direction related to the problems
and development of digital technology to provide consistency in of implementation and automation of quality management
the media of normative documents of the quality management systems, technology analysis of business processes of various
system). organizations is developing.
Keywords—automated control system; quality management; Turning to statistics, we can note that the starting point of
document structure; formal language management development in Kazakhstan are the 90's, when
the management situation and the market system have
I. INTRODUCTION changed dramatically, and there was a need to study the
The economic potential of Kazakhstan is directly related foreign experience of management, to apply it in the country.
to the development of business, and the management More and more interest in the implementation of the QMS and
environment affects the efficiency and effectiveness of the automation of business processes of enterprises is noted. For
enterprise. By joining the WTO, Kazakhstan has increased example, the query "automation of quality management" in
competition between enterprises to occupy its niche in the the information base of scientific electronic library
market and strengthen its position, which requires the release eLIBRARY.RU received 43085 results [3-4], which indicates
of higher quality products to the market, capable of satisfying the interest in quality management systems and its automation
the end users [1]. in the scientific community.
The enterprise adapts to changes in the external In an industrialized country, it is necessary to meet the
environment, which determines its organizational component challenges of implementing and keeping up-to-date modern
of management, which includes many links (units), classified management systems, where competition, knowledge-
by appropriate strategies, relationships (relationships), and intensive, innovative and technologically sophisticated
documented requirements. production is developed.
Today, the effective result of the organization of all
functions of the enterprise: planning, organization,
- 50 -
Since the beginning of 2002, the organization, design and II. THE IMPORTANCE OF BUILDING A FORMAL LANGUAGE
implementation of QMS of organizations in Kazakhstan has
been noticeably strengthened [5-6].
A formal language is treated as a set with all strings under
In the field of quality management the orders of Prime a finite number of alphabets. The formal language theory
Minister of the Republic of Kazakhstan № 28-r dated states that the study of all families is done by formal methods.
06.02.2004 and № 175-r dated 27.06.2006 on the soonest This formal method is mainly used to develop and verify
rearrangement of enterprises according to ISO standards have software and hardware systems. It is usually described for
been adopted, so on achieving the set goals in the country the formal languages and automata theory. Typically, formal
proper material and technical support, normative and languages deal with a set of strings and are defined in two sets
methodological base for implementation of international of rules. The first rule deals with syntax, which specifies how
standards is created [7,8]. to use symbols, while the second rule deals with semantics,
The regulatory framework of the Republic of Kazakhstan which tells the meaning of symbols and legal expressions
[7,9] includes 36 state standards, where international ISO [13].
standards were taken as a basis and adopted as state standards Constructing a formal language that characterizes
of the Republic of Kazakhstan. normative documents is an important tool for ensuring
Development and implementation of standards is part of accuracy and clarity in describing the rules and requirements
the standardization plan of the Republic of Kazakhstan each contained in normative documents. Regulatory documents
year in the field of management. are used in various fields of activity, from jurisprudence to
technical documentation, and often have legal significance.
In their work, A. Ateş and U. Bititchi [10] argue that the Therefore, they need to be written in clear and
business environment is constantly changing, and managing comprehensible language so that users can easily understand
change to adapt to an uncertain future is a problem that the requirements they contain.
requires resilience - the ability of an organization to survive, However, normative documents are often complex and
adapt and maintain business sustainability in the face of high confusing, which makes them difficult to understand and
rates of change.
leads to errors in their interpretation and application. Building
Innovative companies are adopting new systems a formal language allows you to develop a common language
management approaches, and enterprise quality managers are to be used to describe documents and establish clear syntactic
achieving greater results for their organizations in less time rules for its use. This will make it easier for the user to
[11,12]. understand the requirements contained in normative
Several problems can be identified that quality documents and reduce the likelihood of errors.
management professionals encounter when using such static In addition, building a formal language can help in
management systems: developing new regulatory documents and amending existing
ones. With the language, you can quickly identify what
1) Unproductive use of work time. Managers have to enter concepts are already used in other normative documents,
the same data several times to collect data from different what syntax rules apply, and how different requirements are
sources and combine it into single reports. described. This can greatly speed up the process of
2) Lack of real-time, relevant information. It takes some developing and amending documents [14].
time to distribute and update information throughout the Thus, in order to increase the speed of creating documents
organization. The data updates slowly in the system, and in and the accuracy of the data contained in them, automated
order to provide an accurate assessment of the situation, the document management systems are being actively developed
most recent changes in compliance data must be processed. and widely used. The development of a universal formal
3) Limiting access to accurate analysis. Facility language will make it possible to create a system of
management relies on managers to perform surveillance of automated formation of documents of any composition and
what is going on in the facility. Since the managers themselves purpose.
are not quality managers-they rely on the latter to analyze and
present data they can understand and read. It takes a lot of time
to create reports, thereby slowing down the feedback process. III. BUILDING A FORMAL LANGUAGE
The construction of a formal language characterizing
4) Preparing monthly reports for most businesses is a normative documents is an important tool for ensuring
duplication of data. When calculating data, it takes time for accuracy and clarity in the description of rules and
managers to submit reports to supervisors. requirements contained in normative documents. Normative
5) Ineffective communication with employees. Creating documents are used in various fields of activity, from
connections with employees and training people is very time jurisprudence to technical documentation, and often have
consuming. Quality managers spend a lot of time on legal significance. Therefore, it is necessary that they be
compliance issues without focusing on other activities. written in a clear and understandable language so that users
can easily understand the requirements contained in them.
6) Lack of databases of documented information and as a
result, no automated decision support system. Lack of prompt However, normative documents are often complex and
notification and response to the facts of deviations of the confusing, which makes them difficult to understand and leads
values of the planned indicators. to errors in their interpretation and application. The
construction of a formal language makes it possible to develop
a single language that will be used to describe documents, and
to establish clear syntactic rules for its use. This will make it
- 51 -
easier for the user to understand the requirements contained in The three-dot (...) denotes other characters of the FL:
normative documents and reduce the likelihood of errors. <Text>, <East Lit>, <<Word>, <Digit>, <Punctuation Mark>,
<Ending Mark>, <Letter>, <Arithmetic>.
A formal language is considered as a set with all strings
which are under a finite number of alphabet. Formal language The normative document can be presented in CT in three
theory states that studies of all families on formal methods. forms:
This formal method is mainly used for developing and
verifying the software and hardware systems. It is typically CD = (KD, BD, ND), (1)
described for formal languages and automata theory. where KD is a set of components reflecting the structural
Generally formal languages deal with the set of strings and elements of the documents; BD is a set of links between the
defined in two set of rules. The first rule is about syntax which structural elements of the documents; ND is a set of nodes
tells that how to use the symbols whereas the second rule is located at the connection points of the structural elements of
about semantics which tells that meaning of the symbols and the documents.
legal expressions [15,16].
In turn, the KD component set contains three subsets:
The system based on the formal language is fundamentally
different from its counterparts in that it implements a new
approach in terms of documenting, which consists in KD = { KDI, KDP, KDZ }, (2)
constructing the document plan from special components that
are computer models of the document's constructive
assemblies. This process is called "document where KDI - set of information converter components; KDI
construction"[17]. - set of information converter components; KDZ - final
Using FL, you can describe the structure of almost any component ("document").
document. The structure of the order in the above example KD topology and coordinate system components are not
(Fig. 1) can be described by a set of the following nonterminal supported by S1 boiler. S1 bylaws have the following types:
symbols of the F span; span; logic; zhol; array; and derek and file [19-22].
Building a formal language can help in the development Components are the source of information about the KDI
of new normative documents and amendments to existing set (Fig. 2) to set the source information and transfer it to other
ones. The language can be used to quickly determine what components or directly to the document. The algorithmic
concepts are already used in other normative documents, what model and process of the information source component are
syntax rules apply, and how various requirements are characterized by the following sequence:
described. This can greatly speed up the process of developing
and amending documents [18].
Fig. 2 - Graphical representation of the source component
1) component parameter values and source information to
be included in the document;
2) the component stores the specified parameter values and
source information in special variables;
3) the component is linked to other components (by means
of information contact n1);
4) the document creation process begins;
5) the component transmits the information previously
stored in the variable to info contact n1 via info channel b1;
Fig.1 - An example of using FL to describe the structure 6) the components connected to the component via
of a document infocontact n1 receive the information, then process it and
send it to the document or send it to the document immediately
The nonterminal symbol <Document 1> denotes the entire [23].
document. It is part (as a fragment) of a more complex
document, indicated in the figure by the symbol <Document Components are transducers of information in the KDP set
2>; (Fig. 3) for receiving source information, transforming it and
transmitting it to other components or directly to the
The <Text> symbol denotes a block of text that can document. Components of this type may have one or more
include a set of all characters from the ASCII (or Unicode) inputs and only one output info-contact.
table;
The symbol <Requisites> expresses a requisites of a
document(name of the position of the person who signed the
document, signature and transcript of the signature).
- 52 -
«Contemporary Issues in Business, Management and Education 2013»,
2014. V.110. - P. 627-637
[3] Olszewska A.M. Research Issues Undertaken within Quality
Management - Overview of Selected Literature and a Knowledge Map
// 7th International Conference on Engineering, Project, and Production
Management, 2017. V.182. -P. 518 - 523
[4] B. Jose, S. Abraham Intelligent processing of unstructured textual data
in document based NoSQL databases, Materials Today: Proceedings
Fig. 3 - Graphic representation of the information- 2021
converter component [5] Utepbergenov I.T., Toibayeva Sh.D., Muslimova A.K., Isabekova L.S.
Nechetkaya model' menedzhmenta kachestva tekhnologicheskikh
The end component "Document" of the KDZ set (Fig. 5) is protsessov pishchevogo proizvodstva// Mezhdunarodnaya nauchnaya
used to The end component "Document" of the set KDZ (Fig. konferentsiya «Informatika i prikladnaya matematika», Almaty, 2017.
4) is used to collect all the information coming to it and to CH.II. - P.374-388.
form a document. [6] Muslimova A.K., Utepbergenov I.T., Sklyarova Y.V., Toibayeva
Sh.D. Formalizatsiya analiza funktsionirovaniya i effektivnosti SMK
dlya ekspertnoy sistemy //Vestnik KazNU. Almaty, 2016. - №3/1 (90).
- P. 87-96
[7] Website of the National Accreditation Center LLP//
http://www.nca.kz/ 02.05.2023
[8] Satbaev R.A., Dugalov G.T. Technical regulation and metrology:
industry and new technologies of the Republic of Kazakhstan. "The
concept of development of management systems in the Republic of
Kazakhstan " - Astana: Ed. KazInST and LLP BaEr ltd. - P. 68-83.
Fig. 4 - Graphic representation of the "Document" [9] Regulatory framework // URL: http://online.zakon.kz/ 02.05.2023
component [10] Ates A., Bititci U., Int. J. Prod. Res, 2011. V.49. -P. 5601-5618
It can also serve to transfer the formed fragment to another [11] Utepbergenov I., Bobrov L., Medyankina I., Rodionova Z. About the
Concept of Information Support System for Innovative Economy in the
document. Using this component makes it possible to form Republic of Kazakhstan// Recent Research in Control Engineering and
complex documents consisting of a number of documents (or Decision Making. ICIT 2019 Conference proceedings - Springer,
their fragments). The "Document" component can have one or Cham, 2019. Vol. 199. - P. 515-526
more input info-contacts (n1...nn) and one output info-contact [12] Medyankina I.P., Bobrov L.K., Toibayeva Sh.D., Nurgulzhanova A.N.
(nn+1). Comparative analysis of the state of information support of innovations
in Russia and Kazakhstan // Bulletin of KazATK, 2018. No. 4 (107). P.
CONCLUSION 239-246
[13] ScienceDirect. Formal Language – an overview | ScienceDirect Topics.
The system developed on the basis of the formal language https: //www.sciencedirect.com/topics/computer-science/formal-
differs from its analogues by implementing a new approach to language. (Accessed on 12/03/2023).
documenting, which consists in forming the document scheme [14] A. P. Sergushicheva, E. N. Davydova. “Development and Application
from special components, which are computer models of of a Software Tool to Support the Teaching of Formal Languages”.
structural elements of the document. This process is called Open Education 25(4):13-22. 2021.
"document construction". [15] Toibayeva Sh.D. A comprehensive methodology for automated quality
management of an enterprise in Kazakhstan, International scientific
Using a formal language for normative documents has journal SCITECHNOLOGY, Latvia, 2019. No. 21. -P.23-27
many advantages. [16] Ekaterina Bakinova , Artem Basharin, Igor Batmanov , Konstantin
Lyubort , Alexander Okhotin , Elizaveta Sazhneva. “Formal languages
Firstly, it reduces the possibility of different over GF(2). Information and Computation”. Volume 283, February
interpretations and interpretations of the document, which 2022, 104672.
avoids various undesirable consequences, such as conflicts, [17] Alberto Pettorossi. Formal Grammars and Languages. “Automata
legal errors, lawsuits, etc. Theory and Formal Languages”. August 2022.
[18] Tomoyuki Yamakami. “Formal Grammars for Turn-Bounded
Secondly, the formal language simplifies the process of Deterministic Context-Free Languages”. In book: Theoretical Aspects
creating and making changes to normative documents, since it of Computing – ICTAC 2022. October 2022.
has a strict logical structure and format, which makes it [19] Panov S.A. Formal language of document structure description and its
possible to automate the process of checking a document for interpretation in the format of the method of component chains //
Papers of Tomsk State University of Control Systems and
compliance with norms and rules [24, 25]. Radioelectronics. - 2014. - №1(31). - С. 197-20
Thirdly, the use of a formal language of normative [20] Lanin V.V., Lyadova L.N. Document management in dynamically
documents increases the efficiency of government bodies, adaptable systems, based on metamodeling. Proceedings of the
International Conference "Intelligent Systems - 2010". - M.: Fizmatlit,
since this language allows improving the process of regulating 2010. - VOL. 1. - P. 510-517.
public relations and ensuring more effective protection of
[21] Gandzha T.V. Panov S.A. Tabular forms generator as a component of
citizens' rights and freedoms. automated documenting system // Science. Technologies. Innovations:
Proceedings of the All-Russian Scientific Conference of Young
Scientists, Novosibirsk, November 29 - December 2, 2012: in 7 p. -
Novosibirsk: Publishing house of NSTU, 2012. - Ч. 3. - С. 88-91.
REFERENCES
[22] Tugengold A.K. et al. Intelligent electronic documentation of
technological The intelligent electronic documentation of
[1] Kazakhstan in WTO: what changes have taken place in 4 years technological objects in the system of PLM // DSTU Vestnik. - 2011. -
//https://strategy2050.kz/en/news/53723/ Т. 11, № 3. - С. 860-867.
[2] Liepiņa R., Lapiņa I., Mazais J. Contemporary Issues of Quality [23] Dmitriev V.M., Gandzha T.V., Gandzha V.V., Panov S.A. Structure
Management: Relationship between Conformity Assessment and and functions of an automated project management system for centers
Quality Management // The 2-dn International Scientific conference of scientific and technical creativity of students // Reports of Tomsk
- 53 -
state university of control systems and radioelectronics. - №2(26), Ч. [25] Veretekhina S. V. V. Methodology of developing interactive electronic
2. - 2012. - С. 200-203. operational documentation for science-intensive products of the
[24] Tugengold A.K. et al. Intelligent electronic documentation of telecommunications and information industry. - MOSCOW: ALL-
technological The intelligent electronic documentation of RUSSIAN RESEARCH INSTITUTE OF PH, 2008
technological objects in the system of PLM // DSTU Vestnik. - 2011. -
Т. 11, № 3. - С. 860-867
- 54 -
On the issue of constructing and identifying the
parameters of the piezometric pattern of pumping
wells in the underground leaching of metals
Bakhyt Mukhanov Zhanar Omirbekova
Department of Automation and Control Department of Automation and Control
Almaty University of Power Engineering Satbayev University
and Telecommunications named after Almaty, Kazakhstan
Gumarbek Daukeyev [email protected]
Almaty, Kazakhstan ORCID 0000-0001-8425-5454
[email protected]
ORCID 0000-0003-1634-6914
Yerbol Orakbayev Gulmira Bazil
Department of Automation and Control Department of Automation and Control
Almaty University of Power Engineering Almaty University of Power Engineering
and Telecommunications named after and Telecommunications named after
Gumarbek Daukeyev Gumarbek Daukeyev
Almaty, Kazakhstan Almaty, Kazakhstan
[email protected] [email protected]
ORCID 0000-0003-3100-7419 ORCID 0000-0001-9563-0352
Abstract — The article is devoted to the problems this paper we conduct practical and mathematical
of underground leaching of metals. An algorithm for experiments based on empirical approach and
the current assessment of the main parameters of the modelling obtained fascinating results hypothesis
piezometric characteristics of pumping wells is being [7-10]
developed. The piezometric profile of the downhole
pump characteristic is given on the example of the M ETHODOLOGY
GRUNDFOS pump. The results of research and
calculations showed that the dependence of the The examination is as follows consider the
increase in the pressure of the liquid column Рst on technological features of pumping wells equipped
the filter in the pumping well on the flow rate should with downhole pumps when lifting production fluids
be used to solve the problem of optimal control of the to the surface. Normally, pumping wells are
in-situ leaching process This methodology allows for combined into a production mud receiving unit
optimal control of the in-situ leaching process by (PMRU), the scheme of which is shown in Fig. 1.
controlling the fluid pressure.
Keywords — mining operation; in-situ leaching;
identifying; optimal control.
I NTRODUCTION
Efficiency of production - by the method of in-
situ leaching (ISL) of metals is largely determined
by piezometric (hydrodynamic) characteristics of
pumping wells. In this connection the development
of algorithms of current estimation of basic
parameters of the given characteristics is actual.
In-situ uranium leaching is one of the most
efficient and, most importantly, relatively least
hazardous industries. The authors [1, 3-5]
thoroughly investigate and discuss the problems in
this area. The authors Mukhanov et al. [2-3] Figure 1. Technological scheme of the extraction well
1 - pressure sensor; 2 - flow meter; 3 - valve.
conducted numerical experiments and modelling. In
- 55 -
In this paper, to approximate the flow-pressure
An important element of a pumping well is a characteristics, it is proposed to use the expressions
downhole pump. Consider the flow-pressure
characteristic of the pump together with the , (1)
characteristic of the well, which has the form, see
fig. 2.
where: - consumption of PR , m 3 / h ; -
pressure created by the pump (M).
The parameters of the model (1) can be
identified from the nameplate characteristics of the
pump.
Model (1) will make it possible to build a
piezometric picture of the extraction well and
determine the piezometric pressure of the productive
solution on the filter (see Fig. 3).
To obtain a piezometric profile of the operation
of a pumping well, the following parameters were
taken: the depth of the well is 200 m, the well pump
is at a depth of 110 m, groundwater starts from a
depth of 85 m.
Figure 2. Flow characteristics
Figure 3. Piezometric profile in a production well with a borehole pump type SP -14
- 56 -
Consider the piezometric profile of the
downhole pump characteristics given for the
GRUNDFOS pump , type SP -14, with a power of
7.5 kW. The flow-pressure characteristic (see Fig. 2)
is represented by the expression:
Profile 1 was built at a flow rate of the
productive solution equal to 8 m 3 / h, while the
pressure sensor (see Fig. 1. - pos. 1) shows 2 atm.
The profile shows that in order to raise the
productive solution to the surface from a depth of
180m, it is necessary to create a pressure of 13 atm,
(11 atm + 2 atm (instrument reading)).
In this example, we consider that 1 atm is Figure 4 . Dependence of the influence of the liquid column on the filter
rounded off by 10 m of a liquid column, the density of the pumping well
of the product solution and the resistance due to the
velocity of the solution in the pipe are not yet taken This dependence should be used to solve the
into account. Thus, the piezometric straight line problem of optimal control of the in-situ leaching
shows the pressure drop in the extraction well, process.
which must be created to transport 8 m 3 / h Pr. At C ONCLUSION
the second stage of the algorithmizing of the process
of constructing a piezometric profile of a pumping An empirical experiment has been carried out
well, we calculate the drop created by the downhole on pumping wells to get the piezometric pro file of
pump, in our case, the straight pumping unit operation. The experiment shows that
line in Fig.3. by regulating the pumping pressure the process of
pumping productive mud can be optimally
At the third stage, we build a straight line (
controlled and the lifetime of the pumping units can
) from a point equal to 9 atm , assuming that the
be prolonged, and the operating cost of the
liquid from the ore body rises to the borehole pump, equipment decreased.
it is necessary to create 9 atm (90 m). The point of The experiment showed that by controlling the
intersection of the straight line with - 0 - the different modes in the pump, it is possible to
coordinate of the piezometric profile is marked with successfully overcome the obstacle of groundwater,
a dot . At this point, it can be seen that the which also adds an additional obstacle in the
filter (ore body) has a liquid column equal to 70 m, extraction of productive mud by the borehole
that is, a pressure of 7 atm is created. method. This control method effectively distributes
Profile 2 - reflects the consumption Pr equal to control problems to the process field. The strength
6 m 3 / h. This flow is created by valve 3, see fig. 1. of the experiment is the instantaneous hanging
As can be seen from the piezometric profile, the effect of the productive fluid and the ability to
pressure after the pump is created at 15 atm (straight effectively control the entire process. The weakness
line ). According to equation 2, the pressure is the experiment is performed on a specific well
drop across the well pump is calculated, it is equal and requires large-scale production.
to (direct ). point of R EFERENCES
intersection of the line c -0- coordinate of the [1] Mukhanov B. K., Pogrebnyak V. V., Omirbekova Zh. Zh.,
piezometric profile is marked with the letter (δ), it Orakbaev E. Zh. To solve the problem of optimal
shows that the liquid column presses with a force of distribution of underground leaching // Bulletin of
Automation. - 2017. - No. 4. - S. 10-13.
6 atm. [2] Mukhanov, B. K., Omirbekova, Z. Z., Orakbayev, Y. Z.,
Thus, it is possible to plot the dependence of Alimanova, M. U., & Pogrebnyak, V. (2019). Numerical
the increase in the pressure of the liquid column modeling of ORE body and development hydrodynamic
models of in-situ leaching process. 14th International
on the filter in the pumping well on the flow Conference on Electronics Computer and Computation,
rate and it will look like, see Fig.4. ICECCO 2018.
https://doi.org/10.1109/ICECCO.2018.8634795 .
[3] Sun, Z., Ma, W., Liu, Y., Liu, J., & Zhou, Y. (2021).
Research progress on groundwater contamination and
remediation in in situ leaching uranium mines. In Earth
Science Frontiers (Vol. 28, Issue 5).
https://doi.org/10.13745/j.esf.sf.2021.2.11
[4] Tan, K., Li, C., Liu, J., Qu, H., Xia, L., Hu, Y., & Li, Y.
(2014). A novel method using a complex surfactant for in -
situ leaching of low permeable sandstone uranium deposits .
Hydrometallurgy, 150.
https://doi.org/10.1016/j.hydromet.2014.10.001
- 57 -
[5] Li, H., & Zhou, Y. (2019). Progress and Prospect of system. // Status, trends and problems of development of
Research on Solute Transport during In-Situ Leaching of the oil and gas potential of Western Siberia: Proceedings of
Uranium. In Xiyou Jinshu/Chinese Journal of Rare Metals the All-Russian Scientific and Technical Conference. -
(Vol. 43, Issue 3). Tyumen: FSUE " ZapSibNIIGG ", 2009 - pp. 356-360.
https://doi.org/10.13373/j.cnki.cjrm.XY18120015 [9] Rogov E.I. , Yazikov V.G., Horns A.E. Hydrodynamic
[6] Zhao, L., & Li, P. (2021). Relationship between chamosite model of underground leaching of uranium // Mining
alteration and Fe-plugging in sandstone pores during acid information and analytical bulletin (scientific and technical
in situ leaching of uranium. Minerals, 11(5). journal) . - M.: 2000. - No. 5. - S. 40 - 42.
https://doi.org/10.3390/min11050497 [10] Solovyov I. G., Vedernikova, Yu. A. Development and use
[7] Zhang, C., Tan, K., Xie, T., Tan, Y., Fu, L., Gan, N., & of hydrodynamic models of well systems equipped with
Kong, L. (2021). Flow microbalance simulation of installations of submersible electric centrifugal pumps //
pumping and injection unit in in situ leaching uranium Bulletin of Cybernetics. Tyumen : Publishing House of
mining area. Processes, 9(8). IPOS SB RAS , 2002. No. 1, p. 85-91.
https://doi.org/10.3390/pr9081288
[8] Fomin V.V., Vlasov D.A., Govorkov D.A. Technology of
visual-graphical analysis of hydrodynamics of the well
- 58 -
Analysis and synthesis of single-bit adders for
multi-bit adders with sequential transfers
Sakhybay Tynymbayev Assel Mukasheva Ibragimov Kuanyshbek
PhD, Professor of the Department of PhD, Associate Professor of the PhD doctoral student of the Faculty
Information Systems Department of Information Systems of Information Security System of the
and Cybersecurity And Cybersecurity Kazakh National University named
Almaty University of Power Engineering Almaty University of Power Engineering after Al-Farabi
and Telecommunications and Telecommunications Almaty, Kazakhstan
Almaty, Kazakhstan Almaty, Kazakhstan [email protected]
[email protected] [email protected] orcid.org/0009-0008-2126-3420
orcid.org/ 0000-0002-9326-9476 orcid.org/0000-0001-9890-4910
Abstract - The paper provides an analysis of existing single- -on valves "exclusive OR", and multiplexers, where an
digit binary adders from the point of view of their implementations adder with a conditional sum (conditional-sum addition_
on fans built on the basis of MOSFETs. The synthesis of a single- CSA) is implemented.
digit adder with a conditional sum is carried out. The considered Before determining the hardware complexity (the number
adders are compared in terms of speed and hardware complexity
of MOSFETs) and performance (the number of logic
(by the number of MOSFETs). In the final part, a scheme of a
multi-bit adder with consecutive transfers on adders with a elements on which the signal is delayed), the circuits of the
conditional sum is given. An example of performing addition analyzed adders are depicted as logic gates on MOSFETs.
operations is given. Consider the analysis of an adder built on the basis of two-
stage logic AND-OR-NOT [1,5].
Keywords—single-bit adder, multi-bit adder with consecutive The circuit of such an adder is shown in Figure 1, and
transfers, conditional sum adder, multiplexer. Figure 2 shows the circuit of this adder on the logic circuits
I. INTRODUCTION of MOSFETs [3].
The analytical formula of the analyzed adder has the form:
Adders perform arithmetic addition and subtraction of
numbers. Together with other logical operations, adders are Ci = (𝑎𝑖 𝑏𝑖 𝑎𝑖 𝐶𝑖−1 𝑏𝑖 𝐶𝑖−1 )
the core of the circuits of arithmetic logic devices that 𝐶𝑖−1 (𝑎𝑖 𝑏𝑖 𝐶𝑖−1 )𝑎𝑖 𝑏𝑖 𝐶𝑖−1
Si = ̅̅̅̅̅ (2)
implement a number of various operations (multiplication,
division, etc.), which are an indispensable part of various Figure 1 shows the functional diagram of the adder
processors. constructed according to the formula (2).
Important parameters of adders are their hardware Figure 2 shows its logic circuit on MOSFETs.
complexity and performance, therefore, many variants of
single-bit and multi-bit adders with serial, parallel and
combined transfers have been developed [1].
The listed adders are synthesized based on the truth table.
The analytical expressions of the sum (Si) and transfer
functions Ci have the form:
Si =𝑎̅𝑖 𝑏̅𝑖 𝐶𝑖−1 𝑎̅𝑖 𝑏𝑖 ̅̅̅̅̅
𝐶𝑖−1 𝑎𝑖 𝑏̅𝑖 ̅̅̅̅̅
𝐶𝑖−1 𝑎𝑖 𝑏𝑖 𝐶𝑖−1
Ci = 𝑎𝑖 𝑏𝑖 𝑎𝑖 𝐶𝑖−1 𝑏𝑖 𝐶𝑖−1 ; (1)
(1)
where ai and bi are the i–th digits of the numbers A and B
Ci-1 – transfer from the junior category
The formula can be reproduced on various sets of logic
Figure 1. Functional diagram of the adder
elements. AND-NOT, OR-NOT, "exclusive OR", etc. At the
same time, the hardware complexity and performance may be
different.
II. MATERIALS AND METHODS
The analysis of single-digit adders on elements is
discussed below:
- two-stage logic AND-OR-NOT [1];
- two "exclusive OR" valves and OR-NOT and AND-
NOT circuits [2];
-three-way valve "exclusive OR" and circuits AND-NOT;
- 59 -
Figure 4. Functional diagram of the Si calculator based on two addition
schemes modulo two.
Figure 4 shows that the implementation of Si will require
Figure 2. Logic circuit of the adder on MOSFETS
four inverters and six two-input circuits AND-NOT. Which
will require (4x2) +(6x4) = 30 transistors.
Figure 2 shows that the adder consists of six circuits
In total, to build an adder on two circuits modulo 2, you
AND-NOT two inputs each. To implement them, 6x4 = 24
will need (18+32)=50 transistors. The second adder can be
transistors will be required. It has nine circuits of NOT (2x9
built on the basis of an adder modulo two of three variables.
= 18) transistors and a gate AND-NOT for three inputs (6
In our case, ai ,bi and Ci-1come in as input variables. Then the
transistors), a gate OR-NOT for three inputs (6) transistors
truth table for Si has the form as shown in table 1.
and a gate OR-NOT for four inputs (8) transistors. In total, 62
MOSFETs will be required to implement the adder.
TABLE I. THE TRUTH TABLE FOR THE SUM OF SI
The time of formation of the transfer tci = 4 л.э..
The time of formation of the sum Si = 7 л.э.. Ci-1 ai bi Si
The second single-digit adder [1,2], which is subject to 0 0 0 0
analysis, refers to the formula: 0 0 1 1
0 1 0 1
Si = 𝐶𝑖−1 ⊕ 𝑎𝑖 ⊕ 𝑏𝑖 0 1 1 0
1 0 0 1
Ci = ̅̅̅̅̅ 𝑎𝑖 𝑐𝑖−1 ̅̅̅̅̅̅̅̅
𝑎𝑖 𝑏𝑖 ̅̅̅̅̅̅̅̅ 𝑏𝑖 𝑐𝑖−1 (3) 1 0 1 0
1 1 0 0
The functional scheme of the Ci transfer is shown in Figure 1 1 1 1
3, which consists of three two-input and one three-input
circuits AND-NOT. To implement them, you will need (4x3) From this table we have:
+6 = 18 transistors. Time of transfer formation tci = 2 л.э..
̅̅̅̅̅
Si=𝐶 ̅𝑖 𝑏𝑖 ̅̅̅̅̅
𝑖−1 𝑎 𝐶𝑖−1 𝑎𝑖 𝑏̅𝑖 𝐶𝑖−1 𝑎̅𝑖 𝑏̅𝑖 𝐶𝑖−1 𝑎𝑖 𝑏𝑖 (4)
To calculate Si, the formula (4) is transformed as follows:
Si = ̿̿̿̿̿̿̿̿̿̿̿̿̿̿̿̿̿̿̿̿̿̿̿̿̿̿̿̿̿̿̿̿̿̿̿̿̿̿̿̿̿̿̿̿̿̿̿
𝐶𝑖−1 𝑎𝑏𝑖 ̅̅̅̅̅
̅̅̅̅̅ 𝐶𝑖−1 𝑎𝑖 𝑏̅𝑖 𝑖 𝐶𝑖−1 𝑎̅𝑖 𝑏̅𝑖 𝐶𝑖−1 𝑎𝑖 𝑏𝑖 =
̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
̅̅̅̅̅̅̅̅̅̅̅̅̅ ̅̅̅̅̅̅̅̅̅̅̅̅
̅ ̅̅̅̅̅̅̅̅̅̅̅̅ ̅̅̅̅̅̅̅̅̅̅̅̅
̅̅̅̅̅
(𝐶 ̅̅̅̅̅
̅𝑖 𝑏𝑖 )(𝐶
𝑖−1 𝑎 𝑖−1 𝑎𝑖 𝑏𝑖 )(𝐶𝑖−1 𝑎̅𝑖 𝑏̅𝑖 )(𝐶 𝑖−1 𝑎𝑖 𝑏𝑖 ) (5)
The functional scheme for the sum of Si constructed
according to the formula (5) is shown in Figure 5.
To calculate Si, the formula (4) is transformed as follows:
Si = ̿̿̿̿̿̿̿̿̿̿̿̿̿̿̿̿̿̿̿̿̿̿̿̿̿̿̿̿̿̿̿̿̿̿̿̿̿̿̿̿̿̿̿̿̿̿̿
𝐶𝑖−1 𝑎𝑏𝑖 ̅̅̅̅̅
̅̅̅̅̅ 𝐶𝑖−1 𝑎𝑖 𝑏̅𝑖 𝑖 𝐶𝑖−1 𝑎̅𝑖 𝑏̅𝑖 𝐶𝑖−1 𝑎𝑖 𝑏𝑖 =
̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
̅̅̅̅̅̅̅̅̅̅̅̅̅ ̅̅̅̅̅̅̅̅̅̅̅̅
̅ ̅̅̅̅̅̅̅̅̅̅̅̅ ̅̅̅̅̅̅̅̅̅̅̅̅
Figure 3. Functional scheme of Ci transfer ̅̅̅̅̅
(𝐶 𝑖−1 𝑎
̅̅̅̅̅
̅𝑖 𝑏𝑖 )(𝐶 𝑖−1 𝑎𝑖 𝑏𝑖 )(𝐶𝑖−1 𝑎̅𝑖 𝑏̅𝑖 )(𝐶 𝑖−1 𝑎𝑖 𝑏𝑖 ) (5)
The Si value in formula (3) is calculated by adding modulo The functional scheme for the sum of Si constructed
2 (Si1 = ai⊕bi), then Si = Si1 ⊕ Ci-1 is calculated (the first according to the formula (5) is shown in Figure 5
option). Figure 4 shows a functional scheme for calculating
Si based on two addition schemes modulo two.
Si1 =𝑎̅𝑖 𝑏𝑖 𝑎𝑖 𝑏̅𝑖
(4)
𝑖 𝐶𝑖−1 𝑆𝑖 𝐶𝑖−1
̅̅̅
Si =𝑆 1 1 ̅̅̅̅̅
- 60 -
To construct the sum of Si0 on the fans AND-NOT, OR-
NOT and NOT, formula (6) must be reduced to the form
̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅ ̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
Si0 = (𝑎𝑖 𝑏𝑖 )(𝑎 𝑖 𝑏𝑖 ) = (𝑎𝑖 𝑏𝑖 )(𝑎𝑖 𝑏𝑖 ) = (𝑎𝑖 𝑏𝑖 ) + (𝑎𝑖 𝑏𝑖 )
̅̅̅̅̅ ̅̅̅̅̅ ̅̅̅̅̅̅̅ ̅̅̅̅̅̅̅̅̅
(11)
According to the formulas (7), (9), (10) and (11) it is
possible to put the functional sum of the adder of the
conditional sum. At the same time, the functional circuit must
be supplemented with multiplexers that are controlled by
transferring from the lower digit to select one of Ci0and Ci1
and one of Si0 and Si1, forming a transfer to the next digit Ci
and the value of the sum Si.
Taking into account the above, the functional scheme of
the adder has the form as this scheme is shown in Figure 6.
Figure 5. Functional scheme for the sum of Si constructed according to the
formula five.
According to Figure 5, it is not difficult to calculate that
38 MOSFETs will be required for the implementations of the
Si adder.
According to Figure 3, to calculate Ci, 18 transistors will
be required for a total of 56 transistors. For the formation of
Si, a delay on three logical elements will be required, i.e. tsi =
3 л.э..
Recently, the construction of so-called conditional sum
adders (conditional-sum addition_ CSA) has been vigorously
discussed [6,7,8].
The adder with a conditional sum is also built according
to the truth table. Table 2 shows a modified table of the truth
of the adder.
TABLE II. MODIFIED ADDER TRUTH TABLE
Ci-1 ai bi Si0 Si1 Ci0 Ci1
0 0 0 0 - 0 -
0 1 0 1 - 0 -
0 0 1 1 - 0 -
0 1 1 0 - 1 -
1 0 0 - 1 - 0
Figure 6. Functional diagram of the conditional sum adder
1 1 0 - 0 - 1
1 0 1 - 0 - 1
1 1 1 - 1 - 1 The adder consists of two circuits OR-NOT1 , AND-NOT2
three inverters INV1, INV2 and INV3 and a MUX1 and MUX2
From the first part of Table 2, where Ci-1 = 0 we have multiplexer. At the inputs of the circuits OR-NOT1 and AND-
NOT2, the i-th digits ai and bi of the numbers A and B. At the
Si0 = 𝑎̅𝑖 𝑏𝑖 𝑎𝑖 𝑏̅𝑖 (6) output of the valve AND-NOT2, the sum of Si0 is formed,
which is fed to the input "0" of the MUX1 multiplexer and to
Ci0 = 𝑎𝑖 𝑏𝑖 (7) the input of the inverter INV2 and at its output, the sum of Si
is formed, which enters the input "1" MUX1 multiplexer.
From the second part of Table 2, where Ci-1 = 0 we have From the output of the inverter INV1 transfer Ci0 = 𝑎 ̅̅̅̅̅
𝑖 𝑏𝑖 =
𝑎𝑖 𝑏𝑖 enters the input "0" of the MUX2 multiplexer. From the
Si1 = 𝑎̅𝑖 𝑏̅𝑖 𝑎𝑖 𝑏𝑖 (8) output of the inverter INV3 transfer
Ci1 = 𝑎̅𝑖 𝑏𝑖 𝑎𝑖 𝑏̅𝑖 𝑎𝑖 𝑏𝑖 = 𝑎𝑖 𝑏𝑖 (9) Ci1=𝑎 𝑖 𝑏𝑖 = 𝑎𝑖 𝑏𝑖
̅̅̅̅̅̅̅ (12)
From this table it is also easy to see that it is fed to the input "1" of the MUX2 multiplexer. The control
inputs of the multiplexers are transferred from the lower-
Si = ̅̅̅
𝑆𝑖0 (10) order Ci-1.
The adder works as follows. After the bits ai and bi are fed
to the inputs of the gates OR-NOT1 and AND-NOT1, the
- 61 -
value Si0 is formed at the output of the circuit OR-NOT2, and
the value of the sum Si1 is formed at the output of the inverter
INV2. At the same time, the transfer value Ci0 is formed at the
output of the inverter INV1 and the transfer value Ci1 is
formed at the output of the inverter INV3.
After feeding the transfers Ci0 and Ci1 and the sums Si0 and
1
Si at the input of the corresponding multiplexers by
transferring Ci-1, which is fed to the control inputs of the
multiplexers simultaneously, the value of the sum Si is
formed at the output of the MUX1 multiplexer and at the same
time, the transfer C1 is formed at the output of the MUX2
multiplexer, which is fed to the inputs of the multiplexers of
the next highest digit.
Figure 7 shows the functional diagram of the MUX1
multiplexer on the MOSFET transistor and the MUX1 Figure 8. MUX2 Multiplexer function diagram
operation table 3.
TABLE V. MUX2 MULTIPLEXER TRUTH TABLE
Ci-1 ai bi Ci0 Ci1 Ci
0 0 0 0 0 0
0 0 1 0 1 0
0 1 0 0 1 0
0 1 1 1 1 1
1 0 0 0 0 0
1 0 1 0 1 1
1 1 0 0 1 1
1 1 1 1 1 1
The order of operation of the bit adder with a conditional
sum is given in Table 5.
TABLE VI. THE ORDER OF OPERATION OF THE ADDER WITH
A CONDITIONAL SUM
Figure 7. Functional diagram of MUX1 multiplexer on MOSFETS № Ci-1 ai bi Si0 Si1 Ci0 Ci1 Si Ci
1 0 0 0 0 1 0 0 0 0
2 0 0 1 1 0 0 1 1 0
TABLE III. MODIFIED ADDER TRUTH TABLE 3 0 1 0 1 0 0 1 1 0
4 0 1 1 0 1 1 1 0 1
TABLE IV. 5 1 0 0 0 1 0 0 1 0
6 1 0 1 1 0 0 1 0 1
Ci-1 Si0 Si1 Si 7 1 1 0 1 0 0 1 0 1
0 0 1 0 8 1 1 1 0 1 1 1 1 1
0 1 0 1
From Figure 8, it is not difficult to calculate the number
1 0 1 1 of transistors. To form the transfers of Ci0 and Ci1 and the
1 1 0 1
sums of Si0 and Si1 , 18 transistors are required. To build two
multiplexers, 12 transistors are required.
Figure 8 shows the functional diagram of the MUX2
multiplexer and its operation table (Figure 8). Total Vtr=18+12=30 transistors.
To build a CSA adder, a gate is required AND-NOT for
two inputs (4 transistors), two gates OR-NOT for two inputs The maximum delay time for the formation of Si0 and
1
(2x4=8 transistors), three inverters INV1 / INV3 (3x2=6 Si =4л.э, the delay time on parallel functioning multiplexers
transistors), two multiplexers (2x6=12 transistors). is -2л.7.
Everything will be required (4+8+6+12)=30 transistors. The A summary table of the main parameters for the
Si formation time is 4 л.э. and the delay time on multiplexers considered adders is given in Table 6.
is 2 л.э.
TABLE VII. SUMMARY TABLE OF PARAMETERS OF VARIOUS
ADDERS
- 62 -
Adders Number of Time of Formati Delay
transistors transfer on time on
1. a0=0 b0=1 С0=1
(N) formation Ci л.э MUX According to the line 7 S00=1 S01=0 and C00=0 C01=1
л.э л.э At the same time S0=0 and C1=1
Adder on two- 2. a1=0 b1=1 С1=1
stage logic 62 4 л.э 7 л.э - According to the line 5 S10=0 S11=1 and C10=0 C01=0
(CM-1)
Adder on two At the same time S1=1 and C2=0
schemes 3. a2=1 b2=1 С2=0
50 2 л.э 6 л.э -
"Excluding According to the line 4 S20=0 S21=1 and C2р=1 C21=1
OR" (CM-2) At the same time S2=0 and C2=1
Adder on a
three-input
4. a3=1 b3=0 С2=1
circuit 56 2 л.э 3 л.э - According to the line 7 S30=1 S31=0 and C30=0 C31=1
"Excluding At the same time S3=0 and C4= S5=1
OR" (CM-3) SА+В=100102=1810
Conditional
Sum Adder For N bit adders,Тsm= Тsm=nMUX+4л.7.
30 2 л.э 6 л.э 2 л.э
(CSA)
(CM-4) By MUX=2л.7
Table 6 shows that for the construction of a single-digit ТСМ=(2n+4)л.7 (13)
adder, the minimum number of transistors has an adder with
a conditional sum. The same adder has a minimal delay in the
formation of inter-bit transfers. CONCLUSION
In conclusion, let is consider an example of adding The paper analyzes adders built on the basis of two-stage
numbers to four-digit (N=4) adders with a conditional sum. logic AND-OR-NOT and on two and three-input "exclusive
The functional diagram of which is shown in Figure 9. OR" circuits, synthesized a conditional sum adder. The main
parameters (the number of MOSFETs and the speed of
adders) are determined. It is shown that from the point of
view of the number of transistors and the time of formation
of inter-bit transfers, the conditional sum adder is
advantageous. The efficiency of the adder is shown by an
example. The final part shows a functional diagram of a four-
digit adder with consecutive transfers.
REFERENCES
[1] Ugryumov E. P. Digital circuitry: textbook. Handbook for universities. -
2nd ed. - St. Petersburg: BHV-Petersburg, 2005. - 500 p.: ill.
Figure 9. Functional diagram of a multi–bit adder based on a single-bit
[2] Lekhin S. N. L52 Computer circuitry. - St. Petersburg: BHV-Petersburg,
conditional sum adder
2010. - 672 p.: ill.
[3] David M. Harris and Sarah L. Harris Digital Circuitry and Computer
Let A=a3a2a1 = 1 1 0 02 Architecture ed. Morgan Kaufman, 2013. - 1619 p.: ill.
+ [4] D. Patterson J. J. Hennessy. Computer architecture and design of
computer systems. Fourth ed. - St. Petersburg: Peter 2012.
B= b3b2b1 0 1 0 12 [5] Kartsev M.A. Arithmetic of digital machines. Publishing house
C0=1 12 «SCIENCE», Moscow 1969, 676 p.
1 0 0 1 02 [6] J. Sklansky, “Logic of addition with a conditional sum”, IRE Trans,
Electronic Computing Machines, vol.9, No. 6, pp.226-231, June 1960.
[7] Jian-Jun LUO, Fast binary adder with conditional transfer generation
Using Table 3, let's consider the operation of adding numbers IEEE Transaction on a Computer, VOE. 46. No. 2, February 1997.
A and B: [8] Kuo-XING CHENG and SHOU-WEN CHEN improved the 32-bit
a0=0 b0=1 С0=1 conditional sum Adder for low-power high-speed applications
According to the line 7 S00=1 S01=0 and C00=0 C01=1 JOURNAL OF Information SCIENCE AND ENGINEERING 22, 975-
989 (2006).
At the same time S0=0 and C1=1
- 63 -
August 22(TUE) - 25(FRI), 2023
Oral Session 2
14:00~15:40 Wed, August 23, 2023
Room : A432
Chair: Jung-Ju Kim (Hoseo University, Korea)
- 64 -
Analysis for Fault Characteristics of MMC in
DC Distribution System
Jin-Sol Song 1, Jong-Myoung Kim 1, Seung-Whan Bae 1, Arif Mehdi 1, S Jarjees UI Hassan 1, Abraham.
D. Arefaynie 1, Zeeshan Haider 1, Chul-Hwan Kim 1*, Myung-Hwan Min 2, Tae-Pung An 2
1
Sungkyunkwan University, Republic of Korea
2
ENTEC Electric & Electronic, Republic of Korea
Email: [email protected]
Abstract—Due to the increasing integration of renewable submodule's capacitance rapidly discharges due to a fault.
energy for carbon neutrality, traditional AC distribution During this stage, the fault current quickly reaches its peak
systems have reached their limitations. Therefore, there is a value due to the discharge of the capacitance. Once the MMC
demand for a transition to DC distribution systems, considering goes into a blocking state, the discharge of the capacitance
factors such as increased interconnection capacity of renewable stops, and the fault current from the AC system flows through
energy and improved power conversion efficiency. However, the the MMC into the DC distribution network. This stage is
implementation of DC distribution systems is still an area that referred to as the AC fault current injection stage.
has not been fully realized, and several aspects need to be
researched before its deployment. One of these aspects is the Fig. 2 schematically depicts the flow of fault current in the
protection-related issues of DC distribution systems. In HB submodule during the capacitance discharge stage and the
particular, research on the fault characteristics of MMCs AC fault current injection stage. During the capacitance
(Modular Multilevel Converters), which constitute the DC discharge stage, as shown in the figure, the fault current flows
distribution network, is essential. This paper aims to analyze the through the capacitance due to the repetitive switching of
fault characteristics of DC distribution networks based on HB semiconductor switches. However, when the MMC execute
MMC (Half-Bridge Modular Multilevel Converter) and FB the blocking control, the switches are fixed in the off state, and
MMC (Full-Bridge Modular Multilevel Converter). the capacitance no longer affects the fault. As depicted in the
Fig. 2, only the AC-side fault current flows through the
Keywords— DC System, Fault Characteristic, Grounding,
MMC, MVDC, Protection System
freewheeling diodes towards the DC side fault and in this
stage MMC acts as an uncontrollable rectifier. This implies
I. INTRODUCTION that, in the case of HB MMC, fault current continues to flow
through the MMC, even when block control is executed and it
In order to achieve carbon neutrality, the integration of can not interrupt the fault current itself.
renewable energy into the power system is ongoing, and as of
December 2021, the installed capacity of renewable energy
connected to the power system reached 24.9 GW in the
Republic of the Korea. However, the existing AC power
system has reached its limit in accommodating renewable
energy, necessitating a transition to the next-generation power
system, such as a DC distribution system. The conversion to a
DC distribution system offers advantages such as increased
integration of renewable energy and improved power
conversion efficiency. However, there are several challenges
that need to be addressed as prerequisites for the adoption of
a DC distribution system. One of these challenges is related to
the protection of DC distribution system. In particular,
research on the fault characteristics of MMCs (Modular
Multilevel Converters), which constitute the DC distribution
network, is essential. This paper aims to present the analysis
results of the fault characteristics of a MMC-based DC
distribution system regarding to MMC submodule type,
ground resistor magnitude fault type and fault location. Fig. 1. The Sturucture of MMC and Submodule
II. FAULT CHARACTERISTICS FOR MMC BASED DC
DISTRIBUTION SYSTEM
A. Half-Bridge MMC
The MMC can be classified into two types: Half-Bridge
(HB) MMC and Full-Bridge (FB) MMC, depending on the
type of submodule, as shown in Fig. 1. For HB MMC, its fault
characteristics can be broadly divided into two stages [1]. The
first stage is the capacitance discharge stage, where the
- 65 -
illustrates a 4MVA MMC-based 20kV DC distribution system,
where a DC side fault is simulated at 4 seconds. And there are
three grounding location ① AC grounding of the transformer
star point, ② AC grounding with a star point reactor, ③ DC
grounding. To analyze the grounding method, several
grounding cases are simulated. Table I represents the
simulation cases. The fault current and voltage are simulated
for different submodule type, fault type, fault location and
grounding method to analyze how each factor effect the fault
current and voltage in DC and AC system. The simulation
Fig. 2. Fault Current Flow in the HB Submodule
cases are modeled and executed by the MATALB Simulink.
B. Full-Bridge MMC
FB MMC and HB MMC exhibit similar behavior when a
fault occurs, as the sub-module capacitances discharge,
resulting in a rapid increase in fault current magnitude.
However, the difference between FB MMC and HB MMC lies
in their response after block control in the MMC. In the case
of FB MMC, due to the structural characteristics of the sub-
modules, when block control is executed, the fault current
magnitude rapidly converges to zero. Fig. 3 depicts the flow
of fault current within the FB submodule during block control.
Fig. 4. MMC based DC Distribution System [3]
As shown in the figure, even during block control, fault
current flows through the capacitance via the freewheeling TABLE I. SIMULATION CASES FOR FAULT CHARACTERISTIC OF MMC
diodes, regardless of the direction of the fault current.
However, because the fault current passes through the Simulation Cases
opposite polarity of the capacitance voltage, it rapidly
converges to zero, ultimately resulting in the current being Submodule
HB Submodule, FB Submodule
blocked. In other words, when the MMC is in a block state, no Type
fault current flows from the AC system, and FB MMC is
capable of self-blocking faults [2]. Fault Type Pole-to-Pole Fault, Pole-to-Ground Fault
Fault
Location
1[km], 5[km], 10[km], 20[km]
AC grounding of the transformer star point by 1[Ω]
AC grounding of the transformer star point by 100[Ω]
Grounding
method
AC grounding with a star point reactor by 1[Ω]
AC grounding with a star point reactor by 100[Ω]
DC grounding by 10,000[Ω]
B. Simulation Results for HB MMC
Fig. 5 depicts the capacitance voltage of the HB
submodule when a fault occurs in the DC distribution network.
After the fault occurrence, the capacitance discharges rapidly,
and this discharge continues until the MMC goes into a
blocking state.
Fig. 6 and 7 illustrate the fault voltage and fault current at
the DC and AC sides of the HB MMC-based DC distribution
system under AC grounding of the transformer star point by
different grounding resistance conditions (1 [Ω] and 100 [Ω]).
After the occurrence of a fault, the fault current rapidly
increases and reaches its peak as the capacitance discharges
before the MMC blocks. Subsequently, as the MMC blocks,
fault current flows from the AC side, and the magnitude of the
fault current oscillates around a certain value. The grounding
resistance not only affects the peak value of the fault current
during capacitance discharge but also influences the
magnitude of the fault current injected from the AC side. A
Fig. 3. Fault Current Flow in the FB Submodule During the Block
larger grounding resistance reduces the magnitude of the fault
III. SIMULATION CONDITIONS AND RESULTS current flowing towards the DC side.
A. Simulation Condition
This paper aims to analyze the fault characteristics of HB
MMC and FB MMC-based DC distribution system. Fig. 4
- 66 -
Fig. 8. HB MMC Pole-to-Ground Fault (AC Grounding with a Star Point
Fig. 5. Submodule Voltage during the Fault Reactor by 1[Ω] & 30[H])
Fig. 9. HB MMC Pole-to-Ground Fault (AC Grounding with a Star Point
Fig. 6. HB MMC Pole-to-Ground Fault (AC Grounding of the Transformer Reactor by 100[Ω] & 30[H])
Star Point by 1[Ω])
Fig. 10. HB MMC Pole-to-Ground Fault (DC Grounding by 10,000[Ω])
Fig. 7. HB MMC Pole-to-Ground Fault (AC Grounding of the Transformer
Star Point by 100[Ω]) The grounding resistance also affects the observed
voltages and currents on the AC side. A higher grounding
Fig. 8 and 9 illustrate the fault voltage and fault current at resistance suppresses the magnitude of the fault current
the DC of the HB MMC-based DC distribution system under flowing on the AC side but fails to provide an effective
AC grounding with a star point reactor (30 [H]) by different grounding path, resulting in biasing of the AC voltage and
grounding resistance conditions (1 [Ω] and 100 [Ω]). This ultimately causing overvoltage. Conversely, a smaller
grounding method involves connecting grounding through a grounding resistance allows for a larger fault current on the
large reactor that is wired in a Y configuration in an AC AC side while effectively suppressing overvoltage.
system. As a result, in the event of a fault, the rate of increase
of the fault current on the DC side is significantly slow, C. Simulation Results for FB MMC
leading to a relatively long time for the fault current value to Fig. 11 represents the fault voltage and fault current of an
converge after a momentary transient state. FB MMC-based DC distribution network. Similar to HB
MMC, the fault current rapidly increases as the capacitance
Fig. 10 illustrates the fault voltage and fault current at the discharges, reaching its peak. However, due to the structural
DC of the HB MMC-based DC distribution system under DC characteristics of the FB submodule, once the FB MMC goes
grounding by 10,000 [Ω]. Due to huge impedance of the into a blocking state, it self-blocks the fault, and no further
grounding, the fault current after short transient is very small. fault current flows into the DC distribution network.
- 67 -
The Fig. 14 and 15 illustrate the fault voltage and fault
current at the DC of the FB MMC-based DC distribution
system under AC grounding with a star point reactor (30 [H])
by different grounding resistance conditions (1 [Ω] and 100
[Ω]). When grounding is achieved through a star point reactor,
the fault current magnitude rises very slowly due to the
presence of a large reactor. This slow rise in fault current is
independent of the size of the grounding resistance. As a result,
it can be observed that the fault current does not rise
sufficiently before being interrupted by the FB MMC (Fault
Current Blocking MMC).
Fig. 16 illustrates the fault voltage and fault current at the
DC of the FB MMC-based DC distribution system under DC
Fig. 11. FB MMC Pole-to-Pole Fault
grounding by 10,000 [Ω]. Due to huge impedance of the
grounding, the fault current after short transient is very small
The Fig. 12 and 13 depict the fault voltages and fault like HB MMC.
currents for Pole-to-Ground fault in both DC and AC
measurements under the conditions of grounding resistors of
1 ohm and 100 ohms, respectively. The magnitude of the
grounding resistor had an influence on the observed fault
current in both DC and AC measurements. A smaller
grounding resistor resulted in a significantly larger fault
current. Furthermore, after implementing block control,
similar to Pole- to-Pole faults, the injected current in the AC
measurements was interrupted, causing the fault current to
cease flowing in both DC and AC measurements, and the fault
voltage returned to its pre-fault value.
Fig. 14. FB MMC Pole-to-Ground Fault (AC Grounding with a Star Point
Reactor by 1[Ω] & 30[H])
Fig. 12. FB MMC Pole-to-Ground Fault (AC Grounding of the Transformer
Star Point by 1[Ω])
Fig. 15. FB MMC Pole-to-Ground Fault (AC Grounding with a Star Point
Reactor by 100[Ω] & 30[H])
Fig. 13. FB MMC Pole-to-Ground Fault (AC Grounding of the Transformer
Star Point by 100[Ω])
- 68 -
ground with an appropriate impedance to strike a balance
between these factors.
ACKNOWLEDGMENT
This work was supported by the Korea Institute of Energy
Technology Evaluation and Planning(KETEP) and the
Ministry of Trade, Industry & Energy(MOTIE) of the
Republic of Korea (No. 20225500000110).
This work has been supported by the National Research
Foundation of Korea (NRF) grant funded by the Korean
government (MSIP) (No. 2021R1A2B5B03086257)
REFERENCES
[1] Geon-Woong Kim and Hyo-Sang Choi, “Limiting Characteristics of
Capacitor Discharge Current of MMC-Based System Using the SFCL
Fig. 16. FB MMC Pole-to-Ground Fault (DC Grounding by 10,000[Ω]) on Short Circuit”, IEEE Transactions on Applied Superconductivity,
vol. 32, no. 4, pp. 5600505, 2022.
D. Simulation Results for Different Fault Location [2] Bin Li, Jiawei He, Ye Li and Botong Li, “A review of the protection
for the multi-terminal VSC-HVDC grid”, Protection and Control of
The Fig. 17 depicts the fault currents for Pole-to-Pole fault Modern Power Systems, vol. 4, 2019, no. 21, pp. 1-11, 2019.
in HB MMC based DC system for different fault location (1, [3] CIGRE WG C6/B4.37, “Medium Voltage DC Distribution Systems”,
5, 10, 20[km]). As the fault distance increases, the line 2022
impedance up to the fault location also increases, resulting in
a decrease in the peak value of the fault current due to the
influence of MMC submodule capacitance discharge.
However, in the fault current injection stage from the AC
system after capacitance discharge, the influence of the fault
location is negligible.
Fig. 17. Fault Current by Pole-to-Pole Fault Regarding to Fault Location
IV. CONCLUSION
In this paper, the fault characteristics of HB MMC and FB
MMC were investigated as preliminary tasks for the
introduction of DC distribution networks. Both MMC types
exhibit the characteristic of rapid increase in fault current as
the capacitance discharges during a fault occurrence. However,
HB MMC allows fault current to flow through the MMC even
after the blocking control, while FB MMC has the capability
to block the fault current. These features were confirmed
through simulations. Furthermore, the characteristics of fault
currents were studied in relation to grounding methods, and it
was observed that increasing the fault resistance leads to a
decrease in the maximum value of the fault current.
Additionally, when grounded through a large reactor on the
AC side, it was confirmed that the rate at which the fault
current rises can significantly slow down. The small
magnitude of fault current and its slow rise can help to prevent
damage of the power system equipment and enhance system
continuity. However, detecting and protecting against faults
becomes very challenging, and there is a risk of inducing
overvoltage on the AC side. Therefore, it is necessary to
- 69 -
Assessment of Locational and Total Hosting
Capacity of Distribution Network with Multitype
DGUs by Iterative OPF Method
S Jarjees Ul Hassan 1, Arif Mehdi 1, Jin-Sol Song 1, Abraham D. Arefaynie 1, Zeeshan Haider 1, and
Chul-Hwan Kim 1*
1
Department of Electrical and Computer Engineering, Sungkyunkwan University, Suwon, Republic of Korea
Email: [email protected]
Abstract—The increasing demand for power generation has The hosting capacity is a location-dependent and varies
spurred rapid development in renewable energy sources to meet from one bus to another within the system. Locational Hosting
the growing energy requirements. While distributed generation Capacity (LHC) refers to the hosting capacity at a specific bus,
units (DGUs) offer several advantages, their excessive while total hosting capacity represents the maximum hosting
integration into electrical networks can lead to various capacity of the entire distribution system. Research on hosting
challenges. To maximize the benefits while minimizing capacity can be broadly categorized into assessment and
problems, it is crucial to assess the total hosting capacity of enhancement. Assessment methods involve various analytical
distribution systems for interconnecting DGUs. This research [6], stochastic [7], [8] and heuristic techniques [9], such as
aims to evaluate both the locational and total hosting capacity of
Monte Carlo simulations, Bayesian optimization, particle
a distribution system that incorporates multi-types single and
multiple DGUs while satisfying all relevant constraints. The
swarm algorithm, genetic algorithms, bat algorithms, and
assessment is performed using an iterative optimal power flow robust optimization [10]–[12]. The enhancement techniques
(OPF) method, which is described in detail in this study. The for hosting capacity focuses on network reconfiguration and
IEEE-33 bus radial distribution system is employed as a case reinforcement, reactive power control, voltage control using
study, and the OPF method is implemented using MATLAB on-load tap changer transformers, active power curtailment,
programming. The hosting capacities obtained with different energy storage technologies, and smart inverters [13].
types of DGUs are compared, and the most favorable case is
This paper presents an assessment of the locational and
presented.
total hosting capacity of a distribution system with multiple
Keywords—Locational hosting capacity, total hosting types of DGUs using an Iterative OPF method. The DGUs are
capacity, DGU types, Iterative optimal power flow method, categorized into three types based on their active and reactive
renewable energy resources power generation capabilities. A detailed explanation of the
OPF iterative method for the hosting capacity objective
I. INTRODUCTION function is provided. The proposed method is validated using
The deployment of distributed generation units (DGUs) in the IEEE-33 bus distribution system, and the results
distribution systems has revolutionized the field of power demonstrate the hosting capacities of the system for different
systems. In addition to economic and environmental types of single and multiple DGUs, enabling a comparison to
advancements, DGUs offer various technical benefits, identify the best-case scenario.
including improved voltage profiles, minimized power losses The remainder of this article is organized into the
[1], enhanced system stability, and reliability. However, the following sections. Section II provides an introduction to the
excessive installation of DGUs in distribution systems can base test system and the types of DGUs used in this study. The
have detrimental effects, such as overvoltage, undervoltage, assessment of hosting capacity is then presented in Section III,
equipment thermal overloading, and protection system where the methodology and implementation of the OPF
malfunctions [2], [3]. Hence, it is essential to calculate the iterative method for calculating hosting capacity are described
hosting capacity of the system, which determines the in detail. In Section IV, the simulation results are presented
maximum number of DGUs that can be interconnected and discussed, highlighting the hosting capacities of the
without violating operational constraints. system for different types of single and multiple DGUs. The
An extensive research review over the past decade shows implications and significance of these findings are thoroughly
the importance of hosting capacity in the field of electrical examined. Finally, Section V concludes the paper by
power system. The concept of hosting capacity is complex and summarizing the key findings and suggesting potential
needs special attention to be addressed. The literature review directions for future research in the field of hosting capacity
defines the hosting capacity as, “The maximum DGUs assessment.
penetration in distribution system up to acceptable range,
subjected to operational limits” [4]. The operational limits also
known as limiting factor of hosting capacity are voltage level,
ampacity, and power balance [5]. The concept of hosting
capacity is illustrated in Fig. 1.
- 70 -
Distributed generation units are designed as negative load,
having ability of providing only or both active and reactive
powers into the distribution network from PQ buses. The
hosting capacity of distribution system is calculated with all
these types of categorized DGUs.
III. ASSESSMENT OF HOSTING CAPACITY
The hosting capacity study conducted in this work was carried
out by OPF iterative method. The objective function with
constraints and OPF iterative method detailed description are
presented in this section.
Fig. 1. General concept of hosting capacity The objective function considered for this work is to
calculate the maximum hosting capacity of distribution
II. SYSTEM DESCRIPTION system form bus ‘1’ to ‘n’th bus. Where “i” is the number of
buses starting from 1 to last bus “n” of distribution system.
A. Base Test System
This objective function can be presented mathematically as
The base test system selected for this study is the IEEE-33 (1):
bus radial distribution system, as depicted by a single line
𝑛
diagram in Fig. 2. This test system is widely used in research
and provides a realistic representation of distribution 𝑂𝑏𝑗𝑒𝑐𝑡𝑖𝑣𝑒 𝐹𝑢𝑛𝑐𝑡𝑖𝑜𝑛 = 𝑀𝑎𝑥𝑖𝑚𝑢𝑚 ∑ 𝐻𝐶 (1)
networks. The substation, represented by bus 1, supplies a 𝑖=1
voltage of 12.66 kV to the distribution system. The load buses
within the distribution system contribute a total active power The objective function of hosting capacity is subjected to the
of 3715 kW and a total reactive power of 2300 kVAR [14]. following constraints as (2) and (3).
These values reflect the combined power demands of the
connected loads, representing a typical distribution scenario. 𝑉𝑚𝑖𝑛 ≤ |𝑉𝑖 | ≤ 𝑉𝑀𝑎𝑥 (2)
The IEEE 33-bus radial distribution system serves as a 𝐼𝑚𝑖𝑛 ≤ |𝐼𝑖 | ≤ 𝐼𝑀𝑎𝑥 (3)
suitable testbed for evaluating the hosting capacity of the
distribution system with multiple types of DGUs. By utilizing The OPF iterative method is used for assessment of
this base test system, the impacts of DGU integration and their hosting capacity with objective function as in (1) subjected to
influence on system performance can be effectively studied the operational constraints of voltage and current mentioned
and analyzed. in (2) and (3). It is an iterations-based method that follows a
systematic procedure by increasing the DGUs capacity
penetration at each bus iteratively. The steps of this
methodology for assessment of hosting capacity of
distribution system with single and multiple DGUs
penetration is as follow.
A. Assessment of Hosting Capacity with Single DGU:
Step 1) Start the program with reading line and load date
of distribution system.
Step 2) Start the iteration from selecting the starting bus
from i=1, 2,…,n.
Fig. 2. Single line diagram of IEEE 33-bus network Step 3) Add a single DGU of 1KW capacity as a
negative load on bus (i).
B. Distributed Generation Units
Step 4) Perform load flow calculations to check the
The hosting capacity has also dependency on optimal limits of operational constraints.
placement and sizing of DGUs. The DGUs are categorized
into three types based on their active and reactive power Step 5) If limits of constraints are not violated, go to step
generation capability. 3 with step increase in capacity of DGU. If the
limits are violated go to step 6.
• Type-1: DGUs that injects only real power 'P' into the
distribution system. These units primarily contribute to Step 6) Print the DGU capacity of previous iteration as
meeting the active power demand of the system. a hosting capacity of bus (i).
• Type-2: DGUs that injects only reactive power 'Q' into Step 7) Apply the same procedure for all buses by going
the system. These units are primarily responsible for back to step 2 with updating the iteration and
addressing reactive power requirements, such as end the program if all buses are investigated.
voltage regulation and power factor correction.
• Type-3: DGUs that injects both active and reactive
power into the network. These units provide a
combination of real and reactive power, contributing
to both active power demand and reactive power
requirements of the system.
- 71 -
B. Assessment of Hosting Capacity with Multiple DGUs: DGUs with respect to number of DGUs are calculated using
(4) as:
Step 1) Start the program with reading line and load date 𝑁𝐵
of distribution system. 𝑁𝐶 = (4)
𝑁𝐷𝐺 ! × (𝑁𝐵 − 𝑁𝐷𝐺 )
Step 2) Choose number of DGUs for injection into
distribution system.
IV. RESULTS AND DISCUSSIONS
Step 3) For multiple DGUs, we get multiple buses as This section presents the results of the assessment
option to locate. Therefore, it is necessary to conducted to evaluate the locational and total hosting capacity
make a combination of NB buses with number of of the 33-bus distribution system, considering the presence of
DGUs (NDGU) using (4) and select first iteration all three types of single and multiple DGUs. The obtained
i=1,2, 3,..,n. results provide insights into the system's capability to
Step 4) In the same way, it is also necessary to construct accommodate distributed generation and shed light on the
a matrix of combinations of all possible impacts of different DGU configurations on the hosting
capacities of DGUs by (4) and start with capacity.
k=1,2,3…., n. The assessment of locational hosting capacity considers
Step 5) Follow the step 4 and step 5 as mentioned for the maximum DGU’s capacity integration at each bus within
Single DGU. the distribution system, providing valuable information on the
maximum number of DGUs that can be interconnected
Step 6) Find the maximum in all DGUs capacities which without violating operational constraints. Furthermore, the
did not violate any constraints and print it as total hosting capacity considers the optimal number of buses
“Hosting Capacity”. with maximum capability of DGUs’ capacity integration,
The steps of OPF iterative method for assessment of hosting which represents the overall ability to integrate maximum
capacity are also presented in Fig. 3 (a) and (b) and the DGUs capacity across the entire distribution network.
combination for bus numbers and all possible capacities of
Fig. 3. Assessment of hosting capacity with iterative OPF method.
- 72 -
A. Type-1 DGUs investigation on IEEE 33-bus distribution system with
The OPF iterative method is applied on type-1 DGUs to single DGU integration the bus 2 is proven to be the optimal
get the locational hosting capacity of each bus, and total location with maximum locational hosting capacity of
hosting capacity of distribution system with single and two 9559K with type-3 DGU as compared to type-1 and type-2
DGUs as shown in Fig. 4. DGU which is 8340K and 5660K respectively.
TABLE I. LOCATIONAL HOSTING CAPACITY OF 33-BUS
DISTRIBUTION SYSTEM WITH MULTI-TYPE DGUS
Bus No. Type-1 LHC Type-2 LHC Type-3 LHC
(KW) (KVAR) (KW+KVAR)
1 0 0 0
2 8340 5660 8550+j4300
3 8020 5640 8160+j4080
4 7200 5680 7160+j3580
5 7160 5720 7080+j3540
6 7100 5940 5640+j2820
7 4940 4180 4620+j2310
8 4800 4140 4160+j2080
9 3260 2740 3020+j1510
Fig. 4. Locational and total hosting capacity of 33-bus system with
10 3240 2760 2760+j1380
single and two type-1 DGUs.
11 3180 2760 2680+j1340
12 3140 2740 2560+j1280
B. Type-2 DGUs 13 2700 2780 2120+j1060
The results of locational hosting capacity and total 14 2540 2780 1980+j990
hosting capacity of 33-bus distribution system with type-2 15 2380 2700 1840+j920
single and two DGUs are presented in Fig. 5. 16 2200 2500 1720+j860
17 1940 2100 1480+j740
18 1820 1960 1400+j700
19 2880 2660 2620+j1310
20 2860 2680 2600+j1300
21 2780 2680 2540+j1270
22 2720 2680 2160+j1080
23 3420 2780 3200+j1600
24 3380 2780 3140+j1570
25 3000 2700 2760+j1380
26 3260 3280 3400+j1700
27 3280 3280 3400+j1700
28 3320 3300 3440+j1700
29 3360 3340 3200+j1600
30 3100 3280 2980+j1490
Fig. 5. Locational and total hosting capacity of 33-bus system with 31 3040 2720 2520+j1260
single and two type-2 DGUs. 32 2900 2700 2400+j1200
33 2680 2620 2260+j1130
C. Type-3 DGUs
The tests are also performed on type-3 DGUs to The total hosting capacity of IEEE 33-bus distribution
calculate the locational hosting capacity of every bus, total system with single and two DGUs injection are also
hosting capacity with single and two DGUs. The results are presented numerically in Table 2 for all three types of
as shown in Fig. 6. DGUs. These results indicates that the total hosting capacity
increases with increase in number of DGUs injection at
optimal bus locations. The results clearly shows that the
total hosting capacity of distribution system is maximum
with two DGUs penetration of type-3 that is 9615K as
compared to type-1 and type-2 DGUs which is 8500K and
6200K respectively.
TABLE II. TOTAL HOSTING CAPACITY OF 33-BUS DISTRIBUTION
SYSTEM WITH MULTI-TYPE DGUS
DGU No. of Optimal DGU Total HC
Type DGUs Bus Capacities
1 1 2 8340 8340
Fig. 6. Locational and total hosting capacity of 33-bus system with 2 6, 19 7100, 1400 8500
single and two type-3 DGUs. 2 1 2 5660 5600
2 10,29 2800, 3400 6200
The results of locational hosting capacity with single 3 1 2 8550 8550+j4275
2 4,19 7100, 1500 7100+j3550
type-1, type-2, and type-3 DGU injection are shown 1500+j750
numerically in Table 1. It shows that each bus location has
different locational hosting capacity within the distribution
system which highlights the optimal placement and sizing
of DGUs considering all the operational constraints. By
- 73 -
V. CONCLUSIONS capacity in distribution grids: A review of definitions,
This paper provided a comprehensive analysis of the performance indices, determination methodologies, and
locational and total hosting capacity of a distribution system enhancement techniques,” Energy Sci. Eng., vol. 11, no. 4, pp.
with single and two multi-type DGUs. The investigation 1536–1559, 2023, doi: 10.1002/ese3.1389.
utilized the OPF iterative method to optimize the objective [4] M. Bollen and F. Hassan, Integration of Distributed Generation
function while considering the constraints of the standard in the Power System, no. February. IEEE Press, 2014.
IEEE-33 bus distribution system. The simulation results of
[5] S. Fatima, V. Püvi, and M. Lehtonen, “Review on the PV
the locational and total hosting capacity were presented
graphically and numerically, contributing valuable insights hosting capacity in distribution networks,” Energies, vol. 13, no.
to the field of power system research. 18, 2020, doi: 10.3390/en13184756.
[6] S. J. U. L. Hassan, T. Gush, and S. Member, “Maximum Hosting
1) The hosting capacity is a concept that varies from
Capacity Assessment of Distribution Systems With Multitype
one bus to another within the distribution system.
DERs Using Analytical OPF Method,” IEEE Access, vol. 10,
Therefore, the optimal placement and sizing of DGUs are
no. September, pp. 100665–100674, 2022, doi:
crucial factors to consider. By strategically locating DGUs,
10.1109/ACCESS.2022.3207488.
the hosting capacity of distribution system can be
maximized, leading to efficient integration of DGUs. [7] E. Mulenga, M. H. J. Bollen, and N. Etherden, “Solar PV
2) The results clearly demonstrate that the total hosting stochastic hosting capacity in distribution networks considering
capacity of the distribution system increases with an aleatory and epistemic uncertainties,” Int. J. Electr. Power
increase in the number of DGUs injections. This indicates Energy Syst., vol. 130, no. February, p. 106928, 2021, doi:
the system's ability to accommodate a greater amount of 10.1016/j.ijepes.2021.106928.
distributed generation, resulting in enhanced utilization of [8] Y. J. Liu, C. Y. Lee, W. M. Liu, Y. Der Lee, C. C. Cheng, and
renewable energy resources. Y. F. Chen, “Optimization-Based Stochastic Analysis Method
3) The hosting capacity of the distribution system is for the Assessment of PV Hosting Capacity,” Proc. 2022 8th
maximized when type-3 DGUs, which inject both active Int. Conf. Appl. Syst. Innov. ICASI 2022, pp. 33–36, 2022, doi:
and reactive power, are integrated. The combined 10.1109/ICASI55125.2022.9774450.
contribution of active and reactive power from these DGUs [9] C. Han, D. Lee, S. Song, and G. Jang, “Probabilistic Assessment
enables better support for system stability, voltage of PV Hosting Capacity under Coordinated Voltage Regulation
regulation, and power factor correction. in Unbalanced Active Distribution Networks,” IEEE Access,
In future research, conducting a cost analysis of each case vol. 10, pp. 35578–35588, 2022, doi:
would be beneficial. This analysis would provide insights 10.1109/ACCESS.2022.3163595.
into the economic aspects of integrating DGUs and help [10] H. Wu, Y. Yuan, X. Zhang, A. Miao, and J. Zhu, “Robust
identify the most cost-effective approaches for enhancing comprehensive PV hosting capacity assessment model for
the locational and total hosting capacity of the distribution active distribution networks with spatiotemporal correlation,”
system. Additionally, exploring advanced techniques and
Appl. Energy, vol. 323, no. July, p. 119558, 2022, doi:
technologies to further optimize the hosting capacity, such
10.1016/j.apenergy.2022.119558.
as advanced control strategies, energy storage integration,
and intelligent forecasting algorithms, could be intriguing [11] X. Geng, L. Tong, A. Bhattacharya, B. Mallick, and L. Xie,
avenues for future investigations. “Probabilistic Hosting Capacity Analysis via Bayesian
Optimization,” IEEE Power Energy Soc. Gen. Meet., vol. 2021-
ACKNOWLEDGMENT July, pp. 1–16, 2021, doi:
This work was supported by the National Research 10.1109/PESGM46819.2021.9637907.
Foundation of Korea (NRF) funded by the Korean [12] M. Vellingiri et al., “Maximum hosting capacity estimation for
Government [Ministry of Science, ICT and future Planning renewables in power grids considering energy storage and
(MSIP)] under Grant 2021R1A2B5B03086257 and Korea transmission lines expansion using hybrid sine cosine artificial
Institute of Energy Technology Evaluation and Planning rabbits algorithm,” Ain Shams Eng. J., vol. 14, no. 5, p. 102092,
(KETEP) and the Ministry of Trade, Industry & Energy
2023, doi: 10.1016/j.asej.2022.102092.
(MOTIE) of the Republic of Korea (No. 20225500000110).
[13] V. C. Moro and F. C. L. Trindade, “A Review on Techniques to
REFERENCES Increase Hosting Capacity of Distribution Systems to DERs,”
[1] S. J. Ul Hassan, W. Haider, A. Mehdi, T. Gush, J. S. Song, and 2021 Work. Commun. Networks Power Syst. WCNPS 2021, no.
C. H. Kim, “Optimal Placement and Sizing of Multi-Type Wcnps, pp. 1–6, 2021, doi:
Distributed Generation Considering Voltage Profile 10.1109/WCNPS53648.2021.9626269.
Improvement and Power Loss Minimization,” Trans. Korean [14] W. Haider, S. J. Ul Hassan, A. Mehdi, A. Hussain, G. O. M.
Inst. Electr. Eng., vol. 70, no. 4, pp. 561–567, 2021, doi: Adjayeng, and C. H. Kim, “Voltage profile enhancement and
10.5370/KIEE.2021.70.4.561. loss minimization using optimal placement and sizing of
[2] S. M. Ismael, S. H. E. Abdel Aleem, A. Y. Abdelaziz, and A. F. distributed generation in reconfigured network,” Machines, vol.
Zobaa, “State-of-the-art of hosting capacity in modern power 9, no. 1, pp. 1–16, 2021, doi: 10.3390/machines9010020.
systems with distributed generation,” Renew. Energy, vol. 130,
pp. 1002–1020, 2019, doi: 10.1016/j.renene.2018.07.008.
[3] N. Qamar, A. Arshad, K. Mahmoud, and M. Lehtonen, “Hosting
- 74 -
Optimal Parameter Design of Thyristor-based
Multi-module SSCB in MVDC Network
Jong-Myoung Kim 1, Jin-Sol Song 1, Seung-Whan Bae 1, and Chul-Hwan Kim 1*
1
Sungkyunkwan University, Republic of Korea
Email: [email protected]
Abstract— With the increasing penetration of Photovoltaic system components. Consequently, DCCB require a short
(PV) and other Renewable Energy Sources (RES), the DC interruption time of within 2ms, unlike AC circuit breakers [4].
source in the power system is rapidly increasing. As a result,
there is active discussion regarding the introduction of Medium Therefore, it is crucial to ensure both fast interruption and
Voltage DC (MVDC) distribution networks in Korea. Because consistent interruption time in DCCB to establishing a
DC fault have different characteristics compared to AC fault, protection coordination system for MVDC distribution
research on DC circuit breakers is essential for establishing networks. If there is significant variation in the interruption
protection system in MVDC distribution networks. In this paper, time depending on the type and other condition of the fault, it
research on the optimal parameter design of a SCR-based solid- becomes challenging to set a time delay between the main
state circuit breaker is conducted. Proposed SSCB combines the protection and backup protection. This can potentially lead to
advantages of SSCB with fast interruption time and outage in areas of the network where faults have not occurred.
interrupting method of HCB that generate artificial zero- However, existing research on DCCB has primarily focused
current crossing point. The proposed SSCB is capable of on fast interruption and interruption reliability, while studies
performing the interruption within 50us upon receiving a trip on consistent interruption time are lacking.
signal from the relay. It is expected to contribute to the design
of a reliable protection coordination system. The validation of The Solid-State Circuit Breaker (SSCB) proposed in this
the operation of the SSCB was performed by paper performs interruption by injecting LC resonant current
MATLAB/Simulink. into the main branch to create an artificial zero-crossing point.
The main branch employs a thyristor arrangement, enabling
Keywords—Current Injection, MVDC, PSO, SSCB, Zero- its operation under high voltage and high current conditions.
current Switching Upon the occurrence of the zero-crossing, the Thyristor, as a
semi-controlled device, undergoes a state transition, leading to
I. INTRODUCTION
the disconnection of the fault from the DC network. The
As carbon emissions increase, the climate crisis is auxiliary branch for current injection is configured in a multi-
becoming increasingly severe problem, and countries around module structure to control the injected current for various
the world are showing a movement to replace conventional conditions of DC faults and perform the interruption within
power sources such as coal-fired power generation with the predetermined time. In this paper, considering the faults
Renewable Energy Sources (RES) in response to the climate occurring within the MVDC distribution network, the
crisis. According to the 10th Basic Plan for Power Supply and common capacitance and the inductance of each module of the
Demand announced in January 2023, the share of renewable auxiliary branch are optimized using the Particle Swarm
energy generation is expected to account for 21.6% of the total Optimization (PSO) algorithm. The proposed SSCB is
generation capacity by 2030 [1]. Photovoltaic (PV) is one of validated by MATLAB/Simulink for faults occurring in the
the prominent DC source that accounts for over half of the MVDC distribution network.
share of renewable energy sources. As the integration of
renewable energy sources within the power grid increases,
there is a growing need for Medium Voltage DC (MVDC) II. DC CIRCUIT BREAKERS
distribution networks. Since faults in DC systems have
different characteristics from AC faults, research on A. Type of DCCB
protection systems for DC systems is essential and must be DCCBs can be broadly classified into two main types:
conducted as a prerequisite. Solid-State Circuit Breakers (SSCBs) and Hybrid Circuit
Because of its constant output nature, DC faults are Breakers (HCBs) [2]. SSCBs achieve interruption through the
characterized by the absence of zero current crossing points, switching operation of semiconductor devices, resulting in
unlike AC faults. In conventional AC systems, circuit breakers remarkably short interruption time, typically within tens of
are designed to interrupt the current at zero-crossing points to [3]. This inherent capability of SSCBs helps maintain high
minimize arc formation. However, in the case of DC faults, power quality during fault by minimizing voltage drops. The
separating contacts without zero current crossing points can term "solid-state" comes from the fact that there are no
lead to an increased risk of arc formation. Therefore, in the physical moving parts in the interruption process. This
case of DC Circuit Breakers (DCCBs), an artificial zero- characteristic not only prevents the occurrence of arcs during
crossing point needs to be created to interrupt the current. interruption but also eliminates issues like corrosion and noise
Additionally, DC systems have lower inertia within the that may occur during contact separation. As a result, SSCBs
system, causing fault currents to reach their peak values in a demonstrate an extended lifespan. However, it is important to
shorter time compared to AC systems. The maximum fault note that the semiconductor devices commonly utilized in
current in DC systems is also much higher compared to AC SSCBs, such as MOSFETs, have lower ratings and are not
systems [3]. The rapid current variation and high fault current well-suited for applications in high voltage and current.
within a short duration can lead to breakdown of insulation in Consequently, SSCBs are primarily utilized in LVDC and
- 75 -
lower MVDC systems. Furthermore, the on-state resistance in
semiconductor devices results in significant conduction losses,
leading to overheating. This can potentially cause
performance degradation and reduce the overall lifespan of
SSCBs. Therefore, while SSCBs offer advantages in terms of
fast interruption and enhanced power quality, careful
consideration must be given to their specific application
requirements and limitations.
On the contrary, Hybrid Circuit Breakers (HCBs)
represent a combination of semiconductor devices and
conventional Mechanical Circuit Breakers (MCBs) utilized in
AC systems. Under normal operating conditions, current is
flowed through the MCB to minimize conduction losses.
However, when fault occurs, the semiconductor branch
generates an artificial zero-crossing point, MCB separating
their contacts to interrupt the fault. HCBs effectively address
the limitations of SSCBs, such as high conduction losses and
lower ratings, by integrating them with MCBs. Nevertheless, Fig. 1. DCCB structure (a) conventional SSCB, (b) current commutation
the interruption time in HCBs is influenced by the operation HCB, (c) current injection HCB, (d) Proposed thyristor-based SSCB
of the MCBs, leading to long interruption time. Moreover, the
long interruption time in HCBs may result in lower power
quality during the fault blocking process. Additionally, the
physical separation of contacts in the MCBs can give rise to C. Structure of proposed SSCB
arc generation, affecting power quality and electrical safety. The proposed SSCB in this study utilizes thyristor to the
Furthermore, concerns related to contact corrosion and noise main branch. Thyristors are suitable for high voltage and high
associated with mechanical circuit breakers still need to be current applications, with a high rated voltage range of 6.5kV
taken into account. to 8.5kV and a high rated current of 1kA. Additionally,
thyristors exhibit low conduction losses and offer cost-
effectiveness compared to other semiconductor devices. In
B. Interruption method of HCB this paper, the current injection method, one of the
Hybrid circuit breakers utilize two primary interruption interruption method of HCBs, is utilized to generate the zero-
methods: current commutation and current injection [4]. The crossing point in thyristor and deactivate it.
operational principle of a Current Commutation (CC) method The auxiliary branch, responsible for current injection, is
is commutating fault current to flow auxiliary branch, creating constructed as a multi-module structure. The selection of the
an artificial zero-crossing point in main branch. When a fault appropriate current injection module is based on the fault
occurs, the MCB and the series-connected Load Commutation current's magnitude. Accurate consideration of the fault
Switches (LCS) are turned off, while the parallel-connected current's magnitude is critical for effective current injection.
semiconductor branch is turned on. The fault current is then Injecting a current lower than the fault current may result in
flowed through the semiconductor branch, generating an
failure to deactivate the SCR, whereas injecting a
artificial zero-crossing point in the MCB. As a result, the
significantly higher current than the fault current may induce
MCB separates its contacts, achieving a physical interruption
of the fault. Following the successful separation of contacts, reverse voltage and current in the SCR, potentially causing
the semiconductor branch is turned off to isolate the fault from insulation breakdown. Therefore, prediction of the fault
the system. Remaining energy in the system is discharged current's magnitude is essential to ensure the circuit breaker's
through surge arresters. operational reliability and performance.
Upon turning off the thyristor, the semiconductor devices
The operational principle of a Current Injection (CI) within the current injection module are deactivated to cease
method is injecting current in auxiliary branch to create an current injection. After that, a Surge Arrester (SA) is
artificial zero-crossing point in main branch [5]. When fault employed to discharge the remaining energy within the
occurs, semiconductor devices in the auxiliary branch become system. The current resulting from this energy discharge
on-state, allowing the series connected precharged capacitor
gradually decreases, ensuring that the rate of change of the
and inductor to induce an LC resonant current. Consequently,
the current flowing through the main branch, which comprises current, which is utilized for fault detection by the differential
the MCB, gradually diminishes, leading to the formation of a relay, does not trigger false operations.
zero-crossing point. As the current approaches zero, the MCB
separates its contacts. The LC resonant current continues to The proposed SSCB offers the following advantages:
decrease over time until it is eventually interrupted by the · Suitable for high voltage and high current: It addresses
thyristor. As with the CC method, any remaining energy in the the limitations of typical SSCB by utilizing thyristor as
system is discharged through surge arresters, thereby main branch, And it doesn’t have disadvantages of
completing the interruption process. MCB, such as arc generation and noise
· Fast interruption time: Among the MCBs, the Ultra
Fast Disconnector (UFD) stands out with the shortest
blocking time of 1.7 ms. And the High-Speed Circuit
Breaker (HSCB) have longer interruption time than
- 76 -
UFD, To establish protection coordination system in represents the common capacitance, and is the initial
DC, DC faults should be blocked in a 2ms. Therefore, voltage across the common capacitance. is the equivalent
using HCBs to establish protection coordination inductance of the injection module, and is the on-state
system becomes challenging. However, the proposed resistance of the MOSFET. is the angular velocity of the
SSCB, with its very fast interruption time of 50 , is LC resonance current. The waveform of the LC resonant
well-suited for building a protection coordination current is depicted as shown in Fig. 2. The waveform used for
system in MVDC network. current injection is the waveform up to reaching the maximum
value. The time it takes for the injected current to reach its
· Interruption time control: The proposed SSCB maximum value can be determined by using the derivative of
provides the capability to control the interruption time the injection current.
by selecting the injected LC resonant current. This
attribute facilitates the consideration of interruption
time when devising the CTI curve for relays employed
in forthcoming MVDC distribution networks, thereby
enabling the design of time delays.
III. DESIGN OF PROPOSED SSCB
Fig. 3. LC resonance current waveform
A. Selection scheme of injection current
In this paper, interruption time of the SSCB is set to be
() = [ cos( ) − sin( )] (4)
within 50 , considering both the sampling period of the ×
differential relay and the switching period of the MOSFET. In
order to determine the magnitude of the injection current, it = tan (5)
becomes necessary to predict the magnitude of the fault
current 50 after the current injection. As depicted in Figure
According to equation (2) and (3), as the inductance
00, the initial rate of change of the fault current in DC faults
exhibits a continuous decrease. Hence, it can be expressed the decreases, the injected current increases, while the angular
magnitude of the fault current 50 after the injection using velocity decreases. This implies that a smaller inductance
the following equation. leads to a higher injected current and a shorter time to reach
the maximum value. Additionally, as described in equation (5),
+ 50 = + 50 × 10 ≥ ( + 50 ) (1) as angular velocity decreases, the time takes for the injected
current to reach its maximum value decreases.
C. LC resonance parameter optimization
To optimize the parameters of the current injection branch,
PSO algorithms were utilized for twice. The first PSO
algorithm aimed to find the optimal solutions for the overall
equivalent inductance , common capacitance , and initial
voltage . The injected current reaches its maximum value
when all modules are in the on state, corresponding to the
scenario with the lowest equivalent inductance. Moreover, the
time taken to reach the maximum value is minimized in this
situation. However, to ensure the reliability of the interruption
time, it is necessary for the time taken to reach the maximum
injected current to be longer than the target interruption time
of 50 . This constraint is imposed to secure the desired
interruption time reliability.
Fig. 2. Changing rate of fault current in MVDC network
≥ (6)
By selecting a higher injection current than the predicted
value, the SSCB can achieve a higher level of interruption To ensure the formation of an artificial zero-crossing point
reliability. And the interruption time cannot exceed 50 . in the main branch, the maximum injected current should be
greater than or equal to the maximum fault current. In DC
systems, there are two common fault types: Pole-To-Pole
B. LC resonance current analysis (PTP) fault and Pole-To-Ground (PTG) fault. Since the PTP
fault current is generally larger than PTG fault, the maximum
The LC resonant current injected by the auxiliary branch injected current should satisfy the following condition.
can be expressed by the following equation.
≥ (7)
() = sin ( ) (2)
×
The objective function considering both conditions
can be defined as follows. is reliability coefficient that
= , = , = | − | (3)
√
- 77 -
ensures the maximum injected current sufficiently higher than () ()
() = (13)
the maximum fault current, allowing for the formation of an
artificial zero-crossing point while maintaining the desired
interruption time. should be set to a value greater than 1. In 0 ≤ () ≤
this paper, is 1.03. () = − () () < (14)
() − < ()
= − + × − (8)
The objective function , considering the constraints
The second PSO algorithm is utilized to optimize the mentioned, can be represented as follows.
inductance of each module in the current injection branch. To
select appropriate injection current, it is necessary to = − + ∑ () + ∑ () (15)
determine the inductance of each module. The current
injection branch consists of a total of 5 modules. Table I represents the coefficients utilized in the objective
indicates number of modules. The first condition states that function of PSO algorithm. The magnitude of the injected
the overall equivalent inductance should be equal to the current, obtained through the optimized parameters, is
optimized equivalent inductance obtained from the first presented in Table II. It can be observed that the injected
PSO algorithm. current values are generally evenly distributed. Table III
shows the optimized values of the SSCB parameters.
= ∑
= (9) TABLE I. PSO ALGORITHM COEFFICIENT
2
The second condition involves arranging the injected 1 10
currents for each case in ascending order. The injected 1.03 1
currents from the auxiliary branch are classified into 2 − 0.5 1
1 cases based on the on/off states of each module. Each
module can be represented by a binary state, where '1' TABLE II. INJECTION CURRENT BASED ON MODULE STATE
indicates that the module is turned on, and '0' indicates that the Module state Injection
module is turned off. The state of the injected modules for
current (A)
each case can be denoted by a 5-digit binary number,
representing the combination of on/off states. As the current 0 0 0 0 1 240.97
injection modules are connected in parallel, the equivalent 0 0 0 1 0 418.73
inductance of the activated modules is smaller than the 0 0 0 1 1 648.43
individual module inductances. By arranging the injected 0 0 1 0 0 797.41
currents in ascending order for each case, it simplifies the 0 0 1 0 1 1016.67
determination of the injected currents based on the magnitude 0 0 1 1 0 1178.22
of the fault current.
0 0 1 1 1 1386.70
|()()| 0 1 0 0 0 1631.75
()()
() = (10) 0 1 0 0 1 1826.98
0 1 0 1 0 1970.57
Finally, it is important to maintain a certain range between 0 1 0 1 1 2155.55
the injected currents for each case. If the interval between 0 1 1 0 0 2275.17
injected currents is too small, it indicates a concentration of 0 1 1 0 1 2450.66
currents within a narrow range. Conversely, if the interval is 0 1 1 1 0 2579.52
too large, it can result in excessive current injection, leading 0 1 1 1 1 2745.23
to the generation of excessive reverse voltage across the SCR.
1 0 0 0 0 2935.18
Assuming a uniform distribution of injected currents, the
interval between the injected currents can be calculated as 1 0 0 0 1 3089.00
follows. 1 0 0 1 0 3201.68
1 0 0 1 1 3346.22
= (11) 1 0 1 0 0 3439.29
1 0 1 0 1 3575.23
The interval between the injected currents should satisfy 1 0 1 1 0 3674.57
the following conditions. , represents the lower and 1 0 1 1 1 3801.66
upper bound of the injected currents interval. In this paper, 1 1 0 0 0 3949.34
is 0.5 and is 2.0. 1 1 0 0 1 4065.56
1 1 0 1 0 4150.17
≤ ( + 1) − () ≤ (12) 1 1 0 1 1 4258.01
1 1 1 0 0 4327.00
To normalize the interval between the injected currents, it 1 1 1 0 1 4426.89
can be represented as follows below. And based on the 1 1 1 1 0 4499.67
normalized function (), a penalty function () can 1 1 1 1 1 4591.78
be defined as follows.
- 78 -
TABLE III. OPTIMIZED PATAMETER IN CURRENT INJECTION MODULE is uniformly distributed, the magnitude of the injected current
that can be provided is limited within this range. Consequently,
[] 2047.35 [] 138.78
a relatively significant difference occurs between the fault
[] 1166.28 C 2 current and the injected current, leading to interruption time
[] 598.78 10 less than 40 .
[] 277.08 1 Interruption Time According to Fault Type
60
50
IV. SIMULATION AND RESULT ANALYSIS
A. Simulation design 40
The thyristor-based SSCB was installed in a symmetric
30
monopole structure of an MVDC distribution network. The
MVDC distribution network operates at voltage level ±20
and utilizes a Modular Multilevel Converter (MMC) for the 20
AC to DC conversion. The MMC is composed of Half Bridge
Sub-Modules (HBSMs) and has a 4MW of rated power. The 10
MVDC distribution network consists of a 20km line and DC
loads. To limit the magnitude and rate of rise of fault currents, 0
-0.5 0 0.5 1 1.5
fault limiting reactors are installed at each pole [6]. Fault Type(0-PTP, 1-PTG)
Fig. 5. SSCB interruption time by fault type
Fig. 4. MVDC distribution network structure
To validate the operation of the SSCB under various fault
conditions, simulations were conducted based on fault
scenarios presented in Table IV. The fault types considered
include Pole-To-Pole (PTP) faults and Pole-To-Ground (PTG)
faults. Furthermore, both low-fault-resistance bolted faults
and high-fault-resistance faults with a resistance of 10 ohms
were simulated.
TABLE IV. FAULT CONDITION IN MVDC DISTRIBUTION NETWORK
Fig. 6. SSCB interruption time by fault resistance
Fault type PTP, PTG fault
In Fig. 6, it was confirmed that the SSCB's interruption
Fault resistance [ohm] 0.01(bolted fault), 10
performance is not affected by variations in fault resistance.
Fault location [km] 0, 5, 10, 15, 20 However, for bolted faults, a relatively wide distribution of
Trip signal reception time [] 100, 300, 500 fault interruption times can be observed. This is because
smaller fault resistances result in larger fault currents, leading
to errors between predicted and actual currents.
To examine the SSCB's performance concerning fault
location, faults were simulated at 5km intervals. Lastly,
simulations were performed with varying time settings for
receiving the trip signal from the relay after the occurrence of
the fault.
B. Simulation result analysis
The interruption time according to the fault type are shown
in Fig. 5. A fault type of 0 represents a PTP fault, while a fault
type of 1 indicates a PTG fault. Based on the simulation results,
there were no cases where the interruption execution time
exceeded 50 for either fault type. This confirms the reliable
operation of the SSCB in terms of interruption time. For PTP
faults, the interruption time predominantly locates between 40
and 50 , while for PTG faults, it mainly ranges from 30
to 40 . PTG faults generally have lower fault current
magnitudes compared to PTP faults. As a result, the fault
current remains within a certain range regardless of variations Fig. 7. SSCB interruption time by fault loaction
in other fault conditions. However, since the injected current
- 79 -
The correlation between fault location and interruption V. CONCLUSION
time is weak. Interruption time is evenly distributed for each This paper investigates a SSCB using thyristors for current
fault location, indicating that the fault location does not injection. Compared to AC faults, DC faults have larger fault
influence the SSCB performance. currents and require faster interruption time, making SSCBs a
promising alternative to traditional mechanical circuit
breakers. Thyristor-based SSCBs can be applied even in high-
voltage and high-current network, making them suitable for
MVDC distribution networks. The thyristor-based SSCB
operates by forming an artificial zero current in the thyristor
through current injection using an LC resonant circuit, which
separates the fault from the system. To determine the
appropriate injection current based on the fault current, a
multi-module structure is employed. The parameters of each
module are optimized through two stage of the PSO algorithm.
To help designing protection coordination scheme for the DC
system, proposed SSCB is designed to and perform
interruption within 50 after receiving the trip signal,
considering the sampling period of the relay and the switching
period of power electronics. The thyristor-based SSCB
successfully achieves interruption within 50 under various
Fig. 8. SSCB interruption time by trup signal reception time
fault conditions, including fault type, fault resistance, fault
In Fig. 8, it was observed that a shorter trip signal reception location, and trip signal reception time. Given that the
time after fault occurrence results in a shorter interruption time, proposed circuit breaker considers the characteristics of faults
less than 50 . As depicted in Fig. 2, during the transient state in DC systems, it is expected to be effectively utilized in
of the fault, the rate of change of fault current decreases designing protection coordination schemes for MVDC
significantly, and as time progresses, this rate of change systems.
decreases at a slower pace. With a shorter trip signal reception ACKNOWLEDGMENT
time, there is a larger discrepancy between the predicted fault
current and the actual fault current, leading to a shorter This work was supported by the Korea Institute of Energy
interruption performance time, less than 50 . Technology Evaluation and Planning(KETEP) and the
Ministry of Trade, Industry & Energy(MOTIE) of the
Simulation results show that the thyristor-based SSCB Republic of Korea (No. 20225500000110).
achieves a 100% interruption reliability, as it consistently
performs the interruption within 50 after receiving the trip This work has been supported by the National Research
signal, regardless of variations in fault conditions. However, a Foundation of Korea (NRF) grant funded by the Korean
phenomenon occurs where the interruption time is shorter than government (MSIP) (No. 2021R1A2B5B03086257)
50 under certain conditions. This phenomenon can be
REFERENCES
attributed to two main reasons. Firstly, there is a discrepancy
between the actual fault current and the predicted fault current.
During the early stages of a fault, the fault current undergoes [1] Ministry of trade, industry and energy. 10th basic plan for power
supply and demand (2023)
a rapid rate of change, resulting in a substantial discrepancy
[2] Coffey, Sophie, et al. Review of MVDC applications, technologies, and
between the actual and predicted fault currents. As a result, future prospects. Energies Vol. 14, No. 24, pp 8294 (2021).
when the trip signal reception time is short, the interruption
[3] Rodrigues, Rostan, et al. A review of solid-state circuit breakers. IEEE
time can deviate significantly from 50 . Secondly, a Transactions on Power Electronics Vol. 36, No. 1, pp 364-377 (2020).
significant gap between the predicted fault current and the [4] Wu, Yifei, et al. A novel current injection DC circuit breaker
injected current can cause substantial errors in the interruption integrating current commutation and energy dissipation. IEEE Journal
time. To address this issue, it is necessary to use more modules of Emerging and Selected Topics in Power Electronics Vol. 8, No. 3,
to set the injected current intervals more densely. However, pp 2861-2869 (2019)
increasing the number of modules also raises the cost due to a [5] Kheirollahi, Reza, et al. Ultrafast solid-state circuit breaker with a
higher component count, and it can complicate parameter modular active injection circuit. IEEE Journal of Emerging and
Selected Topics in Industrial Electronics Vol. 3, No. 3, pp 733-743
optimization as well. Therefore, achieving precise interruption (2021)
times within 50 may require striking a balance between the [6] Keshavarzi, Davood, Ebrahim Farjah, and Teymoor Ghanbari. Hybrid
number of modules used, associated costs, and the complexity DC circuit breaker and fault current limiter with optional interruption
of parameter optimization. capability. IEEE Transactions on Power Electronics Vol. 33, No. 3 pp
2330-2338 (2017)
- 80 -
Investigation of Grounding Schemes in Hybrid
(AC/DC) Medium Voltage Distribution Networks
Zeeshan Haider1, Arif Mehdi1, S Jarjees Ul Hassan1, Jin-Sol Song1, Abraham D. Arefaynie1, Chul-Hwan Kim1*
Department of Electrical and Computer Engineering, Sungkyunkwan University, Suwon, Republic of Korea
Email: [email protected]
Abstract—Due to the growing demand for energy, the present systems, as well as the identification of the locations where the
distribution lines' capacity has become insufficient to meet the system is to be grounded and the equipment that is to be
energy demand. The idea is to transform the existing AC employed. In [8] compares distribution system design practices
distribution system into hybrid (AC/DC) medium distribution between the United States and Europe, highlighting the
lines to enhance power transfer capabilities and improve overall similarities and differences that exist. Solid grounding is utilized
system efficiency. Nonetheless, the design and implementation of in 50% of the UK's system, resistance grounding in 40%, and
grounding strategies for such hybrid networks present unique reactor grounding is employed in 10% of cases to control ground
challenges due to the coexistence of AC and DC systems. In this fault current. Belgium and Spain have reactors, whereas France
paper, we investigate multiple grounding strategies for hybrid utilizes resistance grounding. In contrast to the United States,
AC/DC MV distribution systems. We have analyzed the benefits Greece does not employ impedance grounding for neutrals in
and drawbacks of various grounding approaches. We have primary distribution. Secondary distribution within commercial
considered multiple factors, such as fault current, common mode and industrial settings in the United States, on the other hand,
voltages, stray current, system stability, and grounding protection. commonly employs resistance grounding.
By utilizing simulation studies, we assess the effectiveness of
different grounding methods, comprising solid grounding,
ungrounded, low-resistance grounding, and high-resistance
grounding configurations. We evaluate their ability to reduce
common mode voltages and mitigate fault currents, and leakage
currents to ensure the safety of the equipment and the person. The
hybrid AC/DC MV distribution network is designed and simulated
using MATLAB/Simulink software.
Keywords—Hybrid AC/DC system, grounding strategies,
medium voltage distribution, investigation of the grounding
schemes.
Fig. 1. Typical hybrid AC/DC distribution network
I. INTRODUCTION
Previously, various techniques were used for the grounding
The integration of decentralized renewable energy sources
of the MVAC distribution system, but limited research has been
has reshaped modern power transmission and distribution
done so far on the grounding of the MVDC distribution.
networks [1],[2]. Additionally, it has been demonstrated that DC
Grounding is important for the safety of the equipment and
distribution confers a greater number of benefits in comparison
person as defined by the National Electrical Code (NEC). In [9]
to AC distribution [3], [4]. It provides easy integration of DC
basis of the grounding equipment function and grounding
sources with the system and reduces unnecessary DC/AC
requirements for 600V or less are briefly discussed. Moreover,
conversions. The upsurge in the utilization of electric vehicles
[10] has discussed the significance of neutral grounding in low
and other direct current (DC) loads such as laptops, mobile
and medium-voltage industrial power distribution systems. This
phones, and DC air conditioners has resulted in the
study conducts a comparative analysis of three distinct types of
development of hybrid medium voltage distribution networks
grounding systems and results that a customized high-resistance
that merge both alternating current (AC) and DC components
grounding system accompanied by ground fault tripping could
[5]. Fig. 1 shows a typical hybrid AC/DC network having
potentially emerge as the optimal choice for certain industrial
multiple energy sources integrated. This incorporation enhances
applications. In [11] investigates the impact of grounding on AC
power transfer capabilities and system efficiency [6]. However,
distribution systems for fault responses through different case
the design and implementation of grounding strategies for such
studies. Fault current, voltage swell, ground potential rise, and
networks present unique challenges due to the coexistence of
neutral wire voltages and currents are analyzed. Power failures
AC and DC systems.
in industries and commercial installations may result in
The IEEE green book [7] furnishes an outline of the economic and operational losses. As a solution, [12] has
grounding system-related concerns, encompassing a concise analyzed solidly neutral, resistance grounded, and high
account of the benefits of both grounded and ungrounded resistance grounding to avoid grounding faults. In [13] diverse
- 81 -
simulations were conducted to explore the grounding schemes the DC system. In the realm of AC systems, this topology is
in both bipolar and unipolar DC systems. The simulations were commonly referred to as capacitance ground, owing to the
carried out with the use of PSCAD software to investigate the presence of stray capacitances. However, this scheme also has
effects of grounding on the detection of line-to-ground faults, its drawbacks, namely the high common mode voltage and
fault current, leakage current, common mode current, and safety increased voltage stress on the non-faulted phase during faults,
of both personnel and equipment. Similarly, in [14] the which can compromise line insulation integrity. Furthermore,
occurrence of transient short-circuit current and overvoltage in overcurrent protection for this scheme is difficult to detect [16]
the context of monopoly ground fault and various grounding due to very low fault currents.
configurations including ungrounded, grounded through resistor
AC_Bus1 AC_Bus2
high and low, and solid grounding topologies. The finding PCC F1
suggests that ground with a small resistor of 400-800 ohm is CB1 Transformer CB2 CB3 CB4 CB5
efficient for controlling the overcurrent and overvoltage. It is
important to consider the influence of load conditions and
grounding strategies on overcurrent and overvoltage. This paper
[14] also mentions that the grounding mode of the system should
affect the relay protection scheme. The study [15] presents a DC_Bus1
F2 DC_Bus 2
grounding method that is designed to mitigate the incidence of CB1 CB2 CB3 CB4
direct current (DC) stray currents in standard circumstances and
guarantee proper grounding during fault transients. By Transformer Rectifier
integrating a limiter resistor, the grounding system effectively Fig. 2. Ungrounded system
curtails ground fault currents to an acceptable level, thereby
providing a reliable safeguard for the power electronic To analyze the scheme an LG fault on phase A of the AC
interfaces. side at t=0.4-0.7s is imposed and Fig. 3 shows the impact of the
This paper aims to address the grounding challenges in LG fault on the hybrid system. Fig. 3 (c), illustrates the fault
hybrid AC/DC distribution networks. The research focuses on current. The impact of the LG fault for the ungrounded system
exploring different grounding strategies and assessing their on the DC voltage is shown in Fig. 3(d).
advantages and limitations. The investigation considers several
factors such as fault detection, fault current distribution, system
stability, and grounding protection. Additionally, the influence
of grounding techniques on power quality, safety, and
dependability is analyzed, using simulation studies and real-
world examples. The efficacy of various grounding approaches,
including solid grounding, low-resistance grounding,
ungrounded systems, and high-resistance grounding
configurations, is evaluated in terms of limiting fault currents (a) (b)
and ensuring proper fault detection and isolation in AC/DC
networks. Based on the investigative findings of our study, we
have identified optimal grounding techniques for hybrid
networks.
The paper is structured as section II explains the different
grounding schemes with their simulation results. Section III
discusses an overview of grounding topologies with
characteristic tables. The detailed conclusion is provided in (c) (d)
section IV.
Fig. 3. Simulations result for an ungrounded system under LG fault a) AC
II. GROUNDING SCHEMES AND SIMULATIONS voltage, b) AC current, c) AC fault current, d) DC voltage
This segment of the manuscript scrutinizes varied grounding Now for the pole-to-ground fault for the ungrounded fault at
topologies implemented in hybrid AC/DC systems via the t=0.4-0.8s on the DC side of the system and Fig.4 shows the
simulation of MATLAB/Simulink, in conjunction with their effect of pole-to-ground fault for ungrounded on the hybrid
corresponding merits and demerits. Essentially, three grounding AC/DC network.
approaches are commonly employed in the power system,
namely ungrounded, solidly grounded, or grounded through a
resistor, inductor, or capacitor. Every scheme has its unique
benefits and drawbacks.
A. Ungrounded system
The proposed scheme involves the isolation of the
distribution line from the ground, as shown in Fig. 2 where an
ungrounded system is simulated. This approach offers the (a) (b)
benefit of minimizing fault current in the event of single-
phase grounding, while also preventing the occurrence of stray
currents during normal operation that could cause corrosion in
- 82 -
(c)
Fig. 4. Simulations result for an ungrounded system under PG fault a) AC (b) (d)
voltage, b) DC voltage, c) DC fault current
Fig. 6. Simulations result in a solidly grounded system under LG fault a) AC
For the PG fault on the DC side, since there is no physical voltage, b) AC current, c) AC fault current, d) DC voltage
grounding on the system thus the fault current is very low Fig. 4
(c) shows the fault current. Moreover, due to the very low fault The voltage and current waveform of the AC system for the
current, this pole-to-ground fault does not affect the AC system. solidly grounded system can be observed in Fig. 6 (a) and (b),
Besides these, the simulation results show that when a PG fault while Fig. 6 (c) reveals a momentary flash of fault current that
occurs at the ungrounded system the voltage at the un-faulted can potentially cause harm to both equipment and protection
line increases by 2 p.u as we can see in Fig. 4 (b) and the system schemes in place within the system. Furthermore, it should be
is operating under the fault because the LL voltage is not noted that similar to the ungrounded system, this topology has a
affected. minor impact on the DC side of the system, which is evident
B. Solidly grounded system from the simulation results presented in Fig. 6 (d), where a
voltage dip is noticeable between 0.4 to 0.7 seconds.
The given arrangement involves the grounding of the
distribution system with no impedance, as represented by the For the DC component of the system, an analysis has been
single-line diagram of the solidly grounded system in Fig. 5. conducted on the pole-to-ground fault occurring between time
Despite its advantages in terms of facilitating the identification intervals of 0.4 to 0.8 seconds. The simulation outcomes have
of LG faults [17], this topology has significant drawbacks such been depicted in Fig. 7 and aim to elucidate the influence of solid
as the occurrence of arc flash and the possibility of equipment grounding on the system.
damage, leading to high fault current in both the AC and DC
systems. Moreover, the DC system is prone to corrosion due to
the high stray current under normal conditions, but it does offer
the benefit of reducing the common mode voltage and voltage
stress on un-faulted lines [18].
AC_Bus1 AC_Bus2
PCC F1
CB1 Transformer CB2 CB3 CB4
(a) (b)
Grounding
DC_Bus1 DC_Bus 2
F2
CB1 CB2 CB3
Transformer Rectifier
Grounding
(c)
Fig. 5. Solidly grounded system Fig. 7. Simulations result in a solidly grounded system under PG fault a) AC
voltage, b) DC voltage, c) DC fault current
In this analysis, we examine the solid grounding of the
hybrid AC/DC system in light of the LG fault that occurs on the The system is solidly grounded and the fault current
AC side during the time interval t = 0.4-0.7s. Through possesses a direct route to the ground, giving rise to a substantial
simulation, we observe the consequences of this fault on the current that flows through the ground, as depicted in Fig. 7(c).
system when it is subject to a solidly grounded topology. This high fault current leads to disruption in the AC system,
which can be observed in Fig. 7(a). The voltage at the unaffected
line is raised, as demonstrated in Fig. 7(b), but the magnitude in
this particular topology is less than 2 p.u.
C. Grounding through Impedance
The method involves linking the distribution system to the
earth via an impedance element, as illustrated in Fig. 8. This
arrangement is categorized into two classifications, namely
high-impedance grounding and low-impedance grounding. A
(a) (b) significant benefit of this approach is that the fault current can
be restricted to meet specific requirements by modifying the
impedance value. Consequently, this grounding scheme has
been highly endorsed by various scholars in their research
- 83 -
papers, particularly those advocating for high impedance
grounding [10],[13]. In [14] grounding of the system through
resistance of 400-800 ohm is recommended for the reliable
operation of the system.
AC_Bus1 AC_Bus2
PCC F1
CB1 Transformer CB2 CB3 CB4 CB5 (e)
AC_Load
Fig. 9. Simulations result for resistance grounded system under LG fault a)
Resistance
AC voltage, b) DC voltage, c)Fault current for grounded through 1-ohm, d)
Grounding grounded through 5-ohm, e) grounded through 10-ohm
DC_Bus1 DC_Bus 2
F2
CB1 CB2 CB3 CB4
From the simulation results, it can be seen in Fig. 9 (c), (d),
Transformer Rectifier DC_Load and (e) that for the 1-ohm resistance, the arc flashing is high, and
Resistance
when we change the grounding resistance from 1 ohm to 5, the
overcurrent flash is decreasing, and the same behavior is
Grounding recorded for the 10-ohm resistor. Moreover, the fault effect on
Fig. 8. System grounded through a resistor.
the DC system is like the ungrounded and solidly grounded
system.
In this particular topology, an impedance is incorporated Now for the pole-to-ground fault at the DC system, the
between the system neutral and ground to limit the fault current simulation results have shown the effect of this topology on the
in the event of any faults occurring in the system. The magnitude hybrid network. The value of the grounding resistor is taken as
of the fault current is directly proportional to the value of the 1 ohm, 5 ohm, and 10 ohms respectively for the grounding.
impedance employed in the system. For example, a high
impedance value will result in a low fault current, whereas a low
impedance value will result in a high fault current. In an AC
system, an instantaneous overcurrent is encountered when an
LG fault occurs, which may lead to damage to the protection
scheme and equipment. Impedance grounding is implemented to
mitigate this overcurrent arcing. The primary objective of
restricting the fault current is to ensure the safety of both the (a) (b)
equipment and personnel. Moreover, this controlled fault current
is also utilized for detecting line-to-ground faults in the system
and executing the protection scheme.
We analyzed the high-impedance and low-impedance
grounding systems and the simulation results have revealed the
effects of both approaches on the hybrid network. The
simulation results for the phase-to-ground fault on the AC side (c) (d)
of the system grounded with resistances of 1 ohm, 5 ohms, and
10 ohms are presented in Fig. 9.
(e)
Fig. 10. Simulations result for resistance grounded system under PG fault a)
(a) (b)
AC voltage, b) DC voltage, c) Grounded through 1-ohm, d) Grounded through
5-ohm, e) Grounded through 10-ohm
In the DC system, in the event of a pole-to-ground fault at a
small resistance of 1 ohm, there is a high fault current that flows
through the line, resulting in an impact on the AC system as
depicted in Fig. 10(a). Another critical aspect to consider in the
AC system is the reduction of fault current when grounding
resistance is increased, resulting in a corresponding decrease in
(c) (d) the impact on the AC side. The simulation results for the DC
system at varying resistance values, as illustrated in Fig. 10(c),
(d), (e), indicate that an increase in resistance leads to decreased
fault current and longer settling time of the fault current due to
the increased grounding resistor.
- 84 -
III. DISCUSSIONS characteristics of different grounding methods for the hybrid
From the outcomes of the simulation, it is evident that the AC/DC distribution system. Several advantages of high
hybrid AC/DC distribution network is susceptible to faults on resistance grounding can be seen in the table, including fair fault
either AC or DC, which have an impact on both systems. protection, good noise immunity, and good ground loop
However, through the implementation of grounding techniques, immunity. Additionally, it has moderate overcurrent and stray
this effect can be mitigated. Table I provides an overview of the current. Therefore, high-resistance grounding is recommended
features of various grounding schemes for the DC system. for hybrid networks.
TABLE III HYBRID AC/DC CHARACTERISTIC TABLE
TABLE I. DC GROUNDING SYSTEM CHARACTERISTICS
Grounding Solid High-Resistance Ungrounded
Grounding Un- Solidly High Low
Methods grounding Grounding
Strategies grounded Grounded Impedance Impedance
Fault Protection Good Fair Poor
Grounding Grounding
Stray Low High Moderate Moderate Noise Poor Good Excellent
Current Immunity
LG Fault Low High Moderate Moderate Over Current High Moderate Low
Current
Transient High Low Moderate Moderate Stray Current High Moderate Low
Over
Voltage
Ground Loop Poor Good Excellent
Common High Low Moderate/ Low
Immunity
Mode High
Voltage
Insulation High Low High Low
Level IV. CONCLUSION
Service Yes No Yes No
Continuity
Ensuring the protection of hybrid AC/DC networks is a
challenging task that requires the implementation of effective
grounding schemes. The present paper focuses on addressing the
From the simulation results, it has been observed that the
existing grounding challenges by examining various grounding
fault on the DC side is minimal in the ungrounded system while
schemes. The analysis identifies three types of grounding
it is significantly high in the solidly grounded configuration.
schemes, namely ungrounded, solidly grounded, and grounded
Additionally, in a system grounded through resistance, an
through impedance. Recommendations are made based on
increase in resistance leads to a decrease in fault current
parameters such as moderate levels of stray current, common
magnitude, as well as a reduction in the settling time of the fault
mode voltage, fault current, and transient overvoltage. It is
current. On the AC side, the instantaneous flash overvoltage
suggested that the grounding through a high-impedance scheme
during a fault occurrence is comparatively low in the
would be the most suitable for the network. By adjusting the
ungrounded system and high in the solidly grounded setting.
impedance employed in the grounding system, the magnitude of
However, this instantaneous flashover can be mitigated by
the fault current can be effectively curtailed to ensure that it
grounding the system through a resistor.
remains within the designated safety thresholds, thereby
safeguarding both equipment and individuals. Thus, this paper
TABLE II. AC GROUNDING CHARACTERISTICS
provides an overview of the different schemes that can be
Characteristic Resistance Ungrounded Solidly
employed to overcome grounding challenges in future hybrid
Grounding Systems Grounding AC/DC networks.
Fault Current Limited fault Uncontrolled High fault
current fault current current
Future studies may further explore robust grounding
Ground Fault Enables Difficult to Enables ground schemes to strengthen the protection of the hybrid AC/DC
Detection ground fault detect ground fault detection networks.
detection faults
Equipment Minimizes Risk of Risk of V. ACKNOWLEDGMENT
Damage equipment equipment equipment
damage damage damage This work was supported by the National Research
Shock Hazard Reduced risk Risk of Risk of Foundation of Korea (NRF) funded by the Korean Government
of electrical electrical shock electrical shock [Ministry of Science, ICT and Future Planning (MSIP)] under
shock
Grant 2021R1A2B5B03086257 and Korea Institute of Energy
Reliability May require compromised in Improved
additional case of fault system stability Technology Evaluation and Planning (KETEP) and the Ministry
maintenance of Trade, Industry & Energy (MOTIE) of the Republic of Korea
(No. 20225500000110).
Table II presents an overview of the traits exhibited by
diverse grounding mechanisms on the AC side. It is suggested VI. REFERENCES
that resistance grounding be implemented on both the AC and
DC sides due to the benefits it affords in terms of constraining [1] A. Mehdi, S. Jarjees Ul Hassan, and C. H. Kim, “Squaring and
lowpass filtering data-driven technique for AC faults in AC/DC
fault currents, detecting ground faults, and mitigating electric lines,” Electric Power Systems Research, vol. 223, Oct. 2023, doi:
shocks. 10.1016/j.epsr.2023.109581.
[2] J. J. Justo, F. Mwasilu, J. Lee, and J. W. Jung, “AC-microgrids versus
Based on the simulation results of the proposed hybrid DC-microgrids with distributed energy resources: A review,”
AC/DC medium voltage network we have analyzed the effects Renewable and Sustainable Energy Reviews, vol. 24. pp. 387–405,
of the grounding schemes. Table III summarizes the
- 85 -
2013. doi: 10.1016/j.rser.2013.03.067.
[3] S. Parhizi, H. Lotfi, A. Khodaei, and S. Bahramirad, “State of the art
in research on microgrids: A review,” IEEE Access, vol. 3. Institute
of Electrical and Electronics Engineers Inc., pp. 890–925, Jun. 10,
2015. doi: 10.1109/ACCESS.2015.2443119.
[4] A. Meghwani, S. Chakrabarti, S. C. Srivastava, and S. Anand,
“Analysis of fault characteristics in DC microgrids for various
converter topologies,” in 2017 IEEE Innovative Smart Grid
Technologies - Asia: Smart Grid for Smart Community, ISGT-Asia
2017, Institute of Electrical and Electronics Engineers Inc., Jun. 2018,
pp. 1–6. doi: 10.1109/ISGT-Asia.2017.8378426.
[5] Q. Xu, J. He, S. Huang, J. Zeng, C. Fan, and W. Wang, “Researches
on grounding modes of the AC/DC hybrid system with various
distributed renewable energy,” The Journal of Engineering, vol.
2019, no. 16, pp. 3035–3038, Mar. 2019, doi: 10.1049/joe.2018.8387.
[6] W. Zhang, T. Wei, Q. Chen, Y. Cui, W. Liu, and J. Yang, “Study on
grounding modes of AC/DC hybrid distribution system,” in
Proceedings - IEEE International Conference on Energy Internet,
ICEI 2019, Institute of Electrical and Electronics Engineers Inc., May
2019, pp. 42–46. doi: 10.1109/ICEI.2019.00014.
[7] IEEE Industry Applications Society. Power System Technologies
Committee., IEEE recommended practice for grounding of industrial
and commercial power systems. Institute of Electrical and Electronics
Engineers, 1982.
[8] A. P. S. Meliopoulos, J. Kennedy, C. A. Nucci, A. Borghetti, and G.
Contaxis, “Power distribution practices in USA and Europe: Impact
on power quality,” in Proceedings of International Conference on
Harmonics and Quality of Power, ICHQP, IEEE Computer Society,
1998, pp. 24–29. doi: 10.1109/ICHQP.1998.759834.
[9] R. B. West, “Grounding for Emergency and Standby Power
Systems,” IEEE Trans Ind Appl, vol. IA-15, no. 2, pp. 124–136, 1979,
doi: 10.1109/TIA.1979.4503627.
[10] L. Yu and R. L. Henriks, “Selection of system neutral grounding
resistor and ground fault protection for industrial power systems,” in
Record of Conference Papers Industry Applications Society 38th
Annual Petroleum and Chemical Industry Conference, Publ by IEEE,
1991, pp. 147–153. doi: 10.1109/pcicon.1991.162941.
[11] Institute of Electrical and Electronics Engineers., ICHQP 2008 : 13th
International Conference on Harmonics & Quality of Power :
University of Wollongong, Australia, 28th September-1st October
2008. IEEE, 2008.
[12] J. C. Das and R. H. Osman, “Grounding of AC and DC low-voltage
and medium-voltage drive systems,” IEEE Trans Ind Appl, vol. 34,
no. 1, pp. 205–216, 1998, doi: 10.1109/28.658747.
[13] J. Mohammadi, F. B. Ajaei, and G. Stevens, “Grounding the DC
Microgrid,” in IEEE Transactions on Industry Applications, Institute
of Electrical and Electronics Engineers Inc., Sep. 2019, pp. 4490–
4499. doi: 10.1109/TIA.2019.2928278.
[14] Y. Wang, Z. Yu, J. He, S. Chen, R. Zeng, and B. Zhang,
“Performance of shipboard medium-voltage DC system of various
grounding modes under monopole ground fault,” IEEE Trans Ind
Appl, vol. 51, no. 6, pp. 5002–5009, Nov. 2015, doi:
10.1109/TIA.2015.2427119.
[15] Clemson University and Institute of Electrical and Electronics
Engineers, Clemson University Power Systems Conference :
September 4th-7th, 2018, Charleston, South Carolina, USA.
[16] J. Burke and M. Marshall, “Distribution system neutral grounding,”
Proceedings of the IEEE Power Engineering Society Transmission
and Distribution Conference, vol. 1, pp. 166–170, 2001, doi:
10.1109/TDC.2001.971228.
[17] J. P. Nelson, “The grounding of power systems above 600 volts: A
practical view point,” in Record of Conference Papers - Annual
Petroleum and Chemical Industry Conference, 2003, pp. 13–22. doi:
10.1109/pcicon.2003.1242595.
[18] 1653.6-2018 - IEEE Recommended Practice for Grounding of DC
Equipment Enclosures in Traction Power Distribution Facilities.
IEEE, 2018.
- 86 -
August 22(TUE) - 25(FRI), 2023
Oral Session 3
14:00~15:40 Wed, August 23, 2023
Room : A434
Chair: Yemar Amitov (AUPET, Kazakhstan)
- 87 -
Research and development of an information
system for speech recognition based on neural
network algorithms.
Maksat Kanashev Sergiy Gnatyuk Mirambayeva Nazym
MSc student of Power Engineering and Vice Dean of the Faculty of Cybersecurity, Teacher at Gumarbek Daukeev
Telecommunications Computer and Software Engineering at Almaty University of Energy
Almaty University National Aviation University and Communications College
Almaty, Kazakhstan National Aviation University Almaty, Kazakhstan
ma.kanashev @aues.kz Kyiv, Ukraine [email protected]
[email protected]
Assel Mukasheva
PhD, Associate Professor of the Department
of Information Systems and Cybersecurity
Almaty University of Power Engineeringand
Telecommunications
Almaty, Kazakhstan
[email protected]
orcid.org/0000-0001-9890-
4910
Abstract — This article presents a comparison of the speed lines and determine the recognition accuracy. The findings of
and accuracy of two speech recognition models in the context this study can be useful for improving speech recognition
of educational applications and artificial intelligence. The systems and creating more adaptive and efficient solutions for
relevance of the research lies in determining the optimal model audio data processing. The results of comparing the accuracy
for use in educational technologies and the development of of speech recognition models are recorded in JSON format,
intelligent systems, taking into account performance after which, using the matplotlib library, they are visualized as
limitations and hardware limitations. The study is based on a graphs, comparing the speed and accuracy of recognition.
set of audio recordings containing various phrases uttered by Visualization of data in this way contributes to their
different people with different accents and intonations. improvement and more effective integration into educational
Parameters such as recognition speed, audio-to-text conversion systems. In conclusion, the authors emphasize the prospects for
accuracy, and resource consumption are analyzed. Firstly, the further development of speech recognition technologies in the
importance of speech recognition technologies in the context of field of education and artificial intelligence, as well as potential
education and the need for continuous development and areas of future research, including multilingualism and
improvement of models to ensure successful application is speaker emotion analysis.
emphasized. The introduction of more efficient and accurate
speech recognition models into educational technologies and Keywords: Artificial Intelligence, speech recognition,
intelligent systems is an important step in ensuring affordable, accuracy, machine learning, recognition model
high quality and inclusive education for all. To further expand
the potential of speech recognition technologies in education, I. INTRODUCTION
the article proposes to integrate these models with other areas Speech recognition models play a crucial role in
of artificial intelligence, such as machine learning, natural educational applications and artificial intelligence, enabling
language processing and computer vision. Secondly, such
integration can lead to the development of more complex and
advancements in the field of intelligent systems. This article
advanced educational applications that provide a higher aims to compare the speed and accuracy of two speech
quality of education and better support for students. In recognition models within the context of educational
addition, the authors emphasize the importance of taking into technologies. By examining the performance limitations and
account hardware limitations and performance when choosing hardware constraints, this research seeks to identify the
and developing speech recognition models. Such accounting optimal model for integration into educational systems.
ensures a balanced ratio between recognition accuracy and The importance of speech recognition technologies in
resource consumption, allowing educational applications to education is widely recognized. Li et al. provide a
function effectively on various platforms and devices with comprehensive review of speech recognition in educational
different capabilities. Considering these aspects, the article
makes a valuable contribution to the development of
applications, emphasizing the need for continuous
comprehensive, innovative and inclusive solutions that can development and improvement of models to ensure
change and enrich the educational landscape. This article successful application [1]. Liu and Li conduct an empirical
examines the parameters of speech recognition in audio study comparing different speech recognition models
recordings containing a variety of phrases uttered by different specifically tailored for educational applications, shedding
people with different accents and intonations. The aim of the light on their effectiveness and applicability [2]. These
study is to evaluate the recognition speed, the accuracy of studies underline the significance of exploring and
audio-to-text conversion and resource consumption when using implementing efficient and accurate speech recognition
SpeechRecognition and Levenshtein libraries. During the models in educational technologies.
analysis, the above libraries were used to record and recognize
audio files, as well as to calculate the distance between two
Enhancing educational technologies with speech
recognition has been a focus of research and development.
- 88 -
Johnson and Smith discuss the current trends and future However, there are also some limitations associated with
directions of incorporating speech recognition into Google's speech recognition model. One of the notable
educational technologies, emphasizing its potential impact on drawbacks is its reliance on an internet connection for real-
the learning experience [3]. Chen, Liu, and Huang propose time transcription. This dependency on network connectivity
the integration of speech recognition with machine learning may pose challenges in situations where internet access is
to create intelligent educational applications that offer limited or unstable.
enhanced adaptivity and efficiency [4]. Moreover, the privacy concerns related to data storage and
Assessing the performance of speech recognition models usage should be taken into consideration. Since the model
in educational applications is crucial for their optimization. requires sending audio data to Google's servers for
Wang, Yu, and Liu analyze the performance of speech processing, users and developers need to be aware of the
recognition models within educational applications, privacy implications and ensure compliance with relevant
examining factors such as recognition speed and accuracy data protection regulations.
[5]. Garcia and Liu focus on analyzing speech recognition
B. Bing Speech Recognition Model
parameters using the Levenshtein method, providing insights
into improving the accuracy of educational applications [6]. The Bing speech recognition model, developed by
Additionally, Smith and Brown emphasize the visualization Microsoft, offers its own set of advantages and
of speech recognition model accuracy as a means of considerations for speech recognition tasks.
understanding their performance within educational systems One notable strength of the Bing Speech Recognition
[7]. Model is its integration with the Microsoft ecosystem,
Considering the limitations and constraints of hardware including services like Microsoft Azure and Microsoft
and performance is essential when selecting and developing Cognitive Services. This integration enables developers to
speech recognition models for educational applications. Lee leverage the model within a comprehensive suite of tools and
and Park discuss the prospects for speech recognition services, facilitating seamless integration and providing a
technologies in education and artificial intelligence, unified development experience.
highlighting the need to strike a balance between recognition Another advantage of the Bing model is its support for
accuracy and resource consumption [8]. Wang, Chen, and Li real-time transcription, even in offline scenarios. This
examine speech recognition parameters in audio recordings capability is particularly valuable in applications where
using the Levenshtein method, contributing to the immediate and uninterrupted speech recognition is required,
understanding of optimal model selection [9]. Zhang, Li, and such as voice assistants or dictation software [13].
Huang propose adaptive solutions for audio data processing In terms of accuracy, the Bing Speech Recognition
based on speech recognition, addressing the challenges of Model demonstrates competitive performance, achieving
resource optimization in educational applications [10]. reliable transcription results in various scenarios. The model
has been trained on diverse datasets and benefits from
II. BACKGROUD Microsoft's expertise in natural language processing and
A. Google speech recognition model overwiew machine learning.
However, like the Google model, the Bing Speech
Google's speech recognition model is known for its Recognition Model also has limitations. One limitation is its
advanced technology and widespread use in various performance with certain accents and languages, which may
applications. This model offers several advantages, making it not be as comprehensive or refined compared to models
a popular choice for speech recognition tasks. specifically tailored for those linguistic variations [14].
One of the key strengths of the Google Speech Developers and users should consider this factor when
Recognition Model is its high accuracy in converting speech working with specific languages or accents.
to text. The model leverages large-scale training data and Additionally, like any cloud-based speech recognition
sophisticated machine learning algorithms to achieve service, the Bing model requires a stable internet connection
impressive recognition accuracy. This accuracy is for optimal performance. Offline transcription capabilities
particularly beneficial in applications where precise may be limited, and real-time transcription may be impacted
transcription is essential, such as transcribing audio by network latency or disruptions.
recordings or generating subtitles for videos. In conclusion, both Google and Bing speech recognition
Another advantage of Google's speech recognition model models offer unique advantages and considerations. The
is its robustness and adaptability to different accents, Google model excels in accuracy, adaptability, and
languages, and speech patterns. The model has been trained integration with Google services, while the Bing model
on diverse datasets, enabling it to handle a wide range of provides strong integration with the Microsoft ecosystem and
linguistic variations effectively [11]. This flexibility makes it supports real-time transcription. When choosing between
suitable for applications that involve multi-lingual or multi- these models, developers should consider factors such as
accented speech. accuracy requirements, language support, integration needs,
Furthermore, the Google Speech Recognition Model and offline capabilities.
offers excellent integration capabilities with other Google
services and APIs. Developers can easily incorporate this C. Training Algorithms and Model Similarities
model into their applications using Google Cloud Speech-to- Both the Google Speech Recognition Model and the Bing
Text API, allowing seamless integration with various Google Speech Recognition Model utilize sophisticated training
products and services [12]. algorithms to achieve their impressive performance in speech
recognition tasks. These algorithms play a crucial role in
- 89 -
training the models and enabling them to accurately convert algorithms capable of detecting dangerous objects in real-
speech into text. world scenarios. By harnessing the power of advanced
The Google Speech Recognition Model leverages deep machine learning techniques, these algorithms aim to ensure
learning algorithms, particularly recurrent neural networks the safety and protection of critical infrastructures from
(RNNs) and convolutional neural networks (CNNs). RNNs potential threats.
are well-suited for sequential data processing, making them To establish a comprehensive understanding of the
effective in capturing the temporal dependencies in speech broader context, Smith and Johnson [17] provide a
signals. On the other hand, CNNs excel in feature extraction comprehensive review of advancements and applications in
and can capture local patterns in the audio input, enhancing the field of speech recognition. This article serves as a
the model's ability to discern relevant speech information. valuable resource, offering insights into the latest
Similarly, the Bing Speech Recognition Model also developments, challenges, and potential applications of
employs deep learning techniques, including RNNs and speech recognition technology.
CNNs, to process speech data effectively. These algorithms
enable the model to learn complex patterns and structures in IV. MATERIALS AND METHODS
speech signals, contributing to its accuracy and robustness. A. Levenshtein distance method in speech recognition.
Both models share similarities in terms of the training
Universality: The Levenshtein method takes into account
algorithms used. RNNs, with their ability to capture temporal
all types of editing operations (insertion, deletion,
dependencies, are commonly employed in both models to
substitution) and allows working with arbitrary strings. This
model sequential aspects of speech data. This enables the
is especially important in speech recognition, where input
models to effectively recognize and interpret speech patterns
data may contain various phonetic variations and different
and phonetic features.
forms of transcription.
Furthermore, both models make use of CNNs to extract
Flexibility: The Levenshtein method enables measuring
relevant acoustic features from the audio input. CNNs are
the "closeness" or "similarity" between two strings by
particularly adept at capturing local patterns and identifying
determining the minimum number of operations required to
important spectral characteristics, aiding in speech
transform them. In the context of speech recognition, this
recognition tasks.
allows identifying the most appropriate word or phrase based
The combination of RNNs and CNNs in the training
on comparison with the input audio representation.
algorithms of both models demonstrates their shared
Ease of implementation: The Levenshtein algorithm is
emphasis on capturing both temporal and spectral
based on dynamic programming and has a simple structure,
information in speech signals. This integration of different
making it easier to implement and apply in various speech
neural network architectures enhances the models' ability to
recognition scenarios.
handle a wide range of speech variations, accents, and
If the length of word1 is 0, the Levenshtein distance is
languages.
equal to the length of word2. This is because we would need
By utilizing similar training algorithms, the Google and
to perform 'j' insertions to transform an empty string into
Bing speech recognition models exhibit commonalities in
word2.
their underlying architecture and approach to speech
Similarly, if the length of word2 is 0, the Levenshtein
recognition. These similarities contribute to their comparable
distance is equal to the length of word1, as 'i' deletions would
performance levels and accuracy in converting speech to
be required to transform word1 into an empty string.
text.
If the last characters of word1 and word2 are the same
In summary, the Google Speech Recognition Model and
(word1[i-1] = word2[j-1]), we don't need to perform any
the Bing Speech Recognition Model utilize deep learning
additional operations, so the Levenshtein distance is equal to
algorithms, including RNNs and CNNs, to process speech
the distance between the remaining substrings (D(i-1, j-1)).
data and achieve accurate transcription. The shared use of
If the last characters of word1 and word2 are different,
these algorithms highlights the importance of capturing both
three possible operations have to be considered:
temporal and spectral aspects of speech signals. These
Deletion: D(i-1, j) + 1 represents the distance if we delete
commonalities in training algorithms result in similar
the last character from word1 and then find the distance
performance and make both models reliable choices for
between the modified word1 and word2.
speech recognition tasks.
Insertion: D(i, j-1) + 1 represents the distance if we insert
III. RELATED WORKS the last character of word2 into word1 and then find the
distance between the modified word1 and word2.
In this research study [15] delve into the realm of Substitution: D(i-1, j-1) + 1 represents the distance if we
predictive pricing models designed to classify potential substitute the last character of word1 with the last character
customers. Through the utilization of data-driven of word2 and then find the distance between the modified
approaches, the authors explore various techniques to word1 and word2.
effectively analyze customer behavior and preferences. By
The overall Levenshtein distance is then given by the
leveraging machine learning algorithms and extensive
minimum of these three possibilities.
datasets, these models aim to optimize pricing strategies and
enhance customer targeting for businesses.
Addressing the crucial need for heightened security in
critical infrastructures, Azarov et al. present a
groundbreaking approach in their research [16]. The focus (1)
lies on the development of real-time machine learning
- 90 -
B. Google speech recognition model algorithms where Encoder Hidden Statet represents the hidden state
Google's speech recognition model uses many at time step t, Input Featurest represents the input speech
algorithms, one of which are CTC. features at time step t, and Encoder Hidden Statet−1 is the
CTC is a key algorithm used in Google's speech previous hidden state.
recognition models. It allows the model to align the input Attention Mechanism: The attention mechanism allows
audio data with the corresponding transcriptions without the model to dynamically focus on different parts of the input
requiring precise time alignment. CTC works by inserting sequence during decoding. It computes attention weights
blank characters into the transcription, which can then be based on the compatibility between the current decoder
aligned with repeated or missing characters in the input hidden state and each encoder hidden state. The attention
sequence. This allows the model to handle input and output weights represent the relevance/importance of each encoder
sequences of variable length. During training, CTC uses an hidden state at each time step. The context vector, which
iterative optimization process to update the model parameters captures the attended information, is computed as the
based on the alignment between the input and target weighted sum of the encoder hidden states:
transcriptions. Using CTC, Google models can efficiently
learn speech recognition with different durations and
alignments. (6)
To calculate the forward probabilities for each step in the C. Bing speech recognition model algorithms
alignment CTC model use forward algorithm. It is used to
compute the probability of an input sequence given a target Bing speech recognition uses the Connectionist Temporal
sequence. Classification (CTC) algorithm, similar to Google.
Additionally, it incorporates the following algorithms and
techniques.
Deep Neural Networks (DNNs). Bing’s speech
(2) recognition model utilizes DNNs as acoustic models. These
networks learn complex representations of audio features and
where αt(yt) represents the forward probability of label yt capture high-level patterns. The forward pass of a DNN can
at time step t, αt−1(yt−1) is the forward probability at the be represented as:
previous time step, and P(yt|xt) is the probability of emitting
label yt at time step t given the input xt. (7)
When calculating the backward probability for each where hl represents the activations at layer l, Wl is the
alignment step, the backward algorithm of the input weight matrix, bl is the bias vector, and hl−1 represents the
sequence is used, considering the target sequence. activations from the previous layer.
Softmax Activation: The softmax activation function is
(3) often used in the output layer of a DNN for multi-class
classification. It computes the probabilities for each class
where βt(yt) represents the backward probability of label based on the final layer activations. The softmax function is
yt at time step t, βt+1(yt+1) is the backward probability at the defined as
next time step, and P(yt+1|xt+1) is the probability of emitting
label yt+1 at the next time step given the input xt+1. (8)
Label Probability. The probability of a target sequence where: - z is the vector of final layer activations, - C is
given an input sequence is calculated by summing over all the number of classes, - softmax(z)i is the probability
possible alignments at each time step. assigned to class i.
Backpropagation: Backpropagation is used to calculate
(4) the gradients of the network parameters (weights and biases)
with respect to the loss function. The gradients are then used
The formulas associated with sequence-to-sequence to update the parameters during training. The
models, specifically the Listen, Attend and Spell (LAS) backpropagation algorithm consists of two main steps: a)
architecture: Compute the gradient of the loss function with respect to the
Encoder Network: The encoder network processes the output activations.
input speech features and produces a sequence of hidden
states. It can be implemented using recurrent neural networks
(RNNs) such as Long Short-Term Memory (LSTM) or (9)
Gated Recurrent Units (GRU). The hidden states at each time
step are computed as follows:
Encoder Hidden Statet = RNN(Input Featurest ,Encoder Hidden Statet−1) (5)
- 91 -
V. PERFORMANCE TESTS AND RESULTS
A. Reasons for choosing Python programming language
The choice of Python and the specific libraries,
“SpeechRecognition” and “Levenshtein”, was motivated by
several factors. Python is widely recognized for its
versatility and extensive ecosystem of libraries and
frameworks. When it comes to speech recognition and
natural language processing tasks, Python offers a variety of
powerful tools and APIs that simplify the development
process.
Python's rich library ecosystem includes
“SpeechRecognition”, which provides a convenient and
user-friendly interface for working with audio files and
speech recognition models. It supports multiple speech
recognition APIs, allowing for easy comparison and
evaluation of different models. Additionally,
“SpeechRecognition” simplifies tasks such as recording
audio files and extracting text from speech, making it an
excellent choice for this experiment (Fig. 1).
Figure 2. Project structure
A functional model of the project, built on all
elements of the project structure, presents the whole process
visually (Fig. 3).
Figure 1. Recognition process
To enhance the understanding and interpretation of the
experiment's results, the article also incorporates
visualizations using the matplotlib library. These
visualizations include graphs comparing the speed and
accuracy of speech recognition models, providing a clear
representation of the findings.
The overall structure of the project is as follows (Fig.
2):
- Files Directory: This directory includes the audio
files that need to be recognized.
- Sources: The main directory that contains the Figure 3. Functional model
project's code.
- recognition_results: This directory contains a JSON As you could see the whole structure is linear and
file used to record and save the results of the speech by starting the main file, it initiates the others in turn. Figure
recognition process for subsequent visualization. 4 shows what this process looks like in the code itself.
- comparison: This directory includes functionality for
comparing the obtained results.
- data_parse: This directory contains code for parsing
and processing the audio files to prepare them for further
processing.
- main: The main file that serves as the entry point for
the entire project, integrating all the functionalities together.
- recognize: This directory includes the functionality
for speech recognition.
- 92 -
resource requirements, targeted evaluation, and more
concise interpretation of the outcomes.
In figure 5 you can see two graphs visualizing the
recognition results using Google's model, the first graph
shows the speed of recognition, the second the accuracy.
Figure 5. Google recognition model results
Figure 6 shows similar graphs for the Bing
recognition model.
Figure 4. Main code file
On average, the process of speech recognition and
comparison is not the fastest, especially when dealing with
larger volumes of data. Conducting comparisons with a
larger dataset can take more time but often yields a more
stable and reliable result. Alternatively, using a smaller
dataset reduces the processing time and provides a more Figure 6. Bing recognition model results
immediate visual representation of the results.
In this experiment, a smaller volume of data was
used for several reasons, each offering its own advantages CONCLUSION
(Fig. 5): In conclusion, this research conducted a comparison
Time Efficiency: By using a smaller dataset, the between two speech recognition models, employing the
overall processing time is significantly reduced. This allows Levenshtein method to calculate the distance and utilizing
for faster experimentation, iteration, and analysis of the the visualization capabilities of the matplotlib library to
speech recognition models. It is particularly advantageous present the results. The study aimed to evaluate the speed
when there are time constraints or a need for quick feedback and accuracy of recognition in the context of educational
on the performance of the models. applications and artificial intelligence.
Resource Limitations: Working with a smaller By analyzing audio recordings consisting of various
dataset reduces the resource requirements, such as memory phrases uttered by individuals with different accents and
and computational power. This makes it feasible to run the intonations, the study provided insights into the performance
experiment on less powerful hardware or in resource- of the two models. The results showcased the speed and
constrained environments, which can be beneficial when accuracy of audio-to-text conversion, shedding light on the
conducting the experiment on personal machines or less resource consumption of each model.
powerful devices. These findings have important implications for the
Simplification and Focus: A smaller dataset improvement of speech recognition systems and the
simplifies the experiment and focuses the analysis on a development of more adaptive and efficient solutions for
specific subset of speech samples. This can help in isolating audio data processing. The comparison of recognition
and understanding the performance of the speech accuracy in JSON format and the subsequent visualization
recognition models scenarios, such as specific accents or using matplotlib contribute to enhancing the integration of
intonations. It allows for a more focused evaluation of the speech recognition technologies into educational systems.
models' strengths and weaknesses within a constrained Furthermore, the research highlights the significance of
context. considering hardware limitations and performance factors
Clarity of Results: Working with a smaller dataset when selecting and developing speech recognition models.
can provide clearer and more concise results. With a Striking a balance between recognition accuracy and
reduced amount of data, it becomes easier to interpret and resource consumption enables effective functionality of
visualize the outcomes. This facilitates a more educational applications across different platforms and
straightforward comparison of the speech recognition devices.
models, as the differences in performance become more Overall, this study makes a valuable contribution to the
apparent when the dataset is smaller. advancement of comprehensive, innovative, and inclusive
Overall, the decision to use a smaller dataset in this solutions that have the potential to transform and enrich the
experiment offers advantages in terms of time efficiency, educational landscape. The research also identifies potential
resource limitations, experimental focus, and clarity of avenues for future exploration, such as multilingualism and
results. It allows for quicker iteration and analysis, reduced the analysis of speaker emotions, opening exciting prospects
- 93 -
for the further development of speech recognition [9] Wang, H., Chen, Y., & Li, M. (2019). Speech recognition
parameters analysis in audio recordings using the Levenshtein method.
technologies in the fields of education and artificial International Journal of Information and Education Technology, 9(10),
intelligence. 753-758.
[10] Zhang, L., Li, W., & Huang, R. (2018). Adaptive solutions for
REFERENCES audio data processing based on speech recognition in educational
[1] Li, X., Zhang, C., Zhao, Y., & Liu, X. (2020). Speech recognition applications. In Proceedings of the 11th International Conference on
in educational applications: A comprehensive review. IEEE Transactions Educational Data Mining (pp. 543-546). International Educational Data
on Learning Technologies, 13(3), 477-491. Mining Society.
[2] Liu, Q., & Li, H. (2018). An empirical study of speech recognition [11] Google Cloud Speech-to-Text Documentation. (Accessed on
models for educational applications. In Proceedings of the 26th 12.03.2023). Available at: https://cloud.google.com/speech-to-text/docs
International Conference on Computers in Education (pp. 435-444). Asia- [12] Google Cloud Speech-to-Text API. (Accessed on 12.03.2023).
Pacific Society for Computers in Education. Available at: https://cloud.google.com/speech-to-text
[3] Johnson, L., & Smith, R. (2017). Enhancing educational [13] Microsoft Azure Speech Service Documentation. (Accessed on
technologies with speech recognition: Current trends and future directions. 16.03.2023). Available at: https://learn.microsoft.com/en-
Educational Technology Research and Development, 65(5), 1281-1303. us/azure/cognitive-services/speech-service/index-speech-to-text
[4] Chen, Z., Liu, Q., & Huang, Y. (2019). Integrating speech [14] Microsoft Cognitive Services - Speech-to-Text API. (Accessed on
recognition with machine learning for intelligent educational applications. 16.03.2023). Available at: https://azure.microsoft.com/en-
Journal of Educational Technology & Society, 22(4), 143-155. us/products/cognitive-services/speech-to-text/
[5] Wang, Y., Yu, B., & Liu, Y. (2016). Performance analysis of [15] Yedilkhan, D., Mukasheva, A. Predictive pricing models to
speech recognition models in educational applications. In Proceedings of classify potential customers using data-driven approaches. AIP Conference
the 9th International Conference on Educational Data Mining (pp. 621- Proceedings, 2570(1), 040016. (2022). doi:10.1063/5.0099796.
626). International Educational Data Mining Society. [16] Azarov, I., Gnatyuk, S., Aleksander, M., Nazarov, I., Mukasheva,
[6] Garcia, V., & Liu, Q. (2018). Speech recognition parameters A. Real-time ML Algorithms for The Detection of Dangerous Objects in
analysis for educational applications using the Levenshtein method. Journal Critical Infrastructures. CEUR Workshop Proceedings, 3373, 217-226.
of Educational Computing Research, 56(6), 891-912. (2023).
[7] Smith, J., & Brown, A. (2020). Visualizing the accuracy of speech [17] Smith, J., Johnson, R. Speech Recognition: Advancements and
recognition models in educational applications. Journal of Educational Applications. Journal of Artificial Intelligence Research, 29, 123-145.
Technology Systems, 48(3), 354-372. (2023)
[8] Lee, S., & Park, J. (2017). Prospects for speech recognition
technologies in education and artificial intelligence. Journal of Educational
Technology Development and Exchange, 10(1), 39-50.
- 94 -
Research and development of sign language
recognition system using neural network
algorithm.
Yegor Matveyas Arailym Keneskanova Dastan Kambarov
MSc student of Almaty University of Power Senior Lecturer Professor of the Senior Lecturer Professor of the
Engineering and Telecommunications Department of Information Systems Department of Information Systems
(AUPET) named after G.Daukeev and Cybersecurity and Cybersecurity
Almaty, Kazakhstan Almaty University of Power Almaty University of Power
[email protected] Engineering and Telecommunications Engineering and Telecommunications
(AUPET) named after G.Daukeev (AUPET) named after G.Daukeev
Almaty, Kazakhstan Almaty, Kazakhstan
[email protected] [email protected]
Assel Mukasheva
PhD, Associate Professor of the Department
of Information Systems and Cybersecurity
Almaty University of Power Engineering and
Telecommunications (AUPET) named after
G.Daukeev
Almaty, Kazakhstan
[email protected]
orcid.org/0000-0001-9890-
4910
Abstract— Sign language is an important communication LSTM layers, when developing a neural network algorithm for
tool for the hearing-impaired community. Due to technological sign language recognition.
development, it is now possible to develop systems that can
recognize, translate and process sign language into text or Keywords: Sign language, recognition, Python, Keras,
speech, according to the visual representation of gestures. This neural network.
article explores the development of a sign language recognition
system using a neural network algorithm. The Mediapipe I. INTRODUCTION
library was used to determine the key points and orientation of
According to the World Federation of the Deaf, there are
the user's hands and fingers. After that, the software module
transmits the collected data to a sequential neural network that
more than 200 sign languages worldwide, which are used by
includes layers of Long Short-Term Memory (LSTM). To build 70 million deaf people [1].
this type of neural network, the open Keras library was used. Sign languages are natural languages that use visual-
The key feature of the presented neural network model is the gestural communication and are primarily used by deaf and
combination and interaction of convolutional and Recurrent hard-of-hearing individuals to communicate with one another.
neural network (RNN) layers. The considered set of layers Sign languages are distinct from spoken languages and have
provides the ability to track the dependence of data over time,
their own grammar, vocabulary, and cultural nuances. Sign
this is achieved by switching between layers of different types
languages rely on a combination of visual cues, such as hand
and reducing the number of neurons. The LSTM network is
trained using a custom editable dataset based on American Sign
movements, facial expressions, and body language, to convey
Language gestures. The dataset was formed on the recording of meaning. These elements can be combined and modified to
signs. Each sign representation was preprocessed to extract create words, sentences, and complex linguistic structures [2].
three-dimensional landmarks. These collected key points were American Sign Language was used as the basic sign
transmitting to the layers of the LSTM neural network, which language. It is the main sign language in the deaf communities
allowed the model to study the complex relationships between of the USA and English-speaking parts of Canada [3]. The
hand movements and the corresponding gestures. Each sign
choice of this language as the main one was made due to its
sample is represented by a sequence of 24 frames. The
effectiveness of the neural network algorithm is evaluated using
prevalence and availability of materials on this sign language
various indicators, including an accuracy of the model. The [4].
results of the experiment show that the developed software can Real-time sign language recognition is a challenging and
achieve a high level of accuracy in recognizing the gestures of impactful task that aims to facilitate effective communication
the sign language. The relevance of this study is confirmed by between individuals with hearing impairments and the hearing
the possibility of its application in a wide range of areas. In community [5]. The key sides of this study involve collecting
particular, the software has the potential to be used as a service
data, utilizing computer vision techniques, and implementing
or tool for communication between people with disabilities and
gesture recognition and translation. Machine learning
the general public, or as a technology that helps people with
hearing impairments. The authors of the research note that the
algorithms and features were employed to classify signs into
results of the work done demonstrate the possibility and words [6,7], with basic integration of natural language
effectiveness of using a neural network algorithm, including processing and generation for translation purposes, which is
based on the Markov chain [8].
- 95 -
II. MATERIALS AND METHODS
A. Related work
Considering the available tools and technologies used in
the field of human recognition in general and gestures in
particular, several relevant categories can be distinguished [9]:
1. Technologies that use special gloves or bracelets to
read the position and orientation of the hands,
followed by the classification of gestures.
2. Solutions based on the use of an individual user
camera or a combination of such cameras using
additional sensors.
However, often, such a set of hardware is quite expensive
and may not always be available, unlike, for instance, a regular
user webcam. Figure 1. LSTM cell structure.
One of the most important parameters of gesture The states of previous and current LSTM cells are defined
recognition systems is the accuracy of sign classification and as 𝑆𝑘−1 and 𝑆𝑘 respectively.
system performance. It is important to pay attention to the fact Forget gate is represented by the first sigmoid layer, which
that the accuracy of the final classification result can be outputs a number that varies from 0 to 1, aim of this gate is to
affected by both the quality of images, the presence of light make a decision, whenever to keep value or not. The output of
and its intensity in the frame, and the specificity of the this gate is 𝐹𝑘 , which can be described by this equation:
application conditions for which the specific algorithm used is
adapted. In addition, when recognizing a continuous sign 𝐹𝑘 = σ(𝑊𝑘 × [𝐻𝑘−1 , 𝑋𝑘 ] + 𝑏𝑘 )
language, there is a problem of coarticulation, that is, the Input gate consists of two stages. The first one is based on
imposition of gestures in sign languages, which no one has yet the second sigmoid layer and its purpose is to decide which
been able to completely solve [10]. value will be updated. On the next step a vector of new
Determining the position and identification of the hand in potential values is created using the first tanh layer.
space is only the initial stage, followed by the classification of 𝐼𝑘 = σ(𝑊𝑖 × [𝐻𝑘−1 , 𝑋𝑘 ] + 𝑏𝑖 )
the sign. To do this, a conventional convolutional neural
network can be used. It is possible to analyze the entire source 𝑃𝑘 = tanh(𝑊𝑝 × [𝐻𝑘−1 , 𝑋𝑘 ] + 𝑏𝑝 )
image, and to work with the coordinates of the key points of
the fingers. However, in the case of basic convolutional neural The result of input gate is equals to the multiplication of
networks, it is possible to work only with a separately selected 𝐼𝑘 and 𝑃𝑘
frame, but for the recognition of dynamic sings, a whole-time State of the current cell is calculated by the following
interval must be analyzed and, accordingly, a set of frames expression:
must be considered when classifying, each previous one of
which will be used when analyzing subsequent frames. Such 𝑆𝑘 = 𝐹𝑘 × 𝑆𝑘−1 + 𝐼𝑘 × 𝑃𝑘
a task can be implemented using long-term short-term Final output gate is represented by the third sigmoid layer
memory (LSTM) neural networks [11,12]. that defines which part of the final state are going to the
The LSTM layer consists of a repeating block called an output. After this, current state goes through second tanh layer
LSTM cell, which has a more complex structure than and multiplied on the result of previous sigmoid layer.
conventional recurrent neural networks [13]. An LSTM cell 𝑂𝑘 = σ(𝑊𝑜 × [𝐻𝑘−1 , 𝑋𝑘 ] + 𝑏𝑜 )
consists of three main components: input g ate, forget gate,
and output gate (Fig. 1). 𝐻𝑘 = 𝑂𝑘 × tanh (𝑆𝑘 )
In general terms, each of these components controls the B. Neural network structure
flow of information inside the cell. Input gate determines the The neural network model can be described as a directed
source information that will be added to the cell. Forget gate acyclic graph of connected layers, in which input data passes
selects the information that will be forgotten or stored in the through the layers sequentially.
cell. And in the case of the Output gate, it determines what
information will be transmitted from the cell. In addition, the The first step to build a neural network was to define
LSTM cell has an internal state that can be transmitted to the LSTM layers that allow the model to consider the sequence of
next time step, storing information for a specified period of frames under consideration associated with gestures. Each
time [14]. frame in the sequence was transmitted through an LSTM cell,
which updated its internal state and generated an output. After
the LSTM layers, fully connected dense layers have been
added to process high-level features and bind the information
obtained from the LSTM layers. At the end of the model there
was an output layer with the number of neurons corresponding
to the number of classes (signs) in the original sign language
that needed to be recognized. The softmax activation function
- 96 -
was used for the output layer, which allowed to obtain
probabilities for each class.
As part of the current study, a neural network was
compared with a different set of corresponding layers. Based
on the results of efficiency and performance comparisons, a
sequential neural network with the following set of layers was
selected:
Let X be the input tensor of the form (batch_size, 24,
1662), where batch_size is the number of training (test)
samples, 24 is the length of the input sequence, number of Figure 2. Final neural network model.
frames, and 1662 is the number of features at each time step.
C. Data collection and processing
The first layer of the sequential neural network is
At the beginning of the experiment, a balanced data set
represented by an LSTM layer with 128 units, while it returns
was collected. Each sign was represented by a set of 150
the output sequence for the next LSTM layer using the ReLU()
unique samples recorded using a laptop webcam. All the
activation function:
gesture cases are provided as a sequence of 24 frames.
L1 = LSTM(X, ReLU), where L1 is the output tensor of
The Mediapipe library was used to obtain source data for
the form (batch_size, 24, 128).
training and testing. The software implementation initiates the
The second layer applies another LSTM function with 256 recognition process by receiving the input video stream of the
units, returning sequences outputs using the ReLU activation user performing gestures in the sign language. The video was
function: then pre-processed to extract three-dimensional landmarks, as
well as characteristics of movement and hand position [15,16].
L2 = LSTM (L1, ReLU), where L2 is the output tensor of However, in addition to the hands, human pose and face
the form (batch_size, 24, 256). landmarks were detected and collected too. This is necessary
The third layer applies another LSTM function with 64 to combine the sign component as well as human facial
units, without returning sequences, using the ReLU activation expressions, which also plays an important role in most sign
function: languages [17].
L3 = LSTM (L2, ReLU), where L3 is the output tensor of From the coordinates of the key points received from
the form (batch size, 128). Mediapipe, a sequence of data is formed, represented by a
one-dimensional array, where each time step corresponds to a
The fourth layer applies a fully connected (Dense) specific frame of the video stream. It was decided to consider
function with 128 units and a ReLU activation function: only the hands to obtain the initial data, since this set of key
L4 = Dense (L3, ReLU), where L4 is the output tensor of points is optimal and does not cause large deviation with a
the form (batch size, 128). small data set for training a neural network. Each gesture is
represented by a sequence of 24 frames received as part of
The fifth layer applies another fully connected function processing the incoming video stream.
with 64 units and the ReLU activation function:
To prevent duplication and overlay of the previous gesture
L5 = Dense (L4, ReLU), where L5 is the output tensor of when predicting a new one the delay mechanism was realized.
the form (batch size, 64). According to the results of manual testing, the ratio of 36
The sixth layer applies another fully connected function unused frames to 24 frames in use is optimal for latency. This
with 32 units and the ReLU activation function: allows the system to prevent the imposition of gestures in
motion. When processing a video stream, every 60 frames are
L6 = Dense (L5, ReLU), where L6 is the output tensor of selected, of which the first 36 frames act as a delay. These
the form (batch size, 32). frames are not used in any way, they are only needed to wait
for a certain moment before moving or changing the position
The last layer applies a fully connected function with the
of a person's hand. The remaining 24 frames of the sequence
number of units equals to the number of classes in the current
of the video stream are transmitted to the neural network to
classification task using the softmax activation function:
classify the current gesture.
Y = Dense(L6, softmax), where Y is the output tensor of
As an optimal representation of the data, the form of a one-
the form (batch size, number of actions), representing the
dimensional array of numbers normalized in the range from 0
probability distribution of actions.
to 1 was chosen. For each of the hands there are 21 key points
Schematically, the neural network is shown in Figure 2. represented by three coordinates (Fig.3), it means that for one
hand there are 21 * 3 = 63 values. Two hands contain 63 * 2
Batch size is a dynamic value, because of testing was = 126 values. Face landmarks consist of 468 3 dimensional
carried out on different sets and the number of classes for key points. And finally, human pose is represented by 33 4D
gesture classification. landmarks. Adding all these values together, a finite number
of features equal to 1662 was obtained.
- 97 -
TABLE II. TABLE SENTENCES AND PHRASES REPRESENTED
IN THE DICTIONARY
Number Sentence
1 Thank you
2 You are welcome
3 I am fine
4 I am ok
5 How did it happen
6 How are you
7 How do you feel
8 I love you
Based on these phrases and sentences, a matrix of
transitions between words was built, which in the future will
be used when generating text.
III. EXPERIMENT AND RESULTS
The Python programming language was used as the main
language. This research work involved the integration of
Mediapipe and Keras libraries to develop a robust system. The
Figure 3. Mediapipe hand model. first step was to create a script that could effectively collect
data using the Mediapipe library. This script utilized various
After receiving the key points, normalizing their values, modules within Mediapipe to extract critical information such
and converting the result into a one-dimensional array, the as the landmarks of the left hand, right hand, pose, and face
arrays are recorded frame-by-frame into files located in the from video streams (Fig 4).
project dataset directory. There is a separate directory for each
class of the recognizer, which includes several folders, which
amount equals to the number of recorded experiments. In this
case, 150 variations of this gesture were recorded for each
gesture. A total of 1950 samples of signs were collected. Each
folder for all variations of a certain gesture includes a set of
24 documents of .npy format, which corresponds to the
number of frames used to obtain data for training a neural
network. This format for storing numeric arrays is quite
convenient, since the numpy library is used for calculations Figure 4. Function for getting landmarks.
and mathematical operations with arrays, which easily
interacts with the .npy file format. Once the data collection script was in place, a mechanism
was implemented to save the extracted data to .npy files. This
During the learning process, the stored values are parsed allowed for easy storage and retrieval of the recorded data for
from files as part of the cycle through all internal directories further processing. By leveraging the power of NumPy, the
of the dataset. The resulting data set is divided into 2 collected information, including the landmarks and other
categories – training data and testing data, in a ratio of 85% relevant features, was stored in .npy format (Fig 5).
and 15% respectively, so that there is more initial training
data, thereby increasing the accuracy of the trained neural
network.
A list of all American sign language words that were used
in curent classification is shown in Table 1.
TABLE I. TABLE SIGN LANGUAGE WORDS REPRESENTED
IN THE DATASET
Number Word
1 Hello
2 Hi Figure 5. Function for collecting data.
3 Welcome
4 Ok
Separate functions were then developed to parse and work
5 You with the stored data. The previously saved .npy files were
6 I loaded, extracting the necessary information, and applying
7 Fine required preprocessing steps. The preprocessing stage
8 How involved data transformations.
9 It
10 Happen With the data prepared, the project moved on to the
11 Feel creation of different Keras sequential models suitable for sign
12 Love
language recognition. These models were designed and
13 Thank
implemented using the Keras library, which offers a high-
level interface for building deep learning models.
Table 2 demonstrates a set of available sentences and
phrases that can generated according to the sequence of Finally, a script was created to enable real-time testing of
predicted words. the sign language recognition system. This script utilized a
specific trained model to predict sign language gestures in
- 98 -
real-time. By processing the video stream data from a camera, The second model, which was subsequently chosen as the
the system could provide immediate output, such as main one, is a more complex and voluminous version of the
displaying the predicted results on the screen. first model. It has seven layers, which were described in one
of the previous sections. Training on 25 epochs for the second
In total, 2 models were considered. Categorical cross model gave weak results, loss equals to 0.5571 are quite high,
entropy and categorical accuracy were chosen as an evaluation and accuracy, on the contrary, is relatively lower than
of the effectiveness of the neural network models under expected – 0.8564 (Fig. 10.)
consideration, because a large set of initial classes is presented
in the current task of sign classification.
The first model is compact, it includes 4 layers, two of
which are LSTM layers, and the remaining ones are Dense
(Fig. 7).
Figure 7. The first compact model of a neural network.
The first training of this neural network model was
conducted during 25 epochs of training. The results for this
training process were the following results: loss - 0.3248;
accuracy - 0.8944 (Fig. 8).
Figure 10. Results of training the second model for 25 epochs.
However, when training the second model for 75 epochs,
the results improved markedly and surpassed the results of
training the first model: loss - 0.2671; categorical accuracy -
0.9119 (Fig. 11).
Figure 8. Results of training the first model for 25 epochs.
The second training was launched with a duration of 75
epochs and gave the following results: loss - 0.2967; accuracy
- 0.8920 (Fig. 9).
Figure 11. Results of training the second model for 75 epochs.
In this case, increasing the number of units in the LSTM
layers in the second model provides the possibility to increase
the capacity of these LSTM layers and the ability of modeling
complex dependencies in the data.
According to the results of manual testing, the problem of
imposition of gestures was revealed. This led to incorrect
prediction of signs and in this regard, a delay between
predictions was added. The optimal ratio of prediction time
and delay equal to 1 to 1.5 (24 to 36 frames) was established
as a result of testing. This problem appears since sign
recognition begins immediately after the completion of the
previous operation, the user does not have time to move his
Figure 9. Results of training the first model for 75 epochs.
- 99 -
hands at least minimally in space and prepare for the next REFERENCES
gesture. For this reason, adding this kind of delay is necessary. [1] World Federation of the Deaf https://wfdeaf.org/our-work/
In addition to recognizing individual independent words, [2] V. Kimmelman, “Information Structure in Sign Languages”, Sign
Languages and Deaf Communities Volume 10, 2019,
a mechanism for generating sentences described in a special https://doi.org/10.1515/9781501510045
dictionary was created. For this purpose, a function based on
[3] S. K. Liddell, “American Sign Language Syntax”, Approaches to
Markov chains was developed (Fig. 12), which generates Semiotics Volume 52, 1980 https://doi.org/10.1515/9783112418260
sentences and phrases using the results of the classification of [4] J. W. Guido, “Learn American Sign Language: Everything You Need
signs by the neural network described earlier. to Start Signing”, Wellfleet Press, 2015
[5] R Rastgoo, K Kiani, S Escalera, “Sign Language Recognition: A Deep
Survey”, Expert Systems with Applications Volume 164, February
2021, 113794, https://doi.org/10.1016/j.eswa.2020.113794
[6] Didar Yedilkhan, Assel Mukasheva, “Predictive pricing models to
classify potential customers using data-driven approaches”, AIP
Conference Proceedings, 18 August 2022; 2570 (1): 040016.
https://doi.org/10.1063/5.0099796.
[7] I. Azarov, S. Gnatyuk, M. Aleksander, I. Nazarov, A. Mukasheva;
“Real-time ML Algorithms for The Detection of Dangerous Objects in
Critical Infrastructurs”, CEUR Workshop Proceedings, 2023, 3373, pp.
217–226.
[8] T. Almutiri, F. Nadeem, “Markov Models Applications in Natural
Language Processing: A Survey”, I.J. Information Technology and
Computer Science, 08 April 2022,
https://doi.org/10.5815/ijitcs.2022.02.01
[9] M Oudah, A Al-Naji, J Chahl, “Hand Gesture Recognition Based on
Figure 12. Natural language generating function.
Computer Vision: A Review of Techniques” Journal of Imaging. – 23
This function uses a transition matrix that displays the July 2020, https://doi.org/10.3390/jimaging6080073
probabilities of words following each other. Thus, the most [10] S. Albanie, G. Varol, L. Momeni, T. Afouras, J. S. Chung, N. Fox, A.
Zisserman, “BSL-1K: Scaling Up Co-articulated Sign Language
probable word between the last two words in the sentence is Recognition Using Mouthing Cues” European Conference on
determined. After that, a recurrent function call occurs to Computer Vision, 2020, pp 35–53, https://dx.doi.org/10.1007/978-3-
continue constructing a sentence or phrase. 030-58621-8_3
[11] M. L. Amit, A. C. Fajardo, R. P. Medina , “Recognition of Real-Time
IV. CONCLUSION Hand Gestures using Mediapipe Holistic Model and LSTM with MLP
Architecture”, 2022 IEEE 10th Conference on Systems, Process &
A compact model is sufficient for a classifier with a set of Control, 2022, https://doi.org/10.1109/ICSPC55597.2022.10001800
classes of no more than 15 pieces. However, with an increase
[12] A. S. Agrawal, A. Chakraborty, M. Rajalakshmi , “Real-Time Hand
in the number of source classes, this model will not cope with Gesture Recognition System Using MediaPipe and LSTM”,
the task. For a larger number of classes, a more complex International Journal of Research Publication and Reviews, Volume 3,
model is needed, which could consider the features of no 4, pp 2509-2515, April 2022
overlapping gestures that are performed in a similar way. [13] Y. Yu, X. Si, C. Hu, J. Zhang , “A Review of Recurrent Neural
Networks: LSTM Cells and Network Architectures”, Neural
For basic communication, the considered model is quite computation, 2019, https://doi.org/10.1162/neco_a_01199
sufficient. Such a set of communication elements includes 13 [14] R. C. Staudemeyer, E. R. Morris , “Understanding LSTM -- a tutorial
words and 8 complex sentences and phrases. According to the into Long Short-Term Memory Recurrent Neural Networks”, Ralf C.
original training dataset balance was achieved between the Staudemeyer, arXiv 2019, https://doi.org/10.48550/arXiv.1909.09586
capacity of the model and the available resources. Sentences [15] F Zhang, V Bazarevsky, A Vakunov , A. Tkachenka, G. Sung, C.
and phrases were formed correctly during the testing, since the Chang, M. Grundmann , “MediaPipe Hands: On-device Real-time
Hand Tracking”, CVPR Workshop on Computer Vision for
presented dictionary is quite small. Increasing the source Augmented and Virtual Reality, 2020,
dictionary may require choosing a different and more https://doi.org/10.48550/arXiv.2006.10214
progressive approach to generating text in real time. [16] M. Harris, A. S. Agoes, “Applying Hand Gesture Recognition for User
Guide Application Using MediaPipe”, 2nd International Seminar of
Considering the results and effectiveness, for such a set of Science and Applied Technology (ISSAT 2021), 2021,
classes, the accuracy of the model during training was 91.2%, https://doi.org/10.2991/aer.k.211106.017
and for the testing part of initial dataset categorical accuracy [17] E. A. Elliott, A. M. Jacobs, “Facial expressions, emotions, and sign
was equal to 91.1%, which is acceptable indicators. To languages”, Frontiers in psychology, 2013,
improve the accuracy of the second model, it is necessary to https://doi.org/10.3389/fpsyg.2013.00115
increase the number of training samples for each sign. The
number of required epochs for training was increasing
proportionally with the expansion of the model.
- 100 -
Research and Development of a Chatbot Using
Neural Network Technologies to Process
Requests
Kyzyrov Daniyar Ersainuly1, Sapakova Saya Zamanbekovna 2
1
International Information Technologies University, Kazakhstan
2
International Information Technologies University, Kazakhstan
Email: [email protected]
Abstract – The Python programming language has become it will join. Then we add the functions we need. In our case,
the most popular choice for creating chatbots due to its the chatbot acquired the functions of moderation, mute, ban,
effectiveness, affordability, convenience, and the abundance of kick, automatic roles, and prefixes, as well as a couple of
libraries available for natural language processing (NLP) and functions for entertainment. For example, to turn on music in
machine learning. This paper discusses the development of a voice chat, to get information about movies, TV shows, and
chatbot designed for use on the Discord platform using Python, so on. In addition, functions have been added that use neural
specifically utilizing neural network technologies. Chatbots are network technology to process requests, these are Image
becoming increasingly popular on Discord, as they offer Recognition functions, as well as a conversation(dialog)
expanded functionality that can replace certain websites. For
function. All these functions were implemented using the
example, chatbots that provide access to music have become
quite popular. However, the widespread use of chatbots has
Python programming language, its libraries, as well as various
raised concerns about potential harm, particularly when a single APIs, including the TMDB API, and direct interaction with
chatbot using the YouTubeDL library can work on millions of the YouTube search engine.
servers, reducing revenue for YouTube. The development of the For the functions of conducting a dialogue, as well as the
chatbot involved testing a variety of neural network models, Image Recognition functions, as we said earlier, the neural
including VGG, ResNet, AlexNet, Inception, and LeNet, using network models are responsible. In the course of research and
different types of optimizers and normalizers. Ultimately, three development, many variants of neural network models were
different models were created, each using a different optimizer
observed, compared, and the best ones were selected for use
(SGD, Adam, and RMSprop) and undergoing two hundred
epochs with a batch size of five. While all three models
in certain functions.
demonstrated high training accuracies, their testing accuracies II. EASE OF USE
varied, indicating potential overfitting. In addition to its
intended use as a chatbot, the program was extended to include To use the functions of a chatbot, you first need to add it
functionality that allows users to classify photos uploaded to the to your server, give it the appropriate role, and also grant
platform. For this task, convolutional neural network models permissions. After that, with the help of various commands,
were utilized. CNNs are designed specifically for image
we call the functions. In order to use commands, we enter the
processing tasks, such as object detection, image classification,
and segmentation. They consist of several neural layers,
sign ‘!’, after that, we write the name of the function that we
including convolutional, fully connected, and pooling layers. need, so we can use any function of the chatbot.
Several popular CNN models will be introduced and compared, The chatbot has many functions, an example calling the
including AlexNet, ResNet, VGG, Inception, MobileNet, and
‘!ping’ function, the chatbot will give a response to a text chat,
DenseNet. Each model has its own strengths and weaknesses,
and the choice of model ultimately depends on the specific for example, 100ms. This function gives information about the
application. This paper presents the development of a chatbot connection delay between the client and the chatbot.
created using Python and neural network technologies, which The list of files for the chatbot to work can be seen in Fig.
includes a photo classification functionality utilizing 1. Functions are written inside the Python files, data is stored
convolutional neural networks. The development process inside the JSON files, and we can say this is kind of our
involved testing various neural network models and optimizers database, where we store data. The main file is ‘chatbotd1’.
to achieve the desired performance. By discussing the different Here we specify the token and call the rest of the Python files.
components and models utilized in the creation of the chatbot, Inside the rest Python files, various functions are stored.
this paper provides insight into the practical applications of
neural network technologies in the development of
conversational agents.
Keywords – Conversational Agent, Natural Language
Processing, Application Programming Interface, Neural Network,
Artificial Intellegence
I. INTRODUCTION
First of all, to create a chatbot, you need to register on the
Discord website. Switch to developer mode and create an
empty bot template for now. After creating the bot, we specify
its permissions [1], which it will request from each server that
- 101 -
Thus, it becomes possible to use different APIs for different
functions of your chatbot. An example of the search function
can be seen in Fig.3.
Fig 1. Chatbot files Explorer
If there are several chatbots on your server, then the calling
prefix can be changed at any time so that there is no confusion
between chatbots, this is done using the specially written Fig 3. Movie Search
‘setprefix’ [2] function. First, we call it with the default prefix
‘!’, and then we set a new value, let's say a dot. This character An important function of the chatbot is its ability to
is then written to a special ‘prefixes.json’ file, and for identify objects from different images, this function is called
subsequent calls, it will be possible to use the new dot (.) through the find command, where you can upload an image of
character. something, and the chatbot will give you a response in text
format, with the name of the object in the image. The chatbot
III. CHATBOT FUNCTIONALITY also has the function of interacting with YouTube search. An
example can be seen in Fig. 4.
As we have already said, we call all functions using a
special prefix, and the name of the function we need.
Moderation functions that are implemented in the chatbot, are
kick, ban, unban, mute, unmute, commands for working with
prefixes and roles - setprefix, setmuterole, joinrole.
Moderation functions, as you can see from their name, are
needed in order to directly manage server members, exclude
them, add them to the blacklist, moderate their voice chat, and
so on. Also, through the joinrole function, you can specify a
role that all newcomers to the server will automatically
receive.
The function responsible for the dialogue (conversation)
[3] with the chatbot can be called using the ‘!g’ prefix, then
writing your question to the chatbot there. Thus, the chatbot
determines what exactly you wrote to it and, thanks to
machine learning [4], will give you an answer that will match
your question. An example of a dialogue with a chatbot can be Fig 4. Movie Search
seen in Fig. 2. By calling this command, you can listen to audiobooks,
and music in the voice channel, it will be enough to call the
command and write the name of the song or audiobook you
need. The chatbot joins your voice channel, downloads the file
from YouTube, and plays. You can also add music to the
queue, listen to the next song in the queue, pause the song, and
stop the bot, thereby disconnecting from the voice channel.
IV. NEURAL NETWORK MODELS
A. Goldy as a Conversational Agent
For the chatbot to answer our questions (conversational
purpose) [6], a machine-learning model was created and
Fig 2. Conversation with Chatbot trained on a dataset that was defined in a JSON file. This
simple database is divided into three main parts, the first part
Through interfacing with the TMDB API [5], a feature was is our tags, in other words, the general name of the questions
introduced to search for various movies. By calling this that the user can ask. The next part is the questions that users
function using the search command, you can get a lot of ask, based on these questions, our chatbot is trained, and it
information about a particular movie. All this information is gives the answer based on the third and last part of the
taken directly from the website whose API you are using.
- 102 -
database, these are the responses. Answers to user questions third layer had as many neurons as the number of classes to be
are written here, for example, to a request to tell a joke, the predicted and has the softmax activation function. This
chatbot can give out more than 150 different types of jokes. architecture also includes two dropout layers, with a dropout
rate of 0.5. The optimizer used is Stochastic Gradient Descent
Intents and patterns were tokenized and preprocessed (SGD) with a learning rate of 0.01, a momentum of 0.9, and
using the NLTK library in Python. Lemmatization, stemming,
Nesterov momentum enabled.
and removal of unnecessary characters like question marks
and commas were also involved in the preprocessing. The The second model is similar to the first model but with
resulting words were then sorted and saved as a set. larger dense layers. It has three dense layers, with 256 neurons
in the first layer and 128 neurons in the second layer, both with
Next, classes were created based on our tags in the JSON the ReLU activation function. The third layer was the same as
file, as we said earlier, this is the common name for our similar
in the previous one. This architecture also includes two
questions. The data for training our model was selected by dropout layers, with a dropout rate equal to 0.5. The optimizer
looping through each pattern in each of the intents and used is Adam with a learning rate of 0.001.
creating a bag of words for these patterns. A bag of words is a
vector that represents the occurrence of each word in a pattern. The third model was more complex than the previous two
It has been added to the output row corresponding to the tag, architectures. It had four dense layers, with 128 neurons in the
and the output row is then appended to the training data. The first layer, 64 neurons in the second layer, and 32 neurons in
training data was then shuffled and split into the input (bag of the third layer, all with the ReLU activation function. The
words) and output (tag) arrays, or in other words our train_x fourth had an activation function named softmax. This
and train_y. architecture also includes three dropout layers, with a dropout
rate of 0.3. The optimizer used is RMSprop with a learning
A neural network model was then created using the Keras
rate of 0.001. All three architectures had different
API of TensorFlow. The model consisted of three layers - an configurations of layers and optimizers, with varying numbers
input layer with the same number of nodes as the length of the
of neurons and dropout rates. The aim is to experiment with
input array, a hidden layer with 128 nodes and a rectified different architectures to find the best one that gives the
linear unit (ReLU) activation function, and an output layer
highest accuracy on the given dataset.
with the same number of nodes as the length of the output
array and a softmax activation function. The model was Overall, the first model seems to be the most optimal and
compiled with a stochastic gradient descent (SGD) optimizer efficient as it has the fastest training time and the highest
and a categorical cross-entropy loss function. It was trained on training and testing accuracies, therefore, this model, with the
the input and output arrays using a batch size of 5 and an epoch SGD optimizer became the basis for the chatbot.
count of 200. Finally, the trained model was saved as an H5
file using the ‘save()’ method of the Keras API. The resulting Since our dataset is not too large and contains only a small
number of questions and answers (intents), this may not be
model can be used to predict the tag of new user inputs and
enough to train our chatbot. During the training process,
generate appropriate responses.
different models showed different accuracies. In Fig. 5, we
For the chatbot to work correctly, three different models can see the accuracy graph for training and test data for the
were initially created. These models differed in optimizers, the model that used the Adam optimizer. As we can see, the
first model had an SGD optimizer, the second Adam, and the training accuracy shows very good results, after the twenty-
third RMSprop. They all had two hundred epochs and a batch fifth epoch it reaches about 90 percent, and on average shows
size of five. As we can see in Table 1, all three models have above 95 percent accuracy. But at the same time, the test
high training accuracies, but the testing accuracies are accuracy is very different from the training accuracy, which is
different and have a lower value, which may mean some caused by overfitting. This means that our dataset is too small
overfitting. in volume and needs to be expanded.
TABLE 1. MODEL COMPARISON
Model Training Testing Training Inference
Optimizer Accuracy Accuracy Time (s) Time (ms)
SGD 0.98 0.77 23.05 1.16
Adam 0.96 0.55 38.15 1.23
RMSprop 0.95 0.43 54.78 1.34
The first model with the SGD optimizer has the largest
accuracy, then the model with the Adam optimizer and
RMSprop follow in turn. The first model has the fastest
training time, while the third model with the RMSprop
optimizer has the slowest. The inference time is slightly
different.
In addition to optimizers, neural network models have also
been built using other architectures. Three different neural
network architectures for the chatbot were made, each with a Fig 5. Train and Test Accuracy (Model 2)
different set of layers and optimizers.
As we can see in Fig. 6, the situation with the loss is the
The first neural network model had three dense layers, same. By the twenty-fifth epoch, the test loss decreases, but
with 128 neurons in the first layer and 64 neurons in the after that, the decrease stops, and, on the contrary, the increase
second layer, both with the ReLU activation function. The
- 103 -
in loss begins. The train loss still shows good results, by the ResNet101V2, ResNet50V2, and InceptionV3. After
hundredth epoch the value of the loss drops below 0.2. comparing these models, taking into account the accuracy and
loss of each model that can be seen in Table 2, it was decided
to use the InceptionV3 model.
TABLE 2. COMPARISON OF THE CNN MODELS
Model Name Accuracy (Average) Loss (Average)
VGG-16 0.441 1.602
VGG-19 0.617 1.156
ResNet-50 0.654 1.277
ResNet-101 0.718 1.011
ResNet-50V2 0.826 0.559
ResNet-152 0.677 1.091
ResNet-101V2 0.884 0.400
InceptionV3 0.884 0.381
After we found out that the InceptionV3 model performs
well on the Food11 dataset and its loss is lower than ResNet-
101V2 has, it was decided to use this model as the basis for
Fig 6. Train and Test Loss (Model 2) the Image Classification functionality.
Fig. 7 shows our Precision, Recall, and F1 Scores. The Next, we used the same model but already trained based
weighted average value for Precision is 0.45, Recall 0.46, and on the ImageNet database. This database differs from the
0.43 for F1-Score. Precision, recall, and F1 score are Food11 database in its class sets. This database includes 14
commonly used metrics to evaluate the performance of a million different images that are labeled for 21 thousand
classification model. They are calculated based on the different categories of objects. The images were then divided
confusion matrix, which summarizes the number of true into 1.2 million training images, 50,000 validation images,
positives, false positives, true negatives, and false negatives and 100,000 test images. Many state-of-the-art models train
produced by the model. on this dataset for better performance as well as better object
detection. The InceptionV3 model was no exception, and we
will use this model (pre-trained version), which has already
been trained on the dataset, as the basis for our chatbot,
namely for its image detection function.
Fig. 8 shows the structure of the VGG-16 model. It has a
total of sixteen layers, including convolutional layers, pooling
layers, and fully connected layers.
Fig 8.VGG-16 Model Structure
The task of extracting features from an input image is
Fig 7. Precision, Recall and F1-score assigned to the convolutional layers. VGG16 model comprises
B. Image Recognition Feature [10] 13 convolutional layers, each of which has a fixed filter
size of 3x3 and a stride of 1 pixel. The feature map, which
The chatbot has been added with one more functionality captures different aspects of the input image, is generated as
that allows us to classify the photos that the user uploads. For the output of each convolutional layer.
this chatbot function, it was decided to use convolutional
neural network models. Convolutional Neural Networks [7] Pooling layers are implemented to diminish the spatial
(CNNs) are designed for image processing tasks, specifically dimensionality of the feature maps that are generated by the
for object detection, image classification, and segmentation convolutional layers. The Max pooling technique is applied in
[8]. Some models of convolutional neural networks that are VGG16, where a fixed-sized pool of values is selected from
most popular today are AlexNet [9], ResNet, VGG, Inception, the feature map, and the maximum value is chosen. This helps
MobileNet, and DenseNet.chatbot. to prevent overfitting and reduce the number of parameters in
the model.
Initially, it was planned to use a model that we will train
ourselves, based on the Food11 dataset. Food11 is a dataset Fully connected layers are used to transform the extracted
that contains photos of 11 different types of food, namely: features into class scores. VGG16 uses three fully connected
Bread, Dairy products, Dessert, Egg, Fried food, Meat, layers, each of which has 4096 units. The softmax activation
Noodles-Pasta, Rice, Seafood, Soup, and Vegetable-Fruit. function is applied to the output of the last fully connected
The number of images in training data was 9866, and 3430 layer, which transforms the raw class scores into a probability
images for testing data. For this, 8 different architectures were distribution over the possible classes.
taken, VGG16, VGG19, ResNet50, Resnet101, Resnet152,
- 104 -
The ReLU function is the activation function used in effectiveness, as well as its usefulness for the community of
VGG16. It is applied after each convolutional and fully the entire Discord platform.
connected layer. ReLU is a straightforward non-linear Due to the current pace of development in the field of
function that introduces non-linearity to the model and assists artificial intelligence in general, we hope that in the future
in better fitting the data. Each element of the input tensor is chatbots will be used even more actively and will gradually
simply subjected to the function (1) be introduced into all informational, entertainment, and other
f(x) = max(0,x). (1) websites.
Where x – is the input. REFERENCES
It returns x if it is equal or greater than 0, and 0 if input is [1] G. Caldarini, S. Jaf, and K. McGarry, “A Literature
less. Survey of Recent Advances in Chatbots,” 2022.
[2] D. Kyzyrov, “Source Code of the ChatbotD1 (Discord
In summary, VGG16 implements convolutional layers to
Chatbot using Python),” GitHub Repository, 2023. [Online].
extract features from the input image, pooling layers to reduce
the spatial dimensionality of the feature maps, fully connected Available: https://github.com/Kyzyrov
layers to map the features to class scores, and ReLU activation [3] A. R. Ghias, C. Guo, R. Sarikaya, and P. Ponnusamy,
function to introduce non-linearity to the model. “Amazon Alexa, Feedback-Based Self-Learning in Large-
Scale Conversational AI Agents,” Amazon, 2019. [Online].
CONCLUSION Available: https://www.amazon.com.
In conclusion, we can say that chatbots are tools that [4] A. Pal, S. Bhalotia, V. Rathod, S. Bisen, and T. Lalwani,
significantly change the entire system of the implemented “Implementation of a Chatbot System using AI and NLP,” K.
area. Areas such as medicine, sports, entertainment, J. Somaiya Institute of Engineering and Information
education, or the economy. They significantly improve the Technology, 2018.
quality of user service and help to find the information they [5] B. Lutkevich, “Application programming interface
need very quickly, without wasting their nerves, money, and (API),” TechTarget, 2022. [Online]. Available:
time. https://searchapparchitecture.techtarget.com/definition/appli
A functioning chatbot was developed for the Discord cation-programming-interface-API.
platform, the main feature of which is the ability to conduct a [6] R. Regin, S. Suman Rajest, Shynu T, Jerusha Angelene
dialogue with the user. The basis for such a chatbot was a Christabel G, and Steffi R., “An Automated Conversation
neural network model with an SGD optimizer for conducting System Using Natural Language Processing (NLP) Chatbot
dialogues with users, and also a convolutional neural network in Python,” in Central Asian Journal of Medical and Natural
model InceptionV3 for its image recognition function. Sciences, vol. 03, no. 04, pp. 1-5, Jul-Aug 2022.
By combining different neural network models, the [7] G. Tsakiris, C. Papadopoulos, G. Patrikalos, K.-F.
chatbot has become multifunctional. In addition, the chatbot Kollias, N. Asimopoulos, and G. F. Fragulis, “The
has some other features that allow you to use the capabilities development of a chatbot using Convolutional Neural
of the Discord platform more effectively. Networks,” Department of Electrical and Computer
In the future, it is also planned to work with other APIs Engineering, University of Western Macedonia, Kozani,
except for YouTube and TMDb, adding several different Greece, 2022.
functions, as well as further expanding the database, which [8] R. Yamashita, M. Nishio, R. K. G. Do, and K. Togashi,
will allow the neural network model of the chatbot to conduct “Convolutional neural networks: an overview and application
a dialogue even more efficiently. in radiology,” in Insights into Imaging, vol. 9, no. 4, pp. 611-
We found out that chatbots have many different 629, 2018.
categories, can communicate in different ways, and can also [9] L. Alzubaidi, J. Zhang, A. J. Humaidi, A. Al‑Dujaili, Y.
be used to perform different purposes. Duan, O. Al‑Shamma, J. Santamaría, M. A. Fadhel, M.
This paper can later grow into a project, to improve the Al‑Amidie, and L. Farhan, “Review of deep learning:
chatbot of which not a simple dataset of 500 lines will be concepts, CNN architectures, challenges, applications, future
used, but a dataset of various dialogues, communications, directions,” in Journal of Ambient Intelligence and
various questions and answers from call centers, customer Humanized Computing, vol. 12, no. 4, pp. 4021-4054, 2021.
support centers, and so on. [10] Tao, J., Gu, Y., Sun, J., Bie, Y., & Wang, H. (2021).
There are many ways to develop a chatbot, and Research on VGG16 Convolutional Neural Network Feature
accordingly, an unlimited number of different functions can Classification Algorithm Based on Transfer Learning, MDPI,
be implemented in a chatbot. The most important criterion for Advanced Machine Learning and Deep Learning Approaches
the development of our chatbot in the future will be its for Remote Sensing.
- 105 -
A Study on the Effective Multimodal Approach
for Anomaly Detection using Gramian Angular
Difference Field from Heating, Ventilation, and
Air Conditioning (HVAC) System of Railways
Beomjun Kim 1, Younghoon Jeon 2, Jeong Won Kang 3, Chulsu Kim 4, Oh Kuen Kwon 5*
and Jeonghwan Gwak 1,2*
1
Dept. of IT·Energy Convergence, Korea National University of Transportation, Chungju 27469, Korea
2
Dept. of Software, Korea National University of Transportation, Chungju 27469, Korea
3
Dept. of Transportation System Engineering, Korea National University of Transportation, Uiwang 16106, Korea
4
Dept. of Railroad Engineering, Korea National University of Transportation, Uiwang 16106, Korea
5
Dept. Electronic Engineering, Semyung University, Jecheon 27136, Korea.
Email: {zhskfkseh, 0hoon.jeon}@gmail.com, {jwkang, chalskim}@ut.ac.kr, [email protected], [email protected]
Abstract—Heating, Ventilation, and Air Conditioning For instance, a hybrid model-based approach (HyMA) [2]
(HVAC) systems maintain the comfort levels within buildings was developed to predict the system's prognosis and health
or vehicles, thereby directly impacting service quality, management (PHM). This approach was designed to
including customer satisfaction. Consequently, anomalies overcome the limited representation of HVAC datasets for
within the HVAC system are synonymous with service quality predictive model development by employing physics-based
degradation, underscoring the growing necessity for efficient models and CNN-based deep learning algorithms. This
anomaly detection methods. While recurrent deep learning approach, among studies for anomaly detection, was aimed at
models using time-series data from HVAC systems have been overcoming the complexities of such systems.
proposed, the effectiveness of multimodal deep learning Further research conducted for energy efficiency
model learning has been demonstrated across diverse sectors. prediction and outlier detection used LSTM and three
HVAC systems, which already harvest various sensor data, regression models: LASSO, SVR, and MLP [3]. This study
can benefit from performance enhancement through image- highlighted the potential applicability of this deep learning-
type multimodal data generated using the Gramian Angular based approach to real-time state monitoring. Given the
Difference Field (GADF) algorithm. This multimodal multitude of proposed approaches for anomaly detection in
approach retains time dependencies due to the GADF's nature, HVAC systems, it's clear that the demand for, and significance
thereby bolstering the overall reliability and safety of railway of, maintenance for anomaly detection is on the rise. To
systems. enhance the performance of anomaly detection, we have
converted time series data into multimodal data using the
Keywords—Anomaly Detection, GADF, HVAC System of Gramian Angular Difference Field (GADF) algorithm [5, 6].
Railways, Multichannel Image, Multimodal We have also compared and analyzed the results based on the
learning outcomes of existing CNN models[4] and the Vision
I. INTRODUCTION Transformer (ViT) [7].
An HVAC system refers to the heating, ventilation, and air
II. PROPOSED METHOD
conditioning infrastructure that maintains a comfortable
interior environment within a building or vehicle. Its primary A. General Overview
role involves adjusting the temperature, humidity, airflow, and Our proposed methodology synthesizes Gramian Angular
cleanliness -- the four major elements of air conditioning -- to Fields from Time Series data through the Gramian Angular
optimize the surroundings according to their intended purpose. Difference Field (GADF) algorithm. The initial stage creates
However, improper HVAC system operation can result in a set of proportional Gramian Angular Fields using the GADF
increased energy consumption, substandard indoor air quality, algorithm from the provided time series data. Subsequently,
and general discomfort. This can directly correlate to the synthesized set is merged into channel units and
decreased customer satisfaction. reconstructed as multimodal data in the form of a multichannel
According to maintenance strategy literature [1], service image. The ultimate step applies this multimodal data for the
and quality maintenance can largely be classified into three learning of the experimental models, with subsequent
categories: failure-based, preventive, and predictive performance comparison and analysis of each model. For a
maintenance. Each method has its drawbacks, including issues systematic comparative analysis, we first established the time
related to resources, lack of experts, and risk burdens. To series classification performance of existing recurrent-based
address these issues, there is ongoing research into HVAC models [8]. The proposed method's workflow is depicted in
system maintenance methodologies, monitoring devices using Fig. 1, with the rest of this section detailing our proposed
sensors, and abnormality detection through artificial neural model.
networks.
- 106 -
Fig. 1. Overall Architecture of Our Proposed Method with ViT
B. Gramian Angular Difference Field Algorithm III. EXPERIMENTS AND ANALYSIS
The Gramian Angular Field (GAF) represents a method A. Dataset
for generating images that showcase temporal correlations
The experiment employs a dataset collected from real-
based on polar coordinates. As the proposed model requires
world carriages of the railway using sensor devices, including
an image with time dependencies for measuring the normal
temperature sensors. The dataset includes values from
and abnormal probabilities of patch units, GAF is a crucial
temperature sensors, temperature average, an HVAC mode,
technique in the preprocessing phase. This algorithm
and an HVAC setting temperature value, as detailed in Table
normalizes the input time series data and visualizes the matrix
1.
via a transformation based on polar coordinates. There are two
methods of polar coordinate deformation: Gramian Angular TABLE 1. INPUT DATA DETAILS FOR THE PROPOSED MODEL
Summation Field (GASF), which transforms the sum of
angles into cosine, and the Gramian Angular Difference Field Component Value Noted
(GADF), which changes the difference of angles into sine. Our The temperature measured by
Temperature Sensor 1 Float
approach employs GADF. the first temperature sensor.
The temperature measured by
Temperature Sensor 2 Float
∅𝑖 = 𝑎𝑐𝑟𝑜𝑠𝑠(𝑁(𝑥𝑖 )), − 1 ≤ 𝑁𝑜𝑟𝑚(𝑥𝑖 ) ≤ 1 the second temperature sensor.
{ 𝑡𝑖 () Temperature
Float
The average temperature
𝑟𝑖 = , 𝑡𝑖 ∈ 𝑁 Average measured for each sensor.
𝑁
Setting Temperature The temperature set by the
Float
on HVAC HVAC system.
𝐺𝐴𝐷𝐹 = √𝐼 − 𝑥̃
2′ ∙ 𝑥
̃ − 𝑥̃ ̃2
′ ∙ √𝐼 − 𝑥
HVAC Operation Indicates the operation status of
[0, 32, 64]
Mode the HVAC air conditioner.
𝑠𝑖𝑛(∅1 − ∅1 ) ⋯ 𝑠𝑖𝑛(∅1 − ∅𝑛 )
𝑠𝑖𝑛(∅2 − ∅1 ) 𝑠𝑖𝑛(∅2 − ∅𝑛 )
= [ ⋱ ] () The temperature sensor and average temperature values
⋮ ⋮
𝑠𝑖𝑛(∅𝑛 − ∅1 ) ⋯ 𝑠𝑖𝑛(∅𝑛 − ∅𝑛 ) are measured within a real range, while the Setting
Temperature on HVAC represents the temperature set by the
The normalization from -1 to 1 or 0 to 1, expressed in manager. The HVAC Operation Mode signifies the operating
equation (1), accommodates data values that may be overly status of the air conditioner linked to the HVAC system, coded
large or small, with Min-Max Scaling commonly applied. as 0 for stopped, 32 for automatic semi-cooling operation, and
Each normalized signal at distinct time points is converted into 64 for automatic pre-cooling operation. The experiment
polar coordinates using equation (1). Using equation (2), it is utilized a total of 247,800 time series data with these five
possible to create a matrix reflecting temporal correlations. characteristics, and the data distribution is shown in Table 2.
The data generated through GAF primarily maintains time
dependencies, which increase as we move from the top left to TABLE 2. DATASET DISTRIBUTION
the bottom right of the matrix.
Dataset Sampling Number of
Dataset Split
C. Generating Multichannel Images using GADF Type Length Samples
Train: 70%
As GAF is unable to convert multivariate time series Normal 36 123,900 Validation: 15%
collectively, we need an approach that compiles GAF outputs Test: 15%
into singular data to handle multiple-feature HVAC system Train: 70%
data. Therefore, we adopted a method that generates GAF for Abnormal 36 123,900 Validation: 15%
each feature and assembles it into a multichannel image, as Test: 15%
shown in the first step of Fig. 1.
- 107 -
B. Experimental Settings negatives (FN), the Vision Transformer outperformed Bi-
This experiment evaluates the performance of our GRU due to its ability to extract additional spatial features.
multimodal generation technique when trained with existing
CNN models and the Vision Transformer. We then performed
a comparative analysis using the accuracy, F1 Score, Precision,
Recall, and CrossEntropy Loss of the recurrent-based models.
The recurrent-based models chosen for the experiment include
RNN, LSTM, GRU, Bi-LSTM, and Bi-GRU. We also
compared these with Vision Transformer (ViT) and CNN
models, including AlexNet, ResNet50, ResNext50, VGG16,
DenseNet121, InceptionV3, MobileNet, MnasNet,
EfficientNet, and RegNet. For the Vision Transformer, we
used a model structure incorporating a ViT encoder and an
MLP head. Fig. 2. Confusion Matrix of Bi-GRU and ViT
C. Experiment Results
The experiment evaluates the performance of our
proposed multimodal generation technique when trained with IV. CONCLUSIONS AND FUTURE WORK
existing CNN models and Vision Transformer. Details of the Our proposed method leverages Gradient Angular
experiment are available in the second step of Fig. 2. We used Dependency Fields (GADF) to transform time-series data into
each model's accuracy, F1 Score, Precision, Recall, and multimodal data, which is then processed through
CrossEntropy Loss for performance comparison. Table 3 Convolutional Neural Networks (CNN) and Vision
outlines the experimental results for each model. Transformers (ViT). This approach yielded higher
performance results than traditional recurrent-based models,
TABLE 3. EXPERIMENT RESULTS
establishing its suitability for the upkeep of HVAC systems
Model Accuracy
F1
Precision Recall Loss
within railway environments. However, data generated via
Score GADF is essentially a vector reconstructed from the
RNN 53.62% 0.5345 0.5365 0.5360 0.6857 differences in values within a specific time frame.
LSTM 77.68% 0.7722 0.8022 0.7769 0.4032 Consequently, it lacks information regarding the actual values
GRU 87.79% 0.8775 0.8826 0.8779 0.3072
Bi-LSTM 77.13% 0.7681 0.7862 0.7712 0.4957
measured by the sensor. This could potentially lead to
Bi-GRU 91.11% 0.9110 0.9125 0.9110 0.2987 misjudgements, such as deeming all temperature readings
AlexNet 69.25% 0.6854 0.7128 0.6928 0.6422 within a particular time frame as normal, even when abnormal
ResNet50 86.52% 0.8628 0.8914 0.8651 0.3332 values occur in a pattern similar to a consistent normal one.
ResNext50 88.82% 0.8875 0.8972 0.8881 0.2095 Such misinterpretations could have serious implications for
VGG16 81.76% 0.8113 0.8659 0.8175 0.4232 maintenance procedures. To mitigate this, our future work will
DenseNet121 93.67% 0.9367 0.9375 0.9367 0.1759
Inception V3 76.01% 0.7517 0.8003 0.7600 3.4143
focus on the design and experimentation of a deep learning
MobileNet 78.06% 0.7780 0.7935 0.7805 0.6511 model that can utilize both GADF-based multimodal data and
MnasNet 50.00% 0.3333 0.2500 0.5000 0.6939 time-series data gathered from sensors simultaneously.
EfficientNet 94.75% 0.9473 0.9502 0.9474 0.1928
RegNet 91.60% 0.9160 0.9160 0.9160 0.1783 ACKNOWLEDGMENT
ViT 98.92% 0.9896 0.9896 0.9896 0.0476
This work was supported by the Basic Science Research
Recurrent
Average
77.46% 0.7726 0.7840 0.7746 0.4381 Program through the National Research Foundation of Korea
Proposed (NRF) funded by the Ministry of Education (Grant No. RS-
82.67% 0.8090 0.8185 0.8267 0.6329 2023-00248444)
Average
REFERENCES
When training the proposed method with CNNs and [1] J. Ulaganathan and K. M. Sadyojatha, "A Review On Maintenance
Vision Transformer, we observed an average accuracy of Techniques For Industrial Equipment And Its Machine Learning
82.67%. The Vision Transformer achieved an accuracy of Algorithms.", International Journal of Advanced Research in
98.92%, representing the highest performance among all Engineering and Technology (IJARET), Vol. 12, pp. 183-194, April
2021
models. Recurrent-based models showed an average accuracy
of 77.46%, with Bi-GRU recording the highest performance [2] A. Gálvez, D. Galar and D. Seneviratne, "A Hybrid Model-Based
Approach on Prognostics for Railway HVAC.", IEEE Access, Vol. 10,
at 91.11% accuracy. This suggests that our proposed method pp. 108117-108127, October 2022
can improve average performance by about 5%, [3] Y. Wang, C. Yang and W Shen, "A deep learning approach for heating
demonstrating the effectiveness of the GADF-based and cooling equipment monitoring.", 2019 IEEE 15th International
multimodal generation method for HVAC systems. Conference on Automation Science and Engineering (CASE) in
Fig. 2 presents the confusion matrix of Bi-GRU and Vision Vancouver, pp. 228-234, September 2019
Transformer, which showed the highest performance in their [4] A. Elngar, M. Arafa, A. Fathy, B. Moustafa, O. Mahmoud, M. Shaban
respective groups. The superior performance of the Vision and N. Fawzy, "Image Classification Based On CNN: A Survey",
Journal of Cybersecurity and Information Management, Vol. 6, No. 1,
Transformer is likely due to its positional encoding, which pp. 18-50, January 2021
shares similarities with the GADF's time-dependency feature, [5] G. Zhang, Y. Si, D. Wang, W. Yang and Y. Sun, "Automated detection
allowing it to consider both temporal and spatial of myocardial infarction using a gramian angular field and principal
characteristics. Although both Bi-GRU and Vision component analysis network.", IEEE Access, Vol. 7, pp. 171570-
Transformer exhibited higher false positives (FP) than false 171583, November 2019
- 108 -
[6] M. M. Islam and M. M. H. Shuvo, "DenseNet Based Speech Imagery
EEG Signal Classification using Gramian Angular Field.", 2019 5th
International Conference on Advances in Electrical Engineering
(ICAEE), pp. 149-154, September 2019
[7] A. Dosovitskiy, L, Beyer, A, Kolesnikov, D. Weissenborn, X. Zhai, T.
Unterthiner, M. Dehghani, M. Minderer, G. Heigold, S. Gelly, J.
Uszkoreit and N. Houlsby, "An Image is Worth 16x16 Words:
Transformers for Image Recognition at Scale", ICLR 2021, January
2021
[8] K. M. Tarwani and S. Edem, "Survey on Recurrent Neural Network in
Natural Language Processing.", International Journal of Engineering
Trends and Technology (IJETT), Vol. 48, No. 6, pp. 301-304, June
2017
- 109 -
Advancing Railway Track Segmentation:
Harnessing the Power of ResUNet and
Self-Supervised Learning Paradigms
Changjoon Park 1, Jeong Won Kang 2 and Jeonghwan Gwak 1,3,*
1
Dept. of IT·Energy Convergence, Korea National University of Transportation, Korea
2
Dept. of Transportation System Engineering, Korea National University of Transportation, Korea
3
Dept. of Software, Korea National University of Transportation, Korea
Email: [email protected], [email protected], [email protected]
Abstract—The paucity of labeled data specific to railway possess proprietary restrictions, preventing public
components presents a substantial obstacle in the cultivation and accessibility. Thus, scenarios that necessitate raw data
implementation of supervised learning models directed towards collection through rail inspection vehicles and manual
enhancing railway safety and operational efficiency. This study labeling are not only labor-intensive but also financially
introduces a self-supervised learning approach amalgamated burdensome [4]. Additionally, the specifications of fasteners,
with a ResUNet architecture to circumvent this constraint. The sleepers, and rails used to construct railway tracks can vary
proposed model leverages the potent feature extraction based on the type of train, country, and region. This variability
capabilities of ResNet50 within a UNet structure, striking a
introduces a significant diversity in component types across
delicate balance between contextual capture and spatial
jurisdictions, which may pose challenges when applying AI
localization. This approach significantly enhances the
segmentation of railway track components, which is pivotal to
models trained on publicly available datasets to real-world
railway safety. The developed model outperforms previous scenarios.
models in terms of identifying and segmenting rare types of rail
track components, thereby reinforcing the potential of self-
supervised learning strategies combined with ResUNet for fine-
grained image segmentation tasks in safety-critical domains. The
ramifications of this research extend beyond railway track
component segmentation and pave the way for addressing data
limitations in other railway safety analysis fields, contributing
significantly to the comprehensive safety and efficiency of
railway systems.
Keywords— Object Segmentation, Railway Track Components,
ResUNet, Self-Supervised Learning
I. INTRODUCTION
Defects in railway track components exacerbate the
potential for accidents. Integral elements of railway tracks,
such as fasteners, sleepers, and the track's organic connective
structure [1], play a critical role in maintaining stability and
ensuring reliable operations. These components underpin the Figure. 1 Architecture of ResNet50 and ResUNet
track's structure and inhibit derailment or shifts by providing
requisite support [1]. Damage or defect to any of these In this paper, we propose a technique that merges the
elements compromises the organic interconnection, leading to Residual UNet (ResUNet) [5] and Self-Supervised Learning
a discernible degradation in the support capability. This can (SSL) [6] to address these issues. Our method leverages the
result in track instability or deformation, which subsequently nuanced feature extraction capabilities of the ResUNet
increases the risk of accidents. Therefore, comprehensive architecture, which is significantly enhanced using the pre-
monitoring of railway track component defects is paramount trained weights of ResNet50, and couples it with the
for the seamless operation of trains. However, this task comprehensive understanding of overall features gained
presents several challenges. through SSL. ResUNet, an advanced model for railway track
component monitoring, substitutes the encoder of the
The extensive range covered by railway tracks makes it conventional UNet [7] model with a pre-trained ResNet50 [8]
infeasible to establish fixed monitoring structures and feature extraction layer. This substitution facilitates more
inefficient for human-operated manual surveillance. To comprehensive feature extraction and fine-grained
overcome these challenges, modern railway track monitoring segmentation. SSL is a learning strategy that delivers robust
has turned to artificial intelligence (AI) models to ensure performance even with datasets that contain a minimal
robust performance [2, 3]. AI models excel in performance quantity of labeled data. It augments the model's
enhancement when trained on systematically labeled datasets. understanding of limited data by conducting Pretext Task
However, benchmark datasets pertaining to railways, Learning, which trains overall data features using unlabeled
especially those concerning railway track components, are data, and subsequently applies the trained weights to the
scarce [4]. The data used in individual studies or papers often desired model in a downstream process. The dataset for the
- 110 -
proposed technique is constituted by acquiring images of permitting the reflection of both values for a more
railway tracks from relevant authorities in various comprehensive performance assessment.
jurisdictions and conducting small-scale manual labeling for |A∩B|
training purposes. Figure 1 delineates the architecture of both JS = ()
|A∪B|
ResNet50 and ResUNet.
2 × |A ∩ B|
II. PROPOSED METHOD DC = (2)
|A|+|B|
A. SSL Pretext Task using Rotation
III. EXPERIMENTAL RESULTS AND ANALYSIS
To implement SSL, we employed the Rotation technique
as the Pretext Task in the training phase [9]. The unlabeled Initially, we set out to ascertain a foundation for
dataset utilised in the experiment comprised a total of 43,655 contrasting the segmentation performance when SSL is
samples. Images depicting the symmetry of fasteners, sleepers, implemented, premised on the Supervised Learning (SL)
and rails were bisected based on the image's central axis and outcomes of the established UNet model and ResUNet. As
subsequently resized to dimensions of 1024×1024. The evidenced by TABLE I, which presents the Validation Scores,
dataset was apportioned into a ratio of 0.6 for training, 0.2 for the empirical data garnered from our experiments
validation, and 0.2 for testing. The rotation label was instituted substantiates that the optimal performance is achieved when
by randomly selecting an original image, rotating it among a the initial stage of the pretrained ResNet50 is held static or
set of {0°, 90°, 180°, 270°} immediately prior to model 'frozen'.
training, and thereafter assigning the corresponding value as Following this, we undertake a comparative analysis of the
the label. numerical performance and visual segmentation outcomes of
the best-performing model - Pretrained ResUNet (Freeze
Stage 0) - and the ResUNet with applied SSL. The numerical
performance is reported in TABLE II (Test Scores), and visual
segmentation outcomes are exhibited in Fig. 4.
TABLE I. COMPARISON OF THE UNET AND RESUNET MODELS
Fig. 2 Pretext Task Learning Process
Model Pretrained Freeze JS + DC Rank
For Pretext Task Learning, the pre-trained ResNet50 UNet X No Freeze 1.9291 7
model provided by PyTorch was used, with adjustments made ResUNet 1 X No Freeze 1.9385 5
to align the model with the number of Rotation Labels. This ResUNet 2 O No Freeze 1.9486 2
involved transforming the output dimensions from 1024 to
ResUNet 3 O Block 0 1.9545 1
512, followed by a further reduction to 4, enabling
ResUNet 4 O Block 0~1 1.9352 6
classification as illustrated in Fig. 2. Following these
adjustments, we commenced with the training phase. ResUNet 5 O Block 0~2 1.9398 4
ResUNet 6 O Block 0~3 1.9413 3
B. SSL Downstream Task using ResUNet ResUNet 7 O Block 0~4 1.9169 8
Upon the completion of the pretext task learning phase, the
pre-trained ResNet50 model was loaded, the classifier TABLE II. COMPARISON OF THE RESUNET AND SSL-RESUNET MODELS
removed, and the model was downstreamed into the encoder
Accuracy F1 Score JS DC
of ResUNet for subsequent training. The training utilized a
ResUNet 3 0.9864 0.9728 0.9499 0.9728
total of 1474 datasets, apportioned into training, validation,
and testing ratios of 0.7, 0.15, and 0.15, respectively. SSL-ResUNet 0.9871 0.9742 0.9533 0.9742
When comparing the performance of the two models,
although the SSL-ResUNet depicted a marginal improvement,
the numerical discrepancy was not statistically significant,
making a precise analysis challenging. Therefore, our analysis
will primarily concentrate on the visual results.
Fig. 3 Downstream Task Learning Process
Evaluation of the model's performance was conducted
using the Jaccard Score (JS), as defined in (1), and the Dice
Coefficient (DC), denoted as in (2). Early stopping was
employed with a patience value set to 7, thus halting the
model's training process if there was no improvement in the
sum of JS and DC. The rationale for using the sum of JS and
DC derived from their respective capabilities to evaluate the
similarity of segmentation results in unique ways, thus Fig. 4 Comparison of the SSL-ResUNet and ResUNet
- 111 -
Visual analysis results from the segmentation indicate that [5] Y. Zhang, Z. Chen, H. Yu, X. Yao and H. Li, "Feature Fusion for
the prediction outcomes for images of commonly occurring Segmentation and Classification of Skin Lesions," 2022 IEEE 19th
International Symposium on Biomedical Imaging (ISBI), pp. 1-5, 2022,
rail track components within the dataset (used for both SSL doi: 10.1109/ISBI52829.2022.9761474.
and SL) generally appeared similar. However, there was a [6] L. Jing and Y. Tian, "Self-Supervised Visual Feature Learning With
significant difference in the results for the rarely appearing Deep Neural Networks: A Survey," in IEEE Transactions on Pattern
components. The SSL-applied model noticeably had fewer Analysis and Machine Intelligence, vol. 43, no. 11, pp. 4037-4058, 2021,
misdetections compared to the model which showed the doi: 10.1109/TPAMI.2020.2992393.
highest performance in the previous analysis. The quantitative [7] O. Ronbneberger, P. Fischer and T. Brox, “U-Net: Convolutional
analysis provided an objective measure of the overall Networks for Biomedical Image Segmentation,” Medical Image
Computing and Computer-Assisted Intervention – MICCAI 2015, vol.
performance, while the qualitative visual analysis facilitated a 9351, pp. 234-241, 2015, doi:10.1007.978-3-319-24574-4_28
detailed review of the prediction outcomes for each class of [8] K. He, X. Zhang, S. Ren and J. Sun, "Deep Residual Learning for Image
rail track components. These combined analyses suggest that Recognition," 2016 IEEE Conference on Computer Vision and Pattern
the application of ResUNet with SSL may yield performance Recognition (CVPR), pp. 770-778, 2016, doi: 10.1109/CVPR.2016.90.
enhancements in monitoring railway track components. [9] X. Li, X. Hu, X. Qi, L. Yu, W. Zhao, P. A. Heng and L. Xing "Rotation-
Oriented Collaborative Self-Supervised Learning for Retinal Disease
IV. CONCLUSION AND FUTURE WORK Diagnosis," in IEEE Transactions on Medical Imaging, vol. 40, no. 9,
pp. 2284-2294, 2021, doi: 10.1109/TMI.2021.3075244.
This study introduces a novel methodology for enhancing
safety and efficiency in railway operations. By substituting the
encoder of the UNet model with the powerful feature
extraction capabilities of ResNet50 and leveraging the feature
learning ability of SSL for railway track component
segmentation, our technique has shown an ability to
effectively circumvent limitations such as scarcity of labeled
datasets, variability in railway track components across
regions, and high costs associated with dataset labeling. When
juxtaposed with the best-performing model trained on an
identical labeled dataset without SSL implementation, our
technique significantly mitigates the rate of misdetections.
While our research has predominantly been focused on the
segmentation of railway track components, the implications
suggest that our technique can contribute considerably
towards overall improvements in railway safety and efficiency.
Though the present application does not directly extend to the
analysis of other railway components, there exists a promising
prospect for such expansion in future investigations. Future
research will delve deeper into other Pretext Task Learning
methods to further optimize the SSL process and refine the
segmentation process for railway track components. By doing
so, we strive to continue advancing the application of AI in the
realm of railway safety and maintenance, pushing towards a
future characterized by safer and more efficient railway
operations.
ACKNOWLEDGMENT
This work was supported by the Basic Science Research
Program through the National Research Foundation of Korea
(NRF) funded by the Ministry of Education (Grant No. RS-
2023-00248444)
REFERENCES
[1] S. Bhardawaj, R. C. Sharma and S. K. Sharma, “A Survey of Railweay
Track Modelling,” International Journal of Vehicle Structures and
Systems, vol. 11, no. 5, 2019, doi: 10.4273/ijvss.11.5.08
[2] B. Kim, Y. Jeon, J. Kang and J. Gwak, “Multi-task Transfer Learning
Facilitated by Segmentation and Denoising for Anomaly Detection of
Rail Fasteners,” Journal of Electrical Engineering & Technology, vol.
18, pp. 2383-2394, 2022, doi: 10.1007/s42835-022-01347-1
[3] I. Kim, Y. Jeon, J. Kang and J. Gwak, “RAG-PaDiM: Residual Attention
Guided PaDiM for Defects Segmentation in Railway Tracks,” Journal
of Electrical Engineering & Technology, vol. 18, pp. 1429-1438, 2022,
doi: 10.1007/s42835-022-01346-2
[4] M. J. Pappaterra, F. Flammini, V. Vittorini and N. Besinovic, “A
Systematic Review of Architecture Intelligence Public Datasets for
Railway Applications,” Infrastructures, vol. 6, no. 10, 2021, doi:
10.3390/infrastructures6100136
- 112 -
August 22(TUE) - 25(FRI), 2023
Oral Session 4
14:00~15:40 Wed, August 23, 2023
Room : A435
Chair: Seungkwon Shin (Korea Railroad Research Institute, Korea)
- 113 -
Calculating FMEA variables of software failure
risks based on journals and metrics
Denis Valyayev Bolshibayeva Aigerim Dinargul Mukhammejanova
MSc student of Almaty University of PhD, Assistant Professor of the
Power Engineering and Department Computer Engineering and Almaty University of Power
Telecommunications Information Security (CSSE) Engineering and Telecommunications
Almaty, Kazakhstan [email protected] International IT University Almaty, Kazakhstan
orcid.org/0009-0007-9619- Almaty, Kazakhstan [email protected]
1743 [email protected] orcid.org/0000-0002-2693-4130
orcid.org/0000-0003-1191-4249
Assel Mukasheva
PhD, Associate Professor of the
Department of Information Systems and
Cybersecurity
Almaty University of Power Engineering
and Telecommunications
Almaty, Kazakhstan
[email protected] orcid.org/0000-
0001-9890-4910
Abstract—The article describes the joint application of I. INTRODUCTION
some text comparison models, time series analysis and risk
analysis. The technological implementation of these models is
In today's software industry, organizations spend many
based on Data Mining methods. The above models have been resources on dealing with software failures: identifying
studied in depth and have been used for a long time, but their failures, assessing the risks of failures, and analyzing the
joint use for risk assessment raises many questions about their consequences of failures. Software failures can have
capabilities and applicability. In this study, the authors apply catastrophic consequences, both huge financial losses and the
text comparison, time series anomaly detection, and Failure threat to human life, so there is a need for failure prevention
Mode and Effects Analysis (FMEA) models to calculate 3 [1].
software failure risk scores. The purpose of the experiment is
Implementing a software failure risk management
to determine the joint applicability of these models for
assessing the risks of critical software errors, where as a result process at the design stage can help in determining what
an experimental error will be revealed. This error will be improvements are needed to make the final product more
associated with outliers in the time series and an FMEA table reliable and safety [2].
entry for this error will be provided. The result of the The complexity of software failure risk assessment is that
experiment will make it possible to draw conclusions about the it requires interdisciplinary competence and an
possibility of automating the calculation of software failure understanding of software behaviour in response to failures
risk parameters based on data on the application execution
occurring within the system and from the external
using the considered models. The study is based on a data set
that contains: a list of 24-hour application execution log
environment [3]. Software failures are characterized by the
records with timestamps, severity levels and messages obtained possibility of unpredictable effects and the difficulty of
from the Loki log database, a time series for the same period identifying them. Although software behavior is
extracted from the Prometheus time series database. To create deterministic, however a set of factors is difficult to predict.
an initial dataset the test application has been developed which At the same time, the failure of a single element can generate
simulates RAM spikes with an error level entry added to the a cascading failure [4].
log. The experiment was performed using Kotlin and R
There are several types of risk factors, which are
languages. The study used Apache Commons Text package to
distinguished by their properties. This types are certain,
perform text comparison. Time series is passed to the R
runtime using the forecast library, a model is formed to detect
uncertain, dependent, independent [5].
anomalies and predict “normal” values at outlier points. Thus, Following from the practice of many organizations,
a list of points on the time series with anomalous values is analysis and assessment of software failure risks is usually
obtained and they are associated with log records. As a result carried out using subjective assessment methods based on the
of the experiment, a time series graph was obtained with the accumulated experience of a particular professional. This
identified anomalous values and with “normal” values at these approach has a number of drawbacks, such as the different
points. The end result is presented as an FMEA table with
subjective experience of different staff and the difficulty in
calculated risk scores for experimental error. Based on the
results of the experiment, conclusions were drawn about the
transferring this experience between employees [6].
applicability of the considered models. So methods for objective risk analysis and assessment are
relevant to many organizations. The results of this study can
Keywords— Risk assesment, forecast, Time Series, Data help companies to develop models for risk assessment and
Mining, Text Comparsion contribute to further research in this area.
- 114 -
II. BACKGROUND [12]. Many large companies develop their own tools for
anomaly detection based on different algorithms. Microsoft
A. Failure Modes and Effects Analysis has developed its own platform that receives real-time data
One of the most commonly used risk analysis models is through Apache Kafka and analyzes time series based on
Failure Modes and Effects Analysis (FMEA). In its Spectral Residual and Convolutional Neural Network model
traditional form FMEA consists of risk ranking using risk [13]. Most often, anomalies are detected using a predictive
priority number (RPN) based on occurrence (O), detection model, which is trained on the entire time series and
(D) and severity (S) (Fig. 1). If we refer to the Automotive constructs a prediction for the same time period. The
Industry Action Group (AIAG) and the German Association predicted and real values of the series are compared [14].
of the Automotive Industry (VDA), it is possible to make the Purely statistical models are also used to detect anomalies,
risk management system more efficient by using an Action like Interquartile Range, for example for early detection of
Priority (AP) with 3 grades instead of RPN: Low (L), DDoS attacks [15], or for satellite clock corrections [16].
Medium (M) and High (H) [7]. FMEA has difficulties in And an article [17] uses time-series analysis to identify
defining all 3 parameters, and the formula for calculating failures of sensor equipment.
RPN itself is quite controversial [8].
III. RELATED WORKS
A lot of research has been done in the field of software
error and failure detection. The paper [18] describes a
method for clustering failures in the OpenStack cloud
platform based on machine learning models. Using Integer
Linear Programming, namely a fast heuristic algorithm, as
well as failure analysis, they were able to achieve a nearly 2-
fold improvement in Mean Time To Failure [19]. In [20], a
comparative analysis of the application of statistical models
and various machine learning models such as Bidirectional
Long Short-Term Memory, Long Short-Term Memory,
Figure 1. A structure of FMEA model Recurrent Neural Network, for predicting workload failures
in cloud systems is given.
Risk analysis using FMEA is usually built on the
following steps: defining the subject area of the model, There are works on software failure prediction, such as
setting up a risk analysis team, identifying possible failure time series based failure prediction based on bug report data
modes, defining values for variables, defining recommended [21]. And study [22] analyses risk assesment methods based
actions. The FMEA model is actively used in the planning, on a predictive model for identifying defects in software.
design and development stages of software [8].
IV. MATERIALS AND METHODS
B. Text comparison
A. Data collection.
Text comparison models are based on determining the
distance between two texts. These methods are used when it In a typical monitoring and observability setup, Loki and
is necessary to determine how similar two texts are to each Prometheus are commonly used together to collect and
other. analyze logs and metrics, respectively. Loki is a horizontally-
scalable, highly-available, and multi-tenant log aggregation
In this study, text comparison models are applied to system. Prometheus is a popular open-source monitoring and
identify similar errors in journal entries. The Jaro-Winkler alerting toolkit. It collects and stores time-series data, often
similarity model has been chosen, which has been referred to as metrics, from various sources such as
extensively applied in various tasks but also has drawbacks applications, services, and infrastructure components.
in the form of high resource consumption [9].
To extract logs from Loki, you typically need to interact
C. Anomalies detection in time series with its API. The API provides endpoints for querying log
Time series analysis is a complex task, as it involves data based on various filters such as time range, labels, and
identifying and predicting multiple properties of the series, log patterns. Extracting metrics from Prometheus involves
such as trend and seasonality. Trend can be identified using querying its HTTP API or using Prometheus client libraries
regression analysis [10] or Fridman’s super smuother to retrieve metric data. The Prometheus API provides
algorithm (supsmu). In modern systems, the performance of endpoints for querying metrics based on labels, time range,
system services is continuously monitored to determine and aggregation functions.
normal and abnormal behaviour. This kind of time series can Application execution logs are extracted from the Loki
include not only trends, but can also be multi-seasonal with database and metrics from the Prometheus database and
complex characteristics [11]. Models such as supsmu and undergo initial processing before being uploaded to the data
STL decomposition allow the characteristics of the time warehouse. The data analysis algorithms then access the
series to be determined and anomalous values to be identified warehouse and perform processing. (Fig. 2)
from them.
Application performance metrics are time series in which
certain anomalies can be identified. The taxonomy of time
series anomalies includes anomaly classification: point,
context and collective anomalies, as well as many data types:
scalar, vector, matrix, tensor, attributes, fields and properties
- 115 -
However, the seasonal component is optional and can
store different patterns. To determine the trend and
seasonality of the data, the MSTL method for sequential
measurement of the seasonal components of the time series
described in [23] is used.
After determining the components of the time series, the
strength of contribution of the seasonal component is
determined as follows:
var(yt−Tt −St)
F= 1 − var(yt −Tt )
(4)
If F is above 0.6, then the seasonal component is
subtracted from the time series (y*t). Then, for the new series
the trend is calculated using Friedman's super smoother
Figure 2. Process of extracting data and prepare it for analysis (supsmu) method and the remainder is calculated:
B. Identifying the same errors Rt = y*t − Tt (5)
The error data is a time-stamped text entry. Each log The first and third quartiles and the IQR (interquartile
entry appears whenever an error occurs, but may indicate the range) are calculated for the resulting component:
same problem. Accordingly, it is necessary to identify unique
errors in the execution log. IQR = Q3 − Q1 (6)
For the purpose of identification it was accepted that if Finally, all points in the time series are compared with
the error texts match by more than 80% then it is considered the extreme points of the time series adequacy interval. If a
that the record identifies the same error. With the message point does not fall within the interval, then it is considered
itself being more important than the rest of the record, abnormal. The interval is as follows:
weighting factors of 0.7 and 0.3 were added to the similarity Q1 − 3 ∗ IQR > xi < Q3 + 3 ∗ IQR (7)
of the message and the entire record accordingly. The final
formula for determining similarity is as follows: Thus only values that lag well behind the seasonally
constant component of the time series will be defined as
0.7kmsg + 0.3krec > 0.8 (1) anomalous.
The coefficient itself is calculated using the Jaro-Winkler D. Calculating FMEA variables
similarity formula:
The very first step is to identify the failure modes. In this
1 �
k=3 ∗( � +
�
+
�−�
), (2) study, the failure mode is equal to the unique error that was
1 �2 �
identified using the text comparison method.
where m is the number of matching characters in order, s1 To calculate the number of occurrences over a certain
and s2 are string lengths, and t is the number of transpositions: period of time, the number of occurrences is counted. In this
matching characters in different order, divided by 2. study, a value of 48 was taken as the maximum number of
Due to the unique errors detected, the processing occurrences of an error in 24 hours. In other words, it is
algorithm assigns an identifier to the error, calculates the assumed that an error cannot occur more than once every
number of occurrences, the average similarity coefficient for half hour on average. This makes it possible to normalize the
all detected records and stores the time stamps of the error occurrence to an FMEA value between 0 and 10:
occurrences. (Fig. 3) �
O = min(1, 48
∗ 10) (8)
To calculate detection, the average of the similarity
coefficients of the log entries that have been associated with
the current error is taken:
��
D = min(1, (1 − �
) ∗ 10) (9)
The results of anomaly detection in the time series are
used to obtain severity. If there are anomalies in the time
series within a 10 minutes (5 minutes before and after) of
the logged data points, the distance of the predicted normal
values and the actual values are calculated for these points
and normalized to values between 0 and 10.
Figure 3. Association of log entries with an error
In order to calculate this, a clean time series (y_clean)
C. Anomalies detection must be taken as the original time series in which all
To identify anomalies, the time series is broken down abnormal values have been replaced by the predicted normal
into trend, seasonal and remainder components: values within the IQR. The difference between the abnormal
values and the predicted normal values (diff) is then taken,
yt = Tt + St + Rt (3)
- 116 -
with the difference between the other points taken as 0. This After extracting the test application's logs, it was found
calculates the minimum and maximum differences: that 5 error log entries were generated in a 24-hour period.
Using a text-matching algorithm, all entries were associated
max_diff = max(|yt(ti) − y_clean(ti)|); min_diff = min(|yt(ti) with a single bug and given their similarity coefficients to the
− y_clean(ti)|) (10) first occurrence of that error. The FMEA algorithms
Thus severity (S) is calculated as the average of the determined the Occurrence parameter to be 1 and the
normalized distances at anomalous points within a minute: Detection parameter to be 1, as the average similarity
coefficient is 0.992.
|yt(ti) − y_clean(ti)|−min_diff
S=avg( ) ∗ 10 (11)
���_����−���_���� TABLE I. ASSOCIATED LOG ENTRIES WITH SIMILARITY COEFFICIENT
Risk Priority Number (RPN) is calculated as the product # Timestamp Reference Similarity
of the track parameters: 1 2023-06-14T01:30:41.824Z - -
RPN = O ∗ D ∗ S (12) 2 2023-06-14T07:04:52.772Z 1 0.996984
Thus, the final risk rating is made up of all the parameters 3 2023-06-14T14:15:02.902Z 1 0.991499
considered in this study. (Fig. 4)
4 2023-06-14T16:25:11.101Z 1 0.990284
5 2023-06-14T18:36:33.037Z 1 0.991002
B. Results of time series analysis
The time series analysis has been carried out using the R
language and the "forecast" library. This library contains a
function "tsoutliers" that analyses the time series using
MSTL and supsmu algorithms and returns a vector of items
with indexes of the elements of the original time series
where anomalies were detected and the values that are
Figure 4. Contribution of parameters to the final risk rating predicted at those points as normal. Thus, obtaining the
anomalous points and the predicted values is depicted in the
V. IMPLEMENTATION OF THE MODELS AND RESULTS following code snippet. (Fig.5)
The raw dataset was generated on Droplet on the Digital
Ocean cloud platform with Ubuntu 22.04 operating system
and specifications: 4GiB RAM, 2 vCPUs, 4000GiB Transfer,
80GiB SSD. On this virtual machine were deployed: Docker,
Kind (Kubernetes). Within Kind, loki-stack and prometheus
were deployed using free community helm charts of the
latest versions at the time of the experiment.
A simple Java and Quarkus application has been
developed which, at random intervals between 10 minutes Figure 5. Code for obtaining anomalous points in R
and 10 hours, fills a random length list of 1320 character
strings, then outputs an error log to the standard output and The function "tsclean" is also used, which performs the
holds the list in memory for a few seconds. A pointer to this same analysis as "tsoutliers", but also replaces the anomalous
list is then removed and after a period of time the list will be values with predicted values, i.e. it returns a cleared time
removed from RAM using the Garbage Collector of Java series.
Virtual Machine (JVM). For the visualisation, ggplot2 was used. The ggplot2
A. Implementation and results of error identification package is a popular data visualization library in the R
programming language. It provides a powerful and flexible
Error identification as well as data extraction for them are framework for creating high-quality and customizable graphs
implemented in Kotlin, and the Jaro-Winkler similarity and plots. A geom point was used for the time series and a
factor is computed using the Apache Commons Text library. manual colour definition was added to display the legend.
The algorithm takes the first item as the current item and In the original dataset, the timestamp is represented by a
removes it from the list, then runs through the entire list of string and the time series values bytes. For ease of display, a
entries, starting with the second item and comparing it with converted date was added to the dataset using "POSIXct" and
the current item. If the similarity coefficient is greater than the values converted to mebibytes. Time stamps for the
0.8, then a reference to the current item is added to the record, occurrence of the entries in the runtime logs were also
the record is removed from the list and the coefficient value transferred to R. Two graphs were merged for visualization:
for it is stored. When all the elements in the list have been a time series graph with raw and cleaned values. (Fig 6)
traversed and there are still elements in the list, then there are
still unique errors and the algorithm starts again. The result is
an associated list, where the key is the first hit unique entry,
and the value is a list of associated entries.
- 117 -
TABLE II. FINAL FMEA RECORD OF OBTAINED ERROR
Logger Name Occurrence Detection Severity RPN
MEMORY_ERROR 1 1 3 3
loki-0 6 2 0 0
Table II shows that the system identified 2 risks:
MEMORY_ERROR is a test application error, and loki-0 is
a persistent warning loki level error when deployed in Kind.
As the error loki-0 is just a constant noise and doesn't have
Figure 5. Code for visualizing time series analysis in R any effect on memory consumption, the system assigns to it
Severity equal to zero and the final risk rating is also zero.
The time series values of the RAM consumption of the
test application over 24 hours, namely the same date as the The test error that was the main task was to detect it
logged data were unloaded as an initial dataset. This data was occurs very rarely and the message text is always the same,
transferred to R and written to the Data Frame "df". so it has Occurrence and Detection marks equal to 1 in the
risk. But as occurrence of this error in logs almost always
As an intermediate result there are 2 graphs, where on the coincides with a surge in memory consumption, the system
top of the timeline there are black areas, which denote the gave it a Severity rating of 3 and the final rating is also 3.
occurrence of an error and 10 minute frames before and after
the occurrence. In the lower graph there are 2 curves: the CONCLUSION
grey curve denotes the actual data, which was obtained from
Prometheus; the red curve denotes the predicted values. The As a result, it was found that for 24 hours the error that
actual values overlap the predicted values, so the red line is appears in the logs refers to the same error.
only visible at abnormal points in the graph. (Fig. 6) Also anomalies were detected in the obtained time series
of memory consumption by the test application. The
anomalies where sharp "upward" jumps in memory
consumption were found coincided with the time of the error
log entries.
As a result, the system was able to correctly identify the
unique error and calculate the FMEA parameters. And as an
unexpected result, the system was able to identify another
error, which is noise in the data set, and give it a zero rating.
In summary, it can be said that these methods can be
combined to assess software failure risks. This method can
identify the error and identify its severity, and discard the
noise in the execution logs. But it is important to note that
this model has been tested on a small amount of data, so the
model may give different results on larger time intervals. It
Figure 6. Comparison of logging error and RAM consumption time would probably be necessary to divide the entire time series
series graphs with anomalies detected into short sections and analyse them separately.
The graphs show that sharp jumps in RAM consumption The model is far from ideal and needs further
coincide with the occurrence of log errors. As you can see, development. One of the most important improvements is the
one of the log errors coincides with only a small uptick in calculation of Severity on the basis of all available time
consumption, so this spike was not defined as abnormal. series of the application. For example memory consumption,
Also between 17:00 and 22:00 we can see the cumulative CPU consumption, HTTP traffic, network latency and many
effect of two errors: the first error was triggered and the others. If applications in the system communicate through
application held a link to the list and at this time a second middleware, time series of middleware services can also be
error was triggered, which increased consumption even added. The key is to develop and test a model to combine
further. In both cases the graph shows anomalies. results from multiple time series analysis.
C. Final FMEA table calculation results This model can be useful for enterprises to build their
To calculate the final FMEA table, the calculated data own automatic software failure risk analysis system, where
from R was downloaded into Kotlin using the Rserve library monitoring is performed using the loki database and the
and the last Severity parameter for the identified error was Prometheus time-series database.
calculated. The research was conducted as part of project
Kotlin calculates the difference between the elements of AP19576825, 'Developing Methods and Algorithms for
the actual time series and the cleared time series and Machine Learning to Identify the Financing of Terrorist
calculates the highest and lowest values where the difference Activities in the Republic of Kazakhstan'.
is not zero. Then the final Severity is calculated using REFERENCES
formula (11) and the final Risk Priority Number is calculated
[1] W. Eric Wong, Xuelin Li, Philip A. Laplante, Be more familiar with
using formula (12). our enemies and pave the way forward: A review of the roles bugs
- 118 -
played in software failures, Journal of Systems and Software, Volume (eds) Advances in Information and Communication. FICC 2021.
133, 2017, Pages 68-94, ISSN 0164-1212, Advances in Intelligent Systems and Computing, vol 1363. Springer,
https://doi.org/10.1016/j.jss.2017.06.069 Cham. https://doi.org/10.1007/978-3-030-73100-7_60
[2] Christoph A. Thieme, Ali Mosleh, Ingrid B. Utne, Jeevith Hegde, [14] Hansheng Ren, Bixiong Xu, Yujing Wang, Chao Yi, Congrui Huang,
Incorporating software failure in risk analysis––Part 2: Risk modeling Xiaoyu Kou, Tony Xing, Mao Yang, Jie Tong, and Qi Zhang. 2019.
process and case study, Reliability Engineering & System Safety, Time-Series Anomaly Detection Service at Microsoft. In Proceedings
Volume 198, 2020, 106804, ISSN 0951-8320, of the 25th ACM SIGKDD International Conference on Knowledge
https://doi.org/10.1016/j.ress.2020.106804 Discovery & Data Mining (KDD '19). Association for Computing
[3] Hussein David Tafur, Giacomo Barbieri, Carlos Eduardo Pereira, An Machinery, New York, NY, USA, 3009–3017.
FMEA-based Methodology for the Development of Control Software https://doi.org/10.1145/3292500.3330680
Reliable to Hardware Failures, IFAC-PapersOnLine, Volume 54, [15] Jina Kim, Hyeongwon Kang, Pilsung Kang, Time-series anomaly
Issue 1, 2021, Pages 420-425, ISSN 2405-8963, detection with stacked Transformer representations and 1D
https://doi.org/10.1016/j.ifacol.2021.08.047 convolutional network, Engineering Applications of Artificial
[4] Christoph A. Thieme, Ali Mosleh, Ingrid B. Utne, Jeevith Hegde, Intelligence, Volume 120, 2023, 105964, ISSN 0952-1976,
Incorporating software failure in risk analysis – Part 1: Software https://doi.org/10.1016/j.engappai.2023.105964
functional failure mode classification, Reliability Engineering & [16] Rochak Swami, Mayank Dave, Virender Ranga, IQR-based approach
System Safety, Volume 197, 2020, 106803, ISSN 0951-8320, for DDoS detection and mitigation in SDN, Defence Technology,
https://doi.org/10.1016/j.ress.2020.106803 2022, , ISSN 2214-9147, https://doi.org/10.1016/j.dt.2022.10.006
[5] M. Asif and J. Ahmed, "A Novel Case Base Reasoning and Frequent [17] Wei Xie, Guanwen Huang, Wenju Fu, Bao Shu, Bobin Cui,
Pattern Based Decision Support System for Mitigating Software Risk Mengyuan Li, Fan Yue, A quality control method based on improved
Factors," in IEEE Access, vol. 8, pp. 102278-102291, 2020, doi: IQR for estimating multi-GNSS real-time satellite clock offset,
10.1109/ACCESS.2020.2999036 Measurement, Volume 201, 2022, 111695, ISSN 0263-2241,
[6] Chandan Kumar, Dilip Kumar Yadav, A Probabilistic Software Risk https://doi.org/10.1016/j.measurement.2022.111695
Assessment and Estimation Model for Software Projects, Procedia [18] C. -Y. Duan, X. -Q. Chen, H. Shi and H. -C. Liu, "A New Model for
Computer Science, Volume 54, 2015, Pages 353-361, ISSN 1877- Failure Mode and Effects Analysis Based on k-Means Clustering
0509, https://doi.org/10.1016/j.procs.2015.06.041 Within Hesitant Linguistic Environment," in IEEE Transactions on
[7] Linhan Ouyang, Yushuai Che, Ling Yan, Chanseok Park, Multiple Engineering Management, vol. 69, no. 5, pp. 1837-1847, Oct. 2022,
perspectives on analyzing risk factors in FMEA, Computers in doi: 10.1109/TEM.2019.2937579
Industry, Volume 141, 2022, 103712, ISSN 0166-3615, [19] Domenico Cotroneo, Luigi De Simone, Pietro Liguori, Roberto
https://doi.org/10.1016/j.compind.2022.103712 Natella, Enhancing the analysis of software failures in cloud
[8] Marc-André Filz, Jonas Ernst Bernhard Langner, Christoph Herrmann, computing systems with deep learning, Journal of Systems and
Sebastian Thiede, Data-driven failure mode and effect analysis Software, Volume 181, 2021, 111043, ISSN 0164-1212,
(FMEA) to enhance maintenance planning, Computers in Industry, https://doi.org/10.1016/j.jss.2021.111043
Volume 129, 2021, 103451, ISSN 0166-3615, [20] W. Kong, M. Nabi and K. Goossens, "Run-Time Recovery and
https://doi.org/10.1016/j.compind.2021.103451 Failure Analysis of Time-Triggered Traffic in Time Sensitive
[9] Keil, J. M., (2019). Efficient Bounded Jaro-Winkler Similarity Based Networks," in IEEE Access, vol. 9, pp. 91710-91722, 2021, doi:
Search. In: Grust, T., Naumann, F., Böhm, A., Lehner, W., Härder, T., 10.1109/ACCESS.2021.3092572
Rahm, E., Heuer, A., Klettke, M. & Meyer, H. (Hrsg.), BTW 2019. [21] Vitaliy Yakovyna, Natalya Shakhovska, Software failure time series
Gesellschaft für Informatik, Bonn. (S. 205-214). DOI: prediction with RBF, GRNN, and LSTM neural networks, Procedia
10.18420/btw2019-13 Computer Science, Volume 207, 2022, Pages 837-847, ISSN 1877-
[10] A. Mukasheva, Z. Akanov and D. Yedilkhan, "Research of the 0509, https://doi.org/10.1016/j.procs.2022.09.139
Regression Analysis Methods for Predicting the Growth of Patients [22] Umamaheswara Sharma B., Ravichandra Sadam, How far does the
with Diabetes Mellitus," 2021 IEEE International Conference on predictive decision impact the software project? The cost, service
Smart Information Systems and Technologies (SIST), Nur-Sultan, time, and failure analysis from a cross-project defect prediction model,
Kazakhstan, 2021, pp. 1-7, doi: 10.1109/SIST50301.2021.9465975. Journal of Systems and Software, Volume 195, 2023, 111522, ISSN
[11] Kasun Bandara, Rob J Hyndman, Christoph Bergmeir, MSTL: A 0164-1212, https://doi.org/10.1016/j.jss.2022.111522
Seasonal-Trend Decomposition Algorithm for Time Series with [23] Kasun Bandara, Rob J Hyndman, Christoph Bergmeir, MSTL: A
Multiple Seasonal Patterns. arXiv, 2021, 2107.13462. Seasonal-Trend Decomposition Algorithm for Time Series with
[12] J. Gao, H. Wang and H. Shen, "Task Failure Prediction in Cloud Data Multiple Seasonal Patterns. arXiv, 2021, 2107.13462.
Centers Using Deep Learning," in IEEE Transactions on Services https://doi.org/10.48550/arXiv.2107.13462
Computing, vol. 15, no. 3, pp. 1411-1422, 1 May-June 2022, doi:
10.1109/TSC.2020.2993728.
[13] Shaukat, K. et al. (2021). A Review of Time-Series Anomaly
Detection Techniques: A Step to Future Perspectives. In: Arai, K.
- 119 -
Comparative analysis of Hadoop and Spark
performance through different data processing
tasks
Yerulan Suynullayev Sergiy Gnatyuk Didar Yedilkhan
MSc student of the Department of Doctor of Technical Science, Professor PhD, Associate Professor of the
Information Systems and Cybersecurity of the Department of Computer Science Department of Computer
Almaty university of power engineering and National Aviation University Engineering
telecommunications Kyiv, Ukraine Astana IT University
Almaty, Kazakhstan [email protected] Astana, Kazakhstan
[email protected] [email protected]
Aisulu Serik Assel Mukasheva
Senior Lecturer of the Department of PhD, Associate Professor of the
Information Systems and Cybersecurity Department of Information Systems and
Almaty university of power engineering and Cybersecurity
telecommunications Almaty University of Power Engineering
Almaty, Kazakhstan and Telecommunications
[email protected] Almaty, Kazakhstan
[email protected]
orcid.org/0000-0001-9890-4910
Abstract— Big Data processing platforms like Hadoop and would be impossible to see with smaller datasets. For
Spark are in high demand due to the increasing use of large example, in the field of medical technology and health care,
data sets in modern computing. In this paper, the performance big data is used to improve diagnosis, treatment and
of Hadoop and Spark was evaluated for iterative computing prevention of diseases such as diabetes mellitus. [2]
and real-time data processing applications based on metrics
such as runtime, resource utilization, and scalability with Both Hadoop [3] and Spark [4] have their unique
increasing data size. The comparison aimed to provide strengths and limitations when it comes to big data
guidance to researchers, practitioners, and entrepreneurs on processing. Hadoop, with its highly scalable Hadoop
which platform to use depending on their specific Distributed File System (HDFS) [5] and MapReduce
requirements. The study highlighted the strengths and programming model, has for many years been the best
weaknesses of both platforms and provided valuable insights solution for batch processing of large data sets. [6] On the
into optimizing big data applications' performance. The other hand, in-memory processing capabilities and Spark's
comparison used text documents and graphs for Word Count ability to handle streaming data have made it a popular
and PageRank tasks, and performance testing was performed choice for real-time streaming processing and iterative
on datasets of varying sizes. The results showed that Spark machine learning tasks. [7]
outperforms Hadoop in most applications, especially in
iterative computing and real-time data processing, due to its With the growing popularity of both platforms, it has
use of in-memory computation. However, Hadoop is best suited become necessary to compare their performance in different
for batch processing operations that require several steps. It data processing tasks. In this research paper, we aim to
can handle these operations in parallel on multiple cluster provide a comprehensive comparative analysis of the
nodes, ensuring fast handling of large amounts of data. This performance of Hadoop and Spark in different data
comprehensive performance comparison between Hadoop and processing tasks. The analysis will cover a range of use cases
Spark in iterative computing and real-time data processing including batch processing, streaming and machine learning,
applications provides valuable information for researchers, and assess the performance of both platforms based on
practitioners, and enterprises regarding the trade-offs and factors such as speed, scalability, fault tolerance and ease of
benefits of using these platforms for big data processing.
use.
Keywords— Hadoop, Spark, Iterative Computing, Real-Time The results of the study will enable organizations to make
Data Processing, Performance Evaluation informed decisions when choosing the data platform that best
suits their needs.
I. INTRODUCTION
In recent years, big data has become an integral part of II. BACKGROUD
almost every organisation. [1] The sheer scale and A. Overview on Hadoop
complexity of data from a variety of sources makes it
difficult to process and analyse using traditional methods. In The Apache Hadoop framework is designed to process
this scenario, distributed computing systems such as Apache large amounts of data in a distributed environment on
Hadoop and Apache Spark have become powerful tools for clusters of computers. Its main components are MapReduce,
processing large amounts of structured and unstructured data. HDFS, YARN, and. Together, these components provide the
These datasets can be used to analyse patterns and trends that ability to store, process, and analyse large amounts of data in
a distributed environment.
- 120 -
sequential operations are required. This is why Spark was
created
B. Overview on Spark
Apache Spark is a high-performance, open-source
distributed computing system for batch and real-time
processing of large amounts of data. Spark outperforms
MapReduce in iterative machine learning tasks and in 10-20x
execution speed applications [13], by using innovative
techniques such as in-memory computing and lazy
computing to minimize read and write operations to disk.
Figure 1. Hadoop components
An RDD in Spark is a distributed collection of items that
MapReduce is a parallel programming model that allows can be cached in memory and redistributed to cluster nodes
processing large amounts of data on distributed computing in case of failures. Working with RDD includes creating
clusters. [8] It is based on two stages: Map and Reduce, based on input data, transforming with operations like map
which are executed in parallel on different nodes of the or filter, caching in memory, performing actions like reduce
cluster. This model is often used in big data processing or collect, saving results in different formats. RDD is lazy
systems, such as Apache Hadoop, to efficiently process and and immutable, which allows to optimize execution of
analyse large amounts of structured and unstructured data. calculations and minimize reading and writing to disk. (Fig.
(Fig. 2) 3)
Figure 3. How Spark works
To create an RDD, you can either parallelize a data
collection within a driver program or load a dataset from an
external storage system like HDFS or HBase. The
transformations applied to the RDD produce results that are
computed only when the action is invoked. You can store
intermediate results in memory through the use of the cache()
Figure 2. How MapReduce works or persist() method. Ultimately, the output is returned to the
driver program.
MapReduce is performed in two steps: map and reduce.
Each partitioning file corresponds to a map job, which C. Differences between Hadoop and Spark architectures
converts input records into direct records and outputs the A jobClient object is created by the driver program,
result to a circular memory buffer. When the buffer is full, which requests the application ID from the
the mapper begins to flush its contents to the local disk, ResourceManager. Having received the identifier, the
partitioning the data into multiple partitions and sorting them jobClient copies the required resources from HDFS and
by key. The map task is blocked if the buffer is full during sends the application to the ResourceManager. The jobClient
the spill. The reducer then selects a particular partition using then asks the YARN scheduler to allocate a container on the
a data shuffling process that establishes links between all NodeManager. If the job is of small type, the map and reduce
maps and reducer in Hadoop, and merges the partitions. The jobs will run in parallel on the local node. Otherwise the
reduce function processes the merged data and outputs the application wizard will request more containers from
result to HDFS. ResourceManager to perform mapReduce jobs. (Fig. 4)
YARN is a component that manages cluster resources
and allows applications to run on that cluster. [9] It is
responsible for allocating computational resources between
different tasks and applications in a Hadoop cluster, such as
MapReduce, Spark and others. YARN is a key component in
the Hadoop ecosystem and provides more efficient use of
cluster resources for processing large amounts of data.
HDFS is a distributed file system designed to store and
process large amounts of data on server clusters. It is
designed to provide fault tolerance, scalability and reliability
for data handling in a Hadoop cluster. [10].
MapReduce has a high latency due to its execution
model, which requires reading and writing data to disk
between each operation. In particular it has limitations for
iterative algorithms [11] and low latency computations. [12] Figure 4. Hadoop architecture
This can slow down data processing, especially when many A Spark application consists of a central coordinator
(driver) and several worker nodes (executors). The driver
- 121 -
performs two main tasks: converting the user program into IV. MATERIALS AND METHODS
task units and scheduling these tasks on the performers. The
executors, in turn, perform the tasks and store the RDDs in A. Word count sort by keys.
memory. The application creates a SparkContext, which The task of Word count sort by keys is to count the
connects to the cluster manager that allocates resources for number of occurrences of each word in a text document, and
the application. The application code is then sent to the then sort the results alphabetically (by key).
executor and the SparkContext delivers the tasks for For the Word count sort by keys task in Hadoop and
execution. (Fig. 5)
Spark, words are counted from text files that are stored in a
distributed file system (HDFS) or flat data store (RDD). This
is done using mappers, which perform local word counts in
blocks of text, and redrawers, which merge the results of the
mappers and calculate the final number of each word. A
special comparator is used to sort the words by key. The
algorithm can be implemented using APIs for distributed
data and parallel computing, such as MapReduce API in
Hadoop and RDD API in Spark. (Fig. 6)
Figure 5. Spark architecture
III. RELATED WORKS
Figure 6. Sample data on the number of words sorted by key
The paper [14] compares the performance of Apache
Hadoop and Apache Spark on 10GB, 100GB and 1TB Word Count - Sorted by Keys in Hadoop and Spark can
datasets using HiBench benchmarks. The results show that be modeled as follows:
Spark is faster than Hadoop on all datasets with speedups
from 1.5x to 14x due to the in-memory execution model. Suppose the input file contains N words and the number
This allows tasks to be performed without having to read and of nodes in the cluster is M. For simplicity, we assume that
write data from disk, which significantly improves each node has equal processing power.
processing performance. In [15], the performance of a) Hadoop
MapReduce and Apache Spark on Twitter data on a Hadoop
cluster of 10 nodes with a dataset size of 10 GB was The map function processes each word in O(1) time.
investigated. WordCount and Tfidf tasks were considered. Thus, the time complexity of the map assignment is given
The results showed that Spark is faster than MapReduce in by:
both tasks with speedups from 2x to 10x due to the in-
T(Map) = O(N) (1)
memory execution model. This allows the tasks to execute
without having to read and write data from disk, which In the shuffle and sort phase, data is transferred between
significantly improves processing performance. The authors nodes and sorted based on the key. The time complexity of
in [16] performed experiments on a cluster of 4 nodes with this phase is defined by:
different datasets ranging in size from 1GB to 100GB. Spark
𝑁
was faster than Hadoop for small datasets: on a 1-gigabyte T(Shuffle_Sort) = O(N log𝑀) (2)
dataset, the Word Count algorithm executed in 1.5 seconds
in Spark and 3 seconds in Hadoop. However, for larger The reduce function processes all values for a given key
datasets, Hadoop was faster: for a 100-gigabyte dataset, the in O(1) time. Thus the time complexity of the reduce
Word Count algorithm executed in 8 seconds in Hadoop and assignment is given by:
10 seconds in Spark. The authors attribute the performance 𝑁
difference to the fact that Hadoop uses a map-reduce T(Reduce) = O(log𝑀) (3)
framework designed for batch data processing, while Spark
uses a general-purpose engine that is suitable for both batch The total time complexity of the Hadoop job can be
and streaming data processing. The study [17] compared the calculated as follows:
performance of Apache Hadoop and Apache Spark on a T(Hadoop) = T(Map) + T(Shuffle_Sort) + T(Reduce) =
cluster of 16 machines, each with 24 cores and 64 GB of 𝑁 𝑁 𝑁 𝑁
memory. Data sets of 10 GB, 100 GB and 1 TB were used. O(N) + O(N log𝑀) + O(𝑀) = O(𝑀) + (N log𝑀)) (4)
The results showed that Spark was faster than Hadoop on all
datasets with speedups from 1.5x to 14x due to the in- b) Spark
memory execution model. This allows tasks to be performed The map function processes each word in O(1) time.
without having to read and write data from disk, which Thus the time complexity of the transformation is defined
significantly improves processing performance. by:
T(Transform) = O(N) (5)
- 122 -
The reduce function processes all values for a given key The reduce function processes all the values for a given
in O(1) time. Thus the time complexity of the ReduceByKey key in O(1) time. Thus the time complexity of the reduce
transform is defined by: job is given by:
𝑁 𝑁
T(ReduceByKey) = O( ) (6) T(Reduce) = O( ) (10)
𝑀 𝑀
The sortBy function sorts the output RDD based on the The user comparator sorts the output based on count as
key and count in time O(N logN/N). Thus, the time the primary key and word as the secondary key in time O(N
complexity of the secondary sort is defined by: 𝑁
log𝑀).
𝑁
T(Secondary_Sort) = O(N log𝑁) (7) The total time complexity of a Hadoop job can be
calculated as follows:
The total time complexity of a Spark job can be
calculated as follows: T(Hadoop) = T(Map) + T(Reduce) + T(Secondary_Sort) =
𝑁 𝑁 𝑁 𝑙𝑜𝑔𝑁
T(Spark) = T(Transform) + T(ReduceByKey) + O(N) + O(𝑀) + O(N log𝑀) = O( 𝑀 ) (11)
𝑁 𝑁 𝑁
T(Secondary_Sort) = O(N) + O(𝑀) + O(N log𝑁) = O(𝑀) + b) Spark
𝑁
(N log𝑁)) (8) The map function processes each word in O(1) time.
B. Word count sort by values Thus the time complexity of the transformation is defined
by:
Word count sort by values is the task of counting the
number of unique words in a large volume of text data and T(Transform) = O(N) (12)
sorting the results by value (i.e. by the number of times the
word is repeated). The reduce function processes all values for a given key
in O(1) time. Thus the time complexity of the ReduceByKey
Apache Hadoop uses MapReduce framework to solve transform is defined by:
this problem. The process of solving the problem consists of 𝑁
two steps: map and reduce. In the map step, each line of the T(ReduceByKey) = O(𝑀) (13)
file is split into words, which become keys and a unit
becomes a value. Then obtained key-value pairs are passed The sortByValue function sorts the output RDD based
to reduce stage, where the total number of keys for each on the count as primary key and the word as secondary key
𝑁
word is calculated. The result is sorted by frequency of in time O(N log𝑀).
occurrence of the words and written to the output file.
The total time complexity of the Spark assignment can
Apache Spark uses the groupByKey() operation to solve
be calculated as follows:
this problem. First the text file is loaded into RDD (Resilient
Distributed Dataset). Then RDD is converted using flatMap() T(Spark) = T(Transform) + T(ReduceByKey) +
operation which splits lines into words. The resulting key- 𝑁 𝑁 𝑁 𝑙𝑜𝑔𝑁
T(Secondary_Sort) = O(N) + O(𝑀) + O(N log𝑁) = O( 𝑀 )
value pairs are combined using reduceByKey() operation,
which counts the number of words. The result is then sorted (14)
by value using the sortBy() method and written to the output C. PageRank
file.
PageRank is an algorithm used by search engines to
In both cases the result is a list of words sorted in assess the importance of pages on the Internet. It is based on
descending order of the number of occurrences in the text the fact that a web page is considered more important if other
data. (Fig. 7) web pages link to it. In simple terms, PageRank gives each
page a weight based on the number of links that point to that
page and their own weight. The algorithm also takes into
account the number of links per page and the weight of each
of those pages. [18]
The algorithm was originally developed by Larry Page
and Sergey Brin, co-founders of Google. They used it to
Figure 7. Sample word count data sorted by value determine the relevance of a page in search engine rankings.
The formula for calculating PageRank was proposed by
Word count sort by values can be modelled as follows: Google founders Brin and Page in 1998. [19]
Word count sort by values in Hadoop and Spark is O(N 𝑅𝑗
𝑅𝑖 = 𝑑 ∗ ∑ +1−𝑑 (15)
logN), where N is the number of words in the input file. 𝑁𝑗
a) Hadoop where Rj is PageRank value, d is attenuation coefficient
𝑅𝑗
The map function processes each word in O(1) time. (usually 0.85), ∑ 𝑁 is sum of PageRank values for all pages j
𝑗
Thus, the time complexity of the map assignment is given that contain links to page i, divided by their number of
by: outgoing links Nj. The iterative process is repeated until the
PageRank values stabilise, each iteration using the new
T(Map) = O(N) (9) values to calculate the next.
- 123 -
V. PERFORMANCE TESTS AND RESULTS
The performance comparison between Hadoop and Spark
was performed on a VMware Workstation Pro virtual
machine with CentOS 7 operating system. Word Count,
sorting by keys and values and PageRank tasks were used.
These tasks were chosen because they allow us to evaluate
the performance of the two frameworks in near real-world
conditions, as they require parallel processing of large
amounts of data. The PageRank algorithm involves several
iterations of computation, while Spark processes data more
efficiently by storing intermediate data in the cache.
This experiment compared the performance of Hadoop Figure 8. Word count sort by keys result
and Spark on the same cluster machines, using HDFS as the B. Word count sort by values result
file storage system. The examples implemented in both
frameworks were based on the same programming language In an experiment comparing the performance of Hadoop
and algorithm. The code for the tasks was written in Python, and Spark in the Word count sort by values task (Fig. 9) with
which used several libraries such as MrJob, PySpark, increasing amount of processed data, it was found that Spark
NetworkX. significantly outperformed Hadoop. On a 100 MB dataset,
Spark performed 8 times better than Hadoop. When the
The experiment was run 10 times to account for errors amount of data was increased to 500MB, 1GB and 2GB,
caused by unstable network traffic. Errors could occur both Spark continued to perform better, outperforming Hadoop by
when running small jobs for a few seconds and when running a factor of 6. This significant performance difference is due
large jobs for tens of seconds. The results of each run were to the nature of Spark's architecture, which allows for in-
recorded and analysed to get the average performance for memory computation while minimizing reads and writes to
each framework. The results obtained were analyzed and disk. At the same time, Hadoop requires frequent reads and
presented as graphs using Matplotlib library. Based on the writes to disk, which slows down the framework.
data obtained, conclusions were drawn about the
performance of Hadoop and Spark.
A. Word count sort by keys result
Table 1 shows the amount of data used in this and the
following Word count sort by values study.
TABLE I. DATASETS FOR WORD COUNT SORT BY KEYS AND VALUES
Name Size
100M.csv 101.36MB
500M.csv 497.22MB
1G.csv 1001.44MB Figure 9. Word count sort by values result
2G.csv 1.95GB The Word count sort by values and Word count sort by
keys tasks revealed that the main difference in performance
between Spark and Hadoop is the result sorting methods.
The results of the experiment for the Word count sort by Spark uses the sortBy() method to sort results by value,
keys task show that Spark is significantly faster than Hadoop which allows you to specify the sort order and the number of
when the amount of data being processed is increased. concurrent operations. To sort results by key, Spark uses the
sortByKey() method. Hadoop uses the TextComparator class
It can be concluded that the difference in performance to sort results by value and the LongWritable class to sort
becomes noticeable when processing larger amounts of data. results by key. Spark can be faster than Hadoop in the Word
For example, when processing 100 and 500 MB, Spark is 3 count sort by values task due to the sortBy() method and the
times faster than Hadoop, but when processing from 1 GB of ability to perform calculations in memory, which speeds up
data, Spark can complete the task 5 times faster than data processing. In the Word count sort by keys task, both
Hadoop. (Fig. 8) frameworks can potentially show similar performance, but
This difference in processing speed is due to the fact that Spark may be more efficient by using the sortByKey()
Spark uses a more efficient approach to data processing, method.
based on distributed computation in memory, and also C. PageRank result
allowing the results of intermediate operations to be cached.
In contrast, Hadoop uses a more general MapReduce model, TABLE II. DATASETS FOR PAGERANK
which may be less efficient when processing large amounts
Name Size Nodes Edges
of data or performing complex operations.
AMAZON.txt 46.73MB 403,394 3,387,388
GOOGLE.txt 73.61MB 875,713 5,105,039
WEB.txt 107.55MB 685,230 7,600,595
- 124 -
The results of the experiment showed that Spark performance. In this case it is recommended to use additional
significantly outperformed Hadoop in the PageRank task on optimizations such as combiners and data partitioning to
all three data volumes. On average, Spark was 12 times speed up sorting.
faster than Hadoop. (Fig. 10)
Thus, to achieve maximum performance in the word
In the PageRank task, where we have to calculate the count sort by keys task, you should choose the optimal
importance of pages on the web, Spark may be faster than amount of RAM, considering the size of the data to be
Hadoop due to its parallelization ability. PageRank is an processed, and use additional optimizations to speed up
iterative algorithm that continuously processes data until the sorting.
required accuracy is achieved. Spark can easily parallelize
the iteration process across multiple nodes, which can
significantly speed up data processing. While Hadoop can
also be used to perform this task, but due to the limitations of
the distributed file system and slower read/write data, it may
be slower than Spark.
Figure 12. Word count sort by values result for Spark
In this task it was also found that data size had a
significant effect on the speed of the task. As the size of the
data increased, the processing time increased, but the
difference between Spark's performance with increasing
Figure 10. PageRank result RAM remained about the same. (Fig. 12)
D. A performance study of the Spark system when working When sorting by value, it should be taken into account
with limited memory space that this type of sorting can take up additional RAM
For the experiment, a cluster of several machines was depending on the size of the data. When the amount of data
used, on which Apache Spark distributed processing system becomes very large, it is possible that Spark will use
was installed. In the experiment, the value of the additional disk space, which can slow down performance
spark.executor.memory parameter, which defines the significantly.
maximum amount of memory allocated to each executor on In order to maximize performance in word count sort by
the cluster, was changed. values, the optimal amount of RAM should be chosen, taking
into account the size of the data to be processed and the
Investigating the performance of Spark under memory
possible use of extra disk space.
constraints is an important task for optimizing the
performance of cluster computing. In this experiment, we
evaluated the effect of the available memory constraint on
the processing speed of various amounts of data.
Figure 13. PageRank result for Spark
The following conclusions can be drawn from this
experiment. (Fig. 13)
Figure 11. Word count sort by keys result for Spark
Firstly, the more RAM allocated to the task, the faster it
The experiment found that the size of the data had a will run. We can see this in the example of WEB(107.55M)
significant effect on the speed at which the task could be data processing. If we allocate only 1 GB of RAM, the task
performed. With an increase in data size from 100 MB to 2 takes 714 seconds to complete, while if we allocate 2 or 3
GB, processing time increased, but the difference between GB of RAM, the task takes only 166 and 148 seconds
Spark's performance with 1, 2 and 3 GB of RAM remained respectively.
about the same. (Fig. 11)
Secondly, the difference in performance is not as
However, when sorting by key you should be aware that noticeable when processing smaller amounts of data. For
sorting can become a bottleneck and slow down the example, when processing AMAZON(46.73M) data, the
- 125 -
difference in task execution time on 1, 2 and 3 GB of RAM [7] D. Zhu, R. Ma, Y. Chai, B. Cai and L. Han, "Energy big data
was only a few seconds. automatic desensitization model based on Spark parallel computing
framework," 2021 2nd International Conference on Big Data
So, in order to get the best performance when processing Economy and Information Management (BDEIM), Sanya, China,
large amounts of data in Spark, it is recommended to allocate 2021, pp. 89-92, doi: 10.1109/BDEIM55082.2021.00026.
as much RAM as possible to speed up the execution of the [8] A. K. Rai and A. K. Malviya, "Testing MapReduce program using
Induction Method," 2020 IEEE International Students' Conference on
task. However, when processing smaller amounts of data, a Electrical,Electronics and Computer Science (SCEECS), Bhopal,
smaller amount of RAM can also be limited without India, 2020, pp. 1-5, doi: 10.1109/SCEECS48394.2020.178.
significantly degrading performance. [9] T. H. Aung and W. T. Zaw, "Improved Job Scheduling for Achieving
Fairness on Apache Hadoop YARN," 2020 International Conference
CONCLUSION on Advanced Information Technologies (ICAIT), Yangon, Myanmar,
2020, pp. 188-193, doi: 10.1109/ICAIT51105.2020.9261793.
The results found that in most cases, especially when
[10] D. Veeraiah and J. N. Rao, "An Efficient Data Duplication System
dealing with iterative algorithms, Spark performs better than based on Hadoop Distributed File System," 2020 International
Hadoop. Conference on Inventive Computation Technologies (ICICT),
Coimbatore, India, 2020, pp. 197-200, doi:
Spark outperforms Hadoop for iterative algorithms 10.1109/ICICT48043.2020.9112567.
because Spark uses in-memory computation to perform [11] H. Ahmed, M. A. Ismail and M. F. Hyder, "Performance optimization
tasks, whereas Hadoop writes results to disk after each step. of Hadoop cluster using linux services," 17th IEEE International
This means that when dealing with large datasets and Multi Topic Conference 2014, Karachi, Pakistan, 2014, pp. 167-172,
repeatedly traversing them, Spark can significantly speed up doi: 10.1109/INMIC.2014.7097331.
the execution of algorithms such as machine learning and [12] Boduo Li, Edward Mazur, Yanlei Diao, Andrew McGregor, and
deep learning, which often use iterative approaches. In Prashant Shenoy. 2011. A platform for scalable one-pass analytics
using MapReduce. In Proceedings of the 2011 ACM SIGMOD
addition, Spark has a handy API for dealing with iterative International Conference on Management of data (SIGMOD '11).
algorithms, such as GraphX, which provides high-level Association for Computing Machinery, New York, NY, USA, 985–
operations for dealing with graphs and graph-related 996. https://doi.org/10.1145/1989323.1989426
algorithms. [13] Matei Zaharia, Mosharaf Chowdhury, Tathagata Das, Ankur Dave,
Justin Ma, Murphy McCauley, Michael J. Franklin, Scott Shenker,
Thus, Spark is a more efficient tool for dealing with and Ion Stoica. 2012. Resilient distributed datasets: a fault-tolerant
iterative algorithms compared to Hadoop and can abstraction for in-memory cluster computing. In Proceedings of the
significantly reduce runtime for machine learning and other 9th USENIX conference on Networked Systems Design and
iterative applications. Implementation (NSDI'12). USENIX Association, USA, 2.
[14] Nasim Ahmed, Andre L. C. Barczak, Teo Susnjak et al. A
However, it should be noted that Spark is not suitable for Comprehensive Performance Analysis of Apache Hadoop and
applications that do asynchronous fine-grained general state Apache Spark for Large Scale Data Sets Using HiBench, 17 August
2020, PREPRINT (Version 1) available at Research Square
updates and may also consume a significant amount of https://doi.org/10.21203/rs.3.rs-43526/v1
memory. If task speed is not a critical factor and there is
[15] Farhan MN, Habib MA, Ali MA. A study and performance
insufficient memory available, Hadoop would be a better comparison of MapReduce and Apache spark on twitter data on
choice. However, for applications that are speed-sensitive or Hadoop cluster. Int J Inf Technol Comput Sci 2018.
use iterative algorithms and have sufficient memory, Spark https://doi.org/10.5815/ijitcs.2018.07.07
will be the preferred option. [16] Yassine Benlachmi, Abdelaziz El Yazidi and Moulay Lahcen
Hasnaoui, “A Comparative Analysis of Hadoop and Spark
REFERENCES Frameworks using Word Count Algorithm” International Journal of
Advanced Computer Science and Applications(IJACSA), 12(4),
[1] Wang J. et al. Big data analytics for intelligent manufacturing 2021. http://dx.doi.org/10.14569/IJACSA.2021.0120495
systems: A review //Journal of Manufacturing Systems. – 2022. – Т.
62. – С. 738-752. https://doi.org/10.1016/j.jmsy.2021.03.005 [17] Ahmed, N., Barczak, A.L.C., Susnjak, T. et al. A comprehensive
performance analysis of Apache Hadoop and Apache Spark for large
[2] Mukasheva, A., Rakhmetulayeva, S., Astaubayeva, G., & Gnatyuk, S. scale data sets using HiBench. J Big Data 7, 110 (2020).
(2022). Developing a system for diagnosing diabetes mellitus using https://doi.org/10.1186/s40537-020-00388-5
bigdata. Eastern-European Journal of Enterprise Technologies,
5(2(119), 75–85. https://doi.org/10.15587/1729-4061.2022.266185 [18] F. U. Zambuk, A. Y. u. Gital, S. Boukary, F. Jauro and H. Chiroma,
"Evaluation of Iterative Pagerank Algorithm for Web Page Ranking,"
[3] Hadoop Documentation http://hadoop.apache.org/ 2019 4th International Conference on Electrical, Electronics,
[4] Spark Documentation http://spark.apache.org/ Communication, Computer Technologies and Optimization
[5] R. Rathidevi and R. Parameswari, "Performance Analysis of Small Techniques (ICEECCOT), Mysuru, India, 2019, pp. 365-370, doi:
Files in HDFS using Clustering Small Files based on Centroid 10.1109/ICEECCOT46775.2019.9114728.
Algorithm," 2020 Fourth International Conference on I-SMAC (IoT [19] Monica Bianchini, Marco Gori, and Franco Scarselli. 2005. Inside
in Social, Mobile, Analytics and Cloud) (I-SMAC), Palladam, India, PageRank. ACM Trans. Internet Technol. 5, 1 (February 2005), 92–
2020, pp. 640-643, doi: 10.1109/I-SMAC49090.2020.9243418. 128. https://doi.org/10.1145/1052934.1052938
[6] A. K. Rai and A. K. Malviya, "Testing MapReduce program using
Induction Method," 2020 IEEE International Students' Conference on
Electrical,Electronics and Computer Science (SCEECS), Bhopal,
India, 2020, pp. 1-5, doi: 10.1109/SCEECS48394.2020.178.
- 126 -
The Use Of Machine Learning Methods In The Field Of Retail
Kuanyshbek N.S, Sapakova S.Z*
Kazakh national university named al-Farabi
The development of small and medium-sized businesses is the main instrument of economic
growth and socio-political modernization. A particularly important role is played by the
implementation of private property in real life and as a material basis for the formation of the
middle class. Consequently, as the share of small and medium-sized businesses in the economy
increases, our country is getting better, the socio-economic situation of the population is
improving. Retail trade is one of the biggest industries in the world, and it can be considered as
the process of buying goods for the purpose of reselling them to the end consumer.
Recent technological advances have had a significant impact on the retail industry (Shankar et al.,
2021). For example, food retailing has large changes in this century as more and more consumers
buy products online (Kureshi & Thomas, 2019). Over the past few decades, a lot of effort has been
spent on developing and improving forecasting models. In the retail sector, sales forecasting may
assist managers understand customers, forecast their future willings. Then companies can optimize
storage facilities, shelf space and display cases for the preparation of inventory and the
development of product strategies. In this paper, we consider the results of the study in relation to
retail forecasting, delineating the boundaries of the field in order to focus on food.
In particular, this work focused on the applicability of machine learning methods. Demand
forecasting is valuabletask in the retail sector in terms of minimizing costs, establishing the correct
inventory level, optimal use of warehouse space and reducing the problem of lack of goods in
stock. Accurate forecasting of future demand is a difficult task for retailers and wholesalers due to
sudden changes in demand levels, lack of historical data, new trends and seasonal spikes in
demand.
To get good accuracy, you need to choose the right models. There are various models available
for regression analysis, but three regression models are used for this article: Decision Tree,
Random Forest and Gradient Boosting. After implementing these models, we measure accuracy
by dividing the dataset into two parts, which are the training dataset and the test dataset. The
performance of all algorithms was evaluated based
on the mean absolute error (MAE), mean square error (RMSE), mean square error (MSE) and the
coefficient of determination (R2).
The indicators of all R-Square models tend to the 1st and show 98 and 99% accuracy, which
is a good indicator. Based on the analysis of the results obtained, it was revealed that the most
effective model for sales forecasting is the random forest model and the decision tree and Gradient
Boosting shows competitive results. The mean square error of a random forest is 0.083, and the
difference between actual and projected food prices is 0.041.
- 127 -
A study on false defect management of PCB
ICT inspection using failure rate Estimation
Changmok Park 1* and Sangyule Choi 2
Induk University, South Korea
Email: [email protected], [email protected]
Abstract—This paper is about a data analysis method for II. PROPOSED METHOD
managing false defects in PCB ICT (In-Circuit Test) facilities.
In a typical PCB processing line, ICT facilities used to inspect
the electrical properties of final products, and automatic
A. Definition of Fake NG
inspections performed on more than thousands of items. These
facilities consist of fixtures, relay boards and control circuits.
The fixture is composed of numerous contact pins, and the relay Normal measurement data of ICT looks like the figure
board is composed of numerous switching parts. These facilities below.
deteriorate over time and cause false defects. In this study, we
analyze the automatic inspection information data to identify
the cause of false defects. We develop an optimization model to
utilize inspection data as facility maintenance information.
Keywords—PCB, ICT, Inspection, Optimization, false defect
I. INTRODUCTION
PCB ICT is a system that measures electrical characteristic
values (nF, mF, uF, Ko, Mo) and determines NG(No Good) if
they fall outside the preset lower and upper limits. These
facilities are composed of pin and relay parts as shown in
following image, and fake NG occurs due to the aging
phenomenon of pin parts. Efficient facility maintenance is
possible by finding and replacing the pin parts that cause the The result of the inspection test is OK if it falls within the
failure. upper and lower bounds, and NG if it falls outside the bounds.
But sometimes, NG is composed of Real NG and Fake NG. If
the measurement of an inspection item is too large or too small,
as shown in the following image, it can be judged as a Fake
NG.
- 128 -
B. Statistical methodology for finding Fake NG
Data collected over a period of time can be used to
estimate statistical distributions. This statistical distribution
can be used to define outliers. Some of the measurements
determined to be NG can be classified as Fake NG using the III. PRELIMINARY RESULT
outlier criteria.
The developed program performs the following procedures.
The daily Fake NG rate of item is calculated as following.
C. Finding the cause of Fake NG
If multiple inspection items use the same pin, the cumulative
failure rate of the inspection item can be used to estimate the
probability of the pin failure rate. Consider the following
structure of inspection item 1 and 2 and pins 1, 2, and 3.
If you get the following Fake NG Rate by utilizing the data The daily failure rate of pin estimated as following by
for a certain period of time. optimization method.
The following failure rate of pin can estimated by nonlinear
optimization problem.
For pins with consistently high failure rate, maintenance
activities can be performed.
REFERENCES
[1] Pan Hongxia, Huang Jinying, Liu Guangmin," Fault diagnosis of
circuit board based on fault tree," 2008 10th International Conference
on Control, Automation, Robotics and Vision, 2008, pp. 1666-1671.
[2] “Avoiding False Pass or False Fail,”
https://www.evaluationengineering.com/instrumentation/ate/article/13
006510/avoiding-false-pass-or-false-fail
[3] “False Alarm – Radartutorial,”
https://www.radartutorial.eu/01.basics/False Alarm Rate.en.html.
- 129 -
August 22(TUE) - 25(FRI), 2023
Oral Session 5
14:00~15:40 Wed, August 23, 2023
Room : A437
Chair: Chulmin PARK (Korea Railroad Research Institute, Korea)
- 130 -
VIDEO STREAM ANALYSIS ALGORITHMS
BASED ON THE OPENCV LIBRARY USING
THE RASPBERRY P S PLATFORM
Adilzhan Kereyev1, Bauyrzhan Shokanov2*, Lyazzat Kaparova*3, Elubayeva Dina*4
1 Aktobe Regional University named after K. Zhubanov, Aktobe, Kazakhstan
e-mail: [email protected]
*2 Nazarbayev Intellectual school of the Physics and Mathematics direction of Aktobe, Kazakhstan
e-mail: [email protected]
*3 Aktobe Regional University named after K. Zhubanov, Aktobe, Kazakhstan
e-mail: [email protected]
*4 Aktobe Regional University named after K. Zhubanov, Aktobe, Kazakhstan
e-mail: [email protected]
Abstract - This scientific article is devoted to the study and advantages make Raspberry Pi an ideal platform for the
application of video stream processing algorithms based on the implementation of various robotic projects, including video
OpenCV library, including video capture, object detection and stream processing.
object motion detection based on optical flow. To implement OpenCV finds use in many fields; computer vision, robotics,
these algorithms, Python software was used, followed by the
machine learning and etc. Due to its powerful tools and
development of its own system based on the studied ones, and its
further implementation on the Raspberry Pi robotic platform.A flexibility, OpenCV is becoming an increasingly popular tool
natural experiment was conducted to detect objects on a video for processing and analyzing video streams, which makes it
stream. A review of a number of popular methods and important for solving various tasks in the field of robotics.
algorithms for analyzing the video stream from the point of view
of speed, accuracy of definition, and the possibility of II. METHODS AND MATERIALS
implementation on microcomputers is carried out.
Experiments were conducted, as a result of which the Software.
effectiveness of using video stream processing algorithms based
on the OpenCV library for solving problems in the field of
Python is one of the most popular and convenient
computer vision and robotics was confirmed. The advantages of
these algorithms in high accuracy and speed of video processing programming languages that is often used in robotics. Python
were especially highlighted. has extensive capabilities that allow user to control robots and
Thus, this article is of great importance for developers and develop applications for robotics. To develop robotic
researchers working in the field of computer vision and robotics, applications in Python, many developers use specialized
who can use video stream processing algorithms based on the development environments (IDEs).
OpenCV library to solve various tasks. Python IDE is a software that allows programmers to create
and debug Python code. In robotics, the Python IDE is used
Key words: openCV, video processing, object detection, optical to develop robotic applications like robot control systems,
flow, python, raspberry Pi, computer vision, robotics, video
robotic manipulators, and autonomous mobile robots.
capture, motion detection, image processing
One of the most popular Python IDE for robotics is PyCharm.
I. INTRODUCTION PyCharm has many features that allow developers to quickly
create and debug robotic applications. Among the main
Currently, video stream analysis is considered as one of the features of PyCharm the followings also should be noted:
most urgent tasks in the field of robotics and computer vision.
In order to solve this task, various image processing • ROS (Robot Operating System) Support: PyCharm
algorithms are being used to extract information from the has integration with ROS, which allows developers
video stream and use it for decision-making to create ROS-nodes that can be used to control
In this article, we present algorithms for video stream robots and robotic manipulators.
processing and analyzing based on the OpenCV library, using
a two-wheeled robot with a Raspberry Pi processor. As the Debugging: PyCharm allows developers to adjust code using
main tool for image processing, we use OpenCV, which a debugger that allows you to set breakpoints, view variables,
provides a powerful and flexible set of tools for working with and execute code step by step.
images and videos. Dependency Management: PyCharm allows you to manage
The advantages of Raspberry Pi as a platform for robots dependencies that are used in robotic applications, such as
include low cost, small size and power consumption, as well Python libraries and ROS packages.
as a wide range of sensors and modules available. These
- 131 -
Auto-completion: PyCharm has the ability to automatically In general, Raspberry Pi provides many opportunities for
supplement code, which allows developers to quickly create creating and controlling robots, due to its flexibility, low cost
code for robotic applications. and support for various programming languages. In addition,
Integration with Git: PyCharm allows developers to work it has a large developer community and can be easily
with Git repositories, which simplifies version control and integrated with other platforms. All these factors make
combined work on the project. Raspberry Pi one of the best platforms for robot development,
both for beginners and experienced developers.
• Built-in Terminal: PyCharm includes a built-in The use of Raspberry Pi in robotics has already led to the
terminal that allows you to run commands in the creation of a wide range of robots, from self-driving cars to
terminal without having to switch to another controlled drones. This platform is also used in various fields
window. such as education, industry and research.
Despite the fact that Raspberry Pi can be used to create simple
• Support for various platforms: PyCharm can be used and complex robots, it has its limits. For example, it may not
on various operating systems, such as Windows, be too powerful for some complex tasks that require a lot of
macOS and Linux, this provides versatility in the computing power. You may also need to use additional
development of robotic applications. modules and components to implement certain functions.
All in all, Raspberry Pi is a convenient and affordable robot
In addition to PyCharm, there are also other Python IDEs that development platform that can be used in various fields of
can be used to develop robotic applications, such as Visual robotics. Its capabilities continue to expand, and there are
Studio Code, Spyder and Jupyter Notebook. many projects and ideas that can be implemented using this
platform.
Hardware.
III. METHODS AND ALGORITHMS OF VIDEO
Raspberry Pi is a miniature computer that can be used as a STREAM ANALYSIS.
robotic platform for the development of various types of
robots. It was created with the aim of providing accessible Video stream analysis is an important problem in the field of
computer technology and programming to a wide audience. computer vision and image processing. In this paper, we
Raspberry Pi offers many possibilities for creation and robot describe methods and algorithms for video stream analyzing
control, thanks to its flexibility and low cost. using the Python programming language and the OpenCV
Low Cost: One of the main advantages of Raspberry Pi is its library.
low cost. This allows developers to create more affordable To analyze the video stream, we use the following methods:
robots for use in educational establishments, laboratories and
home projects. Capturing a video stream
Flexibility: Raspberry Pi has many features, such as To capture a video stream, we use the Video Capture function
connecting sensors, motors, cameras and other components, from the OpenCV library. This function allows us to open a
which allows user to create different types of robots. In video stream from the camera or from a file and start
addition, the Raspberry Pi can be used as the main controller capturing data.
to control other devices and sensors. For example, the following code shows how to capture a
Operating System Availability: Raspberry Pi comes with video stream from a camera:
Linux operating system, which provides a powerful and import cv2
flexible framework for robot development. Developers can cap = cv2.VideoCapture(0)
use Python, C++ and other programming languages to while True:
develop software and control robots. ret, frame = cap.read()
A broad community of developers: Raspberry Pi has a large cv2.imshow('frame', frame)
community of developers who actively share knowledge and if cv2.waitKey(1) & 0xFF == ord('q'):
experience about the use of this platform in robotics. This break
allows new developers to quickly get help and solve problems cap.release()
that arise. cv2.destroyAllWindows()
Integration with other platforms: Raspberry Pi can be easily
integrated with other platforms such as Arduino, etc. This IV. MOTION DETECTION
provides greater flexibility in the development of robots,
allowing the use of various components and devices. To detect motion, we use an optical flow algorithm. This
ROS Support: Raspberry Pi supports Robot Operating algorithm allows to determine the direction and speed of
System (ROS), which is a popular platform for developing movement of objects in the video.
robotic applications. This allows developers to use ROS for For instance, the following code shows how to detect motion
robot control, data visualization, and many other functions. using an optical flow algorithm:
- 132 -
import cv2 if cv2.waitKey(1) & 0xFF == ord('q'):
cap = cv2.VideoCapture(0) break
ret, frame1 = cap.read() cap.release()
prvs=cv2.cvtColor(frame1, cv2.COLOR_BGR2GRAY) cv2.destroyAllWindows()
mask = np.zeros_like(frame1)
while True:
ret, frame2 = cap.read()
next=cv2.cvtColor(frame2,cv2.COLOR_BGR2GRAY)
flow = cv2.calcOpticalFlowFarneback(prvs, next, None,
0.5, 3, 15, 3, 5, 1.2, 0)
mag, ang = cv2.cartToPolar(flow[..., 0], flow[..., 1])
mask[..., 0] = ang * 180 / np.pi / 2
mask[..., 2] = cv2.normalize(mag, None, 0, 255,
cv2.NORM_MINMAX)
rgb = cv2.cvtColor(mask, cv2.COLOR_HSV2BGR)
cv2.imshow('frame2', rgb)
if cv2.waitKey(1) & 0xFF == ord('q'):
break
prvs = next
cap.release()
Fig. 1. The result of testing the algorithm for capturing a video stream from
cv2.destroyAllWindows() a webcam
V. OBJECT DETECTION We see that the algorithm successfully captures video from
the computer's webcam and does not close until the
To detect objects, we use the Haar cascade classifiers combination of keys 1 and q is pressed in order to complete
algorithm. This algorithm allows to detect objects in the the algorithm.
video. For example, faces, cars, animals, etc. In order to analyze your own video using this algorithm, you
To visualise, the following code shows how to detect faces need to replace the line:
using the Haar cascade classifier algorithm: cap = cv2.VideoCapture(0)
to the line:
import cv2 cap=cv2.VideoCapture("path/to/your/video/file"),
face_cascade=cv2.CascadeClassifier('haarcascade_fronta where "path/to/your/video/file" - is a path to your video file.
lface_default.xml') After replacing this line, the code will read the video file
cap = cv2.VideoCapture(0) instead of capturing video from the webcam, and you will be
while True: able to analyze your own video.
ret, frame = cap.read()
gray = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
faces = face_cascade.detectMultiScale(gray, 1.3, 5)
for (x, y, w, h) in faces:
cv2.rectangle(frame, (x, y), (x + w, y + h), (255, 0, 0), 2)
cv2.imshow('frame', frame)
if cv2.waitKey(1) & 0xFF == ord('q'):
break
cap.release()
cv2.destroyAllWindows()
VI. THE RESULTS OF THE STUDY.
Let's run the video stream capture algorithm on the computer Fig 2. The result of testing the video stream capture algorithm based on an
external video file
using the python IDE and applying the code below to look at
the result.
import cv2 Now, run the algorithm for capturing the video stream and
cap = cv2.VideoCapture(0) motion detection while using the optical stream, applying
while True: the code below and look at the result:
ret, frame = cap.read() import cv2
cv2.imshow('frame', frame) import numpy as np
- 133 -
cap = cv2.VideoCapture(0) Let's run the Haar cascade classifiers algorithm to detect
ret, frame1 = cap.read() objects in the video using the code below and look at the
prvs=cv2.cvtColor(frame1,cv2.COLOR_BGR2GRAY) result.
mask = np.zeros_like(frame1)
while True: face_cascade=cv2.CascadeClassifier('C:/Users/shokanov_b
ret, frame2 = cap.read() .akb/PycharmProjects/pythonProject/venv/Lib/sitepackages/
next=cv2.cvtColor(frame2, cv2/data/haarcascade_frontalface_default.xml')
cv2.COLOR_BGR2GRAY) cap = cv2.VideoCapture(0)
flow= cv2.calcOpticalFlowFarneback(prvs, next, while True:
None, 0.5, 3, 15, 3, 5, 1.2, 0) ret, frame = cap.read()
mag, ang = cv2.cartToPolar(flow[..., 0], flow[..., 1]) gray=cv2.cvtColor(frame,cv2.COLOR_BGR2GRAY)
mask[..., 0] = ang * 180 / np.pi / 2 faces = face_cascade.detectMultiScale(gray, 1.3, 5)
mask[..., 2] = cv2.normalize(mag, None, 0, 255, for (x, y, w, h) in faces:
cv2.NORM_MINMAX) cv2.rectangle(frame, (x, y), (x + w, y + h), (255, 0,
rgb=cv2.cvtColor(mask,cv2.COLOR_HSV2BGR) 0), 2)
cv2.imshow('frame2', rgb) cv2.imshow('frame', frame)
if cv2.waitKey(1) & 0xFF == ord('q'): if cv2.waitKey(1) & 0xFF == ord('q'):
break break
prvs = next cap.release()
cap.release() cv2.destroyAllWindows()
cv2.destroyAllWindows()
CONCLUSION
Fig 3. The result of testing the motion detection algorithm using optical As a result, in this scientific article, algorithms for processing
flow. a video stream based on the OpenCV library, using Python
software and the Raspberry Pi robotic platform, were
reviewed and tested. The tasks of video capture, object
detection and object motion detection based on optical flow
were successfully implemented.
In the outcomes of the experiments, it was revealed that the
algorithms for processing video streams based on the
OpenCV library have high accuracy and speed of video
processing, which makes them an effective tool for solving
problems in the field of computer vision and robotics.
Thereby, the obtained results indicate the possibility of using
video stream processing algorithms based on the OpenCV
library in various fields of science and technology where high
accuracy and speed of video processing are required.
Fig. 4 The result of testing the motion detection algorithm using an optical
stream based on an external video file REFERENCES
- 134 -
[1] Kaushik Das Sharma, Gargi Datta, and Sanjoy Kumar
Saha. "Real-time surveillance system using Raspberry Pi and
OpenCV." Procedia Computer Science 58 (2015): 561-568.
[2]Kamil Ferenc, Paweł Ksieniewicz, and Piotr Wieczorek.
"Application of the Raspberry Pi platform for object
detection in video streams." Proceedings of the 11th
International Conference on Computer Recognition Systems
CORES 2019, Advances in Intelligent Systems and
Computing 989 (2020): 201-211.
[3]Zhang, Z. (2018). "A review on recent advances in
pedestrian detection using deep learning." International
Journal of Computer Vision, 128(1), 3-40.
[4]Ahmed, A. E., & Moniri, M. (2017). "A survey on video-
based traffic flow analysis." IEEE Access, 5, 7329-7345.
[5]Daniel Herrera, Manuel Ruiz-Marín, and Alfonso García-
Fernández. "Optical flow-based detection and tracking of
vehicles in video streams." Journal of Real-Time Image
Processing 15, no. 5 (2019): 1049-1067.
[6]Garg, R., & Kaur, J. (2020). "A Review of Real-Time
Object Detection Techniques." In 2020 IEEE 8th
International Conference on Computational Intelligence and
Computing Research (ICCIC) (pp. 1-5). IEEE.
[7]Redmon, J., & Farhadi, A. (2018). "YOLOv3: An
Incremental Improvement." arXiv preprint
arXiv:1804.02767.
[8]Dalal, N., & Triggs, B. (2005). "Histograms of oriented
gradients for human detection." In 2005 IEEE Computer
Society Conference on Computer Vision and Pattern
Recognition (CVPR'05) (Vol. 1, pp. 886-893). IEEE.
[9]Simon, L., & Rodner, E. (2020). "A survey on bias and
fairness in computer vision." arXiv preprint
arXiv:2005.01982.
[10]Liu, W., Anguelov, D., Erhan, D., Szegedy, C., Reed, S.,
Fu, C. Y., & Berg, A. C. (2016). "SSD: Single Shot MultiBox
Detector." European conference on computer vision (pp. 21-
37). Springer, Cham.
- 135 -
Internet of Things attack and security analysis
Aigul Adamova Tamara Zhukabayeva
Department of Computer Engineering Faculty of Information Technology
Astana IT University LN Gumilyov Eurasian National
Astana, Kazakhstan University
[email protected] Astana, Kazakhstan
[email protected]
Abstract—Everyone knows that the Internet of Things homes, smart cities, e-government, robotics, etc. [1,2].
has dramatically affected the life of the community and The problem of security in the networks of the IoT is a
represents many different technologies. Active universal problem of all countries, since all devices
implementation in everyday life makes us think about the collect information, analyze it and make decisions. At the
confidentiality of data used in the interaction of the Internet
of things. At the same time, the Internet of Things is limited
same time, the Internet of Things is limited by resources,
by resources, and the analysis of algorithms that contribute and the analysis of algorithms that contribute to the
to the preservation of data confidentiality in the process of preservation of data confidentiality in the process of
interaction of the Internet of Things is a very relevant area. interaction of the IoT is a very relevant area [1].
This work is an overview article on the interaction Interaction with each other IoT is one of the main
architecture of the Internet of Things, on the protocols used goals and serves to exchange information, things, etc.
regarding the architecture, security algorithms, simulations anywhere, anytime using the right network or service. In
of various types of attacks and attack detection schemes. this paper, the IoT connected in a wireless sensor network
Keywords— Internet of Things, attack, interaction, will be considered. Wireless sensor network (WSN) -
security, wireless sensor network deployed over a large area with a large number of
wireless sensor nodes that are used to monitor the system
I. INTRODUCTION [3-5]. There will also be an emphasis on maintaining the
The Internet of Things (IoT) is a network of various confidentiality of data when exchanging messages.
devices that are able to detect, connect, and transmit Confidentiality is a protection against unauthorized
information. Thanks to the IoT, it is possible to represent disclosure of information. Confidential data and
many different smart technologies that will be used in information can only be accessed by authorized persons
everyday life and will affect the quality of services. The in an authorized manner [6].
application of the IoT is extensive, including smart
Figure 1. Research methodology
scientific publications and literature in the Scopus
II. REVIEW OF WORKS database. The search was conducted on the keywords
The bibliometric analysis methodology is shown in “IoT security”, as a result, 5,335 review articles, research
Figure 1. The bibliometric analysis consists of 6 stages, articles, book chapters and other works were found for
where the first two stages include an extensive analysis of 2022. At the next stage, since the work is devoted to the
analysis of attacks, the keyword “IoT attack” was added
- 136 -
and the article type was set to research, as a result, 3,862 today we can note new technologies, such as blockchain
papers were received. At the third stage, keywords were technologies, software-defined networks, holochain
added and a varied search was performed, which resulted technologies [9-11]. In the presented work, algorithms
in 2,778 works. At stage 4, queries from the second and that contribute to the preservation of data confidentiality,
third stages are combined. At the fifth stage, all articles possible attacks during IoT interaction, depending on the
are filtered by content and works are selected for further data flow and various functions, will be considered.
bibliometric analysis. In total, 121 articles were selected,
33 of which became the basis for the presented work.
General statistics on publications of works in Google
Scholar for 2022 is shown in Figure 2. The search for
works was carried out using the following keywords:
“IoT privacy”, “IoT encryption privacy”, “IoT
lightweight encryption privacy”, “IoT physical layer
security privacy". As a result of the search, it is important
to note that maintaining confidentiality is an urgent task
and many scientists are conducting research in this area.
Figure 4. Real world IoT/OT attack examples
There are several types of architecture of the Internet
of Things, such as three-, four-, five-layer architectures.
Table 1 lists scientists whose research is being carried out
on various architectures.
TABLE I. IOT ARCHITECTURE
Type of work Authors University/city, Country
architecture
Yuan Ze University,
Three-layer [12] I. Mashal et al.
Taiwan
architecture
[13] B. Paul Shillong, India
[14] C. Zhong et al. Foshan,China
Figure 2. Statistics on publications for 2022 in Google Scholar
The University of
[15] J. Gubbi et al. Melbourne, Vic - 3010,
According to the Microsoft report for 2022, China, the Four-layer
Australia
architecture
USA, India, Korea, Taiwan and Russia are the leaders in Taiyuan University of
the source of malware infection in the IoT interaction [16] J. Li et al. Technology, Taiyuan,
China
process (Figure 3) [7].
Munster Technological
[17] Mirani, AA
Five-layer University, Ireland
architecture IIT Delhi, New Delhi,
[18] P. Sethi et al.
India
III. COMMUNICATION PROTOCOLS ACCORDING TO
ARCHITECTURE LEVELS
A. Sensor level
Physical interfaces connecting various devices, such
as RJ45, for an asynchronous interface - RS-485, RS-232,
and USB, Wireless, SPI, etc. [19]. In particular cases, it is
possible to use a gateway that converts any physical
connector into a wireless one.
Figure 3. Percentage of Malware Infection Sources in IoT Interaction B. Network layer
Today, modern enterprises are striving to modernize Communication protocols are used for data transfer:
workplaces, by automating routine work to reduce the 3G, 4G, 5G, LTE, Dash, Wi-fi, LPWAN, Zigbee,
burden on staff, all these factors justify the real need for Bluetooth, RFID, Lora, LoRaWan, etc. Network
security and data confidentiality. Figure 4 shows vivid communication protocols IPv4, IPv6, WoT, 6LoWPAN,
examples of IOT attacks around the world for the period UDP, RPL [15]. IEEE 802.15.4, NFC, NB-IoT, SigFox,
from 2010 to 2022 [7]. Zigbee, LoRaWAN, Z-Wave, IEEE 802.11 family, BLE
In the review paper [8] M. Rana et al. discuss the are targeted for use in IoT interoperability.
importance of securing IoT networks, consider various C. Application level
threats depending on the architectural model of IoT. The In its structure, it contains many different protocols
paper also presents lightweight cryptographic algorithms for ultra-large volumes and large networks of things.
and protocols for data protection in the IoT environment. CoAP is an Internet transmission protocol for use in
If we take the direction of IoT network security, then devices with limited resources (similar to HTTP), MQTT
- 137 -
is a simplified messaging-oriented network protocol, TABLE II. SECURITY ALGORITHMS
DDS is a universal protocol that allows you to perform Algorithm year Characteristics structure Possible
almost any action, from starting devices to connecting attacks
high-performance networks , XMPP is an extensible Application layer
messaging and presence information protocol, AMQP is a MQTT 1999 OASIS standard TLS Denial of
CoAP 2014 IETF Standard DTLS service,
message queue management protocol that ensures the AMQP 2003 OASIS standard SASL clipboard
joint operation of different systems and applications [19]. DDS 2001 OMG standard overflow,
CoAP, uBounjour, mDNS, DNS-SD, DDS, AMQP, XMPP 2000 RFC standard SASL, SQL
MQTT, XMPP - applicable in IoT interoperability. The TLS injection
attacks,
outdated FTP, Telnet, SSH protocols perform their phishing
functions decently, but are very demanding on resources attacks, etc.
and are not suitable for the field of IoT. Network layer
To ensure data security, different algorithms are used WPA3 2018 192 bit key GCMP-2 Wormhole
256 bit 56 attack, Sybil
for each level. At the sensor level, it is possible to use Broadcast / attack,
lightweight encryption algorithms. At the network level, Multicast grayhole/sel
authentication methods, cryptographic methods, network Integrity ective
security methods are used, and at the application level, Protocol Galois forwarding
Message (SF) attack,
authorization methods and data privacy protection are Authentication Blackhole
used. Table 2 lists various algorithms, main Code attack,
characteristics and possible attacks [20-21]. AES 2001 128 block size SPN Sinkhole
(Bits) attack,
IV. DIFFERENT TYPES OF ATTACKS IN THE 128 keys (Bits) Replay
INTERCONNECTION OF THE INTERNET OF THINGS 10 rounds attack,
Spoofing
One of the best ways to ensure security in the Internet attack, hello
of Things interaction is to investigate the identified flood attack
Sensor level
attacks by reviewing and analyzing the 2022 literature.
PRESENT 2012 64 block size SPN
This paper presents potential network layer security (Bits)
threats in the form of Wormhole attack, Sybil attack, 80 keys (Bits) Host attack,
Grayhole/Selective forwarding (SF) attack, Blackhole 32 rounds damage,
attack, Sinkhole attack, Replay attack, Spoofing attack, SIMON 2013 64 block size Balanced code
(Bits) Feistel injection
hello flood attack and countermeasures against them 96 keys (Bits) network attacks,
(Table 3). 26 rounds replay
Wormhole attack - with creates a false a channel for TWINE 2013 64 block size GFS attack,
false data transmission, violating data confidentiality and (Bits) battery
80/128 key size drain.
can be the initiator of launching additional attacks [22]. (Bits)
Sybil attack - an infected host can impersonate other 2/16/20 rounds
hosts by changing location, thereby violating data
security and negatively affecting resource usage [23]. V. INTRUSION DETECTION SCHEMES
Grayhole/SF attack - a malicious node prevents the There are different schemes for detecting and
flow of information from authorized nodes to the base preventing attacks (Figure 5). By the nature of the
station. So this leads to denial of service attack, which network, the intrusion detection scheme can be
can be converted into a "black hole" attack by attacking centralized and distributed. In a centralized scheme, the
the entire network and restricting the flow of information head node, responsible for collecting data from sensors
from each node to the receiver [24]. connected to it, is the main link that is able to detect and
Blackhole attack - a “black hole” appears in the prevent an attack. In distributed schemes, the main part of
network and absorbs incoming traffic, so the source is not the classter is responsible for detecting and preventing an
aware of the delivery of the message [25]. Sinkhole attack. If we consider schemes on the basis of protection,
attack - a violating node can send out routing messages they are divided into detection, where the scheme is able
and collect all packets from all nodes in its vicinity, going to detect, determine the attack and prevention, where the
to the base station [26]. scheme is able to detect attacks and disable faulty
Replay attack - by eavesdropping on messages, nodes[31-33].
retransmits packets to the host, gaining trust and
launching additional attacks [27].
Spoofing attack - a node masquerades as another node
by falsifying data, thereby violating data confidentiality
[28].
Hello flood attack - a violating node broadcasts or
other information with a sufficiently strong signal,
showing each node in the network that it is their
neighbor, respectively, other nodes transmit their packets
to real neighboring nodes, then they are not accepted by Figure 5. Intrusion detection schemes
them [29].
- 138 -
TABLE III. SIMULATION OF ATTACKS AND WORKS DEVOTED TO THESE ISSUES
Attack simulation Name of attacks work main idea
Wormhole attack discovery algorithm integrated with Node
Zhijun Teng, et al. [22]
Trust Optimization Model (NTOM-DA)
offer a lightweight bloom filter and sybil
Sybil attack Cong Pu, Kim-Kwang
attack detection mechanism based on a
Raymond Choo [23]
physical non-cloneable function (PUF)
X. Huang and Y. Wu an artificial immune system based on a hazard
Grayhole/SF attack
[24] model is proposed,
Ioulianou, PPet al. proposed new security framework for IoT
Black hole attack
[25] networks based on RPL (SRF-IoT)
Arun Kumar Sangaiah a protocol that uses the Multilayer Clustering
Sinkhole attack
et al. [26] Security Protocol (CL-MLSP) with AODV
Al-Shareeda, M. A. et detection and prevention mechanism in
Replay attack
al.[27] SECS/GEM communication system
exploit unique radiation patterns in
M. R. Nosouhiet al. mmWave-enabled devices and propose a
Spoof attack
[28] security mechanism to detect wireless
spoofing attacks
Renya Nath N and et critical security requirements are identified to
Hello flood attack
all [29], Muath A. [30] build advanced secure IoT systems.
- 139 -
[10] Yang G. An Overview of Current Solutions for Privacy in the
VI. CONCLUSION Internet of Things. Front Artif Intelligent. 2022 Mar 3; 5:812732.
doi: 10.3389/frai.2022.812732.
The IoT (Internet of Things) has become a vital part [11] S. Zaman, MRA Khandaker, RT Khan, F. Tariq and K.-K. Wong,
of modern technology, which includes a network of "Thinking Out of the Blocks: Holochain for Distributed Security
devices capable of discovering, connecting and in IoT Healthcare," in IEEE Access, vol. 10, pp. 37064-37081,
2022, doi: 10.1109/ACCESS.2022.3163580
transmitting information. In IoT, the security of data used [12] Mashal, I.; Alsaryrah, O.; Chung, T.Y.; Yang, CZ; Kuo, W.H.;
in the interaction of these devices is an important issue. Agrawal, DP Choices for interaction with things on Internet and
However, the IoT is limited in resources, which requires underlying issues. Ad Hoc Netw. 2015, 28, 68–90.
careful analysis of the algorithms that help maintain the [13] Paul, B. (2022). Internet of Things (IoT), Three-Layer
Architecture, Security Issues and Counter Measures. In: Fong, S.,
confidentiality of data when they interact. The article Dey, N., Joshi, A. (eds) ICT Analysis and Applications. Lecture
provides an overview of the IoT interaction architecture, Notes in Networks and Systems, vol 314. Springer, Singapore.
protocols, security algorithms, various types of attacks, https://doi.org/10.1007/978-981-16-5655-2_3
and attack detection schemes. It also focuses on the [14] C.-l. Zhong, Z. Zhu and R.-G. Huang, "Study on the IOT
Architecture and Access Technology," 2017 16th International
wireless sensor network (WSN) used in IoT interactions Symposium on Distributed Computing and Applications to
and highlights the importance of maintaining data privacy Business, Engineering and Science (DCABES), Anyang, China,
in messaging. The article provides an overview of 2017, pp. 113-116, doi: 10.1109/DCABES.2017.32.
scientific publications, provides statistics of publications [15] Gubbi, J., Buyya, R., Marusic, S., Palaniswami, M.: Internet of
Things (IoT): a vision, architectural elements, and future
and bibliometric analysis, by keywords in the direction of directions. Future. Gener. Comput. Syst. 29(7), 1645–1660 (2013)
security IoT, IoT encryption privacy, IoT physical layer [16] J. Li et al., "A Remote Monitoring and Diagnosis Method Based
security privacy and IOT attacks. Along with this, the on Four-Layer IoT Frame Perception," in IEEE Access, vol. 7, pp.
article discusses various types of IoT architecture, 144324-144338, 2019, doi: 10.1109/ACCESS.2019.2945076.
[17] Mirani, AA; Velasco-Hernandez, G.; Awasthi, A.; Walsh, J. Key
including three-layer, four-layer and five-layer IoT Challenges and Emerging Technologies in Industrial IoT
architectures. Architectures: A Review. Sensors 2022, 22, 5836.
In general, the article provides an insight into the https://doi.org/10.3390/s22155836
security issues of IoT interactions and the need for further [18] Pallavi Sethi, Smruti R. Sarangi, "Internet of Things:
Architectures, Protocols, and Applications", Journal of Electrical
research in this area. and Computer Engineering, vol. 2017, Article ID 9324035, 25
The proposed work is part of a research project that pages, 2017.https://doi.org/10.1155/2017/9324035
will develop a methodology and algorithms for predicting [19] IoT Analytics research 2022. https://iot-analytics.com
the further development and assessing the criticality of [20] Nurlan, Z., Zhukabayeva, T., & Othman, M. (2021). EZ-SEP:
Extended Z-SEP routing protocol with hierarchical clustering
detected attacks in real time. approach for wireless heterogeneous sensor network. Sensors,
21(4), 1021.
ACKNOWLEDGMENT [21] Nurlan, Z., Zhukabayeva, T., & Othman, M. (2019, October).
This research has been/was/is funded by the Science Mesh network dynamic routing protocols. In 2019 IEEE 9th
International Conference on System Engineering and Technology
Committee of the Ministry of Education and Science of (ICSET) (pp. 364-369). IEEE.
the Republic of Kazakhstan (Grant No. AP14973006). [22] Z. Teng, C. Du, M. Li, H. Zhang and W. Zhu, "A Wormhole
Attack Detection Algorithm Integrated With the Node Trust
REFERENCES Optimization Model in WSNs," in IEEE Sensors Journal, vol. 22,
[1] Chanal, P.M., Kakkasageri, M.S. Preserving, Data Confidentiality no. 7, pp. 7361-7370, 1 April1, 2022, doi:
in Internet of Things. SN COMPUT. SCI. 2, 53 (2021). 10.1109/JSEN.2022.3152841.
https://doi.org/10.1007/s42979-020-00429-z [23] Cong Pu, Kim-Kwang Raymond Choo, Lightweight Sybil Attack
[2] Zhukabayeva, T., Oralbekova, Z., Zhartybayeva, M., Detection in IoT based on Bloom Filter and Physical Unclonable
Zhumadillayeva, A., Adamova, A. “Prospects of development of Function, Computers & Security, Volume 113, 2022, 102541, ISSN
technologies in the field of robotics and the stages of design of 0167-4048,https://doi.org/10.1016/j.cose.2021.102541.
mobile robotic complex,” IEEE, pp. 1-4, August 2015. 2015 5th [24] X. Huang and Y. Wu, "Identify Selective Forwarding Attacks
International Conference on IT Convergence and Security Using Danger Model: Promote the Detection Accuracy in Wireless
(ICITCS). Sensor Networks," in IEEE Sensors Journal, vol. 22, no. 10, pp.
[3] K. Gulati, R. Sarath Kumar Boddu, D. Kapila, Sunil L. Bangare, 9997-10008, 15 May15, 2022, doi: 10.1109/JSEN.2022.3166601.
N. Chandnani, G. Saravanan, A review paper on wireless sensor [25] Ioulianou, PP; Vassilakis, VG; Shahandashti, SF A Trust-Based
network techniques in Internet of Things (IoT), Materials Today: Intrusion Detection System for RPL Networks: Detecting a
Proceedings, Volume 51, Part 1 , 2022, Pages 161-165, ISSN Combination of Rank and Blackhole Attacks. J. Cybersecurity.
2214-7853,https://doi.org/10.1016/j.matpr.2021.05.067. Priv. 2022, 2, 124-153. https://doi.org/10.3390/jcp2010009
[4] Nurlan, Z., Zhukabayeva, T., Othman, M., Adamova, A., & [26] Arun Kumar Sangaiah, Amir Javadpour, Forough Ja'fari, Pedro
Zhakiyev, N. (2021). Wireless sensor network as a mesh: Vision Pinto, HamidReza Ahmadi, Weizhe Zhang, CL-MLSP: The design
and challenges. IEEE Access, 10, 46-67. of a detection mechanism for sinkhole attacks in smart cities,
[5] Nurlan, Z., Kokenovna, TZ, Othman, M., & Adamova, A. (2021). Microprocessors and Microsystems, Volume 90, 2022, 104504,
Resource allocation approach for optimal routing in IoT wireless ISSN 0141-9331, https://doi.org/10.1016/j.micpro.2022.104504.
mesh networks. IEEE Access, 9, 153926-153942. [27] Al Shareeda, MA; Manickam, S.; Laghari, S.A.; Jaisan, A.
[6] Sadhu, PK; Yanambaka, V.P.; Abdelgawad, A. Internet of Things: Replay-Attack Detection and Prevention Mechanism in Industry
Security and Solutions Survey. Sensors 2022, 22, 7433. 4.0 Landscape for Secure SECS/GEM Communications.
[7] Microsoft. Cyber signals. December 2022. Sustainability 2022, 14, 15900.
[8] M. Rana, Q. Mamun, R. Islam, Lightweight cryptography in IoT https://doi.org/10.3390/su142315900
networks: A survey, Future Generation Computer Systems, [28] MR Nosouhi, K. Sood, M. Grobler and R. Doss, "Towards
Volume 129, 2022, Pages 77-89, ISSN 0167-739X, Spoofing Resistant Next Generation IoT Networks," in IEEE
https://doi.org/10.1016/j .future.2021.11.011. Transactions on Information Forensics and Security, vol. 17, pp.
[9] Al Hayajneh, A.; Bhuyan, MZA; McAndrew, I. Improving 1669-1683, 2022, doi: 10.1109/TIFS.2022.3170276.
Internet of Things (IoT) Security with Software-Defined [29] Renya Nath N, Hiran V Nath, Critical analysis of the layered and
Networking (SDN). Computers 2020, 9, 8. systematic approaches for understanding IoT security threats and
https://doi.org/10.3390/computers9010008 challenges, Computers and Electrical Engineering, Volume 100,
- 140 -
2022, 107997, ISSN 0045-7906, [32] André Cirne, Patrícia R. Sousa, João S. Resende, Luís Antunes,
https://doi.org/10.1016/j.compeleceng.2022.107997. IoT security certifications: Challenges and potential approaches,
[30] Muath A. Obaidat, Chapter 19 - Threats and vulnerabilities of Computers & Security, Volume 116, 2022, 102669, ISSN
wireless sensor networks in smart cities, Editor(s): Amir H. Alavi, 0167-4048, https://doi.org /10.1016/j.cose.2022.102669.
Maria Q. Feng, Pengcheng Jiao, Zahra Sharif-Khodaei, The Rise [33] Boyu Kuang, Anmin Fu, Willy Susilo, Shui Yu, Yansong Gao, A
of Smart Cities, Butterworth-Heinemann, 2022, Pages 463-491, survey of remote attestation in Internet of Things: Attacks,
ISBN 9780128177846, countermeasures, and prospects, Computers & Security, Volume
https://doi.org/10.1016/B978-0-12-817784-6.00024-2. 112, 2022, 102498, ISSN 0167-4048,
[31] Pamarthi, S. and Narmadha, R. (2022), "Literature review on https://doi.org/10.1016/j.cose.2021.102498.
network security in Wireless Mobile Ad-hoc Network for IoT
applications: network attacks and detection mechanisms",
International Journal of Intelligent Unmanned Systems, Vol. 10
No. 4, pp. 482-506. https://doi.org/10.1108/IJIUS-05-2021-0028
- 141 -
IoT watering robot in technologies of variable
irrigation rationing in precision agriculture
Madina Mansurova1* Baurzhan Belgybaev2 Yerlan Zaitin3
Al-Farabi Kazakh National Al-Farabi Kazakh National Al-Farabi Kazakh National
University University University
Almaty, Kazakhstan Almaty, Kazakhstan Almaty, Kazakhstan
[email protected] [email protected] [email protected]
Askhat Asset4
Almaty University of Energy and
Communications named after G.
Daukeev
Almaty, Kazakhstan
[email protected]
Abstract— Precision farming is a conceptual direction of the • The use of special drippers with pressure
modern agrotechnical strategy. The most important link in regulators increases irrigation costs;
these technologies is solving the problems of shortage and • Loss of yield due to suboptimal drip irrigation;
saving water resources for agriculture during drought. The • The need for a separate drip irrigation network for
paper proposes methods and technologies for creating
different plants;
prototypes of an IoT city tank truck for watering city trees and
a small-sized autonomous IoT sprinkler for intelligent watering • Non-recoupment of capital expenditures due to
of home and garden plants. Functional and circuit solutions lower yields per unit area.
make it possible to replicate IoT device data from available It should be noted that in the conditions of a mini-
electronic sets of blocks and modules offered by the authors . greenhouse and small agribusiness, these shortcomings are
leveled and show positive results with an acute shortage of
Keywords—precision farming, IoT, watering robot, precision water and its high cost [7].
agriculture In the publication and patents, a number of authors
have studied the upgrade of a mini-greenhouse with drip
I. INTRODUCTION
irrigation to the level of an IoT device [8,9,10,11]. Such
Modern technologies of precision farming through the use modernization of mini-greenhouses creates a number of
of space monitoring and the global Internet make it possible technological and social advantages:
to adequately predict the yield of cultivated plants, to • an automated greenhouse with remote access makes it
determine regions with excess or lack of moisture [1,2,3]. possible to develop urban horticulture, when city
Kazakhstan belongs to the region with zones of rainfed dwellers grow salads and early vegetables on home
agriculture [3]. In the southern regions of the country, rice, gardens, roofs of apartment buildings, loggias and
cotton, vegetables, fruits and fodder crops are grown on balconies;
irrigated areas. At present, due to population growth in
Uzbekistan, Kyrgyzstan and in the south of Kazakhstan, • the availability of remote monitoring allows you to
there is an acute shortage of irrigation water for growing manage the technological processes in a mini
agricultural areas [4]. The above regions are the main greenhouse from anywhere in the world;
suppliers of early vitamin products for the northern regions • independent cultivation of salads and vegetables in
of Kazakhstan. The state stimulates with benefits for the urban areas guarantees a fresh and environmentally
distribution of resources for irrigation when applying friendly product;
precision farming technologies. For example, in the
Dzhambul region, a plant was built for the production of • urban residents can go on vacation in any Internet
drip tape and fittings for them using American technology accessible region during the summer months without
[5]. Vegetable growers of our country practice the use of damage to plants due to the lack of agrotechnical
drip irrigation in greenhouses and greenhouses for growing care for them;
tomato crops; this type of irrigation is especially widely • the level of farming culture is increasing among rural
used in the arid Turkestan region [6]. Drip irrigation saves residents, the interest of young people in IoT
water resources, but it has a number of disadvantages: technical creativity, which contributes to their
• Substantial capital investment per unit of irrigated consolidation in the countryside.
area;
• Rapid wear of the drip water supply system and German manufacturers of equipment for drip irrigation
harsh conditions for inlet water pressure; are constantly improving their products. Let us discuss their
• Drop in water pressure along the length of the approaches to solving the problem of pressure drop along
drip tape; the length of the drip pipe. In the old designs of the
irrigation pipe, the droppers were equidistant in steps from 6
- 142 -
cm to 60 cm (see Figure 1a). This was clearly not rational, belongs to one of the greenest megacities in the world (up to
then they began to produce a solid pipe with an attached 10 m2 of green spaces fall on each inhabitant of the city).
special needle for puncture and a dropper without a flow The city has numerous relict century-old oaks and chestnuts.
regulator (as shown in Figure 1 b). The latest modifications They are under strict state registration and protection from
of droppers have a manual flow regulator, which allows, felling, watering of trees is carried out by a ditch system or
with successful selection, to evenly distribute the flow along individually from tankers to the root.
the length of the pipe (Figure 1c). The evolution of the The Akimat of the city, within the framework of the
design of the drip tape and its fittings is shown in Fig.1. Smart City concept, set the task for our university to
robotize and intellectualize the process of watering trees and
flower beds from mobile tankers. Ecologists have proven
that this method of irrigation can significantly save the
limited water resources of glacial mountain water.
а) Fig. 2 shows a scheme for upgrading the functionality of
intelligent watering trees and flower beds using the QR code
of a city tank truck to the level of a neurophysical
device[12].
b)
Fig. 2. Urban IoT mobile water sprinkler based on the KAMAZ ACV-10
UST heavy-duty tanker, where: 1-water tank with a capacity of 10 m3
based on the KAMAZ 43118-50 chassis with an irrigation rail; 2- PC
operator-manager "Zelenstroy"; 3- pump SVN-80 with a capacity of 38
m3/hour; 3-hydraulic four-link manipulator for positioning the water
supply to the root system of a tree or for spraying water over a flower bed;
4-flexible hose from the tank truck pump to the watering can with an IoT
video camera; 5- IoT microcontroller ESP32 Cam with a video camera; 6- a
poster with a QR code, which contains the technological parameters of the
c) plant and its watering modes.
Fig. 1 - sets of equipment for drip irrigation with drippers of
different designs: a) stationary dripper with a constant flow rate; b) pipes, Let us describe the proposed method for watering plants
fittings and droppers built into a drip pipe with three fixed flow rates; c) using a mobile IoT waterer. It should be noted that watering
built-in dripper with manual flow regulator. plants in the city of Almaty is a strictly regulated
technological operation for Zelenstroy employees. Watering
An analysis of existing irrigation equipment from the time is chosen so that the plants are under minimal solar
standpoint of modern digital automation shows that radiation.
automation kits, fittings and pipes for drip irrigation belong In the evening, at night or in the early morning, a tanker
to the analog type of controlling the parameters of a truck travels around its irrigation site on the route map. Each
complex technological process, such as watering cultivated tree or flower bed has unique posters with a QR code. The
plants, taking into account weather conditions. driver-operator from the cab of the car points the
The development and creation of scientific and manipulator with a video camera at a poster with a QR code.
practical foundations for the digitalization of spot irrigation After reading and recognizing this information, it performs
of plants in urban and rural conditions is an important and intelligent watering of the plant[14].
relevant direction in the development of the Internet of The driver-operator of the IoT tanker also sees, through
things. a special web interface, technological information about the
II. MATERIALS AND METHODS place and time of irrigation, the current volume of water
supply, its relation to the optimal parameters of moistening
The improvement of existing urban watering in a given irrigation area on his mobile phone, tuned to the
installations and machines in a metropolis with a hot climate IoT local computer network of the ESP32Cam
to the level of neurophysical devices is a complex and microcontroller. All these operations are performed
complex problem that must be solved separately for automatically, the driver-operator monitors the operation of
stationary and mobile technical watering devices. the scada system[15].
Due to the limited volume of this article, we will Separately, we note the work of the operator-manager of
consider technological and software-hardware methods for the central office of Zelenstroy. On the monitor of his PC,
improving Almaty tank trucks with a large volume of water not only current information is displayed, but also the state
(up to 10 m3 H2O), which are used to water city trees and of moisture of plants in other areas. In case of heavy rainfall
flower beds. It should be noted that the city of Almaty in certain areas of the city, the operator-manager can quickly
- 143 -
change the route map of tank trucks. This preliminary
project was considered and fertilized, the technical and
economic feasibility of this project was proved not only for
innovation, but also by reducing the wages of the forwarder-
sprinkler of the tank truck.
This stage of creating a watering robot is characterized
by the fact that remote automation is carried out only for the
technological part of the irrigation process itself. In
conditions of intensive urban traffic, the creation of an
unmanned watering robot in the form of a tank truck on the
chassis of a heavy truck is a difficult and unsafe problem for
the city population and road users. Therefore, in the work
for urban irrigation, only the process of recognition and
watering of plants is automated, the movement of a tanker
truck along the route is controlled by a human driver.
A promising and safe unmanned watering robot is
proposed to be created for highly automated greenhouses
and urban household plots. The functional diagram of the Fig. 5. Interface of the multifunctional window for controlling
unmanned smart watering robot is based on existing IoT the Xiaomi Mi 360 webcam
solutions for the automation of Xiaomi Mi Home smart
devices is shown in Fig. 3. Fig. 5 shows the current image in front of the camera
with an indication of the projection time, as well as virtual
buttons for turning on the sound, camera, touch joystick for
remote control of the camera centerline in the directions to
the right and left by 360 degrees and up and down by -30
degrees and up +30 degrees .
Thus, the presence of the remote functionality for
changing the axis line of the Xiaomi Mi 360 webcam allows
you to create a remote electromechanical digital device for
remote control of the transport platform as an additional
mobile automation tool for Xiaomi Mi Home. Note that this
approach is a pre-project development. The project does not
attempt to change the circuit solutions in the web camera
Fig. 4. Functional diagram of an unmanned watering robot
chips.
The use of a Xiaomi webcam operating in the Zig Bee III. RESULTS AND DISCUSSIONS
protocol[17,18] and having an online interface in global
Internet networks allows you to create methods for Based on the obtained modules, methods and developed
designing and semi-industrial application of a smart theoretical and practical materials, the following
watering robot operating in greenhouses and private houses. electromechanical robotic device is proposed as a result,
The main feature of the electric transport platform of the which allows remote control of a caterpillar transport
proposed robot is the use of a caterpillar chassis on electric platform and a manipulator with an irrigation hose via the
motors with a manipulator on servo drives. Internet (see Fig. 5). The software and hardware for remote
In technological terms, watering plants is a sluggish control of the manipulator is described in detail in the
process in time; Internet information delays of a few technical literature, this article will cover the issue of remote
seconds are not fatal. Therefore, to control the movement of automation of dosing the amount of water for watering a
the robot, the Xiaomi Mi 360 webcam was used, which has plant using its QR code.
a multifunctional web interface (see Fig. 4).
- 144 -
Let's dwell on new versions of the software for adapting
a webcam based on the ESP 32 CAM microcontroller to
solve practical irrigation problems using the proposed agro-
robot prototype.
Fig. 6. Scheme of the electromechanical method of controlling the IoT
electric motor driver of the watering robot using the Xiaomi Mi 360
webcam Fig. 8. Circuit diagram of connecting the pump to the IoT
microcontroller ESP 32 Cam
To control the transport platform in robotics,
Algorithmic difficulties arose in the process of
microcircuits of the L298N type are used, which, with the
introducing the ESP32QRCodeReader.h program into the
help of H-bridges, change the polarity of the direct current
software and hardware of the agro-robot, in which, with
on the brushes of electric motors [19,20].
standard QR code recognition, a message with the decoding
To code the motor driver, we will use the remote
of the QR code text was displayed on the COM port of the
mobility of the webcam centerline. To move forward, the
Arduino IDE. To water plants using a QR code displayed on
operator on the Internet raises the centerline of the webcam
a mini-poster near the plant, it is necessary not only to
by about + 10 degrees and electromechanically encodes a
decipher the code, but also to perform a series of actions in
digital combination for connectors D3, D1, D4, D2 using
accordance with agrotechnical requirements.
microswitches in the form of a set 0 1 01. As a result, the
Therefore, the program code has been adapted to these
motor driver rotates both tracks clockwise and the robot
requirements. In a QR code, for example, a dwarf tangerine
drives straight ahead.
tree, a number is entered in text form, displaying the amount
Similarly, if the axis line of the webcam is slightly
of water for one irrigation, the separator $ and the name of
turned to the right, the world switches supply the base
the plant.
voltage to the connectors D3, D1, D4, D2 in the form of a
set of 0 1 1 0. Then the left motor rotates clockwise, and the REFERENCES
right motor counterclockwise. The robot makes a tank turn
to the right. In order to stop the robot, it is enough to set the [1] A guide to proper watering for a miracle crop. Russia: E, 2019.-210
webcam exactly on the horizon and along the center line p.)
vertically [2] Evan Christen , Jim Ayars John Hornbuckle, Mark Hickey
The functional diagram of the platform with the Xiaomi .Technology and practice for irrigation in vegetables.- Published by
Mi 360 webcam and the manipulator with the ESP 32 Cam NSW Department of Primary Industries, State of New South Wales
2019.63 p.
camera and the pump, which turns on for the time interval
[3] Features of crop irrigation: furrow irrigation, contour irrigation and
read from the QR code, is shown in Fig. 7. sprinkling irrigation. \Practical guide ... "Center for training,
consultation and innovation" - Bishkek, 2018.-125 p.
[4] Grif M.G., Belgibaev B.A., Umarov A.A. Development of "Smart
Greenhouse" Based on the Model "Plant-Environment-Situation-
- COLLECTION OF SCIENTIFIC WORKS OF NSTU. -
2020. - No. 3 (98). – pp.49–64
[5] Setyo Supratno,Seta Samsiana,Dwi Panji ,GunawanNur Fauzi
Yuliyono Automatic watering and fertilizing with microcontrol
wemos D1 R2 based on IoT (internet of things) .-3-RD
INTERNATIONAL CONFERENCE OF BIO-BASED ECONOMY
FOR APPLICATION AND UTILITY.- January 2023
[6] Smart greenhouse: certificate of state. registration of computer
program No. 7578 / Umarov A.A., Belgibaev B.A., Mansurova M.E.,
Aripkhan Zh.N. - Register. 01/17/2020. of the year
[7] Irrigation regime, methods and techniques for irrigating vegetable and
melon crops in various zones of the Russian Federation \ Guide.-M.,
GNU VNII vegetable growing, 2021.84 p.
[8] Stefano Karpin .RAPID - precision irrigation with robots. URL:
Fig. 7. Layout diagram of an autonomous watering robot on a crawler http://rapid.berkeley.edu/ Accessed May 12, 2023
platform [9] Robot for automatic watering of green spaces. URL:
https://www.smprobotics.ru/applications/robot-dlya-poliva// accessed
May 22, 2023
- 145 -
[10] Chernenko A.B., Chernikov N.., Baginsky N.A., Sysoev M.I. [16] ESP32-CAM Remote Controlled Car Robot Web Server URL:
Features of the use of robotic platforms in agriculture. https://randomnerdtutorials.com/esp32-cam-car-robot-web-
[11] Skvortsov E.A., Vodolazsky F.V., Askerko V.V. The transition of server//Accessed June 22, 2023
agriculture to digital, intelligent and robotic technologies. - Agrarian [17] Aqeel Mohomad , Rayazan Raneem, haraga harmilan esign a Smart
Bulletin of the Urals. - Yekaterinburg. -2018. - p.82-88. mini Robot for Indoor Plant Watering System International Journal of
[12] Reading QR Code using ESP32 Camera Module & Python URL: Sciences^Basic and Applied Research(IJBAR) 61
https://how2electronics.com/qr-code-scanner-with-esp32-cam- [18] M. K. Heinrich, Sebastian von Mammen, Daniel Hofstadler
module-opencv///Accessed June 22, 2023 Constructing Living Buildings: A Review of Relevant Technologies
[13] M. Ghazal, Ali Talezadehlari, Moein Taheri, Mina Nazemi The for a Novel Application of Biohybrid Robotics.- Journal of the Royal
Mobile Robot HILARE: Dynamic Modeling and Motion Society Interface, 2019, 16(156)
SimulationIAES International Journal of Robotics and Automation [19] Xiaomi smart home URL: https://xiaomi-
(IJRA) June 2019 pp150-155 smarthome.ru/bluetoothshlyuzy-xiaomi//Accessed June 22, 2023.
[14] Jamaludin JalaniHo Zi KangAmirul Syafiq SadunZurina Abdul [20] Dom-Florista URL: https://dom-florista.ru/uhod/fazy-
WahabSemi-Auto Dosing Robot with Image Processing Journal of zhiznirasteniy.html// Retrieved June 22, 2023
Electrical engineering/- v.20(2-3),2021, pp.18-2 5) ,2022,p.1-9
[15] Vibhakar Kaushik ,MohtaAdarsh Patnaik, Shivam Kumar Panda
Design of an All-Purpose Terrace Farming Robot .-Springer Nature,
2022
- 146 -
Cache Replacement Policy for Approximate
Computing in Many Core Systems
Juhee Choi
Department of Smart Information Communication Engineering, Sangmyung University, Republic of Korea
Email: [email protected]
Abstract— Approximate computing researchers have
attracted strong attention for cores managing multimedia data,
such as GPUs, DSPs, and NPUs which are found in embedded
systems. In this field, an approximate accuracy level for each
core is important to deal with data shared by many cores
simultaneously without performance hurt. This study proposes
a cache replacement policy which reflects approximate accuracy
and the core characteristics. Understanding that each core has
its own characteristics, the priority used for choosing the
replacement block in the cache should be differently calculated
for approximate blocks and normal blocks. An experimental
model was made to test this approach and it showed the number Fig. 1. Design and implementation of cache coherence protocol[4].
of data replacement decreased by 11.2% without big
performance degradation.
Keywords—Cache Replacement Policy, Approximate
Computing, Many Core
I. INTRODUCTION
Research groups in the approximate computing have
proposed specialized cores for multimedia data like GPUs,
DSPs and neural processing units(NPU) placed in the
embedded systems. However, there are a few studies about
many cores for levels of approximate accuracy that causes
performance drop, when there are shared data among cores[1-
3]. In addition, the researchers have less paid attention the Fig. 2. Approximate cache coherence protocol.
system which includes non-volatile memory (NVM) based
last-level cache (LLC). protocol has M (Modified), S (Shared), E (Exclusive), I
(Invalidate) states. State transitions occur depending on read
The most important role of the traditional cache coherence requests and write requests. This is also distinguished by the
protocol such as MESI protocol is to maintain accurate values type of requestors such as from the core directly connected to
in cache blocks. If the value is changed in a cache block, it is the cache or from outer cores.
indicated as dirty block. When the block is evicted or
forwarded to the other requestors, it influences the transfer The new protocol has AE (Approximate Exclusive), AS
policy. However, NVM based systems also consider the (Approximate Shared), AM (Approximate Modified) states
approximate value in cache block as well as accurate cache instead of the existing M, S, E states when the cache block has
states. Thus, this paper adds extra states that can represent for an approximate value. For optimization of the state transition
approximate values for power-efficiency. diagram, the existing protocol has been modified for state
transitions. in the process of being studied over a long period
Plus, there is another consideration related to the cache of time. Therefore, the new protocol has fewer states
coherence. Even though the state transition diagram of a compared with the traditional method.
general protocol seems that it simply works, implementing the
protocol requires more complicated circuits as shown in Fig. III. SIMULATION RESULTS
1. Fig. 1(a) is the abstract state transition diagram of the MESI To verify the performance and power efficiency of the
protocol and Fig. 1(b) depicts the specific states required to proposed idea, the experiments are performed by the gem5 [5]
implement MESI, indicating the need for more intermediate which is a widely used cycle-accurate simulators. The cache
states and state transitions[4]. Therefore, if extra states are hierarchy consists of an 8-way 256KB unified LLC and 4-way
needed for a new protocol, it implies that more complex states 16KB I/D caches. Table I lists the detailed information of the
also are added to the existing circuits. One of this paper aims experimental parameters. Plus, the benchmark programs are
to minimize states by removing the unnecessary states. selected from the SPEC2006[6].
II. APPROXIMIATE CACHE COHERENCE PROTOCOL Fig. 3 shows the normalized the number of data transfer
This paper proposes an approximate cache coherence among cores. It is divided by read operation (RD_op), linefill
protocol as shown in Fig. 2. Fig. 2(a) shows the existing operation (LF_op), and write operation (WB_op) with the
protocol and Fig. 2(b) is one of the expected diagrams of the MESI protocol. The results showed that the new protocol
newly invented protocol from the research. The existing
- 147 -
achieved a 11.2% reduction in the data transfers without ACKNOWLEDGMENT
performance degradation. This work was supported by the National Research
Foundation of Korea (NRF) grant funded by the Korea
TABLE I. SIMULATION PARAMETERS
government(NRF-2021R1G1A1004340).
Core type Alpha
Cores 8 cores, out-of-order, 2GHz
L1 Cache Private 4-way 32KB I/D Cache REFERENCES
L2 Cache Private 8-way, 256KB [1] Roy, Kaushik, and Anand Raghunathan. "Approximate computing: An
energy-efficient computing technique for error resilient applications."
LLC Shared 16-way, 4MB 2015 IEEE Computer Society Annual Symposium on VLSI. IEEE,
Memory 64bit bus width, 4 read/write ports 2015.
[2] Imani, Mohsen, et al. "Acam: Approximate computing based on
adaptive associative memory with online learning." Proceedings of the
2016 International Symposium on Low Power Electronics and Design.
2016.
[3] Han, Jie, and Michael Orshansky. "Approximate computing: An
emerging paradigm for energy-efficient design." 2013 18th IEEE
European Test Symposium (ETS). IEEE, 2013.
[4] https://www.gem5.org/documentation/general_docs/ruby/MOESI_C
MP_directory
[5] J. Power, J. Hestness, M. S. Orr, M. D. Hill, and D. A. Wood, “Gem5-
gpu: A heterogeneous cpu-gpu simulator,” IEEE Computer
Architecture Letters, vol. 14, no. 1, pp. 34-36, 2015
[6] Henning, J. L. (2006). SPEC CPU2006 benchmark descriptions. ACM
SIGARCH Computer Architecture News, 34(4), 1-17.
Fig. 3. Normalized data transfer.
- 148 -
August 22(TUE) - 25(FRI), 2023
Oral Session 6
16:20~18:00 Wed, August 23, 2023
Room : A428
Chair: Assel Mukasheva (AUPET, Kazakhstan)
- 149 -
ANALYSIS OF CLIENT-SIDE AND SERVER-
SIDE ATTACKS IN THE CONTEXT OF WEB
APPLICATIONS
Adlet Smailov 1* and Saukhanova Zhanat 2
1
L. N. Gumilyov Eurasian National University, Kazakhstan
1
L. N. Gumilyov Eurasian National University, Kazakhstan
Email: [email protected]
I. INTRODUCTION
ABSTRACT
Web applications have become an essential part of our
With the rapid growth of web applications, ensuring their
security has become a critical concern for organizations and
daily lives, from online shopping to social media. However,
individuals alike. This scientific review article aims to analyze the this increase in reliance on web applications has also led to an
various types of client-side and server-side attacks commonly increase in cyberattacks against them . Web application
encountered in web applications, along with their respective attacks occur when a cybercriminal exploits a particular
protection methods. vulnerability or chain of vulnerabilities to gain unauthorized
access to sensitive information or disrupt the normal operation
The article begins by examining client-side attacks, including of an application.
Cross-Site Scripting (XSS), Cross-Site Request Forgery (CSRF),
and Clickjacking. Each attack is described in detail, providing an These attacks can be divided into two categories: client-
understanding of their nature and potential consequences. Concrete side attacks and server-side attacks. Client-side attacks target
examples are presented to illustrate the practical application of these a user's web browser and are initiated by injecting malicious
attacks, highlighting their potential impact on user privacy and code into a web page. Server-side attacks, on the other hand,
system integrity. Furthermore, effective protection measures and target the server hosting the web application and are initiated
best practices are discussed to mitigate the risks associated with by exploiting vulnerabilities in the server software.
client-side attacks.
II. OVERVIEW OF THE MOST COMMON ATTACKS ON WEB
Subsequently, the focus shifts to server-side attacks,
APPLICATIONS
encompassing Server-Side Request Forgery (SSRF), SQL injection
(SQLi), and Remote Code Execution (RCE). The article examines A. Types of client-side attacks: description, example,
the intricacies of each attack, outlining their working principles and method of protection
highlighting real-world instances where these vulnerabilities have
been exploited. The consequences of server-side attacks on system Client-side attacks take place in a user's browser or device
availability, data integrity, and overall security posture are explored, and can be initiated by a user unintentionally visiting a
emphasizing the need for robust countermeasures. Comprehensive malicious website or clicking on a malicious link. An attacker
protection strategies, such as input validation and secure coding can use vulnerabilities in a web application to inject malicious
practices, are recommended to defend against server-side attacks code, steal cookies, or redirect a user to a malicious site.
effectively.
Client-side attacks target an end user's computer or device
In the concluding section, a comparative analysis of client-side by exploiting vulnerabilities in the browser, plug-ins, or other
and server-side attacks is presented, examining various aspects software used to access a web application. The most common
including target, effects, severity, and prevention. This comparison client-side attacks are Cross- Site attacks. Scripting (XSS),
allows readers to understand the distinctive characteristics and risks Cross- Site attacks Request Forgery (CSRF) and Clickjacking
associated with each attack type. Furthermore, it provides insights attacks [1].
into the importance of adopting a multi-layered security approach to
safeguard web applications comprehensively. The article 1. Cross- site Scripting (XSS)
emphasizes the significance of proactive security practices, such as
regular security assessments, patch management, and security Cross Site Scripting (XSS) is a type of client-side attack in
awareness training, to prevent and mitigate the impact of both client- which an attacker injects malicious code into a web page,
side and server-side attacks. which is then executed by the user's web browser. This allows
an attacker to steal sensitive data or perform actions on behalf
In summary, this scientific review article provides a of the user. XSS attacks can be prevented by using input
comprehensive analysis of client-side and server-side attacks in the validation and sanitization techniques, as well as encoding the
context of web applications. By exploring their descriptions, output [2] .
providing real-world examples, and discussing effective prevention
methods, this article aims to enhance the understanding of security Example: A typical example of an XSS attack is a phishing
practitioners, developers, and organizations, enabling them to attack in which an attacker sends a link to a user that is
strengthen the security posture of their web applications and protect vulnerable to Cross - Site Scripting site attack. When a user
against the evolving threat landscape. clicks on a link, malicious javascript is executed code that
Keywords: Web application security, Client-side attacks, intercepts the user's cookies and passes them to the attacker.
Server-side attacks, Comparative analysis With these cookies , an attacker is able to freely connect to a
user's session without knowing the user's login or password.
- 150 -
2. Cross- site Request Forgery (CSRF) 3. Remote Code Execution (RCE)
Cross Site Request Forgery (CSRF) is a type of client-side Remote Code Execution (RCE) is a type of server-side
attack in which an attacker sends a request to a web attack in which an attacker uses a vulnerability in server
application on behalf of a user without the user's knowledge software to execute malicious code remotely. This allows an
or consent. This allows an attacker to perform actions on attacker to take control of the server and access or modify
behalf of the user, such as changing the user's password or sensitive data. RCE attacks can be prevented with modern
committing a transaction. CSRF attacks can be prevented by software and strong access control.
using anti-CSRF tokens and by applying strong access control
[3] . Example. A typical example of an RCE attack is a web
application that launches a vulnerable version of a software
Example. A common example of a CSRF attack is a library. As an example, last year's Log4j vulnerability , also
malicious ad that, when clicked, sends a request to the user's known as CVE- 2021-44228 , is a serious vulnerability in the
bank account to transfer money to the attacker's account. Log4j logging library, which is used in many applications and
servers around the world. The Log4j vulnerability manifests
3. Clickjacking itself in the fact that an attacker can pass a specially
Clickjacking is a type of client-side attack in which an constructed string in the " message " field of a log entry that
attacker tricks a user into clicking on a hidden or disguised contains a link to malicious code. When processing this string,
button or link on a web page. This allows an attacker to the Log4j library downloads and executes malicious code,
perform actions on behalf of the user, such as posting to the which can lead to remote code execution on the machine
user's social media account. Clickjacking attacks can be where the application is running using the affected version of
prevented by implementing clickjacking protection headers Log4j. This vulnerability is especially serious because the
and using frame blocking scripts [4] . Log4j library is used in many critical systems such as financial
institutions, government organizations and large corporations,
Example. A typical example of a clickjacking attack is a which makes it especially attractive to hackers. In addition,
fake login page that looks like a legitimate website but is the Log4j vulnerability is a "chain reaction" - if one system
actually an overlay in disguise. When a user enters their login component uses a vulnerable version of Log4j, this can lead
credentials, the attacker steals them . to a compromise of the entire system, even if other
B. Types of server-side attacks: description, example, components do not use the vulnerable library [7].
method of protection
III. COMPARISON OF ATTACKS: TARGET, EFFECTS , SEVERITY ,
Server-side attacks target a web application server by PREVENTION
exploiting vulnerabilities in the application code, operating
system, or web server software. The most common server-side While both client-side and server-side attacks can have
attacks are Server-Side Request attacks. Forgery (SSRF), SQL serious consequences, they differ in their purpose,
Injection and Remote Code Execution (RCE) attacks. consequences, severity, and methods of prevention. Client-
side attacks target users' browsers and can lead to the theft of
1. Server Side Request Forgery (SSRF) user data or the installation of malware. Server-side attacks
target the web application server or backend code and can
Server Side Request Forgery (SSRF) is a type of server
result in the theft of sensitive data or complete system
side attack in which an attacker sends a request from a web
compromise.
application to another server without the server's knowledge
or consent [5] . This allows an attacker to access sensitive data In terms of severity, server-side attacks tend to be more
or perform actions on the target server. SSRF attacks can be serious as they can compromise the entire system and cause
prevented by using input validation and sanitization widespread damage. However, client-side attacks can also be
techniques and by implementing strong access controls. serious if they result in the theft of sensitive data or the
installation of malware.
Example. A typical example of an SSRF attack is a web
application that allows users to upload files, which are then TABLE I. COMPARISON OF CLIENT-SIDE AND SERVER-SIDE ATTACKS
processed by the server. An attacker can upload a malicious
file containing a malicious request, if configured incorrectly, Attack Target Impact Severity Prevention
the server will process the file and execute the malicious Use input
request contained in the file. validation,
sanitization,
2. SQL injection ( SQLi ) and output
Theft of
encoding to
confidential
SQL injection ( SQLi ) is a type of server-side attack in data,
prevent
which an attacker injects malicious SQL code into a web malicious
User’s unauthorized
application's database, which is then executed by the server Medium code from
XSS web actions on
to High being
[6] . This allows an attacker to access, modify or remove browser behalf of the
injected. Use
sensitive data from the database. SQLi attacks can be user,
a Content
modification of
prevented using prepared statements, parameterized queries, output data
Security
and input validation methods. Policy (CSP)
to restrict
Example. A typical example of an SQLi attack is a login script
form that accepts user input without validation or sanitization. execution
An attacker can inject malicious SQL code into the username
or password field that is executed by the server and bypass
authentication.
- 151 -
TABLE I. (CONTINUED) [5] Portswigger: Web Security Academy // Server-Side Request Forgery.
– URL: https://portswigger.net/web-security/ssrf
Attack Target Impact Severity Prevention
[6] Portswigger: Web Security Academy // SQL Injection. – URL:
Use CSRF https://portswigger.net/web-security/sql-injection
tokens to
[7] UpGuard // Log4Shell: The Log4j Vulnerability Emergency Clearly
validate user
Explained. – URL: https://www.upguard.com/blog/apache-log4j-
requests and
vulnerability
make sure
Unauthorized
they come
actions on
from the
User’s web behalf of the
CSRF High expected
browser user, theft of
source.
confidential
Use the
data
SameSite
attribute on
cookies to
prevent cross-
site requests
Clickjacking
Unauthorized protection
Clickja User’s web actions on headers,
Low
cking browser behalf of a frame
user blocking
scripts
Unauthorized
access to
sensitive data.
An attacker Input
Web can force a validation,
SSRF application server to Medium sanitization,
server connect to strict access
internal control
services in an
organization's
infrastructure.
Prepared
Theft, statements,
Web alteration, or parameterized
SQLi High
application deletion of queries, input
sensitive data validation
methods
Getting full
control over Modern
Web
the server, software,
RCE application High
access to strict access
server
confidential control
data
In conclusion, web application vulnerabilities are a serious
threat that can lead to serious consequences for both users and
owners of web applications. Understanding the most common
vulnerabilities and how they can be exploited through client-
side and server-side attacks is critical to preventing
cyberattacks. It is critical to implement effective prevention
practices and regularly update web applications to minimize
the risk of exploitation. By taking these measures, web
application owners can protect their systems and their users'
data from cyber threats. Ultimately, the responsibility for
securing their systems and ensuring the security of their users'
data rests with web application owners.
REFERENCES
[1] OWASP: Open Web Application Security Project. – URL:
https://owasp.org/Top10/
[2] Portswigger: Web Security Academy // Cross-Site Scripting. – URL:
https://portswigger.net/web-security/cross-site-scripting
[3] Portswigger: Web Security Academy // Cross-Site Request Forgery. –
URL: https://portswigger.net/web-security/csrf
[4] OWASP: Open Web Application Security Project // Clickjacking. –
URL: https://owasp.org/www-community/attacks/Clickjacking
- 152 -
Security of containers and containerization
environments
Amir Abdreev 1* and Assel Nurusheva 2
1
L. N. Gumilyov Eurasian National University, Kazakhstan
2
L. N. Gumilyov Eurasian National University, Kazakhstan
Email: [email protected]
Abstract — Containers have become an increasingly popular In this article, we will delve into the multifaceted nature of
technology for deploying and scaling applications. However, the container security, encompassing both runtime and build-time
security of containers and containerization environments is a security. We will explore the unique attack vectors and
significant concern that needs to be addressed. This review vulnerabilities specific to container environments, including
article analyzes the problem of container security by examining image integrity, container isolation, privilege escalation, and
the various vulnerabilities and threats associated with container escape. Additionally, we will examine the
containerization. effectiveness of existing security measures, such as container
Containers offer many benefits, including portability, image scanning, runtime monitoring, and access control
scalability, and isolation. However, the security of containers mechanisms..
and the underlying infrastructure can be compromised in
various ways, including attacks on the host operating system, II. CONTAINER SECURITY THREAT LANDSCAPE
network attacks, and container breakout attacks. The
A. Overview of security challenges and risks in container
vulnerabilities can be introduced at different stages of the
container lifecycle, from the image creation to the runtime environments
environment. Containerization has revolutionized the deployment and
scalability of applications, enabling organizations to achieve
In this article, we explore the various security challenges
associated with containerization, including the risks associated greater flexibility and efficiency in their software
with the use of public container registries, the importance of development processes. However, the adoption of container
securing the container images, and the need to monitor the environments also introduces a unique set of security
containerized environment continuously. We also discuss the challenges and risks that must be addressed to ensure the
security measures that can be taken to mitigate these risks, such protection of sensitive data and the integrity of applications.
as enforcing security policies, using secure container images,
and deploying network segmentation. 1. Container Isolation
To illustrate the importance of container security, we Container isolation refers to the process of maintaining
provide real-world examples of container security breaches, logical separation between containers running on the same
such as the Docker Hub hack that resulted in the compromise of host, preventing unauthorized access and interference
millions of user accounts and the Kubernetes CVE-2018- between them.
1002105 vulnerability that allowed attackers to bypass
authentication and gain access to the entire cluster.
Example: A security challenge in container isolation is the
“escape to host” vulnerability. If a container is compromised,
This review article aims to provide a comprehensive an attacker may attempt to exploit vulnerabilities in the
understanding of the security challenges associated with container runtime or kernel to break out of the container and
containers and containerization environments. By gain unauthorized access to the host system. This can lead to
understanding the risks and vulnerabilities, organizations can further attacks, such as lateral movement within the host or
take appropriate measures to secure their containerized unauthorized access to other containers.
applications and mitigate potential security breaches.
2. Image and Software Supply Chain
Keywords — containers, containerization environments,
security, vulnerabilities, threats, attacks. The image and software supply chain in container
environments involves the acquisition, verification, and
I. INTRODUCTION deployment of container images and associated software
In recent years, containerization has emerged as a components.
dominant technology in the field of software development and Example: A risk in the image and software supply chain is
deployment, revolutionizing the way applications are the use of compromised or vulnerable container images. If an
packaged and deployed across various platforms. Containers organization relies on untrusted or tampered images from
provide lightweight, portable, and scalable environments that public repositories without proper verification, they may
encapsulate application components and their dependencies, inadvertently introduce malicious code or vulnerable
enabling efficient deployment and management across diverse components into their environment. This can lead to data
infrastructure. However, as container adoption continues to breaches, unauthorized access, or the exploitation of known
soar, ensuring the security of containerized environments vulnerabilities.
becomes paramount. Understanding the intricacies of
container security is crucial for organizations seeking to 3. Container Vulnerabilities
harness the full potential of containerization while
safeguarding their valuable assets and data.
- 153 -
Container vulnerabilities refer to security weaknesses or access to sensitive data or execute arbitrary code within the
flaws within the container runtime, orchestrator, or underlying container.
infrastructure that can be exploited by attackers.
3. Container Misconfigurations
Example: A security challenge related to container
vulnerabilities is the "unpatched container" risk. Containers Misconfigurations are a common cause of security
are often built from base images that may contain outdated breaches in containerized environments. Errors in settings,
software components with known vulnerabilities. If these access controls, or network configurations can lead to
vulnerabilities are not addressed and patched in a timely unintended exposure of sensitive data or services, providing
manner, attackers can exploit them to compromise containers attackers with opportunities to exploit the misconfigured
and gain unauthorized access to sensitive data or the host containers.
system. Example: A misconfigured container might
4. Container Configuration and Orchestration unintentionally expose a management interface or an
unsecured API endpoint, allowing an attacker to gain
Container configuration and orchestration involves the unauthorized access or launch brute-force attacks to discover
proper setup and management of containers, ensuring they weak credentials.
adhere to security best practices and are effectively
4. Privilege Escalation
orchestrated within the container environment.
Example: Misconfigured container security settings can Privilege escalation attacks involve gaining elevated
expose containers to risks, such as "inadequate access privileges within a container to access resources or perform
controls." If containers are not properly configured with actions beyond the intended scope. Attackers exploit
appropriate access controls, attackers can exploit this vulnerabilities in container runtimes or misconfigurations to
misconfiguration to gain unauthorized access to sensitive escalate privileges and potentially compromise the host
resources or execute unauthorized actions within the container system or other containers.
environment. Example: An attacker may exploit a vulnerability in the
5. Container Breakouts container runtime's default configuration, allowing them to
escalate privileges within a container and gain unauthorized
Container breakouts involve attacks or vulnerabilities that access to critical resources or sensitive data.
allow an attacker to escape the confines of a container and gain
unauthorized access to the underlying host system. III. BEST PRACTICES FOR CONTAINER SECURITY
Example: A security risk related to container breakouts is A. Essential Security Practices for Container Deployment
the "kernel vulnerability" exploit. If the kernel running on the and Management
host system has a vulnerability, an attacker who gains control Containerization has revolutionized software development
of a container may be able to exploit that vulnerability to break and deployment, enabling organizations to streamline their
out of the container and gain unauthorized access to the host. processes and improve scalability. However, the rapid
Once inside the host, the attacker can potentially compromise adoption of containers has also introduced new security
other containers or sensitive data. challenges. To mitigate these risks, it is crucial to implement
B. Common Attack Vectors Targeting Containers essential security practices for container deployment and
management.
1. Container Breakout
1. Secure Container Hosts
Container breakout is a critical attack vector where an
attacker attempts to escape the confines of a container to gain The foundation of a secure container deployment lies in
unauthorized access to the host system. This attack usually the security of the underlying host system. It is essential to
leverages vulnerabilities in the container runtime, harden the host operating system by applying security patches,
misconfigurations, or insecure container images. Once disabling unnecessary services, and following best practices
escaped, the attacker can potentially compromise other for system hardening. Employing secure configuration
containers or the underlying host, leading to widespread baselines and utilizing security frameworks like CIS
damage. benchmarks can significantly enhance the security posture of
container hosts.
Example: A container breakout attack could involve
exploiting a kernel vulnerability within a container runtime, 2. Isolation and Segmentation
allowing the attacker to execute arbitrary code on the host and Proper isolation and segmentation are critical for
compromise the entire infrastructure. container environments. Containers should be isolated from
2. Image Vulnerabilities each other and from the host system to prevent unauthorized
access and limit the impact of potential breaches. Utilizing
Container images, which serve as the foundation for container orchestration platforms that enforce strict network
running containers, can harbor various vulnerabilities that segmentation and access control policies is recommended.
attackers exploit to compromise containerized environments. Network policies, such as implementing firewalls and network
These vulnerabilities may include outdated software segmentation using virtual LANs (VLANs) or software-
packages, weak configurations, or even malicious code defined networking (SDN) techniques, can provide additional
inserted into the image. layers of isolation.
Example: An attacker could craft a malicious container 3. Image Provenance and Integrity
image with a vulnerable version of a web server, allowing
them to exploit known vulnerabilities and gain unauthorized
- 154 -
Ensuring the integrity and provenance of container images and their configurations is crucial to prevent the
images is crucial to prevent the deployment of malicious or deployment of vulnerable or malicious components.
vulnerable components. Organizations should establish secure
image repositories and implement mechanisms to verify the 1. Image Hygiene
authenticity and integrity of images. Employing container Maintaining image hygiene is vital to minimize security
signing mechanisms and digital signatures can help guarantee risks associated with container images. It is essential to use
the trustworthiness of images throughout their lifecycle. trusted base images from reputable sources and regularly
4. Vulnerability Management update them to include the latest security patches. Employing
a secure software supply chain and employing container
Regularly scanning container images and the image vulnerability scanning tools can help identify and
underlying host system for vulnerabilities is essential to remediate vulnerabilities in container images.
identify and remediate potential security weaknesses.
Organizations should adopt vulnerability scanning tools and 2. Minimalistic Images
establish processes to track and update software dependencies Adopting a minimalistic approach when creating container
within containers. Automating vulnerability management images can significantly enhance their security. Including
processes and integrating them into the CI/CD pipeline can only necessary dependencies and components reduces the
help ensure continuous security throughout the container attack surface and minimizes the risk of incorporating
lifecycle. vulnerable or unnecessary software. Organizations should
5. Access Control and Privilege Management strive to build lightweight and purpose-specific images,
eliminating unnecessary binaries, libraries, and services.
Implementing robust access control mechanisms is
critical to prevent unauthorized access to containerized 3. Secure Configuration Management
environments. Organizations should enforce the principle of Implementing secure configurations within container
least privilege, granting containerized applications only the images is crucial to prevent misconfigurations that can lead to
necessary permissions and privileges. Implementing strong security vulnerabilities. Following configuration hardening
authentication mechanisms, such as multi-factor guidelines, such as those provided by CIS benchmarks, can
authentication (MFA), and utilizing centralized identity and help establish a secure baseline for container configurations.
access management (IAM) systems can enhance access Additionally, leveraging automation tools to enforce
control and reduce the risk of unauthorized access. consistent configurations and performing regular
6. Logging and Monitoring configuration audits can enhance security.
Effective logging and monitoring are essential for 4. Secrets Management
detecting and responding to security incidents in container Effectively managing secrets, such as passwords, API
environments. Organizations should implement centralized keys, and certificates, within containerized applications is
logging solutions and enable detailed logging within essential to prevent unauthorized access to sensitive
containers to capture relevant security events. Employing information. Organizations should adopt secure secrets
container-specific monitoring tools and integrating them with management practices, such as leveraging vault systems or
security information and event management (SIEM) systems container-specific secret management solutions. Encrypting
can facilitate proactive threat detection and incident response. sensitive data at rest and in transit further strengthens the
7. Container Orchestration Security security of secrets.
Container orchestration platforms, such as Kubernetes, 5. Container Runtime Security
introduce additional security considerations. Organizations Securing the container runtime environment is critical to
should implement secure configuration settings for protect containerized applications during execution.
orchestrators, including network policies, RBAC (Role-Based Employing isolation mechanisms, such as leveraging Linux
Access Control), and secure API server access. Regularly kernel security features like namespaces and cgroups,
updating and patching the orchestration platform, along with enhances container runtime security. Additionally, enabling
monitoring for any security advisories, is crucial to mitigate secure runtime configurations, such as seccomp profiles and
emerging risks. SELinux/AppArmor policies, helps restrict the capabilities
and actions of containers, preventing potential exploits.
8. Incident Response and Disaster Recovery
Preparing for security incidents and having a well- 6. Immutable Infrastructure
defined incident response plan is vital for container Implementing immutable infrastructure principles for
environments. Organizations should establish incident containers can improve security by ensuring that containers
response processes specific to containers, including are not modified during runtime. Instead of making changes
procedures for identifying, containing, and remediating to running containers, organizations should adopt practices
security breaches. Additionally, implementing disaster that create new containers with updated configurations or
recovery mechanisms, such as data backups and replication patches. This approach reduces the risk of unauthorized
strategies, can minimize the impact of security incidents and modifications or persistence of malicious code.
ensure business continuity.
7. Continuous Integration and Delivery (CI/CD)
B. Implementing secure container images and Security
configurations
Integrating security into the CI/CD pipeline is crucial for
Container images serve as the building blocks of ensuring the secure deployment of container images.
containerized applications. Ensuring the security of these
- 155 -
Organizations should include security checks, such as [5] Lipatova S.E., Belov Yu.S., “Cybersecurity practices in Kubernetes”,
vulnerability scanning and image integrity verification, as part E-Scio, pp. 489-497, 2022.
of the CI/CD process. Automating security testing and [6] RED HAT: Kubernetes security concerns escalate, 2022. – URL:
https://www.osp.ru/os/2022/03/13056287
incorporating it into the pipeline ensures that security
[7] Semenov A.M., Khvostov V.A., “Docker and Kubernetes as an object
measures are consistently applied throughout the software of implementation of information security threats and protective
development lifecycle. measures”, Voronezh: Voronezh State University of Engineering
Technologies, pp. 54-55, 2022.
8. Testing and Validation
Thoroughly testing and validating container images and
configurations is essential to identify security weaknesses and
ensure the reliability of containerized applications.
Implementing automated testing frameworks and security-
focused testing tools, such as static analysis and dynamic
scanning, can help identify vulnerabilities, misconfigurations,
and potential runtime issues. Regularly validating images
before deployment helps ensure the integrity and security of
containerized environments.
IV. CONCLUSION
In conclusion, the rapidly evolving landscape of container
security poses significant challenges and risks that must be
addressed proactively by organizations. As containers
continue to gain popularity due to their scalability and
efficiency, it becomes imperative for businesses to prioritize
and implement robust security measures to safeguard their
containerized environments.
The overview of security challenges and risks in container
environments highlighted the vulnerabilities that can be
exploited by malicious actors. From insecure container
configurations to vulnerable images and the ever-present
threat of container breakouts, organizations must remain
vigilant and stay ahead of potential attacks.
Common attack vectors targeting containers were
discussed to shed light on the diverse techniques employed by
attackers. From exploiting unpatched vulnerabilities to
orchestrating container escapes, these attack vectors
underscore the need for comprehensive security strategies that
encompass all aspects of container deployment and
management.
To mitigate these risks, organizations must adopt essential
security practices for container deployment and management.
By implementing these best practices, organizations can
bolster their container security and minimize the likelihood of
successful attacks. Security should be integrated throughout
the container lifecycle, from the development phase to
deployment and ongoing management. Ultimately, container
security is an ongoing process that requires a proactive and
holistic approach.
REFERENCES
[1] Hikov S.P., “MODEL FOR SELECTION AN EFFECTIVE
MEAN OF SCANNING CONTAINERS IMAGES IN KUBERNETES
INFRASTRUCTURE”, pp. 249-253, 2019.
[2] Kosenkov V.V., Bogdanov A.V., “Trade-off between security and
performance of Kubernetes container runtimes”, Moscow: Limited Liability
Company “Publishing House ALEF”, pp. 145-153, 2022.
[3] Ghadeer D., Vorobeva A.A., “Kubernetes monitoring with
Prometheus for security purposes”, Ufa: Limited Liability Company
“Scientific Publishing Center “Vestnik Nauki”, pp. 44-50, 2022.
[4] Gurbatov G.O., Panichev A.D., Ushakov I.A., “Securing
Kubernetes”, St. Petersburg: St. Petersburg State University of
Telecommunications. prof. M.A. Bonch-Bruevich, pp. 282-286, 2021.
- 156 -
Research and development of two-factor
authentication methods using neural networks
Temirkhan Darbek Timur Umarov Sabina Rakhmetulaeva
MSc student of the Department of PhD, Associate Professor of the Associate Professor of the
Information Systems and Cybersecurity Department of Computer Science Department of Information Systems
Almaty university of power engineering and British Management University International University of
telecommunications Tashkent, Uzbekistan Information Technologies
Almaty, Kazakhstan [email protected] Almaty, Kazakhstan
[email protected] [email protected]
Zhanna Suimenbayeva Assel Mukasheva
PhD student of Satpayev University PhD, Associate Professor of the
Almaty, Kazakhstan Department of Information Systems and
[email protected] Cybersecurity
Almaty University of Power Engineering
and Telecommunications
Almaty, Kazakhstan [email protected]
orcid.org/0000-0001-9890-4910
Abstract—This article discusses the development of a two- However, facial recognition has some drawbacks that can
factor authentication method using neural networks to enhance cause authentication failure or increase the risk of data
digital security. The aim of this methodology is to create a leakage. Firstly, high quality and accurate cameras and
secure and convenient authentication system that does not rely software are required to read and recognise faces. Secondly,
on passwords. Current authentication methods such as this method can be fooled by the use of a photo or video of
password and token authentication are not always secure, and the user's face [5].
new methods that combine security and usability are needed.
To achieve this, Python neural network and image processing To eliminate these problems, a second factor can be
libraries including Keras, TensorFlow, OpenCV, etc. were added, the one-time password (TOTP). TOTP is an
used. The proposed two-factor authentication was compared to abbreviation of Time-based One-Time Password [6]. Such a
other security methods like password hashing algorithms and password is generated by a special application and is only
one-time passwords, and the experiment showed that it valid for a certain period of time (e.g. 30 seconds). The use
provided higher accuracy and performance. The result of the of TOTP increases the security level of authentication
experiment is the use of a combination of two authentication because the one-time password prevents possible attacks
factors: face scanning and TOTP (Time-based One-Time using password hijacking or cracking [7].
Password). Therefore, based on the analysis, a two-factor
authentication system was developed that uses face scanning as Using a telegram bot to generate TOTP allows the
the first factor and TOTP as the second factor, which provides creation of individual keys for each user and makes the
a high level of security and usability. The implementation was process available on any device with internet access. This
done using the Python programming language and Pycharm increases the level of security and protection of personal
development environment, and the neural network was trained data. Adding TOTP to the basic facial recognition method
on data sets containing face images and TOTP codes. helps reduce the likelihood of failed authentication and
significantly increases data security, making two-factor
Keywords—Neural networks, Biometric authentication, Deep authentication based on these methods an effective way to
learning, Digital security, Identification protect users' sensitive data. This method can be successfully
I. INTRODUCTION applied in different fields, from the banking sector to social
media and online shopping.
In today's connected world, the security of transmitted
data is a critical issue and the primary means of security is The results of the research have shown that the proposed
the authentication process [1]. In this case, the primary authentication methods provide a high level of information
method of authentication is usually password [2]. However, protection against unauthorized access. Thus, new two-factor
using this method alone may not be sufficient to guarantee authentication methods are becoming increasingly popular in
data security as there are many ways to break it (e.g. packet the modern world, and are a reliable tool for protecting users'
hijacking or social engineering). Adding a second factor in personal information on the Internet.
the form of something that only the user knows (e.g. the
answer to a question) or a device that only the user has can II. BACKGROUD
increase the security of the system. Authentication can take place in different ways,
depending on which authentication methods are used.
One possibility for the second factor is face recognition Usually authentication is done with a username and
[3]. This authentication method can be used on smartphones, password, but other methods such as biometric data
tablets, PCs and other devices that have a camera. It works (fingerprint, facial scanner), one-time codes, keys, etc. can
on the principle of matching a photo of a user's face with a also be used.
pre-stored version of the same face in a database [4].
In the case of username and password [8] the user has to
enter their username and password in the appropriate fields
- 157 -
on the page or application and then the system checks that Figure 2. Authentication methods from SFA to MFA
the data entered is correct and allows the user to access
secure resources if the username and password are correct. III. RELATED WORKS
An article [24] presents a multi-factor authentication
By using biometrics [9], the authentication process can method that combines face verification, one-time password
be automated. For example, to verify a fingerprint, the user and hand gesture recognition. The authors claim that this
simply touches the fingerprint sensor on the device, which combination of factors provides a more reliable and secure
compares their fingerprint with that stored in the system and, authentication method than traditional one-factor methods.
if they match, allows access to secure resources. The results of the study showed that the face verification
One-time codes are one-time passwords generated by the system has an average false positive rate of 1.94% and an
system and sent to the user by SMS or other means. The user average true positive rate of 67.7%. The hand gesture
must enter the code received to complete the authentication recognition system shows an accuracy and recall rate of
process [10]. 96%. The research paper [25] presents a deep learning-based
face recognition system that achieves state-of-the-art results
The authentication process remains the primary means of on the Labeled Faces in the Wild (LFW) dataset. The system
protecting against unauthorised access to devices and uses a convolutional neural network (CNN) to extract
sensitive applications, both offline and online (Fig. 1). In the features from facial images and then uses a support vector
past, the authenticity of transactions was confirmed machine (SVM) to classify the images into different
physically, for example by applying a wax seal [11]. identities. The system achieves an accuracy of 99.63% on the
However, as civilisation has evolved, it has become apparent LFW dataset, which is significantly higher than the accuracy
that sender identification is not always sufficient to ensure of previous systems. In [26], a face recognition system (FRS)
security across the global network [12]. that runs on a cloud computing platform and uses a deep
learning algorithm to extract features from facial images has
been proposed. The system compares these features with a
database of known users, which allows it to authenticate
users with high accuracy. The authors conducted
experiments to evaluate the performance of their FRS
system. They compared the performance of cloud-based FRS
with traditional FRS implemented on a desktop computer
and obtained results that showed significantly faster and
Figure 1. Authentication examples more accurate performance of cloud-based FRS. In [27], they
propose a new two-factor authentication scheme (2FA) for
Initially, only one form of authentication, called Single- the Hyperledger Fabric blockchain that uses the TOTP
Factor Authentication (SFA), was used to verify the identity algorithm to generate a 6-digit authentication code on the
of users. [13] This technique became popular due to its user's device. This code is used to authenticate the user to the
simplicity and user-friendliness, such as entering a password blockchain network. The authors evaluated the effectiveness
or PIN. However, it is the weakest authentication method of their 2FA scheme and found that it is significantly faster
[14] as an attacker can compromise the account by learning and provides a higher level of security than the existing 2FA
the password or conduct a dictionary [15], rainbow table [16] scheme for the Hyperledger Fabric blockchain.
or social engineering attack. [17] Two-factor authentication
(2FA) [18] has been proposed to improve security, which IV. MATERIALS AND METHODS
uses several factors such as password and smart card [19].
Today, three types of factor groups are available [20]: A. Facial recognition .
knowledge factor, ownership factor and biometric factor, Facial recognition is a type of biometric technology that
which help to link the user's identity to the credentials. uses artificial intelligence and machine learning algorithms
[28] to identify individuals based on their unique facial
Further, multi-factor authentication (MFA) has been
features. The process of facial recognition involves several
proposed, which uses more than two categories of credentials
steps, including input frames, pre-processing, face
[21] to provide a higher level of security and continuous
recognition, feature extraction, feature matching by neural
protection of computing devices as well as other critical
nets, face database, face/user identification, and
services against unauthorised access. It is mainly based on
authentication.
biometric techniques [22], which allow automatic
identification of individuals based on their behaviour. This The first step in the facial recognition process is input
significantly increases the level of security as users have to frames. These are images or video frames that contain the
present evidence of their identity [23]. Figure 2 shows the faces of individuals that need to be identified. Once the input
evolution of the authentication methods considered in this frames have been captured, they are pre-processed to
context. enhance the quality of the image and remove any noise or
distortion.
Next, the face recognition stage begins. This involves
using pattern recognition algorithms to identify specific
features of the face such as the eyes, nose, mouth, and chin.
These features are then extracted from the image and
compared to a database of known faces.
The feature extraction stage uses complex algorithms to
extract key features of the face from the input frames. This is
- 158 -
done by identifying unique points on the face, such as the Finally, the similarity between the input face and the
corners of the eyes or the shape of the jawline, and creating a training faces is measured using a distance metric, such as
mathematical representation of these features that can be Euclidean distance or Mahalanobis distance. The formula
used for comparison in the next stage. for Euclidean distance is:
Once the features have been extracted, they are matched D = ‖𝑃 − 𝑄‖ (3)
against known features in a face database. This is where the
neural net comes into play. A neural net is a type of machine The total time complexity of the Hadoop job can be
learning algorithm that can be trained to recognize patterns in calculated as follows:
data. The neural net is trained to recognize the features of a
where D is the Euclidean distance between the input face
particular individual's face and match them against the
features extracted from the input frames. P and the closest training face Q in the eigenspace.
If a match is found between the extracted features and B. TOTP
those in the face database, the user's identity is determined. TOTP stands for Time-based One-Time Password. It is a
This process is known as face/user identification and is type of two-factor authentication that generates a unique one-
typically used for security applications, such as access time password (OTP) for each login attempt. TOTP works by
control systems or border control. combining something the user knows (their username and
password) with something they have (a physical or virtual
Finally, once the user's identity has been established, device that generates OTPs).
authentication can occur. Authentication refers to the
verification of a user's identity to ensure that they are who The TOTP algorithm is based on a shared secret key,
they claim to be. This is typically done by comparing the which is known by both the user and the service provider.
extracted features against a set of pre-registered user features This key is used to generate a new OTP at regular intervals,
to confirm that the user is authorized to access the system or typically every 30 seconds. The OTP is synced between the
service. (Fig. 3) user's device and the service provider using a clock that is
synchronized between the two.
To use TOTP, the user first sets up their account with the
service provider and registers their device. During
registration, the user is provided with a unique secret key
which they must keep secure. The user then downloads an
app or uses a physical device which can generate OTPs
based on the TOTP algorithm.
Figure 3. How face recognition works When the user logs in to the service provider's website or
application, they are prompted to enter their username and
Face recognition is a complex process that involves password as usual. They are then asked to enter the current
several steps, including face detection, feature extraction, OTP generated by their device. The service provider
and face matching. There are different mathematical models validates the OTP based on the shared secret key and allows
used for face recognition, but one of the most common the user access if the OTP is valid.
models is the eigenface model.
One advantage of TOTP is that it does not require a
The eigenface model represents faces as vectors in a network connection during the login process, making it more
high-dimensional space and uses principal component reliable than other forms of two-factor authentication like
analysis (PCA) to extract the most significant features or SMS-based OTPs. Furthermore, since the OTP expires after
eigenfaces. These eigenfaces are the eigenvectors of the a short period of time, it is less vulnerable to replay attacks.
covariance matrix of the training set of faces. (Fig. 4)
The mathematical formula for PCA can be expressed as
follows:
X = U * S * 𝑉𝑇 (1)
where X is the data matrix containing the training
images, U is the matrix of eigenvectors of the covariance
matrix, S is the diagonal matrix of singular values, and 𝑉 𝑇 is
the transpose of the matrix of eigenvectors.
To recognize a new face, the eigenface model projects
the face onto the eigenspace by calculating the coefficients Figure 4. How TOTP works
of the face with respect to each eigenface. The formula for
The TOTP algorithm combines a secret key, which is
this projection is:
known only to the user and the service provider, with the
P = 𝑈 𝑇 * (F - M) (2) current time to generate a unique one-time password that
changes every 30 seconds.
where P is the vector of coefficients, F is the input face
image, M is the mean face computed from the training set, The formula for generating a TOTP code can be written
and 𝑈 𝑇 is the transpose of the eigenvector matrix. as follows:
- 159 -
TOTP = HOTP(K, T) (4) Thus, using two-factor authentication via Telegram bot
with TOTP and facial recognition provides a higher level of
where TOTP is Time-Based One-Time Password, HOTP
security than simple password authentication. This is due to
is HMAC-based One-Time Password, K is Secret key
the fact that to gain access, one must not only know the
shared between the user and the service provider and T is
password, but also have access to the mobile device and
Current time in Unix time format, i.e., the number of
undergo facial scanning. (Fig. 5)
seconds since January 1, 1970, at 00:00:00 UTC.
The HOTP function is defined as: V. PERFORMANCE TESTS AND RESULTS
The Python programming language was chosen to create
HOTP(K, C) = Truncate(HMAC-SHA-1(K, C)) (5) a program for two-factor authentication. This language is
where HMAC-SHA-1 is a keyed hash function that takes known for its simplicity and usability, as well as extensive
as input the secret key K and a counter value C. Truncate is library capabilities, which makes it very popular in the
a function that extracts a specific number of digits from the software development environment.
HMAC-SHA-1 output. For example, if we want a six-digit One of the main libraries used was OpenCV (Open
TOTP code, we would extract the leftmost six digits from Source Computer Vision Library), which provides a wide
the HMAC-SHA-1 output. range of tools for working with images and video, including
face recognition functionality.
C. Two-factor authentication using facial recognition and
TOTP. To implement the telegram bot, the pyTelegramBotAPI
Two-factor authentication (2FA) is a method of library was used, which provides a simple and convenient
protecting user accounts and data that requires two interface for working with the Telegram programming
confirmations: the first confirmation is what the user knows interface. Thanks to this library, it was possible to generate a
one-time password (TOTP) through the telegram bot, which
or has (such as a password), and the second confirmation is
provides an additional level of security when logging in or
what the user has (such as the device on which they are accessing sensitive information.
registered).
Additionally, several other libraries were used in the
In this case, the first factor is facial recognition, which creation process, such as pandas, numpy, pytorch, requests
can be implemented with special software that uses neural and pillow. Pandas and numpy allowed us to work with large
networks. The user will have to go through a procedure of amounts of data, and pytorch was used to train a face
scanning his face so that the system can create his recognition model. Requests was used to interact with the
"biometric template" which will be used for further Telegram API, while pillow allowed us to manipulate images
authentication. in the recognition process and store them in the right format.
The second factor in this case is a time-limited one-time A. Two-factor authentication process based on facial
password (TOTP) that will be issued via a telegram bot. The recognition and TOTP
TOTP is generated based on the secret key provided to the The first authentication factor is face recognition. The
user when setting up 2FA. A special mobile app can be used app uses the device's camera to scan the user's face and store
to generate the TOTP, or bots that work in messengers such it in a database. In order to add their face to the database, the
as Telegram. user must press the 'Sample' button and the app will start the
When attempting to log in, the user will have to go face scanning process. When the scanning is complete, the
through facial recognition, after which a TOTP will be sent user's face is saved to the database for later use in the
to them via a Telegram bot. The user will have to enter this authentication process. A message will be displayed stating
password to confirm their identity and complete the that the user's face has been successfully saved to the
authentication process. database. (Fig. 6)
Figure 6. Face recognition in the first authentication factor application:
Figure 5. Authentication process scanning process and saving the user's face into the database
- 160 -
Once a person has been added to the database, the DISCUSSION OF RESULTS
software must be trained on the basis of the saved faces. Creating a strong authentication system is a key
Training is done by pressing the "Training" button. At this challenge in information security. In recent years, there has
point the software analyses the saved faces and uses them to been an increasing focus on two-factor authentication, which
train the facial recognition algorithms. This allows the improves security by using two or more factors to
software to improve the accuracy of face recognition during authenticate a user's identity. In this article we will look at
the authentication process. If training is successful, the creating two-factor authentication based on facial recognition
message "Image trained" appears on the screen. (Fig. 7) and a TOTP token that is generated via a telegram bot.
The first thing to do is to decide on the first
authentication factor. Face recognition is a sufficiently strong
method of user identification that can be used in conjunction
with another authentication factor. Face detection can be
performed either on the user's device or on the application
server. If face recognition is used on the server, the user's
data can be further protected, as the user will have to be
authenticated before accessing the data on the server.
The second authentication factor is a TOTP token, which
Figure 7. Training of facial recognition algorithms after addition to the is generated through a telegram bot. This method of user
database
authentication is convenient and secure as the generation of
Next, the application uses TOTP as the second the code takes place on the user side and not on the
authentication factor. After successfully recognising the application server. This avoids the transmission of
user's face, the app sends the user's ID to the telegram bot to confidential information through open channels of
generate a temporary verification code. This means that each communication. Moreover, using a telegram bot to generate
user gets his own unique verification code and no one else a TOTP code facilitates the authentication process and makes
can use it instead. The user has to verify that his ID is correct it more convenient for users.
and confirm his identity. If the user enters the wrong ID, they
will not be able to pass the authentication process. (Fig. 8) However, when creating this authentication system,
possible risks and limitations need to be taken into account.
For example, facial recognition may be less reliable in case
of major changes in the user's appearance, such as a change
of hairstyle or growth of a beard. There may also be
problems with facial recognition accuracy in low light or
noise conditions.
Additional methods of user authentication, such as
fingerprint scanning or the use of voice control, can be used
to increase the reliability of the authentication system. In
addition, facial recognition needs to be carefully configured
Figure 8. Using TOTP as a second authentication factor
and TOTP code transmission via a telegram bot needs to be
After entering the ID, the user receives a verification secured.
code on their Telegram. The code is valid for 60 seconds and
So, creating two-factor authentication based on facial
the user must enter it within this time. If the verification code
recognition and a TOTP token that is generated through a
is correct, the software will successfully authenticate the
telegram bot is an effective way to protect personal data and
user. (Fig. 9)
privacy. However, to achieve maximum reliability and
security, the risks and limitations of this authentication
system must be considered and additional user authentication
methods must be used.
CONCLUSION
In conclusion, creating two-factor authentication using
facial recognition and TOTP code generation via a telegram
bot is an effective way to protect data and secure access to
resources and services. Facial recognition is a sufficiently
Figure 9. Entering the verification code during the second authentication strong method of user identification that can be used in
factor
conjunction with another authentication factor.
Overall, two-factor authentication using facial
A Telegram bot that generates a TOTP code allows a user
recognition and TOTP provides a high level of security for
to retrieve the authentication code on their mobile device,
user data. The software stores user face data in a database
which is convenient and secure as the code data is not
that is used to train facial recognition algorithms. This allows
transmitted over open channels. This also facilitates the
the software to recognise users more accurately during the
authentication process and makes it more user-friendly.
authentication process. When TOTP is used, each user
receives their own unique verification code, making the The OpenCV library was chosen to develop a face
authentication process even more secure. recognition program, allowing for efficient image and video
processing. An application capable of recognizing faces in
- 161 -
images and videos and storing them in a database was [12] Danny T. MFA (Multi-Factor Authentication) with Biometrics. 2017.
created. Available online: https://www. bayometric.com/mfa-multi-factor-
authentication-biometrics/ (accessed on 4 Jaurnay 2018).
Tests were conducted that confirmed the effectiveness of [13] Konoth, R.K.; van der Veen, V.; Bos, H. How anywhere computing
the method. The system successfully recognized users' faces just killed your phone-based two-factor authentication. In Proceedings
and generated unique TOTP codes for each authorization, of the International Conference on Financial Cryptography and Data
Security, Christ Church, Barbados, 22–26 February 2016; Springer:
which provided a high level of security and protection Berlin, Germany, 2016; pp. 405–421.
against unauthorized access. [14] Dasgupta, D.; Roy, A.; Nag, A. Toward the design of adaptive
Thus, the implementation of two-factor authentication, selection strategies for multi-factor authentication. Comput. Secur.
2016, 63, 85–116.
including facial recognition and the use of TOTP code via
[15] Wang, D.; Wang, P. Offline dictionary attack on password
Telegram bot, is an effective way to protect access to various authentication schemes using smart cards. In Information Security;
systems and services. This system can be used in various Springer: Berlin, Germany, 2015; pp. 221–237.
areas such as online banking, e-mail and other services [16] Ah Kioon, M.C.; Wang, Z.S.; Deb Das, S. Security analysis of
where a high level of security is required. MD5 algorithm in password storage.
[17] Appl. Mech. Mater. 2013, 347, 2706–2711.
REFERENCES [18] Heartfield, R.; Loukas, G. A taxonomy of attacks and a survey of
[1] Y. Guo et al., "A Security Protection Technology Based on Multi- defence mechanisms for semantic social engineering attacks. ACM
factor Authentication," 2022 IEEE 2nd International Conference on Comput. Surv. (CSUR) 2016, 48, 37.
Mobile Networks and Wireless Communications (ICMNWC), [19] Schneier, B. Two-factor authentication: Too little, too late. Commun.
Tumkur, Karnataka, India, 2022, pp. 1-5, doi: ACM 2005, 48, 136.
10.1109/ICMNWC56175.2022.10032006.
[20] Sun, J.; Zhang, R.; Zhang, J.; Zhang, Y. Touchin: Sightless two-factor
[2] N. Pato. Josep & I. Millett. Lynette, “Biometric Recognition, authentication on multi-touch mobile devices. In Proceedings of the
Chaleenges and Opportunities” NATIONAL RESEARCH Conference on Communications and Network Security (CNS), San
COUNCIL, 2010. Francisco, CA, USA, 29–31 October 2014; pp. 436–444.
[3] Adjabi, I.; Ouahabi, A.; Benzaoui, A.; Taleb-Ahmed, A. Past, Present, [21] Harini, N.; Padmanabhan, T.R. 2CAuth: A new two factor
and Future of Face Recognition: A Review. Electronics 2020, 9, authentication scheme using QR-code. Int. J. Eng. Technol. 2013, 5,
1188. https://doi.org/10.3390/electronics9081188 1087–1094.
[4] Marti, Ni Wayan. (2010). Pemanfaatan GUI dalam Pengembangan [22] Scheidt, E.M.; Domangue, E. Multiple Factor-Based User
Perangkat Lunak Pengenalan Citra Wajah Manusia Menggunakan Identification and Authentication. U.S. Patent 7,131,009, 31 October
Metode Eigenfaces. Seminar Nasional Aplikasi Teknologi Informasi 2006.
2010 (SNATI 2010) ISSN: 1907-5022 Yogyakarta, 19 June 2010.
[23] Frank, M.; Biedert, R.; Ma, E.; Martinovic, I.; Song, D. Touchalytics:
[5] Li, Lei. Feng, Xiaoyi. Xia, Zhaoqiang. Jiang, Xiaoyue. Hadid On the applicability of touchscreen input as a behavioral biometric for
Abdenour. (2018). Face Spoofing Detection with Local Binary continuous authentication. IEEE Trans. Inf. Forensics Secur. 2013, 8,
Pattern. Preprint submitted to Journal of Visual Communication and 136–148.
Image Representation.
[24] Aleluya E. R. M., Vicente C. T. Faceture ID: face and hand gesture
[6] TOTP: Time-Based One-Time Password Algorithm https://www.rfc- multi-factor authentication using deep learning //Procedia Computer
editor.org/rfc/rfc6238.html Science. – 2018. – Т. 135. – С. 147-154.
[7] Plata I. T., Calpito J. L. Application of time-based one time password https://doi.org/10.1016/j.procs.2018.08.160
(TOTP) algorithm for human resource e-leave tracking web app [25] M. Zulfiqar, F. Syed, M. J. Khan and K. Khurshid, "Deep Face
//International Journal of Scientific and Technology Research. – 2020. Recognition for Biometric Authentication," 2019 International
– Т. 9. – №. 3. – С. 4070-4077. Conference on Electrical, Communication, and Computer
[8] Kun, A.L.; Royer, T.; Leone, A. Using tap sequences to authenticate Engineering (ICECCE), Swat, Pakistan, 2019, pp. 1-6, doi:
drivers. In Proceedings of the 5th International Conference on 10.1109/ICECCE47252.2019.8940725.
Automotive User Interfaces and Interactive Vehicular Applications, [26] Pawle A. A., Pawar V. P. Face recognition system (FRS) on cloud
Eindhoven, The Netherlands, 28–30 October 2013; ACM: New York, computing for user authentication //International Journal of Soft
NY, USA, 2013; pp. 228–231. Computing and Engineering (IJSCE). – 2013. – Т. 3. – №. 4. – С.
[9] J. Anand Babu, H. P. Neha, K. S. Babu and R. N. Pinto, "Secure Data 189-192.
Retrieval System using Biometric Identification," 2022 IEEE [27] W. -S. Park, D. -Y. Hwang and K. -H. Kim, "A TOTP-Based Two
International Conference on Data Science and Information System Factor Authentication Scheme for Hyperledger Fabric Blockchain,"
(ICDSIS), Hassan, India, 2022, pp. 1-4, doi: 2018 Tenth International Conference on Ubiquitous and Future
10.1109/ICDSIS55133.2022.9915968. Networks (ICUFN), Prague, Czech Republic, 2018, pp. 817-819, doi:
[10] Acharya, S.; Polawar, A.; Pawar, P. Two factor authentication using 10.1109/ICUFN.2018.8436784.
smartphone generated one time password. J. Comput. Eng. (IOSR- [28] Azarov I., Gnatyuk S., Alexander M., Azarov I., Mukasheva A., Real-
JCE) 2013, 11, 85–90. time ML Algorithms for The Detection of Dangerous Objects in
[11] Balloon, A.M. From Wax Seals to Hypertext: Electronic Signatures, Critical Infrastructures, CEUR Workshop Proceedingsthis link is
Contract Formation, and a New Model for Consumer Protection in disabled, 2023, 3373, pp. 217–226
Internet Transactions. Emory Law J. 2001, 50, 905.
- 162 -
Design and Implementation of an Electronic Voting System
Using the IBM Blockchain Platform
Tolegen Aydynov*, Olzhas Sabiev, Dina Satybaldina
L.N. Gumilyov Eurasian National University
In this paper we present the electronic voting or e-voting system based on the IBM
Blockchain platform. IBM Blockchain platform is integrated with the Hyperledger Fabric
framework, which provides blockchain-as-a-service and offers several edges together with
decentralization and transparency (an audit ability, performance, and measurability. We have
design secure and robust electronic voting system for university elections to elect student leader
from the different faculties candidates.
The details of a blockchain-based e-voting system designing using a commercial
blockchain platform and UML model of the business logic of voting processes are considered.
Visual Studio Code 2022 IDE is selected for software implementation of a client-server application
on JavaScript. Some libraries and frameworks such as VueJs, Webpack, NodeJs, npm, Express,
IBM Blockchain Platform Extension are used. The architecture of the system being developed
consists of a server and a client part and offers the participation of two types of users: a blockchain
operator and a voter. The blockchain operator installs and configures the IBM Blockchain
Platform. On the IBM Blockchain Platform: The Hyperledger Fabric network is created in the
IBM Kubernetes service; the blockchain operator installs and instantiates the smart contract on the
network. The backend (written in Node.js) uses the Fabric SDK to interact with the deployed
network on the IBM Blockchain Platform. It creates an API for use on the client side. The client
side (written in Vue.js) uses the Node.js application API to interact with the web. The voter
interacts with the Vue.js web interface to cast their vote and query the voting status to see their
vote, current and final voting results.
Hyperledger cloud system is a public platform within which all transactions committed are
held on in an exceeding blockchain. The chain grows endlessly as new blocks are added.
Therefore, we will also analyze the performance and the system requirements of our proposed
approach for deploy a scalable and a secure e-voting system in any size university campus. The
performance of the e-voiting system based on the IBM Blockchain platform with an increase in
the number of voters was studied taking into account various blockchain platform configuration
parameters, such as block size, consensus protocol type, mining algorithm complexity, the
transaction throughput and latency.
- 163 -
Study of Network Virtualization Security
in Public Network
Junwon Lee 1 and Juhee Choi 2*
1
Samsung SDS, South Korea
Email: [email protected]
2
Sangmyung University, South Korea
Email: [email protected]
Abstract— Mobile devices such as smartphones, tablets, and This paper identifies the problems and requirements for
wearables are increasingly being used in business environments, applying network virtualization in public networks and
while business culture such as telecommuting, agile, and describes the related technologies, network architecture, and
DevOps is also spreading rapidly. In this environment, users authentication procedures. Additionally, future works for
need a network that easily connects their devices to the network, evaluating network transmission and encryption performance
while also providing the same network environment regardless are proposed.
of user location to ensure continuity of applications and
collaboration with remote users. While IPSec or SSL(TLS) II. PROBLEM ANALYSIS
based VPNs have been applied as a solution to meet these
requirements, they have limitations such as VPN appliance A. Layer-2 security of Virtualized Network
construction on the user network or VPN agent installation on As the VXLAN network expands and covers more devices,
the user PC, as well as communication restrictions to layer-3
the risk of security threats also increases. An attacker can use
networks. Therefore, a layer-2 based virtual network is required
the same ARP attack method that is used to compromise
for seamless communication on an IP-based network. This
paper proposes an architecture to provide a virtualized network
VLANs to attack VXLAN. VXLAN itself does not provide a
environment like inside the campus on the public network. To security mechanism to prevent malicious ARP spoofing.
implement this virtualized network, specific improvements are Moreover, due to the large number of devices that can be
suggested, including systems to support dynamic virtualized connected to VXLAN, there is a need for countermeasures to
networks, trustworthy encryption methods, and measures for prevent security threats that may come from malicious users
SDN vulnerabilities. Future works are also discussed, outlining or devices on the extended VXLAN. Therefore, it is important
the necessary steps to implement this architecture and overcome to implement additional security measures and protocols to
the challenges of creating a secure and scalable virtualized secure VXLAN networks and prevent such security threats.
network in a public network environment.
B. Vulnerability of SDN
Keywords—Network Overlay, VXLAN, L2TP, MACsec, SDN It is important to note that the lack of TLS implementation
in OpenFlow switches and controllers can be a significant
I. INTRODUCTION security risk in a public network environment. Without
Mobile business environment, enterprise users require encryption, attackers can potentially intercept and manipulate
access to internal systems regardless of their location. communication between the controller and switches, leading
However, this access must be restricted to protect the campus to various security threats such as eavesdropping and MITM
network from external security threats [1]. Currently, users attacks. As a result, it is crucial for both software and hardware
must be assigned a new IP when they change campus, and vendors to fully implement the security specifications,
access control policies for IP need to be modified. To address including TLS, in their OpenFlow products to ensure a secure
these challenges, companies require a logically separated SDN deployment in a public network environment.
network in the same physical network, which can reduce costs
and improve business efficiency. Additionally, if the logically III. RELATED WORK
separated network supports the same layer-2 network A. Layer-2 Network Virtualization Protocol
regardless of location, user convenience can be increased [2].
The early adoption of overlay network technology has
To meet these requirements, Network Virtualization and
paved the way for its implementation in scalable and
Software Defined Network (SDN) technologies are needed.
manageable cloud environments. As Software-Defined
However, security concerns must be considered as layer-2
Networking (SDN) based on Network Function Virtualization
network coverage expands, and before applying OpenFlow in
(NFV) continues to evolve, overlay network-based network
Public Network, careful consideration must be given to MITM,
virtualization is becoming increasingly integrated into user
communication encryption, switch authentication, and DoS
network environments [4]. A notable solution in this regard is
[3]. In a public network environment, sufficient virtual
CISCO SDA, which utilizes the dedicated Cisco Meta Data
network pools must be provided to create an isolated network
(CMD) protocol to provide 65,000 security groups [5].
environment for each user, and user authentication procedures
However, the limited number of user groups and the network
are required to block unauthorized access and provide
segmentation between remote users pose challenges to
authorized access to the user. Encryption must be applied to
network virtualization. To address these challenges and
ensure the confidentiality of data transmitted over the virtual
accommodate various protocols and services, layer 2 network
network.
- 164 -
virtualization is deemed appropriate, and technologies such as technology, with additional security measures including
VXLAN, STT, NVGRE, and LISP listed in Table 1 are 802.1X port-based authentication, MACsec for layer-2
utilized [6,7,8]. transport encryption, and virtual network (VN) gateways. As
TABLE 1. Comparison of Network Overlay Technologies
shown in Figure 1, the physical network infrastructure is
divided into ISP and campus areas. The ISP provides layer-2
Category VXLAN STT NVGRE LISP transmission from user PCs to the Virtualized Network Node
Overlay (VNN) built in the public cloud, and relays authentication
24bit 64bit 24bit 24bit
identification services to enable users to connect to the campus network as
LISP Layer if they were inside the campus.
3 draft RFC
Network MTU
should be None; STT
proposes The user authentication process begins with the user
Draft RFC two logging in with their "id@domain," which is then sent to the
Fragmentation increased 50 uses the
proposes using methods for PE controller for verification with the Authentication Server
after bytes to avoid interface MTU
path MTU handling
encapsulation fragmentation and TCP
discovery LISP [11]. After authentication, the VNN Controller orders the
of VXLAN segmentation VNN Archive to start a VNN instance, followed by layer-2
packets that
packets
exceed tunneling of L2TP between the host and VNN, and VXLAN-
MTU EVPN tunneling between VNN and VN Gateway. For
Forwarding of confidential transmission, MACsec is applied to
Layer-2 Encapsulation Encapsulation Encapsulation
broadcast, uses IP uses IP uses IP
Use of communication from the host to VN Gateway, due to its high
underlay encryption speed compared to IPsec. If layer-2
multicast, and multicast as multicast as multicast as
multicast communication is supported, MACsec is a recommended
unknown destination IP destination IP destination IP
unicast encryption protocol.
LISP
leaves open the mapping Overall, our proposed architecture offers a secure and
choice of
data plane
system, scalable solution for virtualized network environments in
control plane, supporting public networks, while ensuring efficient and confidential
Address data plane learning,
keeping it encoding of transmission of data.
learning based learning control plane
separate from instance ID
based, etc.
the data plane and MAC
encapsulation address 2- Fig I. ARCHITECTURE OF VIRTUALIZED NETWORK THROUGH ISP
tuple
B. SDN
SDN Separates control plane and data plane functions.
Control Plane provides the forwarding table to the Data Plane,
and the Physical Switch only performs the Data Plane function
to forward the packets. The centralized control plane
configuration, the centralized controller function can
recognize the network status more quickly than in a distributed
TABLE 2. Feature Detail of Elements
environment, so that the control function can be reflected
dynamically and appropriately. Control Plane's Program Element Function Remarks
Function, Program-capable Controller function is the most PE ∙ Authentication Info relay to Controller OpenFlow
important feature of SDN. You can share hardware, but you (Provider Edge) ∙ Switching, Flow table update Switch
can apply different policies to separate virtual network ∙ 802.1X Authenticator
OpenFlow
environments. SDN provides a centralized control plan for PE Controller ∙ Perform authentication after checking
Controller
user domain
communicating with data planes and applications. Currently,
VNN ∙ L2TP Connection with Host
OpenFlow is used as a representative Southbound API. OpenFlow
(Virtualized ∙ VXLAN-EVPN to VN Gateway
Compared to the Southbound API, the Northbound API Switch (VM)
Network Node) ∙ VNN Resource Monitoring/Reporting
standard is underdeveloped. There is also a need for an East- ∙ VNN generation request, VNN scale
OpenFlow
West API for communication between different controllers [9]. VNN Controller in/out request
Controller
∙ Transfer the VNI of User to VNN
C. 802.1AE (MACsec) VNN Archive ∙ Storage of Customers‘ VNN Image
∙ Node included Gateway IP of VN
MACsec guarantees confidentiality and integrity between
VN Gateway ∙ Connection with Campus Backbone
LAN-connected systems in layer-2 communication. It also Switch
provides strong validation against the reply attack. The ∙ User authentication
Authenticator
MACsec security mechanism does not affect the upper layer, ∙ Assign VNI based on id, mac, and
Server
so there is no need to modify the user application [10]. certificate
IV. VIRTUALIZED NETWORK USING VXLAN
While there is existing research on network structures to V. CONCLUSION
increase data transmission efficiency and guarantee Currently, the network virtualization service used to
availability in virtualized network environments, there is a access internal systems from outside the campus is a layer-3
lack of research on virtualized network architectures that VPN based on IPsec and SSL communication. However, this
provide confidentiality, scalability, and operational efficiency. service primarily focuses on connecting communication lines
and can cause issues for some IP-based services. To address
Our proposed architecture for a Virtualized Network in a this, a proposed layer-2 based network virtualization solution
public network is based on VXLAN and OpenFlow
- 165 -
would not modify the IP header of the original packet. This [4] Data Center Overlay Technologies (Cisco White Paper),
provides users with a seamless experience for business https://www.cisco.com/c/en/us/products/collateral/switches/nexus-
9000-series-switches/white-paper-c11-730116.html
services, whether they are inside or outside the campus.
[5] Gu, Ruichun, et al. "Enhancing Security and Scalability in Software
To evaluate the performance of the proposed architecture Defined LTE Core Networks." 2018 17th IEEE International
and identify any potential issues on the actual network, a proof Conference On Trust, Security And Privacy In Computing And
Communications/12th IEEE International Conference On Big Data
of concept (PoC) should be implemented. Additionally, future Science And Engineering (TrustCom/BigDataSE). IEEE, 2018.
research will aim to propose a flexible method for [6] Han, Bo, et al. "Network function virtualization: Challenges and
authenticating hosts through previously unidentified opportunities for innovations." IEEE Communications Magazine 53.2
authenticators (such as switches or access points) and enabling (2015): 90-97.
communication with the Virtual Network Node (VNN). [7] Tuladhar, Summit R., Carlos E. Caicedo, and James BD Joshi. "Inter-
domain authentication for seamless roaming in heterogeneous wireless
ACKNOWLEDGMENT networks." 2008 IEEE International Conference on Sensor Networks,
Ubiquitous, and Trustworthy Computing (sutc 2008). IEEE, 2008.
This work was supported by the National Research [8] Ward, Rory, and Betsy Beyer. "Beyondcorp: A new approach to
Foundation of Korea (NRF) grant funded by the Korea enterprise security." (2014).
government(NRF-2021R1G1A1004340). [9] VXLAN Overview: Cisco Nexus 9000 Series Switches,
https://www.cisco.com/c/en/us/products/collateral/switches/nexus-
REFERENCES 9000-series-switches/white-paper-c11-729383.html
[1] Reyes, Guido Pineda, Maarten Dammers, and Maarthen Kastanja. [10] Jain, Raj, and Subharthi Paul. "Network virtualization and software
"Security assessment on a VXLAN-based network." Haettu 10.2017 defined networking for cloud computing: a survey." IEEE
(2014): 2013-2014. Communications Magazine 51.11 (2013): 24-31.
[2] Benton, Kevin, L. Jean Camp, and Chris Small. "OpenFlow [11] IEEE 802.1X, https://en.wikipedia.org/wiki/IEEE_802.1X
vulnerability assessment." Proceedings of the second ACM [12] Innovations in Ethernet Encryption for Securing High Speed WAN
SIGCOMM workshop on Hot topics in software defined networking. Deployments., https://www.cisco.com/c/en/us/products/collateral/ios-
ACM, 2013. nx-os-software/identity-based-networking-services/white-paper-c11-
[3] Overview of TrustSec (Cisco Order Guide), 737544.html
https://www.cisco.com/c/dam/assets/global/pdfs/november- [13] Chowdhury, NM Mosharaf Kabir, and Raouf Boutaba. "Network
security/C07-730151-00_overview_of_trustSec_og.pdf virtualization: state of the art and research challenges." IEEE
Communications magazine 47.7 (2009): 20-26.
- 166 -
Developing An Attack Detection Algorithm
Model and Compared with The K-Means
Algorithm
Uyangaa. Kh 1,2, Odonchimeg. L 1, Dashdorj. Ya 1*, Liji Wu 2 and Heemin Park 3
1
Mongolian University of Science and Technology, Mongolia
2
Tsinghua University, China
3 Sangmyung University, Korea
Email: [email protected], [email protected]
Abstract—This research examines network attacks that The following researchers [5], [6] mainly focused on
occur frequently using real-time traffic data. Due to the malware analysis and signature-based attack detection.
significant computational resources needed to analyze raw real- Nevertheless, intrusion detection systems are still being
time data, the data underwent reprocessing. This involved developed based on the research works of these datasets. For
calculating the number of packets per second and selecting the example, the DARPA (Lincoln Laboratory 1998-99) traffic
nearest second for analysis, revealing a high standard deviation file consists of email, web browser, FTP, Telnet, IRC, and
ratio in packet counts. The research then proceeded using the SNMP traffic, as well as DoS, Guess password, Buffer
reprocessed data. To estimate attack probabilities, a normal
overflow, remote FTP, Syn flood, Nmap, and Rootkit attack
distribution was employed, and probability limits were
traffics. These attack traffic files are outdated, and some are
established using Chebyshev's inequality. An algorithm was
developed using these probabilities to classify attacks, and its
stored on currently used databases.
performance was compared to that of the k-means algorithm. Within the scope of this research, we aim to propose a
The proposed algorithm's effectiveness was evaluated by statistical and anomaly-based intrusion detection algorithm by
estimating the differentiation coefficient. Results indicated an transforming and analyzing based on traffic file datasets.
impressive 96% accuracy rate in distinguishing attacks with the Consequently, we chose the CIC-IDS2017 dataset, which
proposed algorithm. In conclusion, this research provides contains the common attack traffic data widely used in
valuable insights into analyzing network attacks using real-time
advanced model experiments [7]. Although the Canadian
traffic data. The proposed algorithm, incorporating probability
Institute for Cybersecurity Datasets is available to pick the
estimation and classification techniques, demonstrates a high
degree of accuracy in effectively identifying and differentiating latest traffic files, such as 2021, 2022, and 2023, respectively,
attacks. we used the CIC-IDS2017 files for the following reasons.
Our research contribution is grounded in two primary
Keywords— Machine learning, Network attacks, K-means factors. Firstly, utilizing the 2017 traffic file, which
algorithm, Chebyshev's inequality, DDoS LOIT, Portscan encompasses a broad spectrum of network data traffic,
I. INTRODUCTION provides an extensive dataset for our proposed research.
Secondly, we have previously conducted an experiment
By the Digital 2023 Global Overview Report, 8.01 billion described in reference [8], where we implemented three
people were counted in January 2023, and 64.4% use the different attacks using the same traffic file as a basis. Building
Internet today[1]. As internet users grow, attacks and upon this foundation, we continue developing our study and
malicious code spread in the internet environment also analyzing the implementation of these attacks using
increase. After the Pandemic period, significant changes are Chebyshev's inequality, as outlined in reference [9]. By
evolving in the cyber world too. Reference [2] summarizes incorporating these scientific methodologies, we aim to
the cybersecurity statistics, including the top cyberattack types enhance the robustness and accuracy of our research.
such as malware, DDoS, and Ransomware. The article [2]
remarks that 0.3 million malware is created daily, and the This research paper is structured as follows: Section 2
cyberattack happens every 39 seconds on average. Internet provides details about the network topology and experimental
network attacks can be detected by anti-virus software on the setup, while Section 3 discusses the statistical processing of
user's device, host-based intrusion detection software, next- traffic data and presents the outcomes of the k-means machine
generation firewall, and the tools such as intrusion detection learning approach. Section 4 presents the results obtained
systems (IDS) on public networks. The intrusion prevention using Chebyshev inequality, Section 5 introduces the
system (IPS) also detects and protects against internet network proposed traffic classification method and its results, and the
attacks. However, these tools may be ineffective in noticing concluding chapter summarizes the research findings,
certain types of attacks. In this case, researchers continue highlighting key outcomes and implications.
studying and analyzing attack traffic files using various
techniques and struggle to find operative datasets developing
their methods [3], [4]. In some cases, the attack traffic files
contain personal information, making using them challenging
for research purposes. Hence, it is required to change some
info, such as the IP addresses and port numbers of the devices,
for research purposes.
- 167 -
Fig. 1. Overview design of attacks and traffic collection
II. EXPERIMENTS TRAFFIC COLLECTION TOPOLOGY including Botnet ARES, PortScan, and DDoS LOIT,
representing common attack strategies. The DDoS attacks
A. General Topology The Traffic Collection Network utilized the LOIC tool, targeting web servers to disrupt their
Topology Network functioning. The PortScan profile emulated reconnaissance
Figure 1 illustrates the experimental setup comprising two attempts by scanning for open ports using the Nmap tool. The
networks: Network 1 (victim side) includes DNS, FTP, and Botnet attack simulated coordinated and distributed attacks by
web servers, along with a dedicated packet collector server, employing the ARES tool against five Windows OS
while Network 2 (attacker side) consists of Windows, Kali computers, enabling the evaluation of intrusion detection
Linux, and Ubuntu operating systems. This setup replicates a mechanisms. These diverse attack profiles in the CIC-
realistic scenario where the attacker side launches diverse IDS2017 dataset facilitate comprehensive research and
attacks on the victim network, enabling comprehensive traffic analysis in network security.
data collection. By adopting this configuration, the study aims
to closely simulate real-world situations, facilitating an in- III. STATISTICAL CHARACTERISTICS
depth analysis of attack implications on the victim network. This research used the Wireshark tool to collect two traffic
This approach enhances the validity and applicability of the files and save them separately by their frame delta time and
research, contributing to a better understanding of network sending packet size. Since these two parameters can be used
security and various attack strategies' impact. as one general network parameter, statistical processing was
performed on them. As shown in Table 2, the attack packets
Table 1 displays the traffic data collected during the are saved as Day2 files, consisting of both normal and attack
experiment, consisting of two types: Day 1 represents normal packets that pass through the network between 9:00 am and
network traffic from 9:00 am to 5:00 pm, saved as Day1.pcap 5:00 pm. The following file (Day1) contains normal packet
(10GB). Day 2 captured attacks using specific IP addresses, flow between 9:00 am to 5:00 pm. The packets collected were
including Botnet ARES, PortScan, and DDoS LOIT, 9 million and 11 million on Day 1 and Day 2, respectively.
occurring at distinct time periods, with the corresponding logs The standard deviation of attack and normal traffic flows are
stored as Day2.pcap (8.6GB). Further details about the attacks calculated, and the results are displayed in Table 2.
can be found in reference [4]. These data provide the basis for According to the results, the difference between their mean
conducting further investigations and developing effective and standard deviation is minuscule, which is 1.3, resulting in
security measures to mitigate the impact of such attacks on the difficulty showing different features. Thus, it was necessary
network. to reprocess the traffic data. We separated the number of
B. Experimental Setup incoming packets and the period between them within a close
The traffic data was captured and logged using Wireshark, range to a second, then carried out statistical processing,
saving it as *.pcap files for further analysis. The CIC-IDS2017 which resulted in the standard deviation ratio to be changed,
dataset was designed to cover a wide range of attack scenarios, resulting in the standard deviation ratio changing to 4.41.
TABLE I. THE ORIGINAL DATA AND PROCESSING RESULTS OF THE COLLECTED TRAFFIC
Data Packet in bytes Inter arrival time in sec Number of Ratio of
No Packet of Traffic
mean Variance mean variance Packets Variance
1 Day2 891 1765833 0.0029 0.0009169 9997874
1.3
2 Day1 851 1346045 0.0025 0.0003937 11709971
- 168 -
TABLE II. REPROCESSING RESULTS (PACKET NUMBER PER SECOND)
Data Packet in bytes Inter arrival time in sec
Number of Ratio of
No Packet of Traffic
Packets Variance
mean variance mean Variance
1 p_num_day2 14758.9 329703215 60.182 4.41 422
4.41
2 p_num_day1 13438.2 74641183.2 60.063 0.013053 425
Table 2 presents the reduced number of packets processed,
with 422 for normal traffic and 425 for attack traffic. Figure
2(a) displays the number of incoming packets per second for
normal traffic, while Figure 2(b) shows the number of
incoming packets per second for both normal and attack
traffic. The graphs reveal a packet count of 111,400 during the
attack compared to 57,595 in normal traffic, indicating a
significant increase. The standard deviation ratio is 4.41,
demonstrating a notable difference from previous traffic.
Figure 3 demonstrates the cumulative distribution graph for
original data of normal and attack traffic as displayed in Table
2. The cumulative distribution graph for normal traffic and
attack traffic is shown in Figure 3, it can be observed that as
the packet size increases the difference becomes smaller and
it becomes the same once it reaches the top limit. In this case,
if we construct a cumulative distribution graph on the original
data for normal and attacked traffic (shown in Table 2) to
show cumulative distribution, the result can be displayed as
Figure 3. Fig. 4. Attack (a) and normal (b) traffic histograms
As shown in Figure 3, the differences between the three
types of attacks can be clearly observed (Botnet ARES, DDoS
LOIT, and PortScan), and certain limits can be determined.
We conducted a statistical analysis of the files and
observed that the reduced traffic size exhibited minimal
differences compared to the original traffic. Leveraging this
finding, we proceeded to determine the attack packets by
filtering the corresponding probabilities using the Gaussian
normal distribution equation. This approach allowed us to
effectively identify and isolate the packets associated with the
attacks, despite the reduction in traffic size.
In order to conduct a thorough analysis of the network
traffic, statistical analysis was performed on the collected
files. As part of this analysis, the traffic size was reduced while
ensuring that the integrity of the data was maintained.
Fig. 2. The number of incoming packets within a second Remarkably, despite the reduction in size, there was minimal
disparity between the reduced traffic and the original dataset.
Based on this observation, we proceeded to identify the
attack packets by employing a filtering approach. Specifically,
we utilized the Gaussian normal distribution equation to filter
the traffic data and determine the corresponding probabilities
associated with each packet. By leveraging this method, we
were able to effectively isolate the packets that were indicative
of potential attacks within the network traffic.
This approach allowed for systematic and reliable
identification of attack packets, contributing to understanding
and analyzing network security. By leveraging statistical
techniques and probability estimation, our research aimed to
enhance the detection and classification of network attacks,
thereby enabling the development of more effective defense
Fig. 3. Cumulative distribution of original data during normal and attack mechanisms against evolving threats.
periods
- 169 -
IV. RATE USING CHEBYSHEV'S INEQUALITY V. CLASSIFYING TRAFFIC BY CHARACTERISTICS
The famous Russian mathematician Chebyshev assumed In this research work, univariate Gaussian distribution will
that the value of a random variable with an unknown be used to express the number of packets per second by the
distribution, the probability is at least (− ) that it corresponding probability limit for that traffic.
deviates from its mean by no more than standard deviation
[9]–[11]. It can be presented as follows: (𝑥−𝜇)2
1 −
𝑃(𝑥 < 𝑋, 𝜇, 𝜎) = 𝑒 2𝜎2 =𝜀 (3)
√2𝜋𝜎 2
1
𝑃(|𝑋 − 𝜇| < 𝐾𝜎) ≥ (1 − ) ()
𝐾2 Let's find x from formula (3).
Here (1), we calculate the value of from the attack and
normal traffic. The probability is presented in (2) using the 𝑥 = 𝜇 + √(−2𝜎 2 ln(𝜀 ∗ 𝜎 ∗ √2𝜋)) (4)
above inequality (1) for attack and normal traffic within a
period of a second.
With the help of the formula (4), we can find the
1 probability and mean of standard deviation corresponding to
𝑃(𝜇 − 𝐾𝜎 ≤ 𝑋 ≤ 𝜇 + 𝐾𝜎) ≥ (1 − ) () the packet size.
𝐾2
Incoming attacks can be categorized by classifying the
Here (2), µ is the average number of packets for a second,
packets based on the corresponding probabilities, as listed in
σ is the standard deviation of packets for a second, is the Table 4.
deviation that is no more than the standard deviation, and Ρ is
the probability. Figure 5 depicts the detection of attacks using a
classification approach based on probability limits. The
TABLE III. NUMBER OF PACKETS PER SECOND probabilities initially start low for DDoS LOIT attacks and
No K- K- gradually increase for Botnet ARES attacks, indicating
min mean max improved identification of more sophisticated attacks. The
min max
P_num_ application of the k-means algorithm faced challenges in
1 488 14758.94 111400 0.78 5.32
Day2 differentiating Botnet ARES and PortScan NMAP attacks
P_num_ from normal traffic. In contrast, our proposed algorithm
2 1617 13438.2 57595 1.3 5.1
Day1 achieved higher accuracy in distinguishing these attack types
Table 3 displays the results obtained by applying from normal traffic. This improved performance can be
inequality (2) to determine the coefficient K, indicating the attributed to the utilization of probability estimates derived
deviation of extreme values from the mean for the number of from the Gaussian normal distribution and the defined
packets within a second. The calculations reveal that during probability limits based on Chebyshev's inequality. By
the attack, the mean value shifts 5.3σ units to the right from incorporating these probabilistic measures, our algorithm
the initial value, while in normal traffic, the distance from the effectively captures the unique characteristics and patterns
mean is 1.3σ for the first packet and 5.1σ for the last packet. exhibited by different attack types, leading to enhanced
These findings align with the theoretical analysis based on classification accuracy. Figure 5 and Figure 6 provide
Chebyshev's inequality. Figure 4 visually represents these graphical representations of the calculated differential
observations through histograms of the attack and normal coefficient (f_score) for the two classification algorithms: the
traffic, confirming the shifts in the mean values as predicted k-means algorithm and our proposed algorithm. The f_score
by Chebyshev's inequality. This analysis, supported by indicates the effectiveness and accuracy of the classification
experimental data, enhances our understanding of the algorithms in distinguishing between attack types and normal
statistical behavior of traffic and the efficacy of Chebyshev's traffic.
inequality in assessing deviations from the mean. These
insights contribute to the assessment and differentiation of In Figure 5, it can be observed that the f_score for the k-
attack and normal traffic, benefiting network security and means algorithm falls below 50%. This suggests that the k-
intrusion detection mechanisms. means algorithm struggles to differentiate between attack
types accurately, resulting in a lower performance level than
our proposed algorithm.
In contrast, Figure 6 demonstrates a significantly higher
f_score of 96% for our proposed algorithm. This indicates our
algorithm's performance and accuracy in classifying and
distinguishing different attack types from normal traffic.
TABLE IV. CLASSIFICATION PROBABILITY LIMITS
No Probability limits Classification
1 𝑃(𝑥 < 𝑋) > 𝜀0 Normal Traffic
2 𝜀1 > 𝑃(𝑥 < 𝑋) > 𝜀2 Botnet ARES
3 𝜀3 > 𝑃(𝑥 < 𝑋) > 𝜀4 PortScan
4 𝜀5 > 𝑃(𝑥 < 𝑋) > 𝜀6 DDoS LOIT
Fig. 5. Classification of normal traffic and attacks
- 170 -
The high f_score demonstrates the effectiveness of our ACKNOWLEDGMENT
proposed algorithm in accurately identifying and This research was supported by KOICA(Korea
differentiating attacks, contributing to enhanced intrusion International Cooperation Agency) through the "Capacity
detection and network security. The notable disparity between Building Project for the School of Information and
the f_scores of the two algorithms highlights the advantage of Communication Technology at Mongolian University of
our proposed algorithm over the k-means algorithm. The more Science and Technology in Mongolia" (Contract No. P2019-
performance of our algorithm, as evident from the higher 00124)
f_score, reinforces its potential for real-world application and
its significance in improving the accuracy and reliability of REFERENCE
intrusion detection systems.
VI. CONCLUSION [1] www.wearesocial.com, “The Changing World of DIGITAL in
2023,” 2023. Accessed: Jan. 26, 2023. [Online]. Available:
In this research, we proposed an attack detection https://wearesocial.com/uk/blog/2023/01/the-changing-world-of-
algorithm that leverages statistical analysis and anomaly digital-in-2023/
detection techniques. The algorithm was developed and [2] Nivedita James, “160 Cybersecurity Statistics 2023 [Updated],”
evaluated using three common types of attack files and normal 2023. https://www.getastra.com/blog/security-audit/cyber-
security-statistics/#:~:text=FAQS-
traffic files to assess its ability to differentiate attacks ,How%20many%20cyber%20attacks%20happen%20per%20year
accurately. A comparative analysis was conducted with the k- %3F,hacker%20attack%20every%2039%20seconds (accessed
means algorithm, and the resulting differential coefficient was May 02, 2023).
calculated. Particularly, our proposed algorithm achieved an [3] X. Liu and J. Liu, “Malicious traffic detection combined deep neural
notable differential coefficient of 96%, signifying its higher network with hierarchical attention mechanism,” Sci Rep, vol. 11,
performance in differentiating attacks compared to the k- no. 1, p. 12363, Jun. 2021, doi: 10.1038/s41598-021-91805-z.
means algorithm. This outcome verifies the capability to [4] I. Sharafaldin, A. Habibi Lashkari, and A. A. Ghorbani, “Toward
Generating a New Intrusion Detection Dataset and Intrusion Traffic
accurately separate attacks compared to the k-means Characterization,” in Proceedings of the 4th International
algorithm, exploiting the principles of Chebyshev's inequality. Conference on Information Systems Security and Privacy,
SCITEPRESS - Science and Technology Publications, 2018, pp.
Further research will be carried out to compare not only k- 108–116. doi: 10.5220/0006639801080116.
means algorithm but also other machine learning algorithms. [5] A. Shiravi, H. Shiravi, M. Tavallaee, and A. A. Ghorbani, “Toward
During this research, statistical processing was done on the developing a systematic approach to generate benchmark datasets
number of packets per second. It reduced the packet size and for intrusion detection,” Comput Secur, vol. 31, no. 3, pp. 357–374,
made it easier to calculate, making it possible to distinguish May 2012, doi: 10.1016/j.cose.2011.12.012.
the characteristics of the attacks. [6] J. O. Nehinbe, “A Simple Method for Improving Intrusion
Detections in Corporate Networks,” 2010, pp. 111–122. doi:
The proposed algorithm's ability to accurately separate 10.1007/978-3-642-11530-1_13.
attacks hold promise for enhancing real-world security [7] M. Rodríguez, Á. Alesanco, L. Mehavilla, and J. García,
systems and inspiring further investigations into the “Evaluation of Machine Learning Techniques for Traffic Flow-
development of the previous research and reliable intrusion Based Intrusion Detection,” Sensors, vol. 22, no. 23, p. 9326, Nov.
2022, doi: 10.3390/s22239326.
detection algorithms. As a result, this study provides a solid
[8] B. Bujmaa and Y. Dashdorj, “Development and Design of an Attack
foundation for further investigations and advancements in Detection System, in mongolian,” in InfoNet&Sec-2020 , MUST,
network security. Mongolia, 2020.
[9] J. G. Saw, M. C. K. Yang, and T. C. Mo, “Chebyshev Inequality
with Estimated Mean and Variance,” Am Stat, vol. 38, no. 2, pp.
130–132, 1984, doi: 10.2307/2683249.
[10] H. Assimi, O. Harper, Y. Xie, A. Neumann, and F. Neumann,
“Evolutionary Bi-objective Optimization for the Dynamic Chance-
Constrained Knapsack Problem Based on Tail Bound Objectives,”
vol. 325, Feb. 2020, [Online]. Available:
http://arxiv.org/abs/2002.06766
[11] P. P Bocharov and A. V Pechinkin, Probability Theory. Math
statistics, in Russian. M. : Gardarika, 1998.
Fig. 6. Attack cluster classification using the k-means algorithm
- 171 -
August 22(TUE) - 25(FRI), 2023
Oral Session 7
16:20~18:00 Wed, August 23, 2023
Room : A432
Chair: Jung-Ju Kim (Hoseo University, Korea)
- 172 -
Development of mobile application for copper
wire molten mark classification
Seong-jun Hong1* and Dong-hyun Tae1, Jin-sik Lee 1
1
Korea Electrical Safety Corporation Electrical Safety Research Institute, Republic of Korea
Email: [email protected]
Abstract— The molten traces of copper wires found at fire molten scar appearance discrimination system using AI
sites can be divided into molten traces caused by electrical technology.
causes and molten traces created by melting by external flames.
Currently, copper wire molten traces are used to predict
ignition points during fire investigations or to check whether
electricity is supplied at the time of a fire. In this paper, the II. ANALYSIS OF MORPHOLOGICAL CHARACTERISTICS OF
characteristics of molten scars created by short circuits among COPPER WIRE MELTING TRACES
molten scars created by external flame and molten scars created Melting marks created by flame and melting marks created
by electrical causes were extracted. In addition, using AI by short circuit have a common point of being melted by high-
technology, image learning of copper wire molten traces was temperature heat. It is melted by Joule heat at the contact area
performed, and a mobile application was developed to instantly between wires. The morphological shapes and surface patterns
classify copper wire molten traces at the fire site. that appear due to the difference in the force acting during the
formation of these formation mechanisms and molten scars
Keywords—Copper wire, short circuit Arc site, Arc melting,
Globule, Arc bead
were analyzed.
I. INTRODUCTION (HEADING 1)
Over the past 10 years, the number of fires in Korea has A. Flame fusion scars Appearance characteristics
reached about 430,000. Among them, fires caused by
electrical causes account for more than 17% [1].
Representative causes of ignition by electric energy include
poor contact between wires, tracking, overcurrent, and short
circuits, and in these situations, molten copper wires are found
with a high probability. Currently, most fire investigators
judge electrical fires based on molten traces. Therefore,
accurate analysis of the molten trace plays an important role (a) (b)
in revealing the cause of the fire caused by electricity. Fig 1. pointed bulbous shape.
In general, molten marks found at fire sites are classified
into primary, secondary, and tertiary [2-5]. The primary
melting mark refers to the melting mark that causes a fire and
is created due to short circuit, poor contact, and tracking at
room temperature. The secondary molten trace has a similar
principle to the primary molten trace, but it is created after a
fire occurs, so the ambient temperature at the time of creation
(a) (b)
is high. Tertiary melting marks refer to melting marks
generated by a heat source such as a flame. Currently, fire Fig 2. Bubble-shaped flame fusion marks formed on a part
investigators investigating the cause of a fire do not have a of the wire.
specific criterion for classifying molten traces found at a fire
site by appearance, so there are cases in which the tertiary
molten trace is mistaken for a fire caused by an electrical cause.
Currently, when fire investigators judge the appearance of
molten traces, they are made based on subjective judgment
without objective criteria, and the discrimination results vary
according to the fire investigator's skill level. In addition, the
molten scar identification method is dependent on manpower (c)부착력 (d) 90 도 연선 4SQ 27
and equipment, which requires excessive time and cost, and Fig 3. Flame fusion traces formed under the influence of
the number of precision evaluations that can be performed adhesion.
each year is limited. In this paper, we develop a cloud-based
- 173 -
Melting marks created by flame and melting marks created by angle created by the short-circuit current at the intersection of the
short circuit have a common point of being melted by high- two wires, an asymmetry of the magnetic field density occurs and
temperature heat. It is melted by Joule heat at the contact area the Lorentz force acts in a specific direction. Fig. 5(b) shows the
between wires. The morphological shapes and surface patterns that contact surface of Fig. 5(a). In area B in Figure 5(b), melting traces
appear due to the difference in the force acting during the formation are more widely distributed than in area A, and an extending shape
of these formation mechanisms and molten scars were analyzed. is confirmed. It is observed that the direction in which the molten
Figure 1 shows the flame fusion trace in the form of a bulb with a traces spread is the direction where the density of the magnetic field
pointed tip. Flame fusion marks are melted from the outside of the is low, and it spreads longer in the direction where the density is
wire by an external flame. The liquefied copper on the outer shell is relatively low.
united by cohesive force and moves in the direction of gravity. At
this time, the copper in the solid state that is not melted inside is
exposed to the outside, and a bulb-shaped flame melting mark with
a sharp end is formed. In Figure 2(a) and (b), it can be confirmed
that the molten copper wires are combined by cohesive force and
moved in the direction of gravity, and the unmelted copper wires
inside are exposed and show a sharp shape. Figure 2 shows the
bubble-shaped flame fusion marks generated on a part of the wire.
The flame fusion marks in the form shown in Figure 2 are formed (a) (b)
by cohesion when the surface of the copper wire is melted and Fig 5. Orientation in the form of short-circuit melt scars
liquefied by an external flame. It can be seen that the periphery of
the molten blob showing a circular shape is thin. These forms are The second feature of short-circuit melting scars is the cut shape.
formed in a high temperature environment where sufficient time is The cut shape refers to a shape in which a boundary where a melted
given for the liquid to migrate. The flame fusion marks in the form portion and an unmelted portion of the copper wire meet is sharply
of Figure 3 appear when the liquefied copper wire is attached to the bent. The cut form is created by the movement of the molten copper
solid copper wire around it. This form occurs more easily as the wire under the influence of the Lorentz force, and is mainly located
temperature of the unmelted solid-state copper wire part approaches on the power supply side. In addition, the cut shape has a smooth
the melting point of the copper wire. Figure 4 shows the melting surface without curves compared to the surrounding melting
distinguishable pattern features appearing on the surface of the surface. Figure 6 shows the shape of the cut in the short-circuit
flame fusion scar. Fig. 4(a) shows the shrunken surface appearing in molten trace. Figure 6(a) is a melting trace formed in the middle of
the flame fusion scar. Figure 4(b) shows the stripes formed the wire, and a cut shape is observed on the power source side.
perpendicular to the direction of the cross-section of the wire. These Figure 6(b) is a cut-shaped melting mark formed at the end of the
stripes appear at the base of the wire and adjacent to the fusion zone. wire. When the contact part of the wire is deeply melted and
Fig. 4(c) shows the spine-like projections appearing in the flame disconnected, it often appears on the wire on the power side. In the
fusion scars. Fig. 4(d) shows melting traces in the form of cut-shaped molten trace formed at the end of the wire, a small ball
independent protrusions. The protrusion is formed when a part of the is often observed at the edge of the wire end.
surface of the copper wire rises when heat is applied to the copper The third characteristic of short-circuit melting scars is the
wire. covered form. The covered form refers to a form in which the molten
copper wire spreads over the surface of the unmelted copper wire.
The covered shape is formed by moving the molten copper wire
under the influence of the Lorentz force or the arc repulsive force,
and extends toward the load side where the strength of the magnetic
field generated by the short circuit current is weak. Figure 7 shows
the covered form. In Fig. 7(a), melting traces formed in the middle
of the wire are observed on the load side. Figure 7(b) is a covered
(a) (b) molten trace formed at the end of the wire, with molten copper
covering the surface of the unmelted wire.
The fourth characteristic of short-circuit melting scars is the shape
of a ball. The bulbous form is found in both short-circuit and flame-
melting scars, but there are differences. The most distinctive feature
of short-circuit melting scars is the shape of wrinkles. The wrinkles
of short-circuit fusion scars are formed by rapid solidification before
(c) (d) the wrinkles are straightened by the action of surface tension. If the
Fig 4. Surface characteristics of flame fusion scars ambient temperature is sufficiently high, wrinkles are not formed.
The bulb-shaped wrinkles are formed horizontally around the long
axis of the bulb. Figures 8(a) and (b) show short-circuit melting scars
B. Short circuit arc side Appearance characteristics in the form of wrinkled bulbs. Wrinkle-patterned bulbs are observed
Due to the unique generation mechanism, short-circuit melting in all solid, stranded, and corded wires, and are accompanied by cut
marks have been repeatedly observed in a similar form in single, or covered features around the bulbs.
stranded, and cord wires, and have distinct morphological
characteristics distinct from flame melting marks. The first
characteristic of short-circuit melting scars is directionality. When
two wires with potential difference come into contact, the wires are
melted by Joule's heat generated at the point of contact, and the
Lorentz force is received by the magnetic field generated by the
short-circuit current. Figure 5 shows the moving path of the current
and the moving direction of the molten trace according to the contact (a) (b)
angle of the two wires. Figure 5(a) shows the path of short-circuit Fig 6. Cut form of short-circuit melting scar
current when two wires are contacted at an angle of 80°. Due to the
- 174 -
discrimination result, the degree of similarity is expressed as
a percentage.
(a) (b)
Fig 7. Covered form of short-circuit melting scar
Fig 11. Mobile application
(a) (b) C. Fire site verification
Fig 8. Bead shape of short-circuit melting scar Figure 12 shows the demonstration of the mobile
application at the fire site. After taking an image by
connecting a small microscope to a mobile device, an instant
III. COPPER WIRE MELTING TRACE DISCRIMINATION identification is performed through a mobile application. Fig.
SYSTEM 12(a) shows the shooting equipment, and Fig. 12(b) shows the
process of discrimination.
We developed a cloud server system that can identify
melting traces of copper wires using a mobile application at a
fire site. Figure 10 shows the configuration of the cloud server.
(a) (b)
Fig 12. Fire site demonstration
Fig 10. Cloud server configuration for molten trace
determination
IV. CONCLUSION
A. Mobile application In this paper, a system for distinguishing copper wire
melting traces was developed using convolutional neural
In order to apply artificial intelligence technology, as shown i network technique to analyze the causes of electrical fires.
n Table 1, copper wire learning data was constructed by divid These research results are significant in that they can provide
ing into melting traces generated by electrical causes and mel accurate discrimination results in a short time through
ting traces generated by flames. 10,000 pieces of data were le artificial intelligence algorithms for the external
discrimination of short circuit marks and melting marks,
arned, each of short-circuit melting marks and flame melting
which are difficult to distinguish visually by forensics. In the
marks. Materials for learning data were collected through fire future, we plan to continue research to increase accuracy and
scenes and reenactment experiments. expand application to various wires through big data
configuration.
TABLE I. LEARNING DATA COMPOSITION
REFERENCES
Short-circuit arc marks Flame melting marks
Division
(EA) (EA)
Number of [1] National Fire Agency, Fire Statistical Yearbook (2020), Korea
melt marks
1068 1170
[2] D. A Gray, D. D. Drysdale, Lewis, 1983, Identification of Electrical
Number of Sources of Ignition in Fires, Fire Safety J
images
10,000 10,000
[3] T. Seki, H. Hasegawa, S. Imada, Y. Isao, 2000, Determination between
Primary and Secondary Molten Marks on Electric Wires by DAS,
NITE
B. Mobile application [4] E-P Lee, 1989, Discrimination between Primary and Secondary Molten
Figure 11 shows a mobile application for molten trace Marks on Electric Wires by DAS, Applied Fire Science 9
identification. It is configured to transmit the captured molten [5] Richard J. Roby, Jamile McAllister, 2012, Forensic Investigation
Techniques for Inspecting Electrical Conductors Involved in Fire,
trace image to the cloud server and check the result. As for the Combustion Science & Engineeri
- 175 -
Isolated control circuit design for heater power
supply of high-power magnetron
Hyun-Bin Jo 1 and Hong-Je Ryoo 1*
1
Department of Energy Systems Engineering, Chung-Ang University, South Korea
Email: [email protected]
Abstract— Two power sources are used to drive the small and lightweight system design, An integrated design of
magnetron. The first is a high-voltage pulse power supply for a high-voltage pulse power supply generating -40 kV and a
generating a voltage of several tens of kV to the magnetron filament heater power supply with an output voltage of 10 V
cathode, and the second is a heater power supply for heating the and an output current of 15 A (150 W) is in progress. When
filament of the magnetron. The magnetron to be used as a load designing compactly integrated design, the high-voltage
in this paper is a 1.7 MW X-band magnetron, which requires a isolated sensing technology of the filament power supply is an
maximum anode voltage of -38kV and a maximum anode important consideration. In this paper, we discuss the
current of 88A, and outputs a maximum filament voltage and proposed technique for isolation control from -40kV voltage.
current of 10V and 15A. In order to apply to the medical linear
accelerator application field that requires a compact and II. STRUCTURE OF INTEGRATED POWER SOURCE
lightweight design, we designed a compact modulator structure FOR DRIVING MAGNETRON AND DESIGN OF
that integrates the high voltage pulse unit and the filament
FILAMENT HEATER POWER SUPPLY
power unit and designed a filament power supply with an output
voltage of 10V and an output current of 15A, which is located in A. Configuration of integrated power supply for driving
the high voltage section directly connected to the high voltage magnetron
pulse modulator. the high-voltage isolated sensing technology of
the filament power supply is an important consideration. In this
paper, a simple high-voltage isolation control method of the
filament power supply is described. Filament heater
Power supply
Keywords— High Power Magnetron, Filament, High voltage
Pulsed power
I. INTRODUCTION
High Voltage
Magnetrons are vacuum tube devices used to generate
microwaves in a wide range of applications, including radar Pulsed Power Magnetron
systems, microwave ovens, and linear accelerators [1]-[3]. Supply
High-power magnetrons are designed to handle power levels
ranging from several hundred watts to several megawatts,
depending on the specific application requirements. These
magnetrons consist of a cathode, an anode, and a resonant Charging &
cavity that supports the generation of microwave oscillations. Control part
Two power sources are used to drive the magnetron. The first
is the filament heater power supply. This is a power source Fig. 1. Diagram of an integrated power supply for driving a magnetron
that heats the filament in order to emit a sufficient amount of
free electrons from the filament of the magnetron, and can be Fig. 1 shows a diagram of an integrated power supply for
supplied with AC or DC voltage. The second required power driving a magnetron under development. The lower part
source is a high voltage pulse power source. This applies a consists of a high-voltage pulse power charging unit, a heater
voltage of several tens of kV to the magnetron cathode to power charging unit, and an integrated control unit. The high-
accelerate the electrons by applying a high electric field. In voltage pulse power source is located in the center, and the
this paper, the driving of a 1.7 MW X-band magnetron is filament heater power source is located at the top. On the right
targeted. The pulse power supply [4]-[5] requires a maximum side of the figure, the upper side of the magnetron, the cathode
anode voltage of -38kV and maximum anode current of 88A, is connected with the heater power and the output of the high
and the filament heater power supply requires a maximum voltage pulse power.
voltage and current of 10V and 15A. Since the output terminal
of the heater power source is connected to a high voltage, B. Design of filament heater power supply
sensing and controlling the output voltage and output current Fig. 2 shows the circuit diagram of the designed filament
of the heater power source is very challenging. In order to power supply. A single-phase 220V input is used, and a half-
apply to medical linear accelerator applications that require a bridge inverter is used for compact implementation of the
charging part, and an LC series resonance circuit (Lr : 100 uH,
This work was supported in part by the National Research Foundation Cr : 44nF) is applied. Next, the transformer requires 40kV
of Korea(NRF) grant funded by the Korea government(MSIP)
(No.NRF2020R1A2C2099663) and in part by Institute of Information &
insulation between the primary side and the secondary side,
communications Technology Planning & Evaluation (IITP) grant funded by
the Korea government(MSIT) (No.2022-0-00100, Development of High
Power Electromagnetic Wave based Anti-Drone System)
- 176 -
U1
N2
220Vac
100uH
N3
Ls
N1
44nF
U2 Cs Vse n
sensing
Linear Filamen t
N4
regulator Con troller
Isen
Vse n
Filamen t Isen
Con troller
Fig. 2. Integrated circuit diagram of filament power supply
and for this purpose, an insulation cable with 60kV
dielectric strength was adopted to the primary winding. The
secondary side of the transformer has a center-tap structure. High voltage side
The transformer has a turn ratio of 10:1:1:4. The output of the
filament heater power supply has a wide load range from 1.5W N2
to 150W, and a switching frequency range of 100 kHz to 800 N3
kHz. The filament heater power source uses two controllers.
Among these controllers divided into low voltage and high N1
voltage side, the high voltage side controller receives insulated 4~36V
power through the 4th winding of the transformer. The low-
Linear
voltage controller supplies the switching signal to the half- N4
regulator
bridge module. Vcc
Isolation
C. Insulation control of proposed heater power supply Isolated
Isen + Curren t V to F
Although the power circuit design of the filament heater Isen - Sen sor Con ver ter
power source is not a relatively difficult technology, it is
difficult to sense and control the output of the filament power Isolation
source located on the high-voltage side of several tens of kV. Vse n + V to F
Vse n -
In this control circuit design, stable power was supplied to the Con ver ter
high voltage side control circuit in a simple insulated manner.
The linear regulator that outputs 5V of the high-voltage
controller must operate from the output voltage of the heater
power supply to 1V. For this purpose, the number of turns of
the multi-winding transformer was appropriately set. The 4th
winding magnetically coupled to the 2nd winding of the center Low voltage side
tap transformer, which is the power part, was used. The ratio Vreal
of turns between the 2th and 4th winding is 1:4. The linear PI F to V
Con troller Con ver ter
switching regulator has a 6.5-36 Vdc input range and outputs Vref
Gate
a constant voltage of 5V regardless of input voltage driver
fluctuations. It is designed so that the voltage applied to the IC
Ireal
regulator input does not exceed the voltage rating at the PI F to V
Con ver ter
maximum output voltage of 10V. ACS722, which has high- Con troller
Iref
frequency insulation characteristics, was utilized as the
current sensor of the high-voltage control part. The minimum
Fig. 3. Diagram of proposed insulation control structure of filament heater
operating voltage of the current sensor and the IC elements in power supply
the high-voltage controller is 3.6V or higher. Accordingly, a
small linear regulator was used for stable power supply. Since it is designed to operate in constant voltage control mode
the actual magnetron load has a relatively low specific when it exceeds a certain voltage. Voltage and current have
resistance value in an initial cold state, a relatively high DC type sensing information. After converting it to frequency,
current flows when a voltage reaching the rated voltage is it is transmitted to the low-side controller using an optical
initially applied, which causes damage to the magnetron signal transmitter and receiver. And on the low voltage side,
filament. To prevent this, it is controlled very slowly from low the frequency information is converted back to dc form. The
voltage to rated voltage through constant current control, and difficulty of control requiring 40kV insulation was resolved
- 177 -
through a simple voltage-frequency-voltage conversion
technology without increasing the physical size.
III. EXPERIMENTS AND RESULTS
Fig. 6. the experimental waveform of the rated output of the filament heater
power supply
Fig. 4. Experimental configuration of the fabricated filament heater power
supply CONCLUSION
In this research paper, we propose a technology that
enables the sensing and control of a filament power supply
connected to high voltage using a simple isolation method.
While designing the power supply circuit for the filament
heater is not particularly challenging, it becomes difficult to
sense and control the output when dealing with a filament
power supply operating at high voltages of several tens of kV.
Our control circuit design addresses this challenge by
providing stable power to the high voltage control circuit
through a simple insulation method. To overcome the
difficulty of achieving 40kV insulation for control purposes,
we developed a voltage-frequency-voltage conversion
technology that does not require an increase in physical size.
The power and control characteristics of this design were
validated through an experiment involving a halogen lamp
load, which served as a model for a filament. Furthermore, we
have future plans to test the insulation performance in
connection with high-voltage power sources.
REFERENCES
[1] H. Obata, N. Tsuji and K. Furumoto, "Frequency Bandwidth
Fig. 5. Experimental waveform lighting a halogen lamp load Narrowing Technology for Pulsed Magnetrons," in IEEE Transactions
on Electron Devices, vol. 56, no. 12, pp. 3191-3195, Dec. 2009, doi:
Fig. 4 shows the experimental environment of the fabricated 10.1109/TED.2009.2031004.
filament heater power supply. The right side of the picture is [2] S. -M. Park, S. -H. Song, H. -B. Jo, W. -C. Jeong, S. -R. Jang and H. -
the charging and control part, and it is the low voltage part. In J. Ryoo, "Solid-State Pulsed Power Modulator for 9.3 GHz 1.7 MW X-
the center is the high voltage part. A transformer and an optical Band Magnetron," in IEEE Transactions on Industrial Electronics, vol.
cable are used to connect the high-voltage part and the 68, no. 2, pp. 1148-1154, Feb. 2021, doi: 10.1109/TIE.2020.2967728.
charging part. In order to model the same filament [3] S. Wang, Y. Shen, C. Liao, J. Jing and C. Liu, "A Novel Injection-
Locked S-Band Oven Magnetron System Without Waveguide
characteristics as the X-band magnetron load targeted in this Isolators," in IEEE Transactions on Electron Devices, vol. 70, no. 4, pp.
paper, halogen lamps of vehicles were configured in series and 1886-1893, April 2023, doi: 10.1109/TED.2023.3243041.
parallel. A 10V 15A rated output experiment was conducted. [4] H. -B. Jo, J. -B. Ahn, W. -C. Jeong, J. -Y. Lee, M. -K. Choi and H. -J.
Fig. 5 shows experimental waveform lighting a halogen lamp Ryoo, "Compact Design of 40 kV 100 A High-Voltage Pulsed-Power
load. A filament load has a low specific resistance value in an Modulator for Driving X-Band Magnetrons," in IEEE Transactions on
initial cold state and is controlled with a constant current Power Electronics, vol. 38, no. 6, pp. 7598-7609, June 2023, doi:
10.1109/TPEL.2023.3244410.
through a constant current control mode. In addition, when the
[5] S. B. Ok, H. J. Ryoo, S.R. Jang, S.H. Ahn, and G. Goussev, “Design
voltage gradually rises and exceeds a certain voltage, it of a high-efficiency 40-kV, 150-A, 3-kHz solid-state pulsed power
operates in constant voltage mode. Fig. 6 is the experimental modulator,” IEEE Trans. Plasma Sci., vol. 40, no. 10, pp. 2569–2577,
waveform of the rated output of the filament heater power Oct. 2012.
supply. Here, the output voltage and output current are 10V
and 15A, respectively, and the resonant current and half-
bridge inverter output waveform are shown, respectively, and
the stable rated operation test result is confirmed.
- 178 -
August 22(TUE) - 25(FRI), 2023
Oral Session 8
16:20~18:00 Wed, August 23, 2023
Room : A434
Chair: Jeonghwan Gwak
(Korea National University of Transportation, Korea)
- 179 -
Simulation of computer vision systems with
artificial intelligence
Perizat Rakhmetova1*, Makabbat Zhamuratova2, Vinera Baiturganova3, Beibit Shingissov4,
Laura Tasbolatova5, Symbat Akparova6
1,2,3
Satbayev University, Kazakhstan
4,5
Academy of Logistics and Transport, Kazakhstan
6
Ultra ALM, Kazakhstan
Author Emails: [email protected], [email protected]
[email protected], [email protected],
[email protected], [email protected],
Abstract—In recent years, the rapid development of artificial decision-making and automation capabilities across various
intelligence technologies has led to significant advances in the field sectors [4, 5]. However, developing and testing these intricate
of computer vision, dedicated to extracting valuable information systems in real-world scenarios presents a multitude of
from visual data. This article presents a comprehensive analysis of challenges, including high computational demands and
computer vision systems modeling, focusing on the integration of potential risks in error-prone environments [6, 7]. Therefore,
AI techniques to enhance their performance and capabilities. a study is intended to draw attention to and consider the role
Through an extensive literature review, this research investigates the and impact of various modeling tools and applications on the
latest methodologies, algorithms, and architectures employed in creation of AI-based computer vision systems. Artificial
computer vision systems to accomplish tasks such as object
intelligence computer vision systems perceive, analyze and
detection, recognition, tracking, and segmentation. The article first
provides an overview of computer vision systems and their
use important insights from visual data, thereby improving
significance in various domains, including autonomous vehicles, decision-making and outreach capabilities in various sectors
surveillance, robotics, and medical imaging. It then delves into the [8]. However, developing and testing complex systems in the
advancements of AI, particularly deep learning techniques such as face of a wide range of problems, including high
convolutional neural networks and recurrent neural networks, which computational requirements and potential risks in
have revolutionized the field of computer vision. The article environments, involves erroneous errors [6].
highlights the effectiveness of AI models in improving the accuracy
The complex coordination and control required for robotic
and efficiency of computer vision tasks, showcasing their
arms can be greatly improved by incorporating AI and
superiority over traditional image processing techniques.
Furthermore, this research explores the challenges and limitations computer vision techniques, which have applications ranging
faced in modeling computer vision systems with AI. Modeling of from precision assembly in industrial settings to delicate
computer vision systems with elements of artificial intelligence is surgical procedures in healthcare [9, 10]. AI offers a robust
carried out in the Matlab/Simulink. Issues such as dataset bias, methodology for modeling the kinematics and dynamics of
scalability, interpretability, and robustness are discussed, along with manipulator systems, as shown in the work of López et al.
proposed solutions and ongoing research efforts. Finally, the article [11]. Neural network models, in particular, provide a powerful
concludes with a comprehensive outlook on the future of computer framework for capturing the complex relationships between
vision systems modeling with AI. It highlights emerging trends and various system parameters. Additionally, AI methodologies
research directions, including multimodal fusion, real-time like neuro-fuzzy systems and teaching learning-based
processing, explainable AI, and edge computing. By understanding optimization techniques have been employed to improve the
the current state-of-the-art and identifying future possibilities, this performance of these robotic systems [12, 13].
research contributes to the advancement and development of more
powerful and intelligent computer vision systems. Computer vision, on the other hand, provides a critical
sensory input for these systems, enabling real-time visual
Keywords— computer vision systems, artificial intelligence, servoing and facilitating robust interaction with the
deep learning, neural networks, object detection and recognition, environment [10]. In particular, convolutional neural
image segmentation, multimodal fusion, edge computing. networks have shown remarkable proficiency in visual data
processing, thereby aiding precise control of manipulators.
I. INTRODUCTION Simulation of these computer vision-based AI systems offers
Technological advances in intelligence are causing an a controlled environment for system testing, optimization, and
interesting and fast connection between artificial intelligence validation [14, 15]. In this context, AI algorithms have shown
(AI) [1] and computer vision detection [2], thus showing the a remarkable ability to track various trajectories of robot
way to numerous breakthroughs in the field of computer manipulator arms efficiently [14]. Simultaneously, advanced
science. Modern AI-enhanced vision systems have begun to control strategies like force sensorless admittance control have
raise alarms across many computing industries, from been developed for the teleoperation of uncertain robot
autonomous vehicles and security monitoring to heightened manipulators, leveraging the capabilities of neural networks
health hazards and more [3]. AI-incorporated computer vision [15].
systems have the capacity to perceive, analyze, and extract Interestingly, modern optimization and learning
essential insights from visual data, thereby improving algorithms have also been incorporated with traditional AI
- 180 -
methodologies to enhance the performance of these systems. segments that collectively cover the entire image or a
For instance, the artificial bee colony optimization algorithm set of contours extracted from the image.
has been successfully combined with fuzzy theory for real-
time machine learning control of articulated robotic Recognition: This process assigns a label to an object
manipulators [16, 17]. based on its descriptors, identifying objects in an
image or a sequence of images.
As AI and computer vision technologies continue to
evolve, novel methodologies such as weighted multiple neural Interpretation: This final process gives meaning to an
network boundary control and adaptive neural network fuzzy array of recognized objects. For instance, in an
sliding mode control have emerged, which demonstrate automated traffic surveillance system, the
enhanced control of manipulator systems [18, 19]. Alongside interpretation process would generate a traffic report
these advancements, there's been a growing interest in the from the input video sequence.
development of intelligent controllers and impact test systems Object Detection, a technique in computer vision, locates
based on six-axis force sensors [20]. instances of objects in images or videos, often leveraging
This paper "Simulation of computer vision systems with machine learning or deep learning. The paper [24] introduces
Artificial Intelligence" explores the synergistic integration of a novel face detection method for security monitoring
AI and computer vision in the context of robotic manipulator systems, leveraging video processing, computer vision, and
systems. Leveraging insights from a wealth of cutting-edge artificial intelligence to enhance efficiency and overcome
research, this investigation aims to illuminate the nuances of traditional limitations like low detection accuracy. The
this complex domain, facilitating better understanding, and proposed system, tested in various complex environments,
fostering further innovation in this exciting technological demonstrated high accuracy, low false alarm rates, and robust
frontier [21]. adaptability, marking it as a promising solution for large-scale
security monitoring applications.
Summarizing the existing work of the authors, there are
problems of simulation computer vision systems with This technique, often powered by machine learning or
artificial intelligence, which include achieving real-time deep learning, involves several key components:
performance while maintaining accuracy and efficiency, the Bounding Box: This is a box drawn around the object
lack of standardized evaluation metrics, and the complexity of interest to localize it within the image. The box is
and parameter tuning of artificial intelligence methods or defined by the coordinates of its top-left and bottom-
architectures. The feasibility, effectiveness, and practical right corners.
implementation of tools for modeling and developing
computer vision systems with AI capabilities also need to be Class Label: Each bounding box typically includes a
explored. class label, indicating the category of the object within
the box, such as 'car', 'pedestrian', or 'bicycle' in a self-
II. METHODS driving car application.
A. Classification of Artificial Intelligence Object Localization: This task involves predicting the
Artificial Intelligence (AI) involves the simulation of location of an object in the image, typically by
human intelligence processes by machines, particularly predicting the coordinates of a bounding box that
computer systems. It includes learning, reasoning, and self- encompasses the object.
correction, and is used in various fields such as medical
Object Recognition: After localizing the object, the
diagnosis [22], stock trading, and scientific discovery.
next step is to assign a class label to it. This is usually
Computer Vision, a subfield of AI, focuses on enabling
done using a classifier trained on a large dataset of
computers to understand and interpret visual data [23]. It
labeled images of the object.
involves tasks like image processing, object detection, and
image segmentation (Figure 1). Image Processing manipulates Non-maximum Suppression: This technique is used
images using algorithms to enhance the image or extract after the object detection algorithm runs on an image
useful information. This type of signal processing takes an and finds multiple bounding boxes for the same
image as input and produces an enhanced image or features object. It keeps only the most probable bounding box
associated with that image as output. The process involves for each object and removes all the predicted
several sub-processes: bounding boxes that have a lower probability of
detecting an object and significantly overlap with a
Preprocessing: This initial step improves the image to
box that has a higher probability.
increase the success rate of subsequent processes. It
includes simple tasks like noise reduction and contrast Scale Invariance: Object detection algorithms need to
enhancement, as well as complex tasks that be scale-invariant to detect objects of various sizes, as
standardize the image's coordinate system. objects in the real world can appear at different scales
in the image.
Feature Extraction: This process identifies a unique set
of features or parameters, such as color, size, shape, Image Segmentation is a process that divides a digital
and texture, that can uniquely identify an object. It image into multiple segments [25], making the image easier to
simplifies the resources needed to accurately describe analyze and more meaningful. It's commonly used to identify
a large dataset. objects and boundaries within images. Key components of
image segmentation include:
Segmentation: This process assigns a label to every
pixel in an image so that pixels with the same label Thresholding: This simple method of image
share certain characteristics. The result is a set of segmentation creates binary images from grayscale
- 181 -
images. A threshold value is selected, and all pixels A rigid body transformation of this point can be
with gray level values below the threshold are represented as:
converted to black, while those above the threshold are
converted to white. 𝑇 ∗ 𝑃ℎ = 𝑃ℎ′
Region Growing: This pixel-based image
segmentation method starts with seed points and where 𝑇 is a 4 × 4 transformation matrix of the form:
groups together neighboring pixels with similar
properties. It's often used in medical imaging to 𝑅 𝑡
𝑇=[ ]
segment structures like bones or lungs. 0 1
Clustering Methods: These segmentation methods Here, 𝑅 is a 3 × 3 rotation matrix, 𝑡 is a 3 × 1 translation
partition the image into clusters based on similar vector, and 0 is a 1 × 3 vector of zeros. The transformed
attributes. K-means is a popular method for image point 𝑃ℎ′ is the point 𝑃 after rotation and translation.
segmentation due to its simplicity and scalability.
Dynamic modeling, on the other hand, involves the study
Edge Detection: This method segments an image into of forces and torques that cause motion. In computer vision,
regions of discontinuity by identifying the boundaries dynamic models are used to predict the system's response to
of objects within images. It works by detecting various inputs and disturbances [14]. These models are often
discontinuities in brightness. more complex than kinematic models and may involve
Watershed Transformation: This technique is useful differential equations that describe the motion of objects over
for image segmentation, especially when the image has time. The exact form of these equations can vary depending
low contrast. It treats the grayscale image like a on the specific application and the complexity of the problem
topographic map, with the intensity of each pixel being addressed.
representing the height, and finds the lines that run Kinematic and dynamic models are used in a variety of
along the tops of ridges [26]. computer vision applications, including object tracking,
Semantic Segmentation: This process links each pixel gesture recognition, and 3D reconstruction. For example, in
in an image to a class label, such as person, car, dog, object tracking, a kinematic model can be used to predict the
cat, etc. Unlike in object detection, pixels are classified future position of an object based on its past trajectory. In 3D
individually in semantic segmentation. reconstruction, kinematic models can be used to infer the 3D
shape and motion of objects from 2D images or video
Machine Learning, another subfield of AI, allows software sequences. Dynamic models, meanwhile, can be used to
applications to become more accurate in predicting outcomes simulate the physical behavior of objects and environments,
without being explicitly programmed. It includes supervised which can be useful in applications such as virtual reality and
learning, unsupervised learning, and reinforcement learning. video game design [19].
These processes collectively allow computers to process
visual data in a way that mimics human vision, with more Despite the significant progress that has been made in the
precision, consistency, and less subjectivity. field of kinematic and dynamic modeling in computer vision,
several challenges remain. These include the difficulty of
accurately estimating the parameters of kinematic and
dynamic models from image data, the computational
Artificial Intelligence
complexity of these models, and the need for robust methods
Computer Vision (CV) Machine Learning that can handle occlusions and changes in lighting conditions.
Image Processing Object Detection Image Segmentation Supervised Learning Unsupervised Learning
Future research in this field is likely to focus on addressing
Preprocessing Bounding box Thresholding Reinforcement Learning
these challenges, as well as on developing new models that
can better capture the complex and dynamic nature of real-
Feature extraction Class label Region growing
world visual scenes.
Clustering
Recognition Object localization
methods
Interpretation: Object recognition Edge detection
III. RESULTS
Non-maximum Watershed This article employed the Сomputer Vision Tools from the
suppression transformation
Semantic
Matlab/Simulink simulation environment for the purpose of
Scale Invariance
segmentation
modeling computer vision systems, as shown in the Figure 2.
Figure 1 – Structure diagram of Artificial Intelligence The toolbox of Computer Vision is a comprehensive suite that
offers algorithms, functions, and apps for designing and
testing computer vision, 3D vision, and video processing
B. Kinematic and Dynamic Modeling of Computer Vision systems. It enables object detection (Figure 3) and tracking,
Systems image segmentation (Figure 4), feature detection, extraction,
and matching, and automates calibration workflows for
Kinematic and dynamic modeling are fundamental aspects
various camera types. The toolbox supports 3D vision tasks
of computer vision systems, providing a mathematical
like visual and stereo vision, structure from motion, and point
framework for analyzing and interpreting the motion of
objects in visual data [9, 17]. In computer vision, kinematic cloud processing. It also facilitates the training of custom
models are used to describe the geometric and time-dependent object detectors using deep learning and machine learning
algorithms, and provides object detection and segmentation
aspects of a system's operation. For instance, given a point 𝑃 algorithms for large images. The toolbox supports C/C++
in 3D space with coordinates (𝑥, 𝑦, 𝑧), we can represent it in code generation for integration with existing code, desktop
homogeneous coordinates as 𝑃ℎ = (𝑥, 𝑦, 𝑧, 1)
- 182 -
prototyping, and embedded vision system deployment, and field of computer vision. The effectiveness of AI models in
can be accelerated on multicore processors and GPUs. improving the accuracy and efficiency of computer vision
tasks is emphasized, demonstrating their superiority over
traditional image processing methods. The paper further
explores the challenges and limitations that come with
introducing AI to computer vision systems, including issues
Figure 2 – Simulation of computer vision systems with AI in
related to dataset bias, scalability, interpretability, and
Matlab/Simulink robustness. Proposed solutions to these problems and current
research initiatives are also detailed. The assimilation of
artificial intelligence components into computer vision
systems is carried out using Matlab/Simulink.
REFERENCES
[1] Perizat Rakhmetova, Asyl-D.Bazarbek. (2023). Analysis of parallel
manipulators with artificial intelligence. The Bulletin of KazATC,
124(1), 237–244. https://doi.org/10.52167/1609-1817-2023-124-1-
237-244.
[2] Gani Balbayev, Aigerim Mussina, Algazy Zhauyt, Beibit Shingissov,
and Marina Kalekeyeva, "Neural Control for Image Stabilisation Using
a Reference Model," International Journal of Mechanical Engineering
and Robotics Research, Vol. 10, No. 1, pp. 17-21, January 2021. DOI:
10.18178/ijmerr.10.1.17-21.
[3] LeCun, Yann & Bengio, Y. & Hinton, Geoffrey. (2015). Deep
Learning. Nature. 521. 436-44. 10.1038/nature14539.
[4] Russakovsky, O., Deng, J., Su, H. et al. ImageNet Large Scale Visual
Recognition Challenge. Int J Comput Vis 115, 211–252 (2015).
Figure 3 – Simulation of object detection in Matlab/Simulink https://doi.org/10.1007/s11263-015-0816-y.
[5] M. Adel, S. M. Ahmed and M. Fanni, "End-Effector Position
Estimation and Control of a Flexible Interconnected Industrial
Manipulator Using Machine Learning," in IEEE Access, vol. 10, pp.
30465-30483, 2022, doi: 10.1109/ACCESS.2022.3157817.
[6] Krizhevsky, Alex & Sutskever, Ilya & Hinton, Geoffrey. (2012).
ImageNet Classification with Deep Convolutional Neural Networks.
Neural Information Processing Systems. 25. 10.1145/3065386.
[7] S. Liang, X. Jianchun and Z. Xun, "An Extraction and Classification
Algorithm for Concrete Cracks Based on Machine Vision," in IEEE
Access, vol. 6, pp. 45051-45061, 2018, doi:
10.1109/ACCESS.2018.2856806.
[8] N. Liu et al., "Modeling and Simulation of Robot Inverse Dynamics
Using LSTM-Based Deep Learning Algorithm for Smart Cities and
Factories," in IEEE Access, vol. 7, pp. 173989-173998, 2019, doi:
10.1109/ACCESS.2019.2957019.
Figure 4 – Simulation of image segmentation in Matlab/Simulink
[9] A. T. Vo and H. -J. Kang, "Neural Integral Non-Singular Fast Terminal
Synchronous Sliding Mode Control for Uncertain 3-DOF Parallel
Robotic Manipulators," in IEEE Access, vol. 8, pp. 65383-65394,
Thus, as shown in Figures 3, object detection results and 2020, doi: 10.1109/ACCESS.2020.2984891.
Figure 4, image segmentation greatly affect AI computer [10] F. Tokuda, S. Arai and K. Kosuge, "Convolutional Neural Network-
vision modeling, affecting performance evaluation, training Based Visual Servoing for Eye-to-Hand Manipulator," in IEEE Access,
and validation, benchmarking, model selection and vol. 9, pp. 91820-91835, 2021, doi: 10.1109/ACCESS.2021.3091737.
optimization, system integration, dataset creation, and [11] E. J. López, D. Servín De La Mora-Pulido, R. Servín De La Mora-
algorithm development. These results serve as critical Pulido, F. J. Ochoa-Estrella, M. Acosta Flores and G. Luna-Sandoval,
"Modeling in Two Configurations of a 5R 2-DoF Planar Parallel
benchmarks for the performance of the system being Mechanism and Solution to the Inverse Kinematic Modeling Using
simulated and guide further advances in AI-assisted computer Artificial Neural Network," in IEEE Access, vol. 9, pp. 68583-68594,
vision. 2021, doi: 10.1109/ACCESS.2021.3073402.
[12] W. Pan, M. Lyu, K. -S. Hwang, M. -Y. Ju and H. Shi, "A Neuro-Fuzzy
IV. CONCLUSION Visual Servoing Controller for an Articulated Manipulator," in IEEE
Access, vol. 6, pp. 3346-3357, 2018, doi:
This research paper provides an in-depth study of the 10.1109/ACCESS.2017.2787738.
current strategies, algorithms, and architectures used in
[13] P. -N. Le and H. -J. Kang, "Robot Manipulator Calibration Using a
computer vision systems to perform tasks such as object Model Based Identification Technique and a Neural Network With the
detection, recognition, tracking, and segmentation. Paper Teaching Learning-Based Optimization," in IEEE Access, vol. 8, pp.
provides a comprehensive overview of computer vision 105447-105454, 2020, doi: 10.1109/ACCESS.2020.2999927.
systems and their important role in various fields, including [14] M. Elsisi, K. Mahmoud, M. Lehtonen and M. M. F. Darwish, "An
autonomous vehicles, surveillance systems, robotics, and Improved Neural Network Algorithm to Efficiently Track Various
medical imaging. Particular attention is paid to advances in the Trajectories of Robot Manipulator Arms," in IEEE Access, vol. 9, pp.
11911-11920, 2021, doi: 10.1109/ACCESS.2021.3051807.
field of artificial intelligence, especially deep learning
[15] C. Yang, G. Peng, L. Cheng, J. Na and Z. Li, "Force Sensorless
methods such as convolutional neural networks and recurrent Admittance Control for Teleoperation of Uncertain Robot Manipulator
neural networks, which have caused a significant shift in the Using Neural Networks," in IEEE Transactions on Systems, Man, and
- 183 -
Cybernetics: Systems, vol. 51, no. 5, pp. 3282-3292, May 2021, doi:
10.1109/TSMC.2019.2920870
[16] H. -C. Huang and C. -C. Chuang, "Artificial Bee Colony Optimization
Algorithm Incorporated With Fuzzy Theory for Real-Time Machine
Learning Control of Articulated Robotic Manipulators," in IEEE
Access, vol. 8, pp. 192481-192492, 2020, doi:
10.1109/ACCESS.2020.3032715.
[17] Zhanibek Issabekov and Nakhypbek Aldiyarov, “Developing the
Mathematical Model of the Bipedal Walking Robot Executive
Mechanism” International Journal of Advanced Computer Science and
Applications(IJACSA), 12(12), 2021.
http://dx.doi.org/10.14569/IJACSA.2021.0121240.
[18] Y. Zhang, Q. Li, W. Zhang, Y. Liu and Z. Xue, "Weighted Multiple
Neural Network Boundary Control for a Flexible Manipulator With
Uncertain Parameters," in IEEE Access, vol. 7, pp. 57633-57641, 2019,
doi: 10.1109/ACCESS.2019.2914077.
[19] Li X, Gao X, Sun L, et al. The adaptive neural network fuzzy sliding
mode control for the 3-RRS parallel manipulator. Advances in
Mechanical Engineering. 2022;14(9).
doi:10.1177/16878132221126112.
[20] Guoxin Hua, Fei Wang, Jianhui Zhang, "Study on Impact Test System
of Manipulator Based on Six Axis Force Sensor and Intelligent
Controller", Journal of Sensors, vol. 2022, Article ID 7869019, 14
pages, 2022. https://doi.org/10.1155/2022/7869019
[21] Li, Z., Li, S.: Recursive recurrent neural network: a novel model for
manipulator control with different levels of physical constraints. CAAI
Trans. Intell. Technol. 1– 13 (2022). https://doi.org/10.1049/cit2.12125
[22] B. Dauren, N. Zhetenbayev, G. Balbayev, B. Shingissov, N. Arailym,
and Y. Saltanat, “Development and simulation of a device for upper
limb rehabilitation,” Vibroengineering PROCEDIA, Vol. 48, pp. 29–
35, Feb. 2023, https://doi.org/10.21595/vp.2023.23167
[23] Y. -T. Su et al., "A Fully Automatic Calibration for Vision-Based
Selective Compliance Assembly Robot Arm and Its Application to
Intelligent Wafer Inspection Scheduling," in IEEE Access, vol. 10, pp.
50100-50113, 2022, doi: 10.1109/ACCESS.2022.3172953.
[24] Z. Dong, J. Wei, X. Chen and P. Zheng, "Face Detection in Security
Monitoring Based on Artificial Intelligence Video Retrieval
Technology," in IEEE Access, vol. 8, pp. 63421-63433, 2020, doi:
10.1109/ACCESS.2020.2982779
[25] W. Yang, "Adaptive Recognition of Motion Decomposition Image
Based on Adaptive Principal Component Extraction Algorithm," in
IEEE Access, vol. 8, pp. 224108-224122, 2020, doi:
10.1109/ACCESS.2020.3044371.
[26] Gloria B.G., Oscar D.S., José Luis E.A., Jesus S.T., Ismael S.G., Noelia
V.E., “Learning Image Processing with OpenCV”, ISBN 978-1-78328-
765-9, Published by Packt Publishing Ltd., 2015, p 232.
- 184 -
Integration of Penalty Functions and
Performance Criteria for Enhanced
Optimization in Optimal Control Problems using
Artificial Immune System Algorithms
Olga Shiryayeva 1*,
1
Satbayev University, IICT, Kazakhstan
[email protected]
Timur Samigulin 2
2
Kazakh-British Technical University, Kazakhstan
[email protected]
Dmirtyi Panchenko 1
1
Satbayev University, Kazakhstan
[email protected]
Abstract—The article explores the construction of optimal enhances the optimization process, enabling the attainment of
control systems using Artificial Immune System (AIS) optimal control solutions that satisfy desired quality
algorithms. By identifying the optimal parameters of the typical assessments. The objective is to minimize the integral
controller that minimize the integral quadratic error (ISE) quadratic error (ISE) of optimal control systems while
integrated with the penalty function, the proposed approach considering penalty functions. Specifically, this study focuses
aims to enhance the performance and efficiency of the optimal on the utilization of AIS as optimization algorithms in the field
control system for the technical system under consideration. The of AI [2]. The proposed approach aims to enhance the
optimization process should consider both the integral efficiency and performance of control systems by integrating
quadratic error (ISE) and the penalty function, ensuring that
AI algorithms and penalty functions into the optimization
the controller parameters yield a control solution that satisfies
process. The use of AIS algorithms has gained significant
the desired quality assessments while minimizing the overall
error. The results of this study are to develop an effective popularity as a mechanism for solving optimization problems
approach that combines the power of artificial intelligence [3]. In the realm of optimal control systems for technical
methods with the integration of penalty functions. Solving this systems and technological objects, AIS algorithms are
problem will contribute to the advancement of optimal control considered intelligent search algorithms inspired by natural
in technical systems by providing a systematic methodology to phenomena.
determine the controller parameters that strike an optimal
The problem at hand involves the optimal control of a
balance between performance and adherence to specific quality
technical system using AIS methods. In this context, the
assessments.
primary objective is to determine the parameters of a typical
Keywords—artificial intelligence, artificial immune system, controller that minimize the ISE while incorporating a penalty
optimal control system, penalty function, typical controller. function. To achieve optimal control, it is crucial to strike a
balance between minimizing the ISE, which measures the
I. INTRODUCTION) deviation between the system's output and the desired
Artificial intelligence (AI) has revolutionized various response, and incorporating penalty functions that capture
domains, including technical systems, by providing advanced specific quality assessments and system requirements. These
optimization algorithms. These algorithms aim to improve penalty functions enable the inclusion of constraints and
system performance and efficiency. Widely used AI considerations such as time regulation, overshoot, rise time,
algorithms for optimal control include Genetic Algorithms and other relevant factors that affect the system's behavior [4].
(GA), Particle Swarm Optimization (PSO), and Artificial The challenge lies in optimizing the parameters of the typical
Immune System (AIS) algorithms, among others [1]. These controller within the framework of the AIS methods.
algorithms possess exceptional characteristics and approaches This article presents the results of evaluating the
that contribute to the effective resolution of optimal control effectiveness of intelligent information technologies for
problems in technical systems. However, while intelligence solving optimization problems in technological processes
algorithms have shown promise, several challenges hinder based on AIS algorithms. Based on the AIS algorithm, which
their effectiveness. These algorithms often focus on is an intelligent optimization method, conditions for solving
minimizing the objective function without considering optimal control problems in technological processes with
specific quality assessments required in optimal control, such performance assessment are presented. The main components
as settling time, overshoot, and rise time. To address this of AIS are discussed, including problem modeling, solution
limitation, penalty functions are proposed as a solution. The representation, evaluation function, and parameter selection.
integration of penalty functions and performance criteria
- 185 -
The importance of adapting AIS algorithms to changing optimization problem of the technological process. It
conditions is also emphasized. is necessary to adjust these parameters to achieve the
best results.
Algorithms based on artificial immune systems have
several advantages for optimization problems, such as the In accordance with the first condition, to enhance the
ability to handle complex problems with a large number of efficiency of population-based algorithms, including
parameters, adapt to changing conditions, and find globally AIS, it is proposed to introduce penalty functions into
optimal solutions. Examples of effective algorithms based on the quality criterion [4]:
artificial immune systems include Cellular Programming,
Clonal Selection Algorithms (CLONALG), Antibody T
Optimization Algorithms, and many others [5]. IE e(t ) Penaltyet min
0
However, it should be noted that intelligent information
technologies for optimization problems based on artificial Where IE is the integral quality criterion, e(t) is the error
immune systems are an active area of research, and their at time t (the difference between the setpoint, SV, and the
application may depend on the specific problem and context actual, PV, value of the system), Penalty(e(t)) is the penalty
of use. The AIS algorithm has indeed gained wide popularity function that depends on the error e(t) and may include various
recently for solving optimization problems, database components, such as penalties for violating constraints or
processing, and decision making. However, like all search undesirable conditions.
algorithms, certain conditions specific to the problem
formulation need to be considered [6]. In optimization problems where the goal is to find the best
solution among a set of feasible solutions, quality criteria and
The implementation of AIS algorithms should be based on penalty functions play an important role in evaluating and
the characteristics and requirements of the specific managing the optimization process. Quality criteria are used
optimization problem, with the selection of quality criteria to measure the effectiveness of a solution, while penalty
aligned with the technological process. Therefore, it is functions help in accounting for and managing constraints
necessary to formulate important recommendations for the associated with solution feasibility. This section examines the
application of intelligent information technologies based on influence of quality criteria and penalty functions on the
AIS algorithms in solving optimal control problems in efficiency of AIS algorithms in optimization problems and
technological processes, with performance assessment linked their role in the process of searching for the optimal solution.
to the choice of quality criteria.
The optimization process can be highly sensitive to the
In accordance with this, the results of developing choice of quality criteria and penalty functions. Appropriate
conditions and requirements for evaluating the effectiveness quality criteria and penalty functions contribute to an efficient
of AIS algorithms are presented in the context of their search for the optimal solution, while an incorrect choice can
implementation for the gas purification process in a lead to undesirable results.
distillation column.
For example, if the penalty function imposes excessive
II. EVALUATION OF THE QUALITY CRITERIA WITH penalties for constraint violations, it may overlook potentially
PENALTY FUNCTIONS good solutions. On the other hand, if the penalty function is
Assessing the effectiveness of AIS algorithms, as too weak, solutions may violate constraints, resulting in
population-based algorithms, is an important task when infeasible solutions. The proper adjustment of quality criteria
studying and applying these methods to technological and penalty functions requires an analysis of the problem
processes. AIS algorithms have several advantages for solving requirements and experimental parameter tuning.
technological process problems, such as searching for optimal For gas industry technological processes, a quality
solutions and the ability to adapt to changing conditions. criterion with penalty functions can be represented by the
However, like all search algorithms, it is necessary to tune the Integral Square Error (ISE), Integral Absolute Error (IAE),
algorithm according to the specific process. Integral Time Square Error (ITSE), and Integral Time
To conduct an evaluation of the effectiveness of AIS Absolute Error (ITAE) [4]. The choice between ISE, IAE,
algorithms, certain conditions of their use need to be ITSE, and ITAE depends on the requirements of the specific
formulated: technological process and optimization priorities, and should
result from the modeling and analysis of these criteria.
It is necessary to mathematically formulate not only
the optimization problem of the technological process Based on the obtained modeling results, the following
but also define a quality criterion that needs to be conclusions can be drawn. The ISE criterion is a quality
optimized based on the given assessments of the criterion in which the error between the setpoint (SP) and the
process quality. Choosing appropriate quality criteria actual system value (PV) is quadratically integrated over time
allows for increasing the efficiency of finding optimal (Fig. 1). ISE gives equal attention to both long-term and short-
solutions and comparing them with other alternatives. term errors. The smaller the ISE value, the closer the system
is to achieving the desired result.
The solutions to the optimization problem should be
represented as antibodies that can be modified and
adapted during the operation of the modified AIS
algorithm based on the properties of biplasticity.
Parameter selection: AIS algorithms have several
parameters that can be tuned according to the specific
- 186 -
and the desired behavior of the optimization algorithm. Proper
tuning of penalty weights and penalty level affects the balance
between achieving optimality and complying with constraints.
Incorporating penalty functions into quality criteria allows for
considering constraints and undesirable conditions during
system optimization. Penalties can be applied to the error
itself, without integration, and their value can be taken into
account when calculating the quality criterion.
In accordance with the processes of the oil and gas
Fig. 1. The results of modeling IE. industry's distillation column, penalty functions related to
overshoot and rising time were introduced [7]. Penalty
The IAE quality criterion is also a criterion in which the coefficients are then tuned to be added to the objective
error between the setpoint and the actual system value is function during optimization. The greater the deviation, the
absolutely integrated over time. IAE gives equal attention to higher the penalty. In general, the quality criterion now takes
both long-term and short-term errors. The difference from ISE the form
is that IAE uses the absolute value of the error instead of its
square (Fig. 2). Penalty=kPov (ymax–ySV)2+krise(trise–trisemax)2min (2)
where kPov is the penalty coefficient for overshoot, ymax is
the maximum output value, ySV is the desired output value, krise
is the penalty coefficient for settling time, trise is the settling
time, and trisemax is the maximum allowable settling time.
This approach to modifying the main quality criterion
allows incorporating custom parameters into penalty
functions, complementing the main objective function. The
penalty coefficient for overshoot, kovershoot, and the penalty
coefficient for settling time, krise, are parameters that account
Fig. 2. The result of the modeling IAE. for requirements related to the transient response, i.e., the
speed and accuracy of the system's transition to the desired
The criteria ITSE and ITAE are modifications of ISE and steady-state value.
IAE, respectively, incorporating an additional time
coefficient. These quality criteria penalize long-term errors Increasing these parameters amplifies the penalties for
more heavily than errors occurring in short-term periods (Fig. overshoot and longer settling time, directing the search
3). Therefore, a system optimized using ITSE or ITAE aims algorithm towards finding parameters that reduce overshoot
for faster convergence to the desired value compared to ISE and expedite the settling time.
and IAE.
The maximum allowable settling time, trisemax, is a
parameter that sets an upper limit on the time it takes for the
system to reach steady-state. It allows considering
requirements for system responsiveness and prevents
excessive system complexity. Increasing this parameter may
reduce the penalty for longer settling time, but it can also
increase overshoot and system oscillations.
Overall, incorporating these parameters allows
considering additional requirements and constraints on the
system's operation. They help find more stable and accurate
Fig. 3. The result of the modeling ITSE. solutions that align with the transient response requirements
and avoid system overcomplication.
ISE can be preferable when precise correspondence Formally, the mathematical representation of ISE and IAE
between setpoint and actual value is the primary requirement. with penalty function (2) can be expressed as follows:
IAE can be preferred when reducing system oscillations is
important. ITSE and ITAE can be useful when achieving
faster system convergence to the desired value and ensuring ISE+Penaltymin (3)
short-term stability are necessary.
Penalty functions can be applied in conjunction with these Penaltymin
quality criteria to account for constraints and optimize the
technological process. They allow penalizing constraint By incorporating penalties into the objective function, the
violations or other undesirable aspects of system operation, system aims to minimize both the error between the desired
ensuring compliance with system requirements and and actual values and the magnitude of the penalties for
limitations. constraint violations (Fig. 4). This helps achieve a balance
Penalty functions can be defined in various ways, between control accuracy and adherence to the required
including linear, quadratic, exponential, and others. The operating conditions of the system.
choice of penalty function depends on the nature of constraints
- 187 -
or finds an acceptable solution. The iterative behavior of the
algorithm and the change in the objective function over time
can be taken into account to determine the convergence
speed. Penalty functions can also be used to guide the
algorithm in avoiding infeasible solutions and accelerating
convergence.
By modifying the objective function and adding
optimization parameters to the AIS algorithm, we have
achieved significant improvements in the convergence speed
Fig. 4. The result of modeling the ISE and IAE with penalty functions. of the algorithm. Figure 6 presents the results of implementing
quality criteria without penalty functions.
Penalties can be applied both to the magnitude of the error
and its duration (Fig. 5).
ITSE+Penaltymin
ITAE+Penaltymin
Where t is the time elapsed since the beginning of
optimization.
Fig. 6. The graph of the convergence speed for the quality criterion.
Figure 7 shows the graph of the quality criterion with
penalty functions. The convergence speed has increased, and
by the 5th iteration, the objective function reaches a minimum
with an error of 0.4.
Fig. 5. Results of modeling the ITSE and ITAE with penalty functions.
These penalties can stimulate the model to make more
efficient use of time or facilitate fast learning on specific tasks.
III. EVALUATION OF THE EFFCTIVENESS OF AIS
ALGORITHMS
According to the conducted research on the adaptability
property of the AIS algorithm based on the bistability
property, as well as the research on quality criteria with
penalty functions, it is possible to introduce indicators for
assessing effectiveness.
The effectiveness evaluation of population algorithms,
including AIS algorithms, can be performed using the Fig. 7. The graph of the convergence speed for the quality criterion with
following indicators: penalty function.
To conduct an evaluation of the effectiveness of AIS The conducted research has resulted in obtaining similar
algorithms, certain conditions of their use need to be optimal coefficients for the PI controller, which were tuned
formulated: to achieve the desired quality assessments of the transient
process. However, the convergence speed of the AIS
1) Solution Quality: This indicator evaluates the accuracy
algorithm has improved. This allows for achieving the
of the solutions obtained using the AIS algorithm. Evaluation desired system behavior when transitioning from one state to
can be done by comparing the found solutions with optimal another and ensures stable overall system performance. This
solutions or with other solutions obtained using alternative approach enhances control quality and improves the overall
algorithms. In order to evaluate the effectiveness of the AIS system efficiency. By using the AIS algorithm and
algorithm in synthesizing an intelligent controller for a introducing additional penalty coefficients for overshoot and
complex control system, experiments and comparative settling time, a more accurate and efficient system behavior
analysis of quality metrics such as settling time, overshoot, is achieved according to the specified requirements.
steady-state error, etc., are conducted. 3) Robustness: Robustness evaluates the ability of the
2) Convergence Speed: This indicator assesses how AIS algorithm to provide consistent results under different
quickly the AIS algorithm converges to the optimal solution conditions or datasets. An algorithm that maintains good
- 188 -
performance across different tasks or parameter variations is The use of penalty functions for quality criteria in AIS
considered more robust. The property of biplasticity and algorithms has been proposed to improve their
adaptation in AIS algorithms is another approach to account performance.
for changing conditions or constraints. Both penalty Through the modeling of quality criteria, it has been
functions and the biplasticity property of AIS algorithms concluded that the convergence speed of the AIS
offer methods to consider changing conditions or constraints algorithm in optimizing the ISE objective function is
in computational algorithms. The choice of approach depends increased when penalty functions for overshoot and
on the specific task and requirements, and may require rise time are incorporated.
experimentation and tuning to achieve desired results. The evaluation results demonstrate that the modified
4) Resource Usage: Evaluating the efficiency of AIS artificial immune system (AIS) algorithm based on adaptivity
may include measuring the utilization of resources such as properties is effective for synthesizing intelligent controllers
computational power, memory, or time. Lower resource costs for process control systems. This opens up new possibilities
to achieve good results indicate a more efficient algorithm. for applying AIS in various domains where precise and
This criterion is relevant when modeling a technological adaptive control is required.
process with controllers implemented based on industrial
PLC software, such as well-known companies like ACKNOWLEDGMENT
Honeywell or Siemens, which are prominent in the
This research has been funded by the Science Committee
distribution of automation and control solutions in the oil and
of the Ministry of Education and Science of the Republic
gas industry. Assessing resource usage helps determine how
Kazakhstan (Grant № APO9258508).
efficiently AIS manages the technological process using the
resources available in the industrial PLC system. If AIS REFERENCES
achieves good results with minimal resource utilization, it
[1] D. Simon, Evolutionary optimization algorithms: biologically-Inspired
indicates high efficiency. and population-based approaches to computer intelligence. New
Jersey: John Wiley & Sons, 2013, 776 p.
The evaluation of algorithm efficiency may also include
other factors such as implementation complexity, ease of use, [2] M. Sanchez, L. Magdalena, “Artificial immune systems applied to
optimal control problems”, Engineering Applications of Artificial
and adaptability to changing conditions. Overall, the Intelligence, vol. 60, pp. 106–118, 2017.
efficiency of an algorithm is assessed within the context of a [3] S. Sundaram, S. Ramachandran, “Artificial immune systems for
specific task and requirements and can be determined based process optimization: A review”, Expert Systems with Applications,
on various criteria. When evaluating the effectiveness of AIS vol. 115. pp. 77–89, 2019.
algorithms with penalty functions and the biplasticity [4] O.I. Shiryayeva, T.I. Samigulin, “Development of a SMART-system
property, it is important to consider the specifics of the for a Complex Industrial Object Control based on Metaheuristic
particular technological process, the adequacy of the penalty Algorithms of Swarm Intelligence”, WSEAS Transactions on Power
Systems., vol. 16, P. 231–240, 2021.
functions, and their impact on performance and convergence
[5] H.S. Bernardino, H.J. Barbosa, “Artificial Immune Systems for
of the algorithm. Conducting experiments on different tasks Optimization”, Nature-Inspired Algorithms for Optimization, vol. 193.
and comparing them with alternative methods will also aid in pp. 389–411, 2009.
assessing the efficiency of AIS algorithms. [6] D. Dasgupta, F.A. Gonzalez, “Artificial immune system: Models,
algorithms, and future directions”, Evolutionary Intelligence, vol. 8(3-
IV. CONCLUSION 4), pp. 119–139, 2015.
The following main results have been obtained: [7] G. Kalani, Industrial process control: advances and applications. USA:
Elsevier Science, 2002, 197 p.
Conditions for evaluating the effectiveness of artificial
immune system (AIS) optimization algorithms have
been established by formulating quality criteria that
enhance the efficiency of finding optimal solutions.
- 189 -
A Comparative Study on One-Class
Classification Based Autoencoder Models for
Anomaly Detection in Brake Operating Units
Inki Kim 1, Jeong Won Kang2 Jeonghwan Gwak 1,*
1
Dept. of IT·Energy Convergence, Korea National University of Transportation, Korea
2
Department of Transportation System Engineering, Graduate School of Transportation, Korea National University of Transp
ortation, Korea
3
Dept. of Software, Korea National University of Transportation, Korea
Email: [email protected], {jwkang, jgwak}@ut.ac.kr
Abstract— Railway transportation systems are intricate However, these mechanisms are primarily reactive, coming
networks of technology, involving numerous interlinked and into play after a malfunction has occurred. The present
dependent components. Among these, the braking system, research intends to shift this paradigm to a more proactive
specifically the Brake Operating Unit (BOU), plays a crucial stance, where potential malfunctions, such as delays
role. Its malfunction, defined as a 4-second delay compared to exceeding four seconds in the Brake Cylinder (BC) response
normal operation, could result in catastrophic outcomes, time, are detected and rectified even before they occur. An
underscoring the significance of early fault detection integral part of these railway systems is the braking system.
mechanisms. This paper focuses on the BOU, utilizing pressure The function of a braking system is not just limited to
data from the Brake Cylinder (BC) for anomaly detection. We
controlling the speed of the train and making necessary stops,
propose a novel deep learning approach based on the One-Class
Classification (OCC) concept, where models are trained solely
but it also ensures the overall safety of the railway system.
on normal data, using autoencoder models. The ensuing high Current inspection methods, such as visual inspections and
reconstruction errors are indicative of anomalies, offering a sensor data monitoring, do not provide a comprehensive
proactive approach towards fault detection, thereby mitigating solution for fault detection. Therefore, the introduction of
potential risks. The Long Short-Term Memory Autoencoder machine learning, specifically deep learning techniques,
(LSTM-AE), Bidirectional Gated Recurrent Unit Autoencoder provides a much-needed proactive approach to fault detection
(BiGRU-AE), and Anomaly Transformer models have shown to and prevention. In this paper, we propose a novel method to
outperform other architectures in our comparative analysis, detect anomalies in the brake system of metro vehicles,
with the Anomaly Transformer demonstrating the highest recall specifically the Brake Operating Unit (BOU). We employ the
score, indicating a lower false negative rate which is of utmost concept of One-Class-Classification (OCC) and autoencoder
importance in safety-critical systems like railway brakes. models for this purpose.
However, these results are specific to the dataset used,
necessitating comprehensive testing and evaluation before The previous study [1] demonstrated the effectiveness of
deployment. Our approach is a significant stride towards LSTM-based models for anomaly detection, highlighting their
proactive fault detection and prevention in critical systems. strengths in handling time-series data and their superior
performance compared to 1D-CNN and bi-directional RNN
Keywords— Anomaly Detection, Brake Operating Unit, Metro models. However, it also pointed out areas for improvement,
Vehicle Brake System, One Class Classification, Autoencoder including real-time system compatibility and handling
datasets with fewer anomalies. Building upon this
I. INTRODUCTION foundational work, our current study conducts a
The role of railway transportation in modern urban comprehensive performance analysis of various models,
infrastructure is beyond mere transportation. It forms the including LSTM-AE, GRU-AE, BiGRU-AE, BiLSTM-AE,
backbone that holds the city together, providing reliable and and the Anomaly Transformer, under a fixed threshold of 0.95.
fast connectivity, and contributing to economic growth. This fixed threshold was not optimized, underscoring our
Maintaining the operational efficiency and safety of these primary objective to evaluate the effectiveness and robustness
systems is of paramount importance. Traditional methods of of these models under non-ideal conditions. Our work,
maintaining these systems revolve around regular therefore, contributes to the existing literature by adding a
maintenance checks and fault detection mechanisms. unique dimension of analysis on the performance of these
models.
Fig. 1. A visual representation of a sample from the Brake Operating Unit (BOU) dataset
- 190 -
II. PROPOSED METHOD utilized several architectural models in our study. This
includes the One-Dimensional Convolutional Neural
A. Dataset
Network Autoencoder (1D-CNN-AE) [2], Long Short-Term
Our study utilizes a large and comprehensive dataset Memory Autoencoder (LSTM-AE) [3], Gated Recurrent Unit
gathered from the Brake Operating Unit (BOU) of a metro Autoencoder (GRU-AE) [4], Bidirectional Long Short-Term
vehicle. The data focuses on the pressure readings from the Memory Autoencoder (BiLSTM-AE) [5], Bidirectional
Brake Cylinder (BC). This is a primary component of the Gated Recurrent Unit Autoencoder (BiGRU-AE) [6], and
braking system, and its efficient functioning is paramount to Anomaly Transformer [7]. These models were trained solely
the safety of the metro system. The dataset we have chosen on normal instances, following the OCC concept. The trained
for our research encompasses 14,400 instances of normal models were then tested on the test set, and their
operational data. This is our training dataset, and we use this performances were evaluated based on precision, recall, and
to train our models on what a 'normal' operation looks like. In F1-score metrics.
addition to this, we have 14,673 instances in our test datasets.
This includes 55 instances of anomalies, which we use to test III. EXPERIMENT RESULTS AND ANALYSIS
our models' efficacy in anomaly detection. To better
A. Experimental Environment
understand the data, Figure 1 provides a visual representation
of a sample from Case1 of the BOU dataset used in the We conducted all our experiments on a robust, high-
experiment. It shows a clear distinction between BC and AC performance workstation. This machine was equipped with a
pressure readings. The importance of segregating these 3.50 GHz Intel(R) i9-11900KF CPU and had 64GB of RAM.
readings lies in their distinct operational parameters and An NVIDIA 3090Ti GPU with 24 GB of VRAM
resultant data patterns. This differentiation allows for a more complemented this configuration, providing ample
precise and effective anomaly detection. To better understand computational resources to effectively train and test our
the data, Figure 1 provides a visual representation of a sample models. For implementation of the models, we used Python
from Case1 of the BOU dataset used in the experiment. The 3.8 as the programming language. The deep learning models
importance of segregating these readings lies in their distinct were built and trained using TensorFlow 2.5, a widely-used
operational parameters and resultant data patterns. This open-source library for machine learning.
differentiation allows for a more precise and effective
anomaly detection. B. Performance Evaluation
In our study, we evaluated the performance of each model
B. OCC Autoencoder for BOU Anomaly Detection using three metrics: precision, recall, and F1-score. Precision
The backbone of our anomaly detection mechanism is the is a measure of the correctness achieved in positive prediction
concept of One Class Classification (OCC). In a typical while recall measures the ability of the model to detect all
machine learning model, the model is trained with both positive instances correctly. The F1-score, on the other hand,
'normal' and 'abnormal' data. This allows the model to learn provides a balance between precision and recall. It is the
the distinction between the two. However, in OCC, the model harmonic mean of these two metrics, taking both false
is trained exclusively with normal operational data. This positives and false negatives into account.
allows the model to learn the underlying data patterns and
TABLE 1. PERFORMANCE COMPARISON OF DIFFERENT MODELS
operational norms of a normal operation. Once trained, these
models are capable of accurately recreating the input data. Model Precision Recall F1 Score
However, when an anomaly occurs, these models fail to
1D-CNN-AE 0.6742 0.1701 0.2717
reconstruct it accurately. This failure to reconstruct leads to
high reconstruction error, which can be quantified by the LSTM-AE 1.0 0.7946 0.8855
following formula: GRU-AE 1.0 0.6594 0.7947
1 BiLSTM-AE 0.9471 0.4522 0.6121
𝐿= ∑𝑥 ||𝑥 − 𝑥̂||2 ()
2 BiGRU-AE 1.0 0.6630 0.7973
Anomaly
In this equation, L denotes the reconstruction error, x 0.5409 1.0 0.7021
Transformer
represents the original input data, and x̂ is the reconstructed
data by the autoencoder. In an instance where the Table 1 provides a comparison of the performance metrics
reconstruction error (L) exceeds a pre-defined threshold, the including precision, recall, and F1-Score for each of the
instance is classified as an anomaly. To ensure the model's models studied in this research. Precision measures the
adaptability to dynamic data and evolving system behavior, correctness achieved in positive prediction, whereas recall
the threshold is regularly adjusted based on the following measures the ability of the model to detect all positive
formula: instances correctly. F1-score is the harmonic mean of
precision and recall, providing a balanced metric that takes
𝑇ℎ𝑟𝑒𝑠ℎ𝑜𝑙𝑑 = 𝑇ℎ𝑟𝑒𝑠ℎ𝑜𝑙𝑑 × 0.95 () both false positives and false negatives into account.
Precision Analysis: Precision measures the correctness
This adjustment makes the model more resilient to sudden achieved in positive prediction, where a higher precision
changes in the operational norms. means fewer false positives. From Table 1, LSTM-AE, GRU-
To cover the wide range of potential faults and AE, and BiGRU-AE all demonstrate perfect precision of 1.0,
malfunctions that might occur in the braking system, we have
- 191 -
signifying their impressive ability to accurately classify As a result, the Anomaly Transformer also stands out as a
anomalies in the braking system. However, the Anomaly strong candidate for such applications. However, it's
Transformer, despite its stellar performance in recall, shows important to note that the final choice of the model would
a relatively lower precision of 0.5409, indicating a higher rate depend on the specific requirements and trade-offs in
of false positives, i.e., instances where normal behavior is precision and recall based on the application context.
mistakenly classified as an anomaly. This could potentially Furthermore, comprehensive testing and evaluation on
lead to unnecessary investigations or system interruptions, different datasets are necessary to validate the performance
hampering the efficiency of operations. of these models before they can be deployed in real-world
Recall Analysis: Recall, on the other hand, measures the scenarios. Future work could focus on enhancing the
ability of the model to detect all positive instances correctly, precision of the Anomaly Transformer while maintaining its
where a higher recall means fewer false negatives. In the excellent recall rate. This would provide a model that is not
context of safety-critical systems like railway brakes, high only highly sensitive in identifying true anomalies but also
recall is crucial as false negatives (i.e., failing to identify a precise in minimizing false alarms.
true anomaly) can have potentially disastrous consequences.
From Table 1, the Anomaly Transformer exhibits the highest IV. CONCLUSIONS AND FUTURE WORK
recall, indicating its superior capability in capturing
anomalies. LSTM-AE and GRU-AE also perform strongly in The field of railway transportation systems is a complex
this regard with recall scores of 0.7946 and 0.6594 web of interconnected components, where the safe and
respectively. efficient operation of each part is critical to the overall system.
F1-Score Analysis: F1-Score, the harmonic mean of With an emphasis on the braking system, specifically the
precision and recall, provides a balanced metric that takes Brake
both false positives and false negatives into account. It is
especially important when dealing with imbalanced datasets,
where one class significantly outnumbers the other. In our
study, LSTM-AE performs the best in terms of F1-Score
(0.8855), demonstrating a good balance between precision
and recall. This makes LSTM-AE a reliable model for
anomaly detection, as it minimizes both types of errors.
GRU-AE and BiGRU-AE also deliver competitive F1-Scores
of 0.7947 and 0.7973 respectively.
In conclusion, each model displays distinct strengths and
areas for potential enhancement. LSTM-AE, GRU-AE, and
BiGRU-AE showcase robust performances across all metrics,
demonstrating their potential as robust models for anomaly
detection in railway braking systems. Meanwhile, the
Anomaly Transformer, with its superior recall performance,
provides an invaluable advantage for detecting anomalies in
safety-critical systems, despite its lower precision. This high
recall can be particularly beneficial in applications such as
railway braking systems, where the cost of failing to detect
an anomaly (a false negative) can be potentially catastrophic.
- 192 -
Advancement in Rail Defect Detection Using
YOLO-NAS: A Neural Architecture Search
Approach
Namjung Kim 1 Junhwi Park 2 Jeong Won Kang 3 and Jeonghwan Gwak 1,2 *
1
Dept. of Software, Korea National University of Transportation, Korea
2
Dept. of AI Robotics Engineering, Korea National University of Transportation, Korea
3
Dept. of Transportation System Engineering, Korea National University of Transportation, Uiwang 16106, Korea
Email: [email protected]
Abstract—The structural integrity of railway rails is of proposed YOLO-NAS model takes advantage of these
paramount importance for ensuring the safety and efficiency of distinctive differences to efficiently identify anomalies in images
rail transport. Any defects in the rail structure, such as cracks captured by track inspection vehicles. This research is poised to
or deformations, can lead to severe accidents, emphasizing the contribute significantly to the advancement of more effective,
urgent need for effective and accurate defect detection methods. automated defect detection methods, ultimately enhancing the
Traditional approaches largely depend on manual inspection of safety and reliability of rail transportation.
rail images, obtained via track inspection vehicles, which can be
both labor-intensive and time-consuming. To address this issue, Keywords— Rail Surface Detection, Deep Learning, Anomaly
we propose a deep learning-based technique for identifying Detection, YOLO-NAS
defects in railway rails. The proposed model employs a blend of
computer vision algorithms for rail detection and the YOLO- I. INTRODUCTION
NAS system. In a sound rail structure, certain visual and The 'rail,' a critical constituent of a railway system, forms
geometric consistencies are expected, whereas defective rails the metallic track on which trains operate, directly interfacing
exhibit noticeable deviations from these consistencies due to with the train wheels. This interaction has profound
factors like environmental impact or wear over time. Our
implications for passenger safety and the stable functioning of
Fig. 1. Experiment Overview
- 193 -
trains. Therefore, it’s highly important task to do monitoring image augmentation strategies such as vertical and horizontal
rails. In South Korea, there is the criteria for rail diagnostics flipping to fortify the robustness of our model during training.
to be specified in “Guidelines for Track Maintenance and
Management” provided by Korea National Railway [1]. The
guidelines stipulate that the lifespan of a '50 kg rail' on a C. Experiment
straight line is approximately 500 million metric tons In this experiment, regarding the presence of anomalies
(accumulated tonnage). Rails may also manifest various on rails, various types were mentioned in introduction.
surface defects, including but not limited to, Corrugation, However, for this study, we conducted Bounding box
Flaking, Shelling, Head Checks, and Spalling. labeling specifically for the locations where rail
Among the structural elements of a rail, the Head, being abnormalities occurred and unified all classes as “Defect”.
the primary contact point with the train's wheels, is susceptible This experiment for detecting defect was performed by
to significant damage and wear. Therefore, regular and regions of interest (ROIs) extracted from the raw images. The
comprehensive inspection of rail conditions is essential. width of all raw images is 2048, the height of those, however,
Traditional approaches predominantly rely on the visual and vary individually. Besides, there is a difference in the width
manual inspection of rail images, a process that is both time- of ROIs. For the reason, we resized their width and height to
consuming and prone to oversight. Recent efforts in railway 180 and 800 (i.e., average values in our dataset) each other.
track monitoring have endeavored to employ Artificial The training phase involved pre-trained weights derived
Intelligence (AI) models for defect detection to circumvent from the COCO dataset, with batch sizes of 16, and a patience
these issues [2, 3, 4]. parameter of 50 for early stopping. To assess the performance
To this end, we propose a novel method for defect of our model, we employed YOLOv5 and YOLOv8 as
detection on rail surfaces using a deep learning model that benchmark models.
integrates a computer vision algorithm and the YOLO-NAS
system [5]. YOLO-NAS overcomes the limitations observed III. EXPERIMENT RESULTS AND DISCUSSION
in previous YOLO models by harnessing the power of Neural The performance metrics of the detection models are
Architecture Search (NAS) and quantization. Its in-house summarized in Table I.
NAS system, AutoNAC, autonomously redesigns the model's
architecture to optimize performance concerning speed, TABLE I. PERFORMANCE COMPARISION OF THE DETECTION MODELS
memory usage, and throughput. Quantization is employed to
convert weights, biases, and activations into integer values, Model Precision Recall F1 Score mAP0.5 mAP0.5:0.95
rendering the model more efficient. The model employs YOLOv5 0.8130 0.7950 0.8039 0.8600 0.5540
quantization-aware blocks (such as QARepVGG Block [6]) YOLOv8 0.8220 0.8200 0.8210 0.8820 0.5420
and a hybrid quantization method to maintain an optimal YOLO-NAS 0.9851 0.8684 0.9231 0.9025 0.6346
balance between accuracy and latency. Note: The values in bold represent the highest performance score for each performance indicator.
II. PROPOSED METHOD
A. Dataset
The dataset employed in our experiment is private and
was obtained from a pertinent institution for research
purposes. It comprises 750 images, allocated as follows: 600
for training, 100 for validation, and 50 for testing. Apart from
the testing data, all other images were subjected to random
data augmentation to increase their diversity.
B. Image Preprocessing
Given that our images were in grayscale, we applied pixel
binarization based on contrast to effectively extract the Fig. 2. Prediction results of each models.
features. We employed a computer vision-based contour
algorithm to isolate rail images for the experiment. As hypothesized, the YOLO-NAS outperformed the other
Furthermore, we exploited certain inherent characteristics of models. Analysis of these results led to the following key
rails in the images: insights. Firstly, the superior performance of YOLO-NAS can
be attributed to its integration of QARepVGG Blocks, which
effectively resolved the trade-off between computational
1) The rails are located around the center of the images.
speed and accuracy inherent in previous YOLO models. This
2) Rails are oriented vertically, and their pixels are strategic deployment of QARepVGG Blocks was deemed
generally brighter than the surrounding pixels. instrumental in YOLO-NAS's high-performance scoring.
Secondly, YOLO-NAS utilizes AutoNAC for the
Utilizing these properties, we extracted the relevant development of an optimized architecture, which bolsters both
features denoted by coordinates x, y, w, and h. Our the accuracy and efficiency of the model in terms of
preprocessing was thus focused on the rail portions located processing time. Given that other YOLO models lack the
centrally within the images. Additionally, we incorporated AutoNAC feature, their overall efficiency may suffer in
comparison to YOLO-NAS. This optimized architecture
- 194 -
renders YOLO-NAS a more efficient and effective tool. These [5] J. R. Terven, D. M. Cordova-Esparaza, “A comprehensive review of
findings suggest that the YOLO-NAS's superior performance YOLO: From YOLOv1 to YOLOv8 and beyond,” arXiv preprint, 2023,
arXiv:2304.00501
is inherently tied to its ability to establish the optimal
[6] X. Chu, L. Li and B. Zhang, “Make RepVGG Greater Again: A
architecture for our dataset. Quantization-aware Approach,” arXiv preprint, 2022,
arXiv:2212.01593
IV. CONCLUSION & FUTURE WORK
Traditional methods for detecting rail defects have proven
to be inefficient and labor-intensive. To address this, we have
introduced a deep learning-based model for defect detection.
This work entailed a series of experiments aimed at gauging
the efficacy of anomaly detection in rail images using the
YOLO-NAS model. We then analyzed these results to assess
the model's performance in rail defect detection. In alignment
with our initial expectations, the YOLO-NAS model
demonstrated high numerical performance. Our interpretation
of these results is twofold. First, YOLO-NAS was able to
effectively address the trade-off between computational speed
and accuracy, thanks to the use of QARepVGG Blocks.
Second, the implementation of AutoNAC led to a more
efficient architecture tailored to our dataset. Furthermore,
YOLO-NAS outperformed its quantitative results with
remarkable qualitative performance, as illustrated in Fig. 2.
The model effectively detected a wide range of defect types,
reinforcing the applicability of YOLO-NAS for rail defect
detection. Future work will focus on the development of a
comprehensive dataset to compare the performance of various
detection models. We will also expand our research to
facilitate class-specific detection of rail defects, further
improving the precision and utility of the model.
ACKNOWLEDGMENT
This work was supported by the Basic Science Research
Program through the National Research Foundation of Korea
(NRF) funded by the Ministry of Education (Grant No. RS-
2023-00248444)
REFERENCES
[1] Korea National Railway(KR), “Guidelines for Track Maintenance and
Management”
[2] I. K. Kim, Y. H. Jeon, J. W. Kang and J. H. Gwak, “RAG-PaDiM:
Residual Attention Guided PaDiM for Defects Segmentation in
Railway Tracks,” Journal of Electrical Engineering & Technology, Vol.
18, pp. 1429-1438, 2022, doi: 10.1007/s42835-022-01346-2
[3] B. J. Kim, Y. H. Jeon, J. W. Kang and J. H. Gwak, “Multi-task Transfer
Learning Facilitated by Segmentation and Denoising for Anomaly
Detection of Rail Fasteners,” Journal of Electrical Engineering &
Technology, Vol. 18, pp. 2383-2394, 2022, doi: 10.1007/s42835-022-
01347-1
[4] M. C. Nakhaee, D. Hiemstra, M. Stoelinga and M. van Noort, “The
recent applications of machine learning in rail track maintenance: A
survey,” In Proceedings of the International conference on Reliability,
Safety, and Security of Railway Systems, pp.91-105, 2019, doi:
10.1007/978-3-030-18744-6_6
- 195 -
August 22(TUE) - 25(FRI), 2023
Oral Session 9
16:20~18:00 Wed, August 23, 2023
Room : A435
Chair: Won Sik Moon (Soongsil University, Korea)
- 196 -
A Neural Network Based Approach for
Disease Prediction Using Symptom Data:
Uncovering Hidden Relationships and
Improving Diagnostic Accuracy
Zoljargal Jargalsaikhan1, Zolzaya Dashdorj1*, Stanislav Grigorev2, Tae-Koo Kang3,
Munguntsatsral Dovdondash1, and Erdenebaatar Altangerel1
1
Mongolian University of Science and Technology, Ulaanbaatar, Mongolia
2
Irkutsk National Research Technical University; Irkutsk, Russia
3
Sangmyung university, Seoul, Republic of Korea
*[email protected]
Abstract—In the field of health-care and bio-medical decision trees, random forests, support vector
research, understanding the relationship between the machines, and neural networks, including deep
symptoms of diseases is crucial for early diagnosis and learning models like convolutional neural networks
determining hidden relationships between diseases. The (CNNs) and recurrent neural networks (RNNs), are
study aimed to understand the extent of symptom types
in disease prediction tasks. In this research, we analyze a
employed for disease prediction. These models can
pre-generated symptom-based human disease dataset and learn complex patterns and relationships from
demonstrate the degree of predictability for each disease symptom data, enabling accurate predictions. Disease
based on the Convolutional Neural Network and the prediction models based on network analysis have
Support Vector Machine. The diversity of symptoms and gained attention. These models capture the
diseases is studied using the Principal Component relationships and interactions between symptoms,
Analysis. Our results indicate that machine learning can diseases, and biological entities like genes and
potentially diagnose diseases with 98-99.9% accuracy proteins. Network-based approaches reveal hidden
taking the characteristics of symptoms into account. Our associations and provide insights into disease
result highlights that types of uncommon symptoms are a
good proxy for disease early identification accurately. We
mechanisms and potential therapeutic targets. This
also highlight that uncommon symptoms increase the research work aims to develop a network-based
accuracy of the disease prediction task. approach for disease prediction using symptom data. In
this study, we analyze a disease-symptom relation
Keywords—health-care data analytics, bio-computing, network to understand the characteristics of patients’
machine learning, deep learning, disease network symptoms in terms of occurrence to improve a disease
diagnosis and prediction task. We demonstrate the
I. INTRODUCTION study by estimating the disease predictability and the
The rapid advancement of machine learning and relation between diseases and symptoms using
data analytics has paved the way for innovative machine (deep) learning techniques (SVM and CNN)
approaches to disease prediction. One promising based on word embedding. By constructing a
method is the utilization of a network-based approach symptoms-disease network, we can capture the
that leverages symptom data to predict and understand interconnectedness between symptoms and diseases,
disease outcomes. By integrating diverse data sources unveiling hidden patterns and associations that can
and mapping the relationships between symptoms and enhance diagnostic accuracy and treatment
diseases, this approach offers a comprehensive decision-making.
understanding of the complex interactions underlying
human health. Many applications are developed II. RELATED WORKS
including health-care chatbots, and disease diagnosis The state of the art in disease prediction given
by CT, MRI images. Integration of Multiple Data symptoms involves the application of advanced
Sources: State-of-the-art disease prediction models machine learning techniques to analyze patient data
incorporate diverse data sources, including electronic and provide accurate predictions. One notable
health records (EHRs), genomic data, wearable approach is the utilization of deep learning models,
devices, social media, and environmental factors. By such as recurrent neural networks (RNNs) and
integrating these data, researchers aim to capture a convolutional neural networks (CNNs), which have
comprehensive view of an individual's health status shown promising results in capturing complex patterns
and improve the accuracy of predictions. Machine and relationships in symptom data. A study by Choi et
Learning and Deep Learning Techniques [10-16]: al. [5] introduced the Retain model, an interpretable
Advanced machine learning algorithms, such as
- 197 -
predictive model for healthcare that utilizes a reverse layers were designed with the following
time attention mechanism. The model incorporates hyper-parameters. The 1st layer is a one-dimensional
patient symptoms over time to predict disease convolutional layer of 64 filters, 2 kernels and Relu
outcomes, allowing for effective disease prediction and activation function. We add Dense (16 units),
interpretability. The Retain model has demonstrated its MaxPooling1D and Flatten layers into the model. The
potential in capturing longitudinal information and output layer contains the number of output classes and
providing insights into the decision-making process. 'softmax' activation. The evaluation results of those
Furthermore, the integration of electronic health models are estimated by F1-score, Precision, Recall.
records (EHRs) has emerged as a crucial component in
disease prediction. Researchers, such as Jensen et al.
(2012), have focused on mining EHR data to develop
robust disease prediction models. By leveraging
large-scale EHR databases, these models can harness
comprehensive patient information, including
symptoms, medical history, and clinical measurements,
to enhance disease prediction accuracy. In addition to
Figure 1. Convolutional Neural Network architecture
EHR data, the incorporation of other modalities, such
as genetic data and medical imaging, has gained 3.3. Symptoms dimensionality reduction
attention in disease prediction research. By integrating Given the high-dimensional nature of symptom
multiple data sources, researchers can enhance the data, feature selection and dimensionality reduction
predictive power of models and provide a more techniques are used to identify the most informative
comprehensive understanding of diseases. Overall, the and relevant symptoms. Feature selection methods like
state of the art in disease prediction given symptoms genetic algorithms, recursive feature elimination, and
involves the use of advanced machine learning models, L1 regularization help eliminate redundant or
integration of diverse data sources, and the irrelevant symptoms, enhancing prediction
development of interpretable models. These performance. Diseases can present with a wide range
advancements hold significant potential for improving of symptoms, varying in severity, duration, and
diagnosis and treatment decisions, ultimately leading combination. The diversity of symptoms makes disease
to better healthcare outcomes. Overall, the study by prediction challenging as different individuals may
Zhou et al. [1] provides a comprehensive analysis of exhibit different sets of symptoms for the same
the symptoms–disease network, shedding light on the disease. Using PCA techniques, we reduce the number
relationships between symptoms and diseases. The of symptoms to detect the degree of essential features
findings highlight the potential for utilizing as symptoms for each disease.
network-based approaches in disease diagnosis,
prediction, and understanding of disease mechanisms. IV. DATA COLLECTION
We used a total of 4,442 diseases and 322
III. METHODOLOGY symptoms from PubMed articles with 7,109,429 (about
The following 3-stage research was conducted using 35.5% of over twenty million records)
machine learning. disease/symptom terms in the MeSH metadata field
[1]. This data represents 98.5% of all symptoms and
3.1. Analyze common and unusual symptoms. 95.0% of all diseases in the MeSH vocabulary. TF/IDF
It is crucial to explore hidden links between diseases to was used to estimate the association between
understand their characteristic difference. Symptoms symptoms and disease. Using clinical notes used in
are commonly observed in diseases and are essential to [17], a total of 4,920 patient records were obtained in
diagnose disease primarily. We analyze common and this research.
unusual symptoms in the disease-symptom network
based on the occurrence, and estimate the rate of the
uniqueness of the symptoms.
3.2. Estimate the degree of predictability of disease
based on symptoms.
Based on common and unusual symptoms, we attempt
to diagnose disease in an early state based on Support
Vector Machine and CNN algorithms. In the data
preprocessing stage, we applied a bag of words of
natural language processing methods, representing
symptoms and disease types in free text. The
architecture of CNN is configured as a multi-layer
network that is designed to require minimal data Figure 2. Network of diseases and symptoms
processing, as described in Figure 1.. Sequentially 5
- 198 -
Those research studies obtained a 95.12% accuracy predictability. The network is visualized in Figure 2.
score in disease prediction tasks by employing The average disease-symptom linkage degree of the
Decision Tree, Random Forest and Naive Bayes disease-symptom correlation network is 1.882. The
classifiers. The dataset of 41 types of disease network is modeled to have a diameter of 10, a radius
consisting of 135 symptoms was used, and the degree of 5, and an average shortest path length of 4.2. We
of symptom severity was graded on three levels. Every identified common and uncommon symptoms based on
disease in our dataset is associated with up to 18 the symptoms occurring in diseases, as shown in
symptoms. Figure 3.
V. EXPERIMENTAL RESULTS
We first build a network of diseases and symptoms
from the clinical note dataset to understand the
Figure 3. Common vs Uncommon symptoms across diseases
Uncommon symptoms were not observed in some Contrary to unusual symptoms, some symptoms occur
diseases, such as Chickenpox, Chronic cholestasis, commonly. For instance, a symptom occurring in 17
Heart attack, Jaundice, Malaria, Hepatitis A, C, D, and diseases is 2. More than two occurrences of symptoms
Hyperthyroidism. The occurrence of symptoms over in the disease are almost 50% of the symptoms. That
diseases is estimated in Figure 4. highlights that most symptoms commonly occur over
disease, and ambiguity of diseases is due to the
common symptoms. On average, the rate of unusual
symptoms occurrence for each disease is around
39.2%, except for the diseases without any unusual
symptoms. The diseases with more than 50% of
uniqueness rate of symptoms are predicted relatively
well. On average, the rate of unique symptoms
obtained for each disease is around 39.2% except for
those diseases which do not have any unique
symptoms.
Table 1. Evaluation performance of
machine learning models
Algorithms F1-score Precision Recall
Figure 4. Uniqueness rate by symptoms
SVM (common 98.2% 98.4% 98%
A symptom occurring only in a single disease is 84, a symptoms)
symptom occurring in 2 diseases is 20, and so.
- 199 -
F1-score were not correlated to the uniqueness of
SVM (common 99.2% 99.2% 99.2%
symptoms and the number of symptoms. The reason
+ uncommon
could be related to the insufficient dataset. However,
symptoms)
understanding common and unusual symptoms are
essential in disease prediction tasks; we try to reduce
CNN (common 99% 99% 99%
the number of symptoms by employing Principal
symptoms)
Component Analysis (PCA). Figure 6 shows the
reduced number of symptoms compared to the
CNN 99.9% 99.9% 99.9%
accuracy of the SVM model. At least four types of
(common +
symptoms, regardless of common or unusual
uncommon
characteristics, should be defined for each disease to
symptoms)
have a predictability rate of more than 91% by a k
(k=5) fold cross-validation.
We trained the disease prediction models that employ
CNN and SVM methods in our split data 80/20. The
dataset is well-balanced. The performance of the
models are evaluated using F1-score, Precision, and
Recall. The performance result in macro averaged
metrics presented in Table 1 explains the predictability
of diseases. Given symptoms, predicting a particular
disease is 99.9% of the F1-score. The evaluation
results were relatively good, 98% - 99.9% considering
common and unusual symptoms. These results indicate
that machine learning can potentially diagnose diseases
in the early stage, taking the characteristics of
symptoms into account. Our model outperformed the
other experiments [17] using the same dataset, with a Figure 6. Reduced number of PCA features
3-5% increase. We also highlight that unusual
symptoms increase the accuracy of the disease Figure 7 presents the Silhouette score of K-means on
prediction task. the data containing PCA-reduced symptoms and all
symptoms. PCA reduction improved K-means
performance as the Silhouette score is higher than 0.5
on the dataset containing PCA-reduced symptoms. The
Silhouette score is 0.64 when the optimal cluster
number is 6 for the dataset containing PCA-reduced
symptoms; the Silhouette score is 0.28 when the
optimal cluster number is 7 for the dataset containing
all symptoms. From cluster 27, the silhouette score
curve became flattened at a score of 0.74.
Figure 5. Predictability rate of SVM
Such a result was also validated by estimating the
prediction probabilities of the SVM model on the test
data, as shown in Figure 5. However, the F1-score of
the following diseases is relatively lower, as Acne
disease - 81.5%, Impetigo - 87.5%, Paralysis (brain
hemorrhage) - 88.8%, Paroymsal Positional Vertigo -
91.6%, Urinary tract infection - 93.7%, Hepatitis E -
96.7%, Hepatitis D - 97.7%, Gastroenteritis - 97.8%,
Fungal infection - 98.1%, Psoriasis - 98.7%. The Figure 7. Silhouette score of K-means clustering
predictability percentage increases by the prediction
SVM model considering both common and unusual The result indicates that the dataset containing
types of symptoms. However, unusual symptoms were PCA-reduced symptoms was better classified, and
important to diagnose a particular disease, but the similar diseases were identified. However, the dataset
F1-score prediction for each disease was relatively containing all symptoms got a silhouette score of 0.84
high, around 99.9%. Those diseases having a lower at cluster 40. The result explains that most diseases
- 200 -
were well separated given the symptoms and why the
machine learning method performed well. REFERENCES
Extensive dataset experiment [1] Zhou, X., Menche, J., Barabási, AL. et al. Human
symptoms–disease network. Nat Commun 5, 4212 (2014).
The same methodology was applied to a large dataset
https://doi.org/10.1038/ncomms5212
of about 4,442 unique diseases and 322 symptoms [2] Ting, D.S.W., et al. (2017). "Artificial Intelligence and Deep
extracted from 7,109,429 PubMed articles. We refined Learning in Ophthalmology." British Journal of
the diseases that reveal more than 4 symptoms. In total Ophthalmology, 103(2), 167-175.
[3] Luo, G., et al. (2018). "Deep Learning-Based Disease
3,233 unique diseases were extracted. Adopting the
Prediction using Electronic Health Records." IEEE Access, 6,
Decision Tree model, the accuracy of training data is 65333-65341.
91.47%, and the CNN model achieved 91.47% of [4] Rajkomar, A., et al. (2018). "Scalable and Accurate Deep
accuracy in the training set, as a preliminary trained Learning with Electronic Health Records." npj Digital
Medicine, 1, 18.
model. Therefore, we considered the severity of the
[5] Choi, E., et al. (2016). "Doctor AI: Predicting Clinical Events
symptoms based on the TF/IDF of PubMed via Recurrent Neural Networks." Journal of Biomedical
occurrences for training data. The Decision Tree model Informatics, 63, 322-327.
performed around 98.77%, and the CNN model [6] Choi, E., et al. (2016). "Retain: An Interpretable Predictive
Model for Healthcare using Reverse Time Attention
achieved 98.47% of accuracy in the training set. This
Mechanism." In Proceedings of the 21th ACM SIGKDD
trained model would be used to further predict diseases International Conference on Knowledge Discovery and Data
from any form of medical documents such as clinical Mining, 2017-2026.
notes. But the terms should be aligned with the MeSH [7] Lasko, T.A., et al. (2013). "Computational Phenotype
Discovery Using Unsupervised Feature Learning over Noisy,
Disease Terms and MeSH Symptoms Terms.
Sparse, and Irregular Clinical Data." PLoS ONE, 8(6),
e66341.
VI. CONCLUSION [8] Kavakiotis, I., et al. (2017). "Machine Learning and Data
Mining Methods in Diabetes Research." Computational and
We trained an extensive amount of MeSH disease
Structural Biotechnology Journal, 15, 104-116.
considering MeSH symptoms with an accuracy of [9] Miotto, R., et al. (2017). "Deep Patient: An Unsupervised
98.47% based on the CNN. This trained model is a Representation to Predict the Future of Patients from the
baseline model for predicting disease in any form of Electronic Health Records." Scientific Reports, 6, 26094.
[10] Singh, P., et al. (2019). "Disease Prediction Using Data
biomedical documents. Therefore, we applied the same
Mining Techniques: A Comprehensive Review." International
methodology to clinical notes to estimate predictive Journal of Intelligent Systems and Applications, 11(8), 57-68.
accuracy. We achieved 99.9% testing accuracy based [11] Farid, D.M., et al. (2017). "Disease Prediction Using Machine
on the CNN model by considering uncommon and Learning Techniques: A Comparative Study." Journal of Big
Data, 4(1), 1-17.
common symptoms. We observed that a sufficient
[12] Gholami, M., et al. (2019). "A Comparative Study of Machine
number of symptoms required at least 4 types for each Learning Techniques for Disease Prediction." In Proceedings
disease in predicting a disease with more than 91% of of the International Conference on Artificial Intelligence in
accuracy. We will extend this study by analyzing Data Science, 97-111.
[13] Naik, G.R., et al. (2020). "A Comparative Study on Machine
disease heterogeneity, and disease specificity for better
Learning Techniques for Disease Prediction." Journal of
prediction. The network-based approach has the Ambient Intelligence and Humanized Computing, 11(11),
potential to outperform conventional methods by 5205-5221.
leveraging the collective knowledge embedded in the [14] Mamun, M.A., et al. (2019). "A Review of Machine Learning
Techniques for Disease Prediction." In Proceedings of the
symptom-disease associations. Through this
International Conference on Computing and Communication
network-based approach to disease prediction using Systems, 32-36.
symptom data, we anticipate unlocking new [15] Islam, M.M., et al. (2021). "Machine Learning Techniques for
opportunities for precise and proactive healthcare Disease Prediction: A Review." International Journal of
Advanced Computer Science and Applications, 12(5),
interventions, thereby improving patient outcomes and
356-364.
reducing the burden of disease. [16] Dash, M., et al. (2020). "A Comprehensive Review on
Machine Learning Techniques for Disease Prediction."
ACKNOWLEDGMENT Journal of Ambient Intelligence and Humanized Computing,
11(7), 2943-2966.
This research was supported by KOICA(Korea [17] S. Grampurohit and C. Sagarnal, "Disease Prediction using
International Cooperation Agency) through "Capacity Machine Learning Algorithms," 2020 International
Conference for Emerging Technology (INCET), Belgaum,
Building Project for School of Information and India, 2020, pp. 1-7, doi:
Communication Technology at Mongolian University 10.1109/INCET49848.2020.9154130.
of Science and Technology in Mongolia" (Contract
No. P2019-00124). Also partially supported by
Mongolian Foundation for Science and Technology
(MFST).
- 201 -
System of Intelligent Control of Processes of
Development and Harmonization of Standards in
the Field of Military and Space Activities on the
Basis of Ontological Engineering
*Note: Sub-titles are not captured in Xplore and should not be used
Anar Utegenova1,2 , Gulnaz Yermoldina 2, Zhanna Suimenbayeva1,3*, Alisher Aden1,3, Vitaliy
Naumenko3,4
(Use * to indicate the correspondence author)
1
Almaty University of Power Engineering and Telecommunications named Gumarbek Daukeev, Kazakhstan
2
Institute of Information and Computational Technologies, Kazakhstan
3
INT-SAT Alatau LLP
4
International University of Information Technology, Kazakhstan
Email: [email protected]
Abstract— Technological modernization of all sectors is the To achieve this goal, the following research objectives were
main priority of developing countries. The expansion of implemented:
international relations directly depends on the readiness of - Development of a model using a systematic approach to the
states to introduce international experience in a particular area. methodology of ontological engineering, to form an
One of these areas is the need to harmonize international unambiguous formulation of definitions of harmonized and
standards in the field of military and space activities. The article newly created standards, without losing the connection between
proposes a new approach to solving the problems of processing entities, with the ability to identify contradictions or
large amounts of textual information of regulatory documents. inconsistencies in terminological unity in priorities.
The tasks described in the article involve a unique and - Application of intelligent control methods in the processes
exceptional sequence of a chain of research procedures – from of the operational cycle of standard development in the
research, formulation, and formalization of tasks to the development of the concept of a single representation of the
development of an application solution. For the first time, standard as a document that meets the requirements of a formal
ontological engineering in the field of process automation will be feature
used to harmonize standards according to the National The paper presents: a model of classification of standards, a
Standardization Plan, which consists in partial automation of module for highlighting entities, a categorization module, a
methodological examination of documents. As part of the module for the formation of a thesaurus, a module for
experiment, an independent tool was implemented, without monitoring and analysis. The calculations were carried out
reference to existing repositories of regulatory documents. The using formalization methods, setting the objective function,
approach is based on methods of extracting qualitative data setting the similarity measure, and determining the feature
from textual information, methods of intelligent management, description of the document. The paper demonstrates the
formulation of new scientific and scientific-technical problems theoretical component of the field of application, the design part,
applicable to standardization, models of monitoring and control and the description of the functionality of the system. The results
of processes. obtained were tested by a group of expert specialists competent
The rapid development of new technologies and algorithms in solving this problem, based on standards and regulatory
in recent years contributes to the development of the necessary documents in the field of the aerospace industry.
tools for automating many processes. First of all, we are talking Keywords— national standards, aerospace activities,
about the creation of tools for developers and experts of harmonization of standards, algorithm for identifying keywords,
standards, which will optimize the workflow of managing categorization of standards.
national standards, taking into account the consistency of
information set forth in existing standards and regulatory I. INTRODUCTION
documents. Secondly, it is necessary to trace all participants in Technological modernization of all industries is the main
the development and the main customers. And the most priority of national importance. The development of
significant work is the development of auxiliary tools for Kazakhstan's defense industry and readiness to develop joint
organizing the terminological unity of national military
ventures with other countries will make it possible to reform
standards in the state language, as well as taking into account
the uniformity of measurements.
the military-industrial complexes. In connection with the
The purpose of this study is to develop an intelligent growing need of the military-industrial complexes, it is
automated system for managing, monitoring, and controlling necessary to focus on the regulatory requirements and the
the processes of the operational cycle of standard development legislative component of this industry. Research in the field of
using technological integrations of optimal methods for creating international practices is reduced to the harmonization and
a single repository. development of national standards. But with all the openness
of international standards, it is necessary to take into account
the requirements of existing standards in Kazakhstan. This
- 202 -
practice is mandatory with the growing demand for the Pamanee are based on the identification of key terms for the
transfer of scientific and technical information, the industry standard that are relevant to legal categories [4].
convergence of the level of quality of goods produced in Metrological certification of the algorithm for assessing the
different countries, ensuring the interchangeability of inconsistency of data from key comparisons of national
elements of complex products, and the promotion of standards is given in the studies of a group of researchers
international relations. As practice shows, the construction Burmistrova, N.A., Viktorov, I.V., Chunovkina, A.G. [5].
and launch of industrial facilities takes almost the same time
as the development of the legislative component, regulatory Particular attention should be paid to research in the field
documents, and national standards. In this connection, it is of education, whether national standards are used for the
necessary to carefully study all stages, taking into account the development of curricula and learning outcomes. The
requirements of all interested parties. Most of the military- presented methods make it possible to test the developed
industrial complexes have a dual purpose, and therefore, when documentation on disciplines and get an analysis of the results
developing standards, it is necessary to consider not only [6-7]. Comparison of algorithms for extracting keywords is of
military standards, but also civilian ones. scientific interest to the world community [8-13].
The rapid development of new technologies and III. THE PURPOSE AND OBJECTIVES OF THE STUDY
algorithms in recent years contributes to the development of The purpose of this study is to develop an intelligent
the necessary tools for automating many processes. First of automated system for managing, monitoring and controlling
all, we are talking about the creation of tools for developers the processes of the operational cycle of standard development
and experts of standards, which will optimize the workflow of using technological integrations of optimal methods for
managing national standards, taking into account the creating a single repository.
consistency of information set forth in existing standards and
regulatory documents. Secondly, it is necessary to trace all To achieve this goal, the following research objectives
participants in the development and the main customers. And were implemented:
the most significant work is the development of auxiliary tools - A model has been developed using a systematic approach
for organizing the terminological unity of national military to the methodology of ontological engineering, to form an
standards in the state language, as well as taking into account unambiguous formulation of definitions of harmonized and
the uniformity of measurements. newly created standards, without losing the connection
between entities, with the possibility of identifying
II. ANALYSIS OF MANAGEMENT SYSTEMS FOR THE PROCESS
contradictions or inconsistencies in terminological unity in
OF DRAFTING AND REVISING LAWS AND REGULATIONS ON
priorities.
CIVIL-MILITARY INTEGRATION
Studies of the problems of the use of information systems - Methods of intelligent control in the processes of the
in the design of key technologies of civil-military integration operational cycle of standard development were used in the
rules and standards management systems are reflected in the development of the concept of a single representation of the
works of Huang, T. The author reveals the issues of managing standard as a document that meets the requirements of a
formal feature
the process of drafting and revising laws and regulations of
civil-military integration. According to the author, the use of IV. MATERIALS AND METHODS AND OBJECTIVES
the system improves the quality of standardization and
timeliness of management of rules and standards of civil- The main idea of the project is to develop a unified system,
military integration, increases the accuracy and traceability of an improved mechanism for monitoring and controlling
data on rules and standards. [1] Significant contributions of processes in the development of national standards. The
Ryu, H.E. are known. The author's research provides quite project involves a unique and exceptional sequence of a chain
strong and loud arguments for the harmonization of standards of research procedures – from research, formulation and
and technologies of asymmetric conformity. The purpose of formalization of tasks to the development of an application
the study is to examine the welfare implications of solution. For the first time, ontological engineering will be
harmonization of standards between technologically used in the field of process automation to harmonize standards
asymmetric countries and to identify optimal harmonization in accordance with the National Standardization Plan, which
strategies for countries with an average level of technological consists in partial automation of methodological examination
progress. [2]. An analysis of the actualization of Russian of documents for submission to Kazakhstan Institute of
standards regarding the requirements proposed by Chudnova, Standardization and Certification (KazInst). The solutions
O.A., Potsulin, A.D., Shilov, A.S., recalled the problems of presented in this project have not yet been implemented and
harmonization of standards when integrating domestic have no analogues in Kazakhstan, in the implementation of
enterprises wishing to expand their markets and do business the tasks, the goals of supporting the developers of national
with international partners into the world market. The study standards with a dynamic response to current changes in the
presents the results of a comparative analysis of RF standards field of military standardization are achieved.
in a particular industry based on the GOST classifier, and also In the implementation of tasks, ontological engineering is
shows the evolution of the development and harmonization of used in the field of process automation to harmonize standards
standards in modern economic conditions. The authors according to the National Standardization Plan based on the
introduce a criterion for measuring the level of harmonization results of a comparative analysis of models and methods for
of standards [3] the use of metadata, digital footprint, methods for automating
Scientific and technological needs that justify the text markup, intelligent control methods in the processes of
importance of the planned research characterize new the operational cycle of standard development.
directions in the field of text recognition, such as the studies
of R. Chaisricharoen, W. Srimaharaj, S. Chaising and K.
- 203 -
The ontological representation of standards implements
the task of classifying standards by industry, i.e. for each
document, the degree of its belonging to a particular industry
is determined. When categorizing, two levels are introduced:
at the first level, the belonging of the standard to the
classification group is determined, after that, at the second
level, the standard's belonging to standards that are close in
content is determined.
For each document, the degree to which the document
belongs to a category is determined, which is formed by the
number of matching keywords. (the greater the degree of
affiliation, the more the document fits into this category).
The classification considers the possibility of defining one
standard for several groups within a classification group.
A formal measure of document similarity is organized by
comparing the search images of document keywords. Once
formed, the image of the keywords of the document, as well
as the scale of the feature space, does not change throughout
the life of the current standards.
When forming a list of a standard's belonging to standards Fig. 1. BPMN Process Diagram
that are close in content, the selection of documents is ranked
by priority (the closer the documents are in content, the lower V. RESULTS
its priority). There are no restrictions on the number of
documents under consideration. The analysis of a document within the framework of each
study is an independent creative process, depending on the
A similar classification method is used for users, for form and content of the document itself, the goals and
categorization by industry. The leaders of the expert group see conditions of the study, the wealth of experience and creative
applications for harmonization of standards only those that fit intuition of the researcher. If it turns out that both series of
them in the profile that provides information on the expert's data obtained independently of each other coincide, then this
keywords and publications. means that the information is sufficiently reliable and stable.
At the end of the classifications of documents, a thesaurus Interaction of the file with the program code (Figure 2).
of keywords is formed for an organized classification group of
standards. 1. The file is distributed to the model
The paper presents: a model of classification of standards, 2. Next, the code from the models receives the necessary
a module for highlighting entities, a categorization module, a information
module for the formation of a thesaurus, a module for 3. After processing, displays it to the client on the interface
monitoring and analysis. The calculations were carried out from the server
using formalization methods, setting the objective function,
setting the similarity measure, and determining the feature
description of the document. The paper demonstrates the
theoretical component of the field of application, the design
part and the description of the functionality of the system. The
results obtained were tested by a group of expert specialists
competent in solving this problem, based on experience in the
aerospace industry.
Figure 1 shows the BPMN diagram of the process.
Business process modeling allows you to find a compromise
between a clear description and technical functionality.
Fig. 2. Database tables
The model describes the tables in the database (Figure 2)
• Allgost keeps a list of files uploaded by the admin and
applicants
- 204 -
• CreatedGost keeps a list of requests for creating In the manager's profile, registered users are displayed in
standards, where all the action takes place at runtime the form of a table, and there is a more detailed button to
clarify information about the user.
• CreatedGOSTComments messages from managers,
performers, translators, admins and the applicant When harmonizing international standards, a detailed
during the creation of the standard study and correct conversion of units of measurement is
necessary. The system provides for an automatic search in
• RukovoditelKeyWords stores keywords for the the text of the harmonized standard of existing units of
manager, according to which there is a selection for measurement, a function is provided for replenishing the
recommendation to the applicant library with new units of measurement, to focus the
• RukovoditelWantWork keeps schedule, where the attention of developers and clarify the translation of metric
manager gives a task to the performers and leaves a job systems, systems of natural units of measurement and
application traditional systems of measurement.
• SimilarFile stores files that have the maximum number With harmonized standards, 3 types are defined, identical
of similar terms (ID), modified (MOD) and non-equivalent standards
(NEQ). The presented system allows its application for all
• TranslateFile is similar to the schedule, but for types of harmonization. Particular attention should be paid
translators, which is given after the completion of the to the module for organizing the rules for constructing
work by the performers standards and general requirements for their content. The
module allows you to semi-automatically create a template
• Documents stores diplomas, certificates, etc. of the that meets the general requirements for the construction,
new user for its verification presentation, design and content of standards (Figure 3).
• User keeps a list of users.
The analysis of the document within the framework of
each study is an independent creative process, depending
on the form and content of the document itself, the goals
and conditions of the study, the wealth of experience and
creative intuition of the researcher. If it turns out that both
series of data obtained independently of each other
coincide, then this means that the information is
sufficiently reliable and stable.
The main login window with user authentication provides
access to the use of project resources only to verified
users. Fig. 3. Automatic formatting
The registration window allows you to enter data about the
The main standards used by the system as an example were obtained from
user, such as full name, Login, Password, Role of the the website of the Kazakhstan Institute of Standardization and Metrology
project executor, Documents for subsequent verification. https://www.ksm.kz/public-discussion/standards/discussion-of-draft-of-st-
The function of adding user files (certificates, diplomas, rk/. This resource contains draft standards for discussion, which made it
bank details, etc.) and deleting them is also available. possible to form a base of standards for the repository (Figure 4).
When registering, you can skip this step and make it after
registration.
Upon registration, the user of the system views the data
entered by him. Also in this window you can view
information about the status of the user, verification.
The organization of the work of users with the
environment is carried out using the main page of the
administrator, includes a table about registered users, with
a button for details about the user. The top menu displays
five tabs: Profile, User List, Standards List, Standards
Developer List. There is a function of replenishment of
libraries. Information about user verification is marked in Fig. 4. Search for units of measurement
green, unverified users of the system are marked in red.
Based on the results of processing, managers have the
Full names of unverified users are highlighted in red. opportunity to view the standard for review. At this stage,
During verification, the administrator specifies the user's the applicant can familiarize himself with the current
role in the project. There are four roles in the pop-up menu: leaders on the topic presented, which is formed, according
Applicant, Manager, Executor, and Translator. At the end to the keywords that determine the direction of activity of
of verification, the color of the user's full name changes to a particular manager The main module is developed and
blue. The administrator appoints a manager and verifies consists in quantitative content analysis, the formation of
the user. keywords from the main document and comparing them
with other documents, distributed Key terms in the text of
the document form a database and are stored in the system.
- 205 -
This list is seen only by those managers who are experts in [4] R. Chaisricharoen, W. Srimaharaj, S. Chaising and K. Pamanee,
this field, the database of managers includes keywords for «Classification Approach for Industry Standards Categorization» 2022
Joint International Conference, on Digital Arts, Media and Technology
the selection of experts in the areas (Figure 5). with ECTI Northern Section Conference on Electrical, Electronics,
Computer and Telecommunications Engineering (ECTI DAMT &
NCON), 2022, pp. 308-313, doi:
10.1109/ECTIDAMTNCON53731.2022.9720348.
[5] Burmistrova, N.A., Viktorov, I.V., Chunovkina, A.G., Metrological
Certification of the Algorithm for Estimating Inconsistent Data of Key
Comparisons of National Standards, (2023) Measurement Techniques,
Volume 65, Issue 7, October 2022, Pages 508-514, DOI:
10.1007/s11018-023-02111-1
[6] Tiffanie Ford-Baxter, Kendall Faulkner, Jennifer Masunaga, Situating
Information Literacy: A Case Study Exploring Faculty Knowledge of
National Disciplinary Standards and Local Program Learning
Outcomes, The Journal of Academic Librarianship, Volume 48, Issue
3, 2022, ISSN 0099-1333,
Fig. 5. Standard page https://doi.org/10.1016/j.acalib.2022.102523.
[7] Barthelemy, F., Cochez, M., Dimitriadis, I., Karim, N., Loutas, N.,
Magnisalis, I., Comet, L. M., Peristeras, V., & Wyns, B. (2022).
Towards a standard-based open data ecosystem: analysis of DCAT-AP
use at national and European level. Electronic Government, 18(2), 137-
180. https://doi.org/10.1504/EG.2022.121856
Fig. 6. Algorithm for keywords [8] Eliseo Reategui, Marcio Bigolin, Michel Carniato, and Rafael Antunes
dos Santos. 2022. Evaluating the Performance of SOBEK Text Mining
Keyword Extraction Algorithm. In Machine Learning and Knowledge
ACKNOWLEDGMENT Extraction: 6th IFIP TC 5, TC 12, WG 8.4, WG 8.9, WG 12.9
This study was carried out and funded by the Science International Cross-Domain Conference, CD-MAKE 2022, Vienna,
Committee of the Ministry of Education and Science of the Austria, August 23–26, 2022, Proceedings. Springer-Verlag, Berlin,
Heidelberg, 233–243. https://doi.org/10.1007/978-3-031-14463-9_15
Republic of Kazakhstan (grant No. AR09058441 "Building a
[9] Mehrabi, Elaheh & Mohebi, Azadeh & Ahmadi, Abbas. (2020).
system of intelligent control of the processes of development Improved Keyword Extraction for Persian Academic Texts Using
and harmonization of standards within the framework of RAKE Algorithm, Case Study: Persian Thesis and Dissertations.
interstate and national standardization based on ontological Iranian Journal of Information Processing Management. doi: 37.
engineering"). 10.52547/jipm.37.1.197.
[10] M. S. U. Miah, J. Sulaiman, S. Azad, K. Z. Zamli and R. Jose,
REFERENCES "Comparison of document similarity algorithms in extracting
document keywords from an academic paper," 2021 International
[1] Huang, T.Design and Key Technology Research of Civil-Military
Conference on Software Engineering & Computer Systems and 4th
Integration Regulations and Standards Management System(2021)
International Conference on Computational Science and Information
Smart Innovation, Systems and Technologies, . DOI: 10.1007/978-
Management (ICSECS-ICOCSIM), Pekan, Malaysia, 2021, pp. 631-
981-15-5073-7_21
636, doi: 10.1109/ICSECS52883.2021.00121.
[2] Ryu, H. E. (2021). Standards Harmonization and Asymmetric
[11] Kungurtsev O. et al. Development of information technology of term
Compliance Technology. Journal of Korea Trade, 25(3), 1-20.
extraction from documents in natural language //Eastern-European
doi:10.35611/jkt.2021.25.3.1
Journal of Enterprise Technologies. – 2018. – Т. 6. – №. 2. – С. 44-51.
[3] Chudnova, O. A., Potsulin, A. D., & Shilov, A. S. (2018, Oct 02-04).
[12] Lytvyn V. et al. Development of a method for determining the
Analysis of Updating of the Russian Standards Concerning
keywords in the slavic language texts based on the technology of web
Requirements Harmonization. Paper presented at the International
mining // East European Journal of Advanced Technologies. – 2017. –
Scientific Multi-Conference on Industrial Engineering and Modern
no. 2(2). - P. 14-23.
Technologies (FarEastCon), Vladivostok, RUSSIA
doi:10.2991/iscfec-18.2019.268
- 206 -
August 22(TUE) - 25(FRI), 2023
Oral Session 10
16:20~18:00 Wed, August 23, 2023
Room : A437
Chair: Juhee Choi (Sangmyung university, Korea)
- 207 -
Algorithms of Methodology for Applying
Augmented Reality in Teaching Physics
Beken Arymbekov
Ph.D. student
Department of Physics and Technology
Al-Farabi Kazakh National University
Almaty, Kazakhstan,
[email protected],
ORCID 0000-0002-9670-2407
Abstract—Purpose of this article is to define algorithms frameworks are among the main difficulties. Overcoming
to develop methodology for applying augmented reality in these challenges requires support from the state through
teaching physics. Therefore, I made augmented reality funding and procurement, as well as the conduct of scientific
applications that can be used in the preparation of research and experiments to validate the effectiveness and
educational materials. Applied methodology analysis the pedagogical benefits of augmented reality in training
experience of introducing augmented reality technologies specialists from different fields. To promote the
into the educational process and educational materials as dissemination and application of augmented reality
sources for the latest augmented reality learning software. technology, it is recommended to organize national and
High school teaching process with augmented reality international events focused on this subject. These events
empower to perceive objects in 3D in real timeline and in serve as platforms for sharing knowledge, exchanging
natural environment using simple phones. This kind of experiences, and showcasing the potential of augmented
opportunity will consent students to improve their reality technologies. My research confirms that augmented
knowledge and lab experience. Educational resources with reality technologies have significant potential in the training
augmented reality is becoming more in demand in physics of future specialists in the new technological era. By
education. Augmented reality will allow visualizing invisible leveraging the benefits of augmented reality and addressing
physics processes, to view the physical experimental models the associated challenges, I can enhance the learning process
in detail. Assumptions to use of augmented reality support and better prepare students for physics lab.
has a progressive effect on the quality of education, but its
efficiency for students for different parts of physics is Keywords—augmented reality; physics; education;
different. The greatest effectiveness of augmented reality in pedagogy
the process of teaching students at high school were
revealed. In general, the impact of augmented reality on I. INTRODUCTION
students is still ambiguous. The issue requires further study, Modern society has entered augmented reality age, it is
in particular, in the aspect of stress. As a result, I have
on the threshold of augmented reality based civilization
gathered valuable information on the benefits, challenges,
and potential applications of augmented reality in various
and culture. As core and the engine of a new reality,
industries, particularly in the fields of social interaction, augmented reality started in last couple of decades of
education, tourism, and commerce. From an educational twentieth century, when the information flows due to
standpoint, the introduction of augmented reality mainstream of digital technologies. Since that moment, we
technology in educational institutions is to offer several can see the beginning of shifting cultural paradigm [1].
advantages. It can enhance the learning experience by Gradually over the years, the traditional ways of teaching
providing interactive and visually engaging content, thereby with books is being slowed down. Since the beginning of
increasing students' motivation to learn and improving the 21st century, we can talk about the situation of total
information assimilation. Additionally, the application of
digitalization of learning processes. Because of
augmented reality technologies in education contributes to
the development of these technologies themselves, leading to development of digital technologies, the way of
a growing demand for specialists in augmented reality. To perceiving the world, the way of thinking, and processing
effectively implement augmented reality in education, an of information began to change drastically. The
integrated approach is recommended, taking into account abundance of information and its fast change are not
the specific needs. This approach aims to adapt augmented giving to students the opportunity to see whole picture
reality technologies to various learning environments and about the world. The picture of the world is formed from
ensure that future physics professionals are adequately different disparate illustrations and impressions, received,
prepared for the new technological era. However, there are in in particular at fast shift frames of cartoons, television
challenges associated with the application of augmented
programs and other sources. Today this way of thinking is
reality technologies. Financial constraints, a lack of
specialized professionals, and the need for methodological called clip thinking, and the culture of thinking is called
clip culture. Clips is defined as short time of changing
- 208 -
frames [2]. At such fast frame pictures for students, its digitalization of education is integration in learning
hard utilize systemic knowledge about the subject. process and educational materials programs and platforms,
According to pedagogical researchers, it is very allowing use 3D images, the task of the author was
challenging for features of cognitive activity, including the consideration and classification of such agendas in the
main features of thinking about whole picture of situation. field of physics education [5]. The research methods used
Dominance of visual form of information overlays over in the article are the methods of analysis of primary data.
high speed of information processing against the Analysis government documents were subjected to
background of a decrease in the possibility of its scientific publications of domestic and foreign auto ditch,
generalization and restructuring. Traditional educational Internet sources with characteristics of programs virtual,
models are aimed at perception of information from the augmented and mixed reality.
book with voluminous texts. Textbooks and various kinds
of text materials are always were the main source of
knowledge for the student. However, modern students
with understands long texts with complications if it is
without illustrations; therefore, texts are boring,
uninteresting, poorly digested and memorized for them.
Accordingly, text materials needed revision, supplying
them with already habitual for new generation’s specific
images formulas formed before the eyes of the observer,
movable diagrams, changing images. Digitalization of
education presumes provides for the digitalization of
educational and methodological materials, supplying them Figure 1. The effectiveness of the two types of teaching styles.
with elements of augmented reality. Digital educational on In this study educational materials, in which refer
the contributions will constitute publicly available textbooks and educational and methodical allowances,
knowledge bases, will allow transfer the educational methodological recommendations for practical learning
process to the global network, using augmented reality and self-study work, accompany any type and level of
technologies featuring 3D graphics [3]. Distance learning educational process. Their success depends on their
forms of education more and more distributed in the availability and quality of educational process, learning
educational platforms. Remote forms learning and using outcomes of all categories of students. Learning success
digital learning materials become especially in demand at depends on correspond needs and features perception
a time when mobility students difficult. Many educators contemporary students. Modern students were born and
are devoted to the issues of the development of digital grew up in an era of rapid development of digital
forms and channels of education, the formation of digital technology. Students are faced with computers, tablets,
culture today. We are witnessing change of cultural smartphones, and various kind gadgets. These instruments
paradigm is happening. The new generation students no organically fit in digitalization as necessity as tactics
longer uses the textbooks, because they lives in a digital survival in replaced ordinary books. For students watching
world. New generation perceives traditional culture form animations was much more interesting than read long
with an established value system through digital and texts [6]. Today everybody more distribution the type of
emerging technology consequences such as clip culture, reading disorder becomes strange, referred to as dyslexia.
screen culture, culture of computer games. Internet gives
possibility continued education in conditions restrictions
mobility, obtain the necessary learning outcomes. In the
current situation, there is an increasing need in digital
learning materials. Proceeding from such need, everybody
more research preparing digital learning materials. Special
interest cause publications with experience in the
application of augmented reality in physics education. I
determined 3D programs established on strong phones,
convenient tool for improving the effectiveness of
teaching and learning. Mobile technologies embodied in
the new smart pedagogy and used technologies of
augmented reality, representing large potential for physics
education. This dedicated problems implementation
Figure 2. Students are testing prototype of augmented reality
elements of augmented reality in construction education application.
[4].
A child who has like view disorders psyche, finds it
II. METHODS
difficult develop reading and writing skills. We are facing
Method that I considered about the forms and creation a new epidemic learning disabilities, this phenomenon is a
of new learning materials, which would be interesting and huge problem. The nature of learning disabilities is not
understandable to the modern students. At the same time, enough studied. In scientific terms, it is based on a
it gives systemic knowledge about the subject under the complex of biological and social causes, in which you
dominance of clip awareness. Since the achievement of need to carefully analyze. The brain of children and
- 209 -
perception develop in a special way. One of the reasons
may be too early and long views clips. Perhaps, so
peculiarity contemporary students is what they not able us
scribble long texts. Textbooks without illustrations,
pictures, are disheartened and confused. Students lose
interest in systematic work with textbooks and others
educational paper materials. Featured teacher textbooks,
located in the library are often remain unclaimed while Figure 3. Architecture of augmented reality application.
presentations of lectures, provided with photo has more
interest cause. Students take pictures of presented slides Especially in demand will become educational mate
and then reproduce fragmentary information without with augmented and mixed reality in physics education
coherent ideas about the issue. Students are confident that institutes. Future professionals in these fields need visuals,
they can surf Internet and get the necessary information such as students studying physics, can study the
immediately; therefore, they are not trying to memorize. phenomenon on a virtual image, future astronomers can
This kind of habits of students deprives systemic visit save other planets, while not risking their lives and
assimilation subject [7]. As a result, we need new teaching health. Transfer from two-dimensional (2D) to three-
services, oriented on specifics thinking students, as well as dimensional (3D) measurement depth knowledge and
new channels of distribution and acquisition information understanding issue. Areas of physics suggest formation
retention. Electrical intellectual libraries have millions of practical skills creative thinking, active professional
copies various publications, access to which not limited. behavior. The augmented reality application in
In addition, special digital repositories store of educational educational materials will help students dents to
materials in form, access to which is open at any time and understand the essence of the subject, to master the
from any place of the globe. Digital educational materials technologies, teach equipment, instruments and then how
allow embed elements of augmented reality. Virtual knowledge and on- tricks can be applied. With augmented
Reality is a world similar to the real one, but it artificially reality applications, students can move virtual objects,
created with help technical tools and computer programs. zoom, move, rotate and view from different angles. Such
The person perceives it through the senses such as sight, learning develops spatial thinking, increases affects the
smell, touch, and hearing. degree of assimilation of the material. Virtual equipment
can be disassembled into parts, study its arrangement
III. RESULTS property, carry out repairs. This is much more efficient
Augmented Reality is different from virtual reality than consider the drawings in the textbook, and
because augmented reality on images of real world objects contributes to the best understanding student’s real
virtual elements are superimposed. Simulation of an technological processes. Active education maybe to be
object or phenomenon is carried out here in a positive added tests and quizzes. Tests basis on augmented reality
way; it should be good natural display fundamental can design the way to student chose options response in
reality, helping us to cognize a real object in a virtual nom augmented reality or fulfilled virtual practical exercises.
form. Augmented reality applications demonstrate three- Gamification with augmented reality could also become
dimensional objects such as video, images, animations, one of the active forms of learning [9]. Every teacher
texts, and this makes it possible to directly touch events to could would use any form augmented reality for learning
interact with objects [8]. As some researchers foretell, students process design. Using augmented reality program
traditional teaching methods will invariably become can penetrate inside already inspect the premises. Vuforia
outdated. Mixed forms of education will come to replace is augmented reality platform, which uses computer vision
when part of the subjects will be studied remotely press. technologies and presents instruction and learning
For example, lectures will be transferred to remote on-line software. Vuforia Engine offers software value for
mode, and practical classes will be held in audience with developing augmented reality that can be used at
application opportunities augmented reality. With the help preparing learning material [10].
of supplemented and mixed programs virtual reality,
students can focus their mobile screens and watch
holographic 3D objects, and the teacher will explain
subject with visual effects. This technological innovation
changes the process of teaching and learning. It will
become more interesting, more exciting, and perhaps more
effective. As experience shows, classes using capabilities
of augmented reality programs no motivation to study,
stimulates creative thinking and develops skills problem
solution.
Figure 4. Vuforia user interface during the AR application
development.
Vuforia tied with Unity3D cross-platform environment
for developing computer games. Unity3D allows you to
- 210 -
create augmented reality applications that work in markers before-foot time and in framework complete scale
different operating systems, including personal computers, object. Full reality and receive access to additional body
game consoles, and mobile devices [11]. digital content. Using the QR code, the student can see
with platforms for creating applications of augmented
realities, see illustrating video materials. Using augmented
reality software, undergo learning practice. Similar on
undertakings are more a product of creativity-interested
enthusiasts than a response to the demands ministries
Sciences and higher education. The analysis of the work
of domestic and foreign educational researchers showed
what usage of augmented reality certainly has positive
impact on quality education; however, its impact on
different groups of learners turns out to be different. I
made conclusion that augmented reality renders an
Figure 5. Unity3D user interface during the AR application
development. average impact on students. The greatest the impact of
augmented reality on high school students. From targeted
Technologies develop so fast that it is suggested group’s most strong influence before- full reality has on
glasses gradually replace the familiar smartphone but the schoolchildren. I was used mixed method with sampling
period will not last long to replace glasses will come from 89 students of the 9th grade (177girls, 46 boys). The
lenses with integrated technologies augmented reality and experimental group participated in reading with using AR
then when augmented reality can will embed in human applications, while in control group used traditional
organism . With the help of the listed programs and methods. I discovered what students’ experimental groups
applications can create so-called "smart" books demonstrated more high level on reading comprehension
(textbooks, educational manuals, guidelines for practical and smooth learning outcomes at weekly measurements
classes) - with texts that are involved through mobile how control group. Pupils experienced satisfaction from
device. "Smart" books are not only textbooks where texts his participation in augmented reality based reading and
come to life on a smartphone screen, but encyclopedias, expressed a desire to use vat similar applications and on
reference books, maps, translators and other publications. other subjects. Student has demonstrated short level
The markers in the book are ordinary pages and lines, anxiety at use programs augmented reality. The problem
which recognized install- lazy application for augmented posed requires further teachings. Topical are questions
reality. FROM with the help of an appropriate program impact virtual helmets, glasses for the psyche of students,
are involved text interactivity features. The user himself possible technostress at different categories learning-
can fulfill virtual laboratory work and mutually interact educators. It is also required to study the perception of
with objects that in real life not available. These textbooks teaching providers of teaching materials to schools with
might be of interest. Student, accustomed from childhood elements of virtual and augmented reality, their
to the perception of formations IT technologies, and even preparation and readiness work with them.
define trajectory his future life.
REFERENCES
IV. CONCLUSION
[1] F. Zhou, H.L. Duh, M. Billinghurst, “Trends in augmented reality
Summarizing this research, I find out that an increase tracing, interaction and display: A review of ten years in ISMAR,
in volume and speed up information flows put modern Mixed and Augmented Reality”, ISMAR 7th IEE/ACM
International Symposium, vol.7, pp. 193-202. 2008.
students in a difficult situation of its perception and [2] S. Yuen, G. Yaoyuneyong, E. Johnson, “Augmented reality: An
mastering. Abundance information makes it difficult to overview and five directions for AR in education”, Journal of
concentrate attention. Information overload and its Educational Technology Development and Exchange, vol.4, No.1,
availability reduce consistency and creativity thinking, pp. 119-140, 2011.
analytical thinking becomes not in demand in the era of [3] M. Dunleavy, C. Dede, R. Mitchell, “Affordances and limitations
of immersive participatory augmented reality simulations for
ready-made and accessible clichés. Clip feed of teaching and learning”, Journal of Science Education and
information changes civilization of textbooks onto Technology, vol.18, No.1, pp. 7 – 22, 2009.
civilization visual images. Digital technologies are [4] B. Kye, Y. Kim, “Investigation of the relationships between media
characteristics, presence, flow, and learning effects in augmented
advancing so fast that the sphere of education, as a
reality based learning”, International Journal for Education Media
conservative does not have time to react on any and Technology, vol.2, No.17, pp. 4-14. 2008.
innovations. Augmented reality technologies are being [5] H.E. Pence, “Smartphones, smart objects, and augmented reality”,
introduced faster and effectively into the education. The Reference Librarian, vol.52, No.2, pp. 136-145, 2010
[6] M. Laal, “Positive interdependence in collaborative learning”,
Although most people admits necessity of educating
Procedia Social and Behavioral Science, vol. 93, pp. 1433 - 1437,
materials for new generations of students focused on the 2013.
specifics of perception. They taking into account [7] K. Doymu, “Effects of a Cooperative learning strategy on teaching
achievements in in the field of information technology. and learning phases of matter and one-component phase
However, in reality such materials so far exist only in the diagrams”, Journal of Chemical Education, vol.84, No.11, pp.
1857-1860, 2007.
form of experimental publications, and still not ready for [8] A.A. Gokhale, “Collaborative learning enhances critical
mass implementation. I started development interactive thinking”, Journal of Technology education, vol.7, No.3, pp. 23-
textbooks with augmented reality, which are available in 31, 1986.
[9] D.D Sumadio, D. R. Rambli, “Preliminary Evaluation on User
the mobile application. In mobile application can be
Acceptance of the Augmented Reality Use for Education”, Second
through the camera of a smartphone see on pages textbook
- 211 -
International Conference on Computer Engineering and [11] Z. Baigunchekov, M. Kalimoldaev, M. Utenov, B. Arymbekov,
Applications, ICCEA 2010, IEEE, vol.5, pp. 461-465, March 2010. “Structural and dimensional synthesis of parallel manipulator with
[10] A. Dünser, R. Grasset, M. Billinghurst, “A survey of evaluation two end—effectors”, Mechanisms and Machine Science, vol.37,
techniques used in augmented reality studies”, ACM SIGGRAPH pp. 15–23, 2016
ASIA 2008 courses on - SIGGRAPH Asia, vol.37, No.1, pp. 1-27,
2008
- 212 -
Utilizing Plane Detection Technology based on
Actual Physical Environment for Classroom
Experiments
Yeji Heo+, Woojeong Shim+, and Juhee Choi*
Department of Smart Information Communication Engineering, Sangmyung University, Republic of Korea
Email: [email protected]
+Both authors contributed equally to this manuscript
Abstract—In recent years, the development of AR
technology has attracted attention for its potential educational
applications. However, current AR education systems have
several limitations. Firstly, these systems utilize virtual
backgrounds instead of projecting digital elements into the real
world. Additionally, practicality is decreased because marker
detection technology only works when a specific image is
prepared. This research proposes the utilization of plane
detection technology, based on the actual physical environment,
to maximize the advantages of AR technology. To demonstrate
the effectiveness of our approach, we have conducted classroom
experiments such as those involving antibiotic resistance or the
dissection of an anchovy. Our research indicates that this
approach can contribute to addressing educational inequality as
well as reducing the budget for science experiments.
Keywords— Augmented reality, Science education, Practical
skills Fig. 1. AR session for object placement
I. INTRODUCTION
II. IMPLEMENTATION
Recently, rapid development of mobile and multimedia A. AR Development
ICT tools affects the field of education. Among them,
augmented reality(AR) can serve as an active learning tool by
enabling experiential-based learning[1-3]. It particularly For the development of AR functionality, we utilized the
proves to be more effective in science experiments, where Unity game engine and AR Foundation[7], which provide
direct experience and observation are key. interfaces for implementing AR features. Additionally, for
compatibility with the target platforms, Unity's provider plug-
In this study, our mobile application provides immersive ins were necessary. Specifically, the Google ARCore XR
laboratory experiences based on real-world spatial plane Plug-in was used for Android environments[8], while the
recognition. Users can participate in experiments without any Apple ARKit XR Plug-in was employed for iOS[9].
restrictions, using their own smartphones as an equal
opportunity platform. We develop content for two During the initial phase of the experiment, AR
experiments, namely the antibiotic resistance experiment and Foundation's markerless plane detection feature was used to
the anchovy dissection experiment, which are part of the enable object placement on surfaces without specific markers.
Korean high school curriculum. The scene setup involved placing AR Session and XR Origin,
provided by AR Foundation. AR Session activates AR
Regarding the antibiotic resistance experiment, we functionalities, while XR Origin converts tracked camera
overcome the time constraint of waiting for the occurrence of items into Unity's coordinate system. These functionalities
a general stress response to antibiotics. On top of that, in order were added as game objects to the scene to complete the setup,
to avoid legal and ethical concerns regarding animal allowing the placement of objects for the experiment.
experimentation, participants perform the anchovy dissection
experiment to individually conduct dissection practices, rather High-quality objects in the experiment were modeled
than in team settings. using Blender 3D computer graphics software and exported to
Unity. Collider and Rigidbody components were added to
We conducted the experiments by placing experiment objects used as experiment tools and materials to enable
objects and enabling the progression of subsequent events collision detection and gravity application. User touches and
through collision or touch detection. Through user interactions drags on the mobile screen are determined using Raycast, and
with the mobile screen, the app promotes active participation RaycastHit provides specific collision information, such as the
of students. Moreover, animation effects and interactive tag of the collided object. Such interactions allow for the
review quizzes help bridge the psychological gap between achievement of specific experimental conditions and scene
students and their academic studies[4-6]. These elements transitions to the next stage.
enhance students' immersion.
- 213 -
B. UI Configuration long as there is a minimal plane space around them,
The app consists of two scientific experiment contents and eliminating spatial constraints. Based on the survey results of
quiz contents related to those experiments. Initially, users existing AR education apps, it is determined that our app will
choose an experiment from the main screen. Using the camera, serve as an innovative learning tool that enhances the
users can identify the surrounding surfaces and position the effectiveness of learning experiences. This will generate users'
experiment table at the suitable location. By touching the interest in learning and improve the effectiveness of their
screen while aiming the camera at a flat surface, the app learning experiences.
generates the 3D model of the experiment table. IV. DISCUSSIONS
With the provided instructions and audio guidance, users We proposed a new learning tool by introducing AR
conduct the scientific experiment using the 3D models of the technology into the education system to enhance students'
experiment tools. Users can always return to the main screen interest and engagement. Through a survey case about an AR
or restart the experiment by pressing the pause button during education app, we were able to observe positive results, but
the content progression. Once users complete all the quizzes areas for improvement were also clearly identified. One area
based on the knowledge gained from the experiments, they is the need to provide instructions to users in an easily
can review the experiment results. understandable manner regarding how to use the app.
Furthermore, the limited availability of experimental content
III. RESULTS
hinders the delivery of effective learning experiences,
highlighting the necessity to develop additional experimental
content. Looking ahead, once a stable AR education app is
established, we can explore the integration of VR technology
to create even more immersive learning tools.
ACKNOWLEDGMENT
This research was supported by program for
postPRIME(Program for Industrial needs-Matched Education)
was conducted by the National Research Foundation of
Korea(NRF) funded by the Ministry of Education.
REFERENCES
[1] Gy. Molnár, The role of electronic and virtual learning support systems
in the learning process, In: Szakál Anikó (ed.) IEEE 8th International
Fig. 2. Collider for collision detection 1 Symposium on Applied Computational Intelligence and Informatics:
SACI 2013. New York: IEEE, 2013. pp. 51-54.
[2] K. Bir, G. Molnár, D. Pap, Z. Szűts, "The effects of virtual and
augmented learning environments on the learning process in secondary
school," 2017 8th IEEE International Conference on Cognitive
Infocommunications (CogInfoCom), 2017, 0371-000376
[3] M. Akçayır and G. Akçayır, ‘‘Advantages and challenges associated
with augmented reality for education: A systematic review of the
literature,’’ Educ. Res. Rev., 2017, vol. 20, pp. 1–11
[4] M. B. Ibáñez, Á. Di Serio, D. Villarán, and C. D. Kloos, ‘‘Experiment-
ing with electromagnetism using augmented reality: Impact on flow
stu-dent experience and educational ffectiveness,’’ Comput. Edu.,
2014, vol. 71, pp. 1–13
[5] B. Kye and Y. Kim, ‘‘Investigation of the relationships between media
characteristics, presence, flow, and learning effects in augmented
reality based learning,’’ Int. J. Educ. Media Technol, 2008, vol. 2, no.
1, pp. 1–11
[6] G. Molnár, D. sik, " Smart devices, smart environments, smart students
Fig. 3. Collider for collision detection 2 - A review on educational opportunities in virtual and augmented
reality learning environments," 2019 10th IEEE International
Conference of Cognitive Infocommunications(CogInfoCom), 2019,
According to the cited paper's results, the survey pp.1-4
conducted after the introduction of AR education systems into [7] S. Murrell, F. Wang, E. Aldrich, X. Xu, "MeteorologyAR: A Mobile
real-world education revealed that 60% of users agreed with AR App to Increase Student Engagement and Promote Active Learning
the statement "The AR-based education system provided in a Large Lecture Class," 2020 IEEE Conference onVirtual Reality
stimulation for learning." [4-7] and 3D User Interfaces Abstracts and Workshops(VRW), 2020, pp.
848-849
However, existing AR education systems developed as [8] AR Foundation
precedents rely on marker recognition technology, which https://docs.unity3d.com/Packages/[email protected]/m
requires specific images to be prepared in advance. In the case anual/index.html
of large lecture halls, pre-testing was necessary to ensure [9] ARCore XR Plug-in
marker recognition from various locations within the https://docs.unity3d.com/Packages/[email protected]/manual/
classroom. [10] ARKit
https://docs.unity3d.com/Packages/[email protected]/manual
To overcome these limitations, this study utilizes plane
recognition technology. Users can smoothly run the app as
- 214 -
Cost Comparison and Analysis of VR Tracker
Manufacturing with Low-cost Sensor and High
value Sensor
Beomjun Cho1+, Chanhee Jang1+, Juhee Choi1* and Joonyoung Kim2*
1
Sangmyung Univ, Republic of Korea
2
IMEC, Belgium
Email: [email protected]
+Both authors contributed equally to this manuscript
Abstract— This research proposes to develop a VR (Virtual
Reality) tracker using low-cost sensors. A tracker in the VR
system is a wearable sensor device to capture body movements.
However, it is required to invent more affordable trackers to
broaden the use of VR devices due to the high cost of existing
VR trackers. Therefore, we utilize low-cost sensors to track user
motions in virtual environments and devised an interface. The
results show that a 75% cost reduction compared to high-end
sensors for the tracker.
Keywords— VR, Trackers, IMU, Seonsor, Cost
I. INTRODUCTION
This paper aims to develop cost-effective VR trackers using
low-cost IMU sensors and compare performance and cost with
expensive trackers and motion capture hardware. Virtual reality
(VR) is a technology that creates a specific artificial environment
and situation with a computer, and VR devices are required to
interact with it [1]. VR devices that need to accurately and Fig. 1. Configureration and attachment Positions of VR Tracker
quickly grasp complex movements of the body require high
technology, and many companies such as Meta, HTC, and Sony can measure rotational motion through angular velocity,
are participating in VR device research. With the development acceleration and gravitational acceleration[4]. IMU is widely
of the VR market, research and development of VR trackers are used through sensor miniaturization and mass production in
being actively conducted. A VR tracker is a device that detects a a variety of fields, including aviation drones, airplanes,
user's movement and implements it into a virtual world. These behavior analysis, motion recognition, and artistic creation.
trackers use an IMU (intensity measurement device) sensor to The acceleration sensor uses gravitational acceleration to
accurately track the position and direction of the body. However, determine the effect of gravity on each axis in the vertical
the current VR hardware ecosystem is suffering from market direction and to measure the angular tilt.
contraction due to entry barriers. In order to measure accurate The accelerometer does not return a value when there is no
body motion, high-performance sensors and technology are movement, but when movement occurs, acceleration occurs
required, which are eventually limited to high prices. Therefore, in that direction. However, it should be recognized that
in order to identify the minimum performance of sensors for the measuring an angle using only an accelerometer can result in
use of VR devices, this paper aims to manufacture trackers an immediate error.
through low-cost sensors and compare performance with
A gyroscope, on the other hand, is a sensor that measures
expensive sensors.
angular velocity and represents the rotational speed of the
sensor within a defined time frame. The integration of the
II. VR TRACKER WITH IMU SENSOR gyroscope data allows you to measure the angle.
However, gyroscopes are temperature sensitive and introduce
VR Tracker plays a key role in providing immersion and
continuous errors, and large errors occur over time. Therefore,
interactivity by accurately tracking the user's location and
we will use complementary filters to calibrate the values
direction. Key techniques used in VR trackers include sensor
received from these two types of sensors and improve their
fusion, computer vision, inertial measuring unit (IMU),
accuracy.
optical tracking and electromagnetic tracking. Sensor fusion
The IMU sensor obtains gyroscope measurements of 𝛼𝑥,
technology combines data from various sensors, including
𝛼𝑦, and 𝛼𝑧 and accelerometer measurements of 𝛽𝑥, 𝛽𝑦, and 𝛽𝑧
accelerometers, gyroscopes and magnetometers, for robust
and accurate tracking [2][3]. to track the position of the tracker.
The IMU is a multifunction device that incorporates To obtain real-time orientation values of an IMU sensor using
acceleration, geomagnetic, altimeter, and gyroscope, which quaternion representation [5], we calculate the vector
quaternion value 𝑞𝑠 = (0, 𝛼𝑥, 𝛼𝑦, 𝛼𝑧) using the rotation
quaternion 𝑞 (𝑡) at a specific time t and the accelerometer
measurements. We obtain the accelerometer measurement 𝑞𝑖
- 215 -
in the reference frame by multiplying 𝑞𝑡 , 𝑞𝑠 , and the inverse Each body consists of a microprocessor, an IMU, a charging
of 𝑞𝑡 . The default orientation of the IMU sensor is such that module required for power processing, and a battery. Figure
the x-axis aligns with the direction of gravity, which allows 3 shows a sensor prototype attached to the arm.
us to obtain the inclination angle 𝜑 and the normalized We use Arduino-based Wemos D1 mini as microprocessor
rotation axis 𝛾 in the moving coordinate system to calculate for streaming and processing IMU data, and Wi-Fi
𝑞 (𝑡+∆𝑡) [6]. Furthermore, according to Euler's rotation communication.
theorem, every coordinate transformation can be achieved The IMU sensor chose MPU-6050, a reasonable and
through a single rotation around a specific axis, known as the inexpensive module. The MPU-6050, an IMU module with
rotation axis. Thus, if we rotate coordinate system {a} by an gyroscopes and accelerometers, is a sensor module with three
angle 𝛽 around the rotation axis, we can transform it into types: a three-axis acceleration sensor, a three-axis gyroscope
coordinate system {b}. In this case, the quaternion sensor and a temperature sensor. By combining a three-axis
representation of the rotation from {a} to {b} can be acceleration sensor with a three-axis gyroscope sensor, the
expressed as following: MPU-6050 achieves six degrees of freedom (6DoF).
Operates at 5 V, the gravitational acceleration can be
𝛽 measured up to ±2g, and the rotational angular velocity can
𝑐𝑜𝑠( ) be measured up to ±250°/sec. The test environment was
2
𝑞0 𝛽 configured using HTC vive HMD and controller, and the
𝑞1 𝑝1 𝑠𝑖𝑛( ) model we made can be connected to unity independently
𝑞𝑏𝑎 = [𝑞 ] = 2
2 𝛽 without using other vr devices. The price to produce one of
𝑞3 𝑝2 𝑠𝑖𝑛( ) our trackers after confirming normal operation can be found
2 in the following table.
𝛽
[𝑝3 𝑠𝑖𝑛( 2 )] Table Ⅰ. Cost Analysis of Components comprising a Single
Tracker
Using the above expression, the quaternion can be expressed
as:
∆𝑡𝑙 𝛼 ∆𝑡𝑙 𝛼 ∆𝑡𝑙 𝛼 ∆𝑡𝑙
𝑞∆ = (cos ( ) , 𝑥 sin ( ) , 𝑦 sin ( ) , 𝑧 sin ( ))
2 𝑙 2 𝑙 2 𝑙 2
When calculated by applying a complementary filter, the
formula follows:
(𝑡 + ∆𝑡)
𝑞𝑐 = 𝑞((1 − 𝛿)𝜑, 𝛾)𝑞 (𝑡) 𝑞∆
By using this formula, you can measure the direction of the
tracker in real time.
III. IMPLEMENTATION AND COST EVALUATION
IV. CONCLUSION
Fig. 3. Comparing our model with Motion Capture Hardware and
Fig. 2. VR Tracker Prototype Gaming VR Tracker
We manufactured a tracker body with four tracker bodies and The comparators are Mti-630 from Xsens and Vive Tracker
three additional IMU sensors to make a tracker worn on a from HTC, which are used for motion capture. The cost of
total of 10 parts. In order to express the minimum body in the graph does not include the price of software and
virtual reality, a single tracker body is applied to both elbows additional accessories at the price of only one terminal.
and thighs, and a tracker Our model showed a huge reduction compared to the Mti-
body with an additional IMU sensor is applied to the waist 630 and confirmed a reduction rate of about 75% compared
and pelvis, ankles and feet with relatively close connections. to the Vive Tracker. This rate of decline is meaningful
- 216 -
because we aimed to lower the hardware entry barrier in the [2] Kreylos, Oliver. "Environment-independent VR development."
Advances in Visual Computing: 4th International Symposium, ISVC
VR game market. 2008, Las Vegas, NV, USA, December 1-3, 2008. Proceedings, Part I
4. Springer Berlin Heidelberg, 2008.
ACKNOWLEDGMENT
[3] H. Zhou and H. Hu, “Human motion tracking for rehabilitation–
This research was supported by program for post- survey,” Biomedical Signal Processing and Control, vol. 3, no. 1, pp.
PRIME(Program for Industrial needs-Matched Education) 1– 18, 2008.
was conducted by the National Research Foundation of [4] Ahmad, Norhafizan, et al. "Reviews on various inertial
Korea(NRF) funded by the Ministry of Education. measurement unit (IMU) sensor applications." International Journal of
Signal Processing Systems 1.2 (2013): 256-262.
REFERENCES [5] Mukundan, Ramakrishnan. "Quaternions: From classical
mechanics to computer graphics, and beyond." Proceedings of the 7th
[1] Mohammadreza Akbari, Nghiep Ha and Seng Kok, “A Asian Technology conference in Mathematics. 2002.
systematic review of AR/VR in operations and supply chain
[6] Sandra Griffin, “Quaternions: Theory and Applications”, Nova
management: maturity, current trends and future directions”, Journal of
Science Publishers, pp. 158 – 166, 2017.
Global Operations and Strategic Sourcing, vol.15, no.3, pp. 534 - 565,
2021 [7] Andrew J. Hanson, “Visualizing Quaternions”, Elsevier Science,
pp. 51 - 54, 2006.
- 217 -
Detection and Classification of the Volatile Organic
Compounds from the Petroleum and Petroleum-
Derived Products Using an Electronic Nose and
Machine Learning Algorithms
Olzhas Alseitov1, Aliya Issainova1, Dina Satybaldina1* and Assem Konyrkhanova1
1
L.N. Gumilyov Eurasian National University, Kazakhstan
Email: [email protected]
Abstract—A method for rapid and selective detection and accidental or intentional operational discharges further
classification of the petroleum and petroleum-derived products aggravates the ecological situation and required attention to
using an electronic nose in combination with machine learning the environment problems [8-10].
algorithms is proposed. The electronic nose system consists of a
multisensory array, an analog-digital processing block, a Precise and rapid detection of petroleum or petroleum
classification block with digital pattern-recognition algorithms product spills are beneficial to identify the source and type
software and reference odor databases. We test odors PDP, accurate estimation of oil spread areas, evaluate the level
corresponding to organic vapors emitted by petroleum and of hazard and to develop a response and recovery treatment a
petroleum-derived products (including gasoline, diesel, motor to reduce environmental effects.
oil and used motor oil) under various conditions such as sample
Remote Sensing (RS) from an aircraft (including
gas concentrations, air temperature and humidity, sensor signal
recovery time. The electronic nose demonstrated fast and high unmanned aerial vehicles, aerostats) or from satellites have
correlation of the vector signal with emitted volatile compounds proven in oil spills detection and monitoring [11]. Optical and
with an accuracy of more than 98% and 90.5% in pattern microwave sensors, machine learning (ML) and deep learning
recognition by machine learning algorithms and a deep neural (DL) models have used for image classification and oil spill
network, respectively. trajectory prediction [12, 13]. The accuracy of both sensors is
affected by the presence of biogenic elements, leading to false
Keywords—artificial olfaction, electronic nose, machine positive depiction of petroleum and PDP spills [13]. There is
learning, deep learning, classification of odors, petroleum. no best methodology for detecting and monitoring oil spills
using remote sensing. Therefore, it is necessary to develop
I. INTRODUCTION new approaches for rapid and accurate classification of
petroleum pollution using alternative sensor technologies that
can replace or supplement computer vision based RS technics.
Republic of Kazakhstan is an oil and gas producing
country located in Central Asia and partially in Eastern Crude oil contains some of the hydrocarbons (HCs), which
Europe. Nearly 300 oil fields and over 60 gas condensate have low boiling points and classified as Volatile Organic
fields have been discovered on the Kazakhstan Republic Compounds (VOCs) with well-known detrimental influences
territory [1]. According to the 2021 BP Statistical Review of on human health and the air ecosystem [14]. According to the
World Energy [2], Kazakhstan has 30 billion barrels of proved results of the inventory of atmospheric emissions published in
oil reserves (12th rank in the world in terms of available oil 2010, approximately 258 tons of HCs could evaporate per day
reserves) and 2.3 trillion cubic meters of proved gas reserves from the Deepwater Horizon oil spill in the Gulf of Mexico,
(the 16th greatest gas reserves volume in the world). which indicates the scale of an environmental disaster [15].
An extraction of natural resources plays an important role Such a context has incited the research communities to
in the development of the country’s economy. Over the past work on the development of alternative technologies such as
30 years, Kazakhstan has increased oil production by 3.5 artificial olfaction, which mimics the mammalian olfactory
times, and in terms of production it ranks 13th in the world (86 system and can identify different complex odors comparing
million tons in 2021) [3]. Today revenues from crude oil and the incoming odor with patterns previously learnt [16]. An
gas exports form a significant part of the country's gross electronic nose (e-nose) as a representative example of an
domestic product [4, 5]. artificial olfactory system uses an array of gas sensors
composed with differing relative sensitivities, and the
At the same time, developing hydrocarbon deposits is collective response these sensors corresponds to the odor
accompanied by a strong anthropogenic impact on the pattern, like a fingerprint, allowing the identification and
environment. The drilling and construction works, laying of classification of analytes [16]. Some of the industrial
pipelines and roads causes a change in the chemical and application of e-noses include clinical diagnosis [17, 18],
physicochemical properties of the soil, disrupts the food quality determination [19], environmental monitoring
hydrological regime of territories, leads to changes of [20, 21].
phytocenoses and animal populations [6, 7]. Crude oil, fuel or
other petroleum-derived products (PDPs) spills due to In this work we investigated whether e-nose is able to
This research is funded by the Science Committee of the Ministry of
discriminate between crude oil (petroleum) and petroleum-
Science and Higher Education of the Republic of Kazakhstan (Grant No. derived products (PDPs) on the basis of the VOC patterns. Our
AP14872171). results are preliminary to future research of assessing oil soil
- 218 -
pollution in Republic of Kazakhstan. Crude oil from three from different oil fractions with non-equal amounts of volatile
Kazakhstan oil fields and four PDPs from local producers organic compounds were chosen as samples (see Table 1).
(including motor gasoline, diesel fuel, motor oil and used
motor oil) were chosen as experimental samples and analyzed For future studies of oil pollution of soils, petroleum
by the E-nose established in Our Laboratory. The volatile products were selected, spills of which actually occur both
fingerprints of the petroleum and PDPs registered by the e- during extraction, transportation, and when they are used for
nose were interpreted using principal component analysis mechanized land cultivation in agriculture. Two types of
(PCA), ML and DL models. We demonstrated high sensitivity motor oil were employed: a new synthetic motor oil
of sensors and selective discrimination of petroleum sample’s commercialized by «High Industrial Lubricants & Liquids
VOC patterns at room temperature. The accuracy of the (HILL) Corporation» (HILL Universal SAE 5W-30) and a
classification results was above 98% and 90% for the ML used motor oils (UMO) of same brend collected in a local
algorithms and convolutional neural network (CNN), garage. UMOs are considered as an environmental pollution
respectively. Comparison of sensory responses for motor oil source containing significant concentrations of toxic
and used motor oil indicates a significant increase in the VOCs compounds (polycyclic aromatic compounds and heavy
concentration from used motor oil. The smell fingerprint of metals) [23, 24].
used motor oil is close to that of gasoline and diesel, indicating TABLE I. LIST OF SAMPLES
that used engine oil is also a very hazardous contaminant.
Thus, the suggested approach opens a new avenue for precise Petroleum or Oil field\ Sample’s physico-chemical
and rapid detection of crude oil brend or PDP type for remote PDP sample's Producing characteristics
type company or
oil spill sensing systems and to reduce environmental effects. brand of sample
The rest of this article is structured as follows. Section 2 Crude Oil Alibekmola\ Light (0.835 g/cm3) , sour
Kazakhoil Aktobe (about 2%) oil
presents research methods (materials, of multisensory e-nose LLP
system a description, datasets, research tools, and research Crude Oil Alibek Yuzhny \ Light (0.842 g/cm3), sour (up
process). The experiment results on measuring sensory Joint Stock to 1.33%) oil
responses to the presence of petroleum and PDPs, processing Company "Caspiy
sensory data, features extracting and performance evaluating Neft TME"
of the machine learning algorithms and the convolutional Crude Oil Kardasyn North \ Heavy (0.945 g/cm3), sweet
Sagiz Petroleum (sour 0.2%) oil
neural network are discussed in Section 3. Conclusion and Company LLP
future research are presented in Section 4. Motor Atyrau Oil Сombustible mixture of light
Gasoline Refinery LLP \ A- hydrocarbons with a boiling
II. MATERIALS AND METHODS 92-K5 with 92 point of +33 to +205 ° C
octane number (depending on impurities).
A. Samples Density is about 0.71 ... 0.76
Crude oil (synonymous with crude petroleum, oil) is a /cm³. Gasoline consists of a
homogeneous mixture of
natural mixture of hydrocarbons that exists as a liquid in small, relatively lightweight
underground geologic formations and also include hydrocarbons with between 4
compounds of sulfur, nitrogen, oxygen, metals and trace and 12 carbon atoms per
elements (vanadium, nickel, iron, and copper) at the part per molecule
million concentrations [22]. Petroleum products are obtained Diesel Fuel Atyrau Refinery Special low-temperature
from the processing of crude oil, natural gas, and other LLP \ DT-A-K5 diesel contains additives to
hydrocarbon compounds. List of PDPs include liquefied Altay-45 keep it liquid at lower
temperatures. Petroleum
petroleum gases (LPG), aviation gasoline, motor gasoline, diesel is produced from the
kerosene, fuel oil, diesel fuel, lubricants, waxes, petroleum fractional distillation of crude
coke, asphalt, and miscellaneous products [22]. oil between 200 and 350 °C
at atmospheric pressure,
VOCs are organic (containing carbon) chemicals that can resulting in a mixture of
easily evaporate into the air and its are a source of odors. carbon chains that typically
Petroleum and petroleum products may have the potential to contain between 9 and 25
carbon atoms per moleculeю
release VOCs. Main goal of our research is a development of
an artificial-nose system for differentiation complex odors Motor Oil High Industrial MО are made from a heavy
from oil and petroleum-derived products samples. Smell Lubricants & and thick hydrocarbon
Liquids (HILL) fraction of crude oil, with
derived from VOCs may be composed of a few compounds Corporation \ additives to improve certain
which only one may be the dominant component. But all «HILL Universal», properties. The bulk of a
compounds contribute to the unique characteristics of the SAE 5W-30 typical engine oil consists of
sample’s collective odor, the VOC pattern. hydrocarbons containing
from 18 to 34 carbon atoms
The main goal of our research is to recognition of complex per molecule.
smell from oil and PDPs for oil pollutions monitoring. Odor Used Motor High Industrial Petroleum oil that has been
from organic sources may consist of the several chemical Oil Lubricants & used, and as a result of such
compounds of which only one or two may be the dominant Liquids (HILL) use is contaminated by
Corporation \ Used physical or chemical
components. But the rest of the chemical compounds also properties.
«HILL Universal»,
contribute to the unique odor of the organic source. The effect SAE 5W-30
of even slight changes in the relative amount of chemicals in
the odor mixture can be detected by the electronic nose [20].
Therefore, crude oil from several fields and PDPs produced
- 219 -
B. E-Nose The MOS sensors use a two-electrode system in which the
The electronic nose was created using an array of 8 metal sensitive layer of tin dioxide (SnO2) has a variable resistance
oxide semiconductor (MOS) sensors (see Table 2) and depending on the concentration and type of gases or gas
Arduino Nano module. mixtures being studied. The interaction of the target gas with
the sensitive layer occurs through a reversible redox reaction,
TABLE II. ELECTRONIC FEATURES AND SCOPES OF USED SENSORS as a result of which the electrical properties of the SnO2 layer
change. These electrical properties are translated into
Equipment Detecting scope measurable parameters such as sensor resistances or voltage
Type Electronic Features
Name [25]
Detection range: 200– across a load resistor in a sensor circuit. MOS sensors, SHT75
It is suitable for temperature and humidity sensor were grouped on a separate
10000 ppm; heating
detecting H2,
Smoke gas
consumption: less than
LPG, CH4, CO,
board and connected via an analog switch to an analog-to-
MQ-2 800mw; operating digital converter (ADC) on the Arduino module. The ADC has
sensor Alcohol, Smoke,
voltage: 5V±0.1V; 16-bit resolution and converts input voltages from 0 to 5 volts
i-butane,
operating temperature: -
20–70 ◦C
Propane. into integers from 0 to 65535 (216-1).
Detection range: The sensor array and Arduino Nano module were installed
0.05mg/L—10mg/L
Alcohol; heating High sensitivity
in a single metal housing to enable constant airflow supported
Alcohol gas consumption: less than to alcohol and by a ventilation fan. The e-nose was powered by means of a
MQ-3 USB interface connected to a PC, also used for data
sensor 750mw; operating small sensitivity
voltage: 5V±0.1V; to gasoline. acquisition.
operating temperature: -
10–50 ◦C C. Data sets
Detection range: 300–
10000 ppm; heating
The data sets were collected during a series of experiments
Methane High sensitity to with crude oil from three Kazakhstan oil fields and four PDPs
consumption: less than
(CH4) Gas Methane, also to from local producers, which were analyzed using a homemade
MQ-4 900mw; operating
Sensor Propane and
voltage: 5V±0.1V; e-nose device. The liquid samples of oil and petroleum
Butane
operating temperature: - products (a volume of 100 µl) were introduced into a glass viol
10–50◦C
connected to the e-nose measuring chamber. Each experiment
Suitable for
Detection range: 200– detecting of was replicated three times. The signal outputs from sensor
10000 ppm; heating LPG, natural gas, array, UADC, were measured at 1 Hz sampling rate over 1 20
Liquefied
Petroleum
consumption: less than town gas, avoid min of total experiment time for each sample (see Table 1),
MQ-5 800mw; operating the noise of which include exposure of the e-nose for 30 min to the air,
Gas (LPG)
voltage: 5V±0.1V; alcohol and then to air passed through the petroleum or PDP sample 60
sensor
operating temperature: - cooking fumes
210–50◦C and cigarette
min and then switched the flow to the air for 30 min.
smoke Experimental data were record in separate files (txt format).
Detection range: 20– They are used in Each file consists of above 7000 row with 10 attributes,
2000 ppm; heating gas detecting namely year-month-day data and time (hour-minute-second),
Carbon consumption: less than equipment for responses from 8 sensors, temperature (degree Celsius),
monoxide MQ-7 350 mw; operating carbon
humidity (%). The dataset contains no missing data nor any
(CO) voltage: 5V±0.1V; monoxide (CO)
operating temperature: - in family and significant outliers.
20–50◦C industry or car.
The total number of records was above 140, 000. We
Detection range: 100– High sensitivity
10000 ppm; heating to Hydrogen divided the dataset into two parts for this study: the training
Hydrogen consumption: less than (H2). Small set (80%), and the test set (20%). The training set was divided
(H2) gas MQ-8 800mw; operating sensitivity to on two part again - 70% used to train the model and rest for
Sensor voltage: 5V±0.1V; alcohol, the model validation.
operating temperature: - LPG,cooking
10–50◦C fumes D. Data analysis
Detection range: 20ppm-
2000ppm carbon To analyze and classify the multivariate data (n = 8) of
Used in gas sensor responses to VOCs from the crude petroleum and
Gas Sensor monoxide, 500ppm- detecting
for Carbon 10000ppm CH4, equipment for
PDPs, we applied pattern recognition approaches, such as
Monoxide 500ppm-10000ppm principal component analysis (PCA), machine learning
carbon
(CO), MQ-9 LPG; heating monoxide and algorithms and convolutional neural networks.
Methane consumption: less than
CH4 ， LPG in
CH4 and 340 mw; operating PCA is a statistics method to reduce the dimensionality of
family and
LPG voltage: 5V±0.1V; data [26]. The approach allows one to discard less significant
industry or car.
operating temperature: -
20–50◦C
features in sensor responses by directing the axes of the
Used in air principal components along the maximum scatter of the data
quality control variance. Thus, it helps to display the space of high metrics in
Detection range: 10ppm-
300ppm NH3; heating
equipments for the Cartesian coordinate system [13].
buildings/offices
Gas Sensor consumption: less than In this study, we trained and tested Machine Learning
MQ- , are suitable for
for NH3, 800mw; operating
CO2, NO2
135
voltage: 5V±0.1V;
detecting of classifiers (Decision Trees, Logistic Regression, Random
NH3,NOx, Tree Forest, k-nearest neighbor classifiers, KNN) using code
operating temperature: -
alcohol,
10–45◦C
smoke,CO2, etc.
written in Python 3.9.15, and Scikit-learn open source
machine learning package [27].
- 220 -
Along with this, a deep learning model with convolutional matrix, with the x‐axis representing 83.46% and the y‐axis
neural network architecture was designed and tested to find 14.73% of the whole covariance. This method only lost 1.81%
the most accurate models based on performance metrics (the of the information, which is a very good level. This is a certain
confusion matrix, accuracy, precision and f-score) [13]. degree of simplification; however, it enables differentiated the
volatile fingerprints of all of the samples at various time points
III. RESULTS AND DISCUSSION during the experiment.
A. E-nose results Since the PCA analysis differentiated the sample volatile
For interactive calculations and experimental data fingerprints at various time points during the experiment, we
processing, a Python application has been developed (in ipynb checked whether these differences were significant enough to
format) containing source codes, input data, and calculation allow the ML and DL models to distinguish between the
results in numerical and graphical representation. petroleum from different oil fields, as well as the type of
petroleum products (see Table 3).
Figures 1-2 show examples of visualization of
experimental results in the form of time-dependent electrical
characteristics of sensors. The figures show that the sensors of
the gas analysis system react differently to experimental
samples of crude oil and petroleum products, which allows
them to differentiate.
An analysis of the experimental data indicates the presence
of the correlation between the physicochemical properties of
crude oils (see Table 1) and the characteristics of sensory
responses to crude oil samples (see Fig.1). The volatile
fingerprints of the petroleum-derived products registered by
the e-nose are demonstrate the correlation between the it's
physicochemical properties and amplitudes and the forms
sensor responses.
Motor gasoline, being a mixture of light hydrocarbons, (a)
corresponds to the lightest fraction of oil, has both the richest
odor and the fastest evaporating. This is confirmed by the high
amplitude sensory responses of the electronic nose to the
presence of VOCs from motor gasoline and the rapid drop in
sensory response due to the high evaporation rate (see Fig.2a).
Diesel fuel corresponds to the average fraction of oil, has
less VOCs compared to gasoline, and the sensory responses of
the electronic nose in the case of diesel fuel are less intense
(see Fig.2b), it’s shapes indicates the average rate of
evaporation and weathering.
At a petroleum refinery, fractional distillation separates a
motor oil fraction from other crude oil fractions, removing the
more volatile components, and therefore the oil's density and
viscosity the density of the oil product increases, and its smell
decreases. And experimental electrical characteristics of MOS
sensors for motor oil have a low intensity and a gentle shape (b)
(see Fig.2c).
Remarkable results have been obtained with used engine
oil. Smell pattern of the oil samples indicate the presence of a
significant amount of VOCs in the used oil (see Fig.2d). The
closeness of the used oil sensor data to the gasoline and diesel
fuel VOCs fingerprints of indicates the possibility of:
- malfunctions in the engine that lead to the ingress of
gasoline or diesel fuel into the motor oil;
- or an increase in toxic components in oil during using.
B. Classification of the petroleum and PDS’s patterns by
using machine-learning analyses and deep learning
model
(c)
Results of the PCA analysis pertaining to the volatile
fingerprints of the crude oils from Alibelmola oil field under Fig. 1. The time-dependent electrical characteristics of MOS sensors of e-
examination are presented in Figure 3. Using PCA, we nose for crude oil from three Kazakhstan oil fields: (a) Alibekmola; (b)
reduced the eight‐dimensional data space (the multisensor Alibek Yuzhny; (c) Kardasyn North.
array of e-nose) to a two ‐ dimensional data covariance
- 221 -
(a) (b)
(c) (d)
Fig. 2. time-dependent electrical characteristics of MOS sensors of e-nose for petroleum products: (a) Motor galazoline; (b) Diesel fuel; (c) Motor oil; (d)
Used motor oil.
TABLE III. PERFORMANSE RESULTS OF PATTERN RECOGNITION MODELS TO CLASSIFY CRUDE OILS AND PDPS SAMPLES ON EACH STAGES
Petroleum or PDP Performance metrics
sample's type Logistic Regression Random Forest KNN Decision Trees CNN
Crude Oil Accuracy: 96,3% Accuracy: 98.32 % Accuracy: 97.76 % Accuracy: 98.32 % Accuracy: 87.8 %
Alibekmola F1 SCORE: 96.27 % F1 SCORE: 98.29 % F1 SCORE: 97.72 % F1 SCORE: 98.29 % F1 SCORE: 87.57 %
PRECISION: 96.15 % PRECISION: 98.53 % PRECISION: 98.06 % PRECISION: 98.53 % PRECISION: 87.9 %
RECALL: 96.43 % RECALL: 98.11 % RECALL: 97.49 % RECALL: 98.11 % RECALL: 87.37 %
Crude Oil Alibek Accuracy: 99.83 % Accuracy: 100.0 % Accuracy: 100.0 % Accuracy: 100.0 % Accuracy: 99.62 %
Yuzhny F1 SCORE: 99.82 % F1 SCORE: 100.0 % F1 SCORE: 100.0 % F1 SCORE: 100.0 % F1 SCORE: 99.62 %
PRECISION: 99.81 % PRECISION: 100.0 % PRECISION: 100.0 % PRECISION: 100.0 % PRECISION: 99.57 %
RECALL: 99.84 % RECALL: 100.0 % RECALL: 100.0 % RECALL: 100.0 % RECALL: 99.67 %
Crude Oil Kardasyn Accuracy: 82.64 % Accuracy: 87.8 % Accuracy: 85.69 % Accuracy: 87.8 % Accuracy: 65.48 %
North F1 SCORE: 82.55 % F1 SCORE: 87.57 % F1 SCORE: 85.47 % F1 SCORE: 87.57 % F1 SCORE: 65.48 %
PRECISION: 82.49 % PRECISION: 87.9 % PRECISION: 85.64 % PRECISION: 87.9 % PRECISION: 68.5 %
RECALL: 82.83 % RECALL: 87.37 % RECALL: 85.35 % RECALL: 87.37 % RECALL: 68.48 %
Motor Gasoline Accuracy: 99.39 % Accuracy: 100.0 % Accuracy: 100.0 % Accuracy: 100.0 % Accuracy: 99.66 %
F1 SCORE: 99.37 % F1 SCORE: 100.0 % F1 SCORE: 100.0 % F1 SCORE: 100.0 % F1 SCORE: 99.65 %
PRECISION: 99.28 % PRECISION: 100.0 % PRECISION: 100.0 % PRECISION: 100.0 % PRECISION: 99.71 %
RECALL: 99.47 % RECALL: 100.0 % RECALL: 100.0 % RECALL: 100.0 % RECALL: 99.6 %
Diesel Fuel Accuracy: 100.0 % Accuracy: 100.0 % Accuracy: 100.0 % Accuracy: 100.0 % Accuracy: 99.41 %
F1 SCORE: 100.0 % F1 SCORE: 100.0 % F1 SCORE: 100.0 % F1 SCORE: 100.0 % F1 SCORE: 99.41 %
PRECISION: 100.0 % PRECISION: 100.0 % PRECISION: 100.0 % PRECISION: 100.0 % PRECISION: 99.45 %
RECALL: 100.0 % RECALL: 100.0 % RECALL: 100.0 % RECALL: 100.0 % RECALL: 99.38 %
Motor Oil Accuracy: 96.62 % Accuracy: 96.14 % Accuracy: 95.66 % Accuracy: 96.22 % Accuracy: 95.59 %
F1 SCORE: 96.59 % F1 SCORE: 96.09 % F1 SCORE: 95.6 % F1 SCORE: 96.18 % F1 SCORE: 95.53 %
PRECISION: 96.9 % PRECISION: 96.64 % PRECISION: 96.25 % PRECISION: 96.71 % PRECISION: 96.2 %
RECALL: 96.41 % RECALL: 95.84 % RECALL: 95.31 % RECALL: 95.93 % RECALL: 95.25 %
Used Motor Oil Accuracy: 71.11 % Accuracy: 72.96 % Accuracy: 72.45 % Accuracy: 72.96 % Accuracy: 70.28 %
F1 SCORE: 70.48 % F1 SCORE: 72.58 % F1 SCORE: 72.1 % F1 SCORE: 72.58 % F1 SCORE: 70.2 %
PRECISION: 73.19 % PRECISION: 76.54 % PRECISION: 76.31 % PRECISION: 76.54 % PRECISION: 72.95 %
RECALL: 77.05 % RECALL: 80.52 % RECALL: 80.15 % RECALL: 80.52 % RECALL: 74.03 %
- 222 -
[7] T. Alimbaev, K. Yermagambetova, S. Kabyltayeva, A. Issayev, Z
. Kairat, & Z. Mazhitova, (2020). Environmental problems of the oil
and gas industry in Kazakhstan. In E3S Web of Conferences
International Scientific Conference on Biotechnology and Food
Technology (BFT-2020), vol. 215, p. 03008, December 2020.
[8] M. Barron, D. Vivian, R. Heintz, & U. Yim, “Long-term ecological
impacts from oil spills: comparison of Exxon Valdez, Hebei Spirit, and
Deepwater Horizon,”Environmental Science & Technology, vol.
54(11), pp.6456-6467, April 2020.
[9] Z., Asif, Z. Chen, C. An, & J. Dong, “Environmental impacts and
challenges associated with oil spills on shorelines,” Journal of Marine
Science and Engineering, vol.10 (6), p. 762, May 2022.
[10] M. Rusin, J. Gospodarek, & A. Nadgórska-Socha, “Time-delayed
effect of petroleum-derived products in soil and their bioremediation
on plant–herbivore interaction,” Archives of Environmental
Protection, vol.47(3), pp.71-81, October 2021.
[11] M., Fingas, C. Brown, & J. Daniels, “An update on oil spill remote
sensing,” in Advances in Environmental Research, vol.80, J.A.
Daniels, Eds. New York: Nova Science Publishers Inc., 2021, pp. 95-
137.
Figure 3. Two‐dimensional principal component analysis (PCA) plots for [12] T. Shamsudeen, “Advances in remote sensing technology, machine
(a) the volatile fingerprints of air and the crude oil from Alibekmola oil fields learning and deep learning for marine oil spill detection, prediction and
under examination. The points represent the averaged sensor data for all vulnerability assessment,”Remote Sensing, vol. 12(20), p.3416,
experimental data for that sample. October 2020
[13] R., Al-Ruzouq, M. Gibril, A. Shanableh, A. Kais, O. Hamed, S. Al-
IV. CONCLUSION Mansoori, & M. Khalil, “Sensors, features, and machine learning for
oil spill detection and monitoring: A review,” Remote Sensing,
Based on the results described above, it can be concluded vol.12(20), p. 3338, October 2020.
that the E-Nose system is able to identify not only the [14] H. Dai, Sh. Jing, H.Wang, Y.Ma, L.Li, W. Song, H. Kan, “VOC
presence/absence of the studied crude oil and PDPs samples characteristics and inhalation health risks in newly renovated
but it can to differ the oil's brend and type of the petroleum residences in Shanghai, China,” Science of The Total Environment,
product. vol. 577, pp. 73-83, January 2017.
[15] H. Rajabi, M. Mosleh, P. Mandal, A. Lea-Langton, & M. Sedighi,
The proposed approach is planned to be used in future “Emissions of volatile organic compounds from crude oil processing–
studies to determine the source and type of oil pollution on soil Global emission inventory and environmental release,” Science of The
samples. Total Environment, vol. 727, p.138654, July 2020.
[16] K., Persaud, & G. Dodd, “Analysis of discrimination mechanisms in
ACKNOWLEDGMENT the mammalian olfactory system using a model nose,”
Nature, vol.299(5881), pp.352-355, 1982.
The authors thank Fedor Fedorov, Senior Researcher at the [17] M. Farraia, J. Rufo, I. Paciência, F. Mendes, L. Delgado, & A. Moreira,
Laboratory of Nanomaterials of the Skolkovo Institute of “The electronic nose technology in clinical diagnosis: A systematic
Science and Technology (Russian Federation, Moscow) for review,” Porto biomedical journal, vol.4(4), July 2019.
for valuable comments. The authors thank Yerlan Abishev [18] B., Lu, L., Fu, B., Nie, Z., Peng, & H. Liu, “A novel framework with
(JSC “Caspiy Neft TME”) for providing crude oil samples high diagnostic sensitivity for lung cancer detection by electronic
from of Kazakhstan’s oil fields. nose,” Sensors, vol.19(23), p.5333, December 2019.
[19] J., Tan, & J. Xu, “Applications of electronic nose (e-nose) and
REFERENCES electronic tongue (e-tongue) in food quality-related properties
determination: A review,”Artificial Intelligence in Agriculture, vol.4,
[1] B. Bealessio, N. Alonso, N. Mendes, A. Sande, & B. Hascakir, “A pp.104-115, 2020.
review of enhanced oil recovery (EOR) methods applied in
Kazakhstan,” Petroleum, vol. 7 (1), pp.1-9, March 2021. [20] M. Taştan, & H. Gökozan, “Real-time monitoring of indoor air quality
with internet of things-based E-nose,” Applied Sciences, vol. 9(16),
[2] BP statistical review of world ener, accessed 8 July 2021, 70th Edition 3435, August 2019.
https://www.bp.com/content/dam/bp/business-
sites/en/global/corporate/pdfs/energy-economics/statistical-review/bp- [21] A. John, K. Murugappan, D. Nisbet, & A. Tricoli, “An outlook of
stats-review-2021-full-report.pdf recent advances in chemiresistive sensor-based electronic nose systems
for food quality and environmental monitoring,”Sensors, vol.1(7),
[3] BP statistical review of world ener, accessed 28 June 2022, 71th Edition 2271, March 2021.
https://www.bp.com/content/dam/bp/business-
sites/en/global/corporate/pdfs/energy-economics/statistical-review/bp- [22] J. G. Speight, Handbook of petroleum product analysis. Hoboken, NJ:
stats-review-2022-full-report.pdf John Wiley & Sons Inc., 2015.
[23] R. Vazquez-Duhalt, "Environmental impact of used motor oil,"
[4] A. Syzdykova, A. Abubakirova, D. Kelesbayev, A. Omarova, G.
Science of the total environment, vol.79.1, pp. 1-23. February 1989.
Amaniyazova, B. Saubetova, & D. Anshayeva, “The effect of export
and imports on national income in Kazakhstan: Econometric analysis,” [24] A. Mishra, H. Siddiqi, U. Kumari, I. Behera, S. Mukherjee, & B.
Revista Espacios, vol. 40 (35), pp. 22-36, October 2019. Meikap, “Pyrolysis of waste lubricating oil/waste motor oil to generate
high-grade fuel oil: A comprehensive review,” Renewable and
[5] G. Azretbergenova, A. Syzdykova, “The dependence of the
Sustainable Energy Reviews, vol.150, p.111446, October 2021.
Kazakhstan economy on the oil sector and the importance of export
diversification,” International Journal of Energy Economics and [25] HANWEI ELETRONICS CO.,LTD. Techical data MQ gaz sensor
Policy, vol. 10 (6), pp. 157-163, August 2020. features http://www.hwsensor.com
[6] Т. Alimbaev, Z. Mazhitova, B. Omarova, K. Yermagambetova, K. [26] J. E. Jackson, A user's guide to principal components. Toronto, Canada:
Atanakova, & A. Zhumanova, “Ecology of the Western region in John Wiley & Sons Inc., 2005.
Kazakhstan: State and main directions of improvement,” In E3S Web [27] J. Hao, & T. Ho, “Machine learning made easy: a review of scikit-learn
of Conferences, International Scientific and Practical Conference package in python programming language,” Journal of Educational and
“Environmental Risks and Safety in Mechanical Engineering” Behavioral Statistics, vol.44(3), pp.348-361, March 2019.
(ERSME-2020), vol. 217, p. 04006, December 2020.
- 223 -
LoRaWAN Optimization for monitoring of
randomly fluctuated parameters
Kim D.K. 1 *, Turlikov A.M. 2, Markovskaya N.V2.
1
Narxoz University, Kazakhstan
2
HSE University, Russia
Email: [email protected], [email protected], [email protected]
Abstract— We consider a communication network as low- Most of the works investigating LoRaWAN optimization
cost monitoring systems (LoRaWAN devices), where the device do not take into account the specifics of the transmitted data
itself initiates transmission. The network consists of (see [2] and references in [3]), with the exception of a few
nodes, each of which sends messages to the base station with of them [4].
information about the current value of some randomly
Then the work is organized as follows: Section II
fluctuated parameters. The messages from each node are sent
at random time intervals and independently of each other. If formulates a message transmission model, which assumes
the airtimes of messages from two or more nodes overlap, a different message sending parameters for various nodes, and
collision occurs and none of the messages reaches the operator. main result. The solution to the optimization problem is
It is necessary to monitor the behavior of the parameters over found in Section III for particular case with a Wiener
time to avoid their extreme values. We study the problem of process.
choosing parameters of message transmission to uniformly
minimize the risk functional of such tramsmission. The work II. SYSTEM MODEL AND MAIN RESULTS
precisely formulates the optimization problem and proposes an
algorithm for its solution in particular cases. In this section we will assume that all nodes use the
same spreading factor (SF) with airtime of one message
Keywords— LoRaWAN, random multiple access, ALOHA, and the same frequency [3]. If during the time
monitoring, probabilistic approach, optimization problem. interval ( ) from the moment one node sends a
I. INTRODUCTION message, the other nodes do not send messages, then the
message is successfully delivered to the operator. Otherwise
Long Range Wide-Area Networks (LoRaWAN) a collision occurs and the messages are lost.
technologies might be used to monitor the results of This assumption is fulfilled in practice when all nodes
measurements at objects located over a large area and are approximately the same distance from the base station.
enable low-cost data transmission and collection systems. We have nodes, each of them sends messages with
We consider LoRaWAN devices of Class A [1] that initiate intesities . It means that intervals
transmission itself and send packets as they have been between messages of node have exponential distribution
generated at random intervals. If during the sending of a with parameter We suppose that they are
message from one node an another node transmits its independent for each node.
message, a collision occurs and both messages are lost Assume that devices in each node measure readings of
(ALOHA type of protocol). some parameter (e.g. voltage) with stochastic dynamics in
We assume that information from different nodes have time, i.e. each node at random times measures values of
various “importance”. Our goal is to use different message random parameter. We assume that the parameter might be
sending intensities depending on the importance of the node describes as random process ( ) where
to optimize some functional that characterizes the – time.
communication system. For example, to monitor voltage Based on these assumptions and by performing a series
fluctuations in the low-voltage electrical grid, the voltage in of arguments [5] we consider a simplified model for the
the nodes near the transformer changes “more slowly” than total stream of successfully delivered messages
in the remote nodes. In order to assess the state of the LV
grid by monitoring the voltage at the nodes, information
from the nodes remote (according to the grid topology) from
the transformer must be received more frequently than from and use the properties of the Poisson process. According to
other nodes. We can take this into account by varying the this model the stream of successfully delivered messages
intensity of sending messages from different nodes: if the from th node might be defined as Poisson with intensity
voltage in a node behaves more predictably, the intensity of where and total intensity of all successfully
sending messages may be low, and vice versa – nodes with delivered messages is equal to
large voltage fluctuations send their messages with a higher
intensity.
- 224 -
Put as exponentially distributed random variable with and
intensity . As a functional that characterizes the
uncertainty of the information from node let us take the
variation ( ( )) ∑
( ( )) This expression might be used for optimal SF allocation.
Now we formulate the optimization problem: to find IV. CONCLUSION
( ) at which risk functional is minimal:
We study communication system, which works based on
{ } LoRaWAN technology. The system consists of nodes.
Every node initiates tensmission with some information and
Remark that very high total message sending intensity from send messages to the base station at random time intervals
all nodes can occupy the transmission medium and thus independently of each other. Every node belongs to one
increase the number of message collisions. The same can spreading factor with some airtime. If nodes have the same
happen when the total message sending intensity is low, i.e. spreading factor and airtimes of messages from two or more
when all nodes rarely send up-to-date information. nodes overlap, a collision occurs and none of the messages
It means that we have to find maximal value of the reaches the operator.
intensity of total stream of successfully delivered messages. According to our model, each node measures some
Consider the optimization problem: parameter, which behaves stochastically in time. Thus we
have randomness associated with the time of measurement
of the parameter and randomness associated with the
The maximal value of the intensity of total stream of parameter itself. We describe the dynamics of the parameter
successfully delivered messages over time with a random process and study its variance over
time between successfully delivered messages. The task is
(1) to find the optimal average lengths of measurement
intervals, so that the variance for all nodes is minimal. In
this work we formulate an optimization problem and find its
III. EXAMPLE WITH ( ) AS A WIENER PROCESS particular solution when the parameter is described by a
Wiener process.
Let ( ) be Wiener process with initial state
( ) and variance : ACKNOWLEDGMENT
Andrey Turlikov and Natalya Markovskaya prepared this
( ) ( ) ( ) work within the framework of the Basic Research Program at
HSE University.
where ( ) – Wiener process. Then
REFERENCES
[1] “What is LoRaWAN,” https://lora-alliance.org/about-lorawan/, 2022.
( ( )) [2] O. Gusev, A. Turlikov, S. Kuzmichev and N. Stepanov, Data Delivery
Efficient Spreading Factor Allocation in Dense LoRaWAN
Find such that all variances are equal to a constant: Deployments, in 2019 XVI International Symposium “Problems of
Redundancy in Information and Control Systems”
(REDUNDANCY), 2019, pp. 199-204.
[3] F. S. D. Silva, E. P. Neto, H. Oliveira, D. Rosário, E. Cerqueira, C.
Both, S. Zeadally and A. V. Neto. A Survey on Long-Range Wide-
Area Network Technology Optimizations, IEEE Access, vol. 9, 2021,
pp. 106079–106106.
It is easy to see that [4] L.-C. Hwang, C.Chen, T. Ku and W. Shyu, A bridge between the
smart grid and the Internet of Things: Theoretical and practical roles
of LoRa, International Journal of Electrical Power & Energy
Systems, 113, 2019, pp. 971–981.
and [5] Kim D., Georgiev G., Markovskaya N. A Model of Random Multiple
Access in Unlicensed Spectrum Systems , in 2022 Wave Electronics
and its Application in Information and Telecommunication Systems,
WECONF, 2022.
∑
It follows from (1) that
From this we obtain that
- 225 -
August 22(TUE) - 25(FRI), 2023
Oral Session 11
14:00~15:40 Thu, August 24, 2023
Room : A428
Chair: Gani Balbaev (Academy of Logistcs and Transport, Kazakhstan)
- 226 -
Enhancing Public Bus Line Classification
and User Route Estimation through Deep
Learning: A Comprehensive Analysis and
Predictive Modeling Approach
Tsetsentsengel Munkhbayar1, Zolzaya Dashdorj1*, Tae-Koo Kang2, Tae-Koo Kang2,
Kyoung-Geun2 Cho, Taekgwon Nam2, and Erdenebaatar Altangerel1
1
Mongolian University of Science and Technology, Ulaanbaatar, Mongolia
2
Sangmyung university, Seoul, Republic of Korea
*[email protected]
Abstract — Bus line classification is a crucial task in transportation speed to 10km/h on weekdays, which is
optimizing public transportation systems. Traditional almost close to the walking speed. Therefore, 51% of the
classification methods may not effectively capture temporal modal share is public transport in the capital city,
dependencies and sequential patterns in bus line data. In Ulaanbaatar, see Figure 1, 2.
this study, we propose the use of Long Short-Term
Memory Networks (LSTM) for bus line classification. We
collected a dataset consisting of various attributes, such as
route length, and number of stops for a set of bus lines. In
this study, we conduct bus line classification by LSTM as
well as the optimal user’s route is estimated. The findings
of this study highlight the potential of LSTM for bus line
classification, providing transportation authorities and
planners with a powerful tool to enhance route planning,
resource allocation, and service improvements, leading to
improved commuter satisfaction and optimized urban
mobility.
Keywords — bus transportation network, intelligent
transportation systems, graph network, deep learning, route Figure 1. Modal share in Ulaanbaatar
optimization
I. INTRODUCTION
Public transportation plays a vital role in modern
urban living, providing a convenient and sustainable
means of commuting for millions of people worldwide.
As cities expand and transportation networks become Figure 2. Ridership in Ulaanbaatar
more complex, optimizing the efficiency and user
experience of public transportation systems becomes Public transportation route classification involves
increasingly important. One crucial aspect of achieving grouping routes based on various factors such as
this optimization is the classification of public distance, frequency, purpose, service level, and
transportation routes. By categorizing routes based on geographic coverage. By assigning distinct categories to
specific characteristics, authorities and commuters alike routes, transportation authorities can efficiently manage
can benefit from improved planning, streamlined and allocate resources, optimize scheduling, and tailor
operations, and enhanced passenger satisfaction. services to meet the diverse needs of commuters. This
Passengers are not familiar with public transport routes, research aims at analyzing the diversity of bus lines by
or do not know the best way to reach their destination bus line classification based on the CNN model.
faster. At this point, it is difficult to know which buses to Therefore, we conduct the user's route estimation using
take and how to transit. For instance, in Ulaanbaatar, the Dijkstra algorithms based on the public transportation
capital city of Mongolia, there were several vehicles 3 network.
times in the last 10 years that reduced public
- 227 -
II. RELATED WORKS payment system (Umoney AFC system) in Ulaanbaatar's
The state of the art in bus line classification public transportation service has been implemented in
involves the application of advanced machine learning Mongolia since 2015.
techniques and data-driven approaches to improve the
accuracy and effectiveness of classification models.
Related works in the field of bus line classification have
explored various approaches and techniques to improve
the accuracy and efficiency of classification models. Liu
et al. [1] proposed a clustering ensemble approach to bus
line classification by combining multiple clustering
algorithms. The ensemble method enhanced the
accuracy and stability of classification results, enabling
more effective bus line categorization. Tang et al. [2]
investigated the use of Support Vector Machines (SVM)
Figure 3. Bus transportation network in Ulaanbaatar city
for bus line classification. It explored different kernel
functions and parameter settings to optimize the
SVM-based classification model, achieving promising
results in classifying bus lines. Feng et al. [3] proposed a
bus route classification method based on K-means
clustering and hierarchical clustering. By analyzing the
spatial characteristics of bus routes, the approach
effectively grouped similar routes, aiding in route
planning and optimization. Guo et al. [4] introduced a
hybrid approach that combined decision tree algorithms
with a genetic algorithm for bus line classification. The
genetic algorithm optimized the decision tree structure
and attribute selection, leading to improved accuracy in
classifying bus lines. Zhao et al. [5] investigated the use
of Deep Neural Networks (DNN) for bus line Figure 4. Bus transportation network in Erdenet city
classification. By utilizing deep learning architectures,
Within the project, a total of 1300 buses were installed
such as multi-layer perceptrons, convolutional neural
with GPS devices, Umoney card readers, and CCTV.
networks, and recurrent neural networks, the study
The data types include the followings:
achieved higher accuracy in classifying bus lines
compared to traditional machine learning methods. Chen ● Bus routes information - Parking routes, their
et al. [6] proposed an ensemble learning framework for order, and the distance between the parking lot
bus line classification, combining multiple classification (length, length of road, number of parking
algorithms, including decision trees, SVM, and neural places)
networks. The ensemble approach enhanced the
accuracy and robustness of the classification model, ● Parking information - The ID, name, and
providing more reliable results. By leveraging these location of each parking lot, parking lot,
approaches, researchers aim to improve the accuracy, latitude and longitude (geographic location)
interpretability, and efficiency of bus line classification
Erdenet city: We collected 6 bus line data of 36 public
models, contributing to better planning, management,
buses and 146 bus stops with the lat/lon which run in
and optimization of public transportation systems.
Erdenet, Mongolia by 2 operator companies. The GPS
III. DATA PROCESSING trajectories are also extracted for each bus every 15 secs.
Ulaanbaatar city: We collected 91 route information of The bus lines are visualized in a transport network graph
1300 public buses and 931 bus stops with the lat/lon in Figure 4.
which run in the capital city of Ulaanbaatar Mongolia, IV. METHODOLOGY
by 21 operator companies. The routes are visualized in a
We propose bus line classification by machine
transport network graph in Figure 3. Those types of data
learning and user route estimation by Dijkstra’s
were downloaded from Umoney AFC system survey
algorithm.
report, which is updated daily. A project, the Electronic
- 228 -
1. The bus line classification: V. EXPERIMENT RESULTS
The architecture of the bus line classification is as As cities continue to grow and transportation demands
evolve, route classification will play a crucial role in
follows. We propose to use LSTM as a deep learning
creating well-organized and user-centric public
model compared to Decision Tree as a conventional
transportation systems, fostering sustainable and
model.
accessible urban mobility for all. We first transform our
Decision Tree: Apply the decision tree algorithm, bus line dataset by dividing it into a sequence of 5 bus
such as ID3 (Iterative Dichotomiser 3), C4.5, or CART stops to encourage LSTM input.
(Classification and Regression Trees), to construct the Table 1. Learning performance on bus line data
decision tree. The algorithm recursively partitions the
dataset based on the selected attribute at each internal F1-score Accuracy
node, creating decision nodes and leaf nodes. At each
node, determine the splitting criterion that best separates Decision Tree 2.05% 2.05%
the data based on the target variable (bus line
classification). Decision Tree 98.3% 98.3%
(including line
information)
LSTM: This typically involves defining the number
of LSTM layers, the number of LSTM units in each LSTM 68.0% 68.0%
layer, and the output layer.
LSTM (including line 98.7% 98.7%
information)
To estimate the characteristics of bus line diversity, we
classify the sequences of all the bus lines. Table 1
summarizes the accuracy and F1-score of the learning
performance in the training dataset.
Figure 5. LSTM layer architecture
We configured the architecture of LSTM. The first
layer is a one-dimensional LSTM layer of 256 units,
with a group-out rate of 0.1, and a recurrent dropout of
0.1. We add a dense (256 units) with a Relu activation
followed by dropout regularization (0.1) and a flattened
layer into the model. The output layer contains 'softmax' Figure 6. Number of overlapped sections of connected
activation and the number of output classes. bus stops over bus lines/bus routes
The results show that LSTM outperformed the
2. The user’s route estimation: conventional learning model, with an accuracy of 98.7%
considering the line information in the sequence of bus
The A* (A-star) algorithm can be used for estimating a stops. Therefore, we highlight that the LSTM
user's bus route by finding the most efficient path from a performance applied in the sequences of bus lines was
given starting point to a destination. A* is guaranteed to relatively small around 68%. This explains that bus lines
find an optimal path without processing any node more are required to be optimized in terms of the diversity of
than once and A* is equivalent to running Dijkstra's bus lines. Figure 6 shows the overlapped section of
algorithm with the reduced cost d'(x, y) = d(x, y) + h(y) connected bus stops over all the bus lines. For instance,
− h(x). a section of 16 sequentially connected bus lines overlaid
- 229 -
over 11 bus lines. But a section of 25 sequentially
connected bus lines overlaid over 3 bus lines. This
explains that bus line planning needs to be optimized,
reducing the overlaps. The diversity of bus lines is poor
due to the overlaps.
We also applied A* algorithm for estimating the optimal
route for user’s demand given starting and destination
bus stops. We represent the bus network as a graph,
where each bus stop is a node, and the connections
between stops (bus routes) are the edges. Assign
appropriate weights to the edges, such as travel time,
distance, or cost. We validated our approach on Figure 8. Alternative routes driven by the A* algorithm
randomly chosen 89 travel routes for passengers. With
starting and destination bus stops, we were able to find We built our search tree of bus routes with weighted
the possible routes that were ordered by the total travel (cost) nodes using three parameters: total distance of
time, total travel length, and the number of bus stops. bus routes, number of bus stops, and total travel time.
The distribution of the number of possible routes in For instance, we search starting bus stop as “Зүүн 4
random pairs of starting and destination locations were зам/хойд” and goal bus stop as “Тэнгис кино театр”
described in Figure 7. More than 1 route has been and three possible routes were found with the estimated
estimated for 45% of the random pairs. The rest of distance, travel time and number of bus stops.
random pairs is only one route recommended. The bus Table 2. Comparison of search performance and results
stops were connected given the routes which were by three parameters
received from the Umoney AFC system survey report. Start and Travel Travel time Number of bus
destination distance stops (passing)
However, the length of the interval of parking is bus stops
constant. Зүүн 4 3.02km 3.29km 3.29km
зам/хойд/->Тэ 14:45 14:06 14:06
нгис кино 5 bus stops 4 bus stops 4 bus stops
театр
Баянзүрх 7.45km 7.61km 7.5km
дүүрэг-> 35:49 33:32 34:04
Өргөө кино 12 bus stops 13 bus stops 11 bus stops
тщатр
Офицерууды 9.73km 10.25km 10.65km
н ордон-> 48:31 46:24 48:46
Парадокс 17 bus stops 17 bus stops 16 bus stops
Бөхийн 5.69km 5.69km 5.69km
өргөө-> 24:21 24:21 24:21
Мишээл 8 bus stops 8 bus stops 8 bus stops
Баянзүрх 8.03km 8.08km 8.08km
дүүрэг-> iii, 38:01 36:04 36:04
iv хороолол 12 bus stops 11 bus stops 11 bus stops
Figure 7. Distribution of number of possible routes in random pairs of 5 шар -> iii, iv 9.16km 9.16km 9.23km
starting and destination locations хороолол 45:11 45:11 46:12
15 bus stops 15 bus stops 14 bus stops
We also compared the best routes generated by the
We demonstrated our approach on a mobile application
search with 3 different parameters (total travel time,
for Erdenet city, as an output of this research and the
total travel length, and the number of stations) in order
search algorithm works on the server side and produces
to know if the best routes are the same, as shown in
results by user's crossover and displays the OpenLayers
Figure 8. In total, 66% routes were the same, 30% routes
Map API. The UI of the application is described in
were 2 different and there were no 3 different routes.
Figure 9.
This shows that the search algorithm could choose the
best route differently depending on the parameters.
- 230 -
Transportation Research Part C: Emerging Technologies,
72, 38-49.
[3] Feng, Y., Wu, Q., & Luo, Y. (2017). Bus Route
Classification Using K-means Clustering and Hierarchical
Clustering. Journal of Transportation Systems
Engineering and Information Technology, 17(5), 149-156.
[4] Guo, J., Liu, Y., & Li, M. (2014). Bus Line Classification
Based on Decision Tree and Genetic Algorithm. In 2014
International Conference on Transportation Information
and Safety (ICTIS) (pp. 422-427). IEEE.
[5] Zhao, X., Han, B., & Huang, Y. (2018). Bus Line
Classification Using Deep Neural Networks. Journal of
Advanced Transportation, 2018, 4601374.
[6] Chen, H., Lu, H., & Wang, C. (2019). Bus Line
Figure 9. User’s route estimation for Erdenet city Classification Using Ensemble Learning. In Proceedings
of the 2019 2nd International Conference on Image,
ACKNOWLEDGMENT Video and Signal Processing (pp. 22-27). ACM.
This research was supported by KOICA(Korea [7] Wu, L., & Qiu, J. (2018). Bus Line Classification and Its
International Cooperation Agency) through "Capacity Application Based on K-means Clustering. Journal of
Building Project for School of Information and Coastal Research, Special Issue 85, 1391-1395.
Communication Technology at Mongolian University [8] Dai, M., He, B., & Zhang, C. (2017). Bus Line
of Science and Technology in Mongolia" (Contract No. Classification Based on DBSCAN Clustering. In
Proceedings of the 2017 IEEE International Conference
P2019-00124). Also partially supported by Mongolian
on Intelligent Transportation Systems (pp. 1-5). IEEE.
Foundation for Science and Technology (MFST).
[9] Du, S., Chen, Z., Chen, J., & Zeng, Z. (2019). A Hybrid
VI. CONCLUSION Model for Bus Line Classification. In Proceedings of the
2019 5th International Conference on Control,
Public transportation route classification offers Automation and Robotics (pp. 86-90). ACM.
significant benefits in terms of service planning, [10] Yin, X., & Wang, C. (2015). Bus Line Classification
resource allocation, user experience, and data-driven Based on Machine Learning. In Proceedings of the 2015
decision making. LSTM based bus line classification IEEE International Conference on Computer and
performed relatively well, with an accuracy of 98.7% Communications (pp. 1219-1223). IEEE.
considering bus line information in the training set. We [11] Zhang, J., Zhang, L., Zheng, Z., & Wang, S. (2017). Bus
also identified the overlapped sections of connected bus Line Classification Using Machine Learning Techniques.
In 2017 5th International Conference on Logistics,
stops over bus lines. For instance, up to 25 connected Informatics and Service Science (LISS) (pp. 1-5). IEEE.
bus stops were observed over 3 bus lines. Bus line [12] Wang, J., Luo, W., & Liu, Z. (2019). Bus Line
diversity is low due to those overlaps. By systematically Classification Based on Improved K-means Algorithm
categorizing routes based on specific characteristics, and SOM Neural Network. Journal of Physics:
transportation authorities can optimize operations, Conference Series, 1168(3), 032009.
improve passenger satisfaction, and ensure efficient [13] Zhang, H., Wang, Y., Wang, H., & Zhu, Z. (2018). Bus
utilization of resources. Our proposed user’s route Line Classification Based on Feature Engineering and
estimation approach is the ideal way to implement when Deep Learning. In 2018 13th IEEE Conference on
Industrial Electronics and Applications (ICIEA) (pp.
the bus stops are limited. We validated our approach
388-392). IEEE.
which relies on the shortest path by time, distance and
[14] Chen, Y., Fu, Y., & Zhang, W. (2020). Bus Line
number of bus stops. We further study identification of Classification Based on Extreme Learning Machine.
Express vs. Local Routes, High-Demand vs. Journal of Intelligent & Fuzzy Systems, 39(4),
Low-Demand Routes, Peak Hour vs. Off-Peak Hour 6015-6025.
Routes, Local Neighborhood Routes vs. Citywide [15] Jiang, Z., Wang, H., & Liu, F. (2017). Bus Line
Routes. Classification
VII. REFERENCES
[1] Liu, X., Yu, H., & Chen, X. (2019). Bus Line
Classification Based on Clustering Ensemble. Journal of
Advanced Transportation, 2019, 6279480.
[2] Tang, L., Lin, Y., Zhou, J., Wang, J., & Li, Y. (2016). Bus
Line Classification Using Support Vector Machines.
- 231 -
Development of Fault Diagnosis Simulator of
Pneumatic Door System for Electric Railway
Vehicles
Gil Hyun Kang1, Young Min Lim2, In Su Chung3 and Chul Su Kim 3*
1
Industry-Academic Cooperation Foundation Korea National University of Transportation, Uiwang, Korea
2
ROTECO, Suwon, Korea
3
School of Railway Engineering, Korea National University of Transportation, Uiwang, Korea
*
Email: chalskim@ ut.ac.kr
Abstract—This study is on the development of a pneumatic methods such as wear measurement of wheels of railway
door failure diagnosis simulator and door durability tester that vehicles [4]. However, in actual commuter trains, the number
can be implemented in the laboratory before applying the of sensors is high due to the large number of doors, so a wire
system to operating vehicles. In particular, as a method of sensor that is a low-cost sensor and easy to calibrate in terms
diagnosing door failure using artificial intelligence, a distance of maintenance was used.
sensor was proposed instead of a pressure sensor or an acoustic
sensor used in previous studies. Thus, the wire sensors, which The purpose of this study is summarized as follows. (1) It
could deliver the distance data, were installed on the door panels is the development of a door failure diagnosis simulator that
of the pneumatic door system. The moving distance of the door uses wire sensor which can replace pressure sensors and
panels could be collected using the A/D converter and DAQ sound sensors for existing pneumatic doors. (2) To this end,
module. Such open and close distance data may be used as it is accurately collected through a simulator developed by
future failure diagnosis data using separate failure diagnosis major failure mode data selected using Failure Modes,
software. Failures that occur during train operation can be Effects and Criticality Analysis (FMECA) [5]. The collected
simulated using test equipment by applying FMECA analysis, data is used for supervised learning so that it can be pre-
and durability evaluation of the parts of door system is available. processed and statistical features can be provided for use in
In addition, the obtained failure data can be used for supervised fault diagnosis using artificial intelligence. (3) In order to
learning during machine learning. Finally, the durability test of apply the simulator to the actual operating vehicle, the
the door engine was performed using this test equipment. As a
durability test of the door engine, including the wire sensor,
result of a total of over 517,205 repeated opening and closing
tests over two months, the door engine confirmed 193,990 B 10
is conducted to verify whether the wire sensor can be
life guarantees at a 95% confidence level. This proved that the commercialized and used in the simulator and future real
current pneumatic door system can be used safely for 3 years, vehicles.
which is the cycle of heavy maintenance for Korean railway
vehicles. In addition, it was found that this displacement
II. DEVELOPMENT OF FAILURE DIAGNOSIS SIMULATOR
measurement system can be used as a sensor for CBM of an A. Pneumatic Door and Failure Diagnosis Simulator
actual vehicle. Figure 1 shows the structure of the pneumatic door system
(left) and the developed fault diagnosis simulator (right).
Keywords—Railway Vehicles, Fault Diagnosis Simulator,
Machine Learning, Durability Test, Pneumatic Door System
I. INTRODUCTION
The door system of commuter rail vehicles is an important
device to be managed to prevent major traffic congestion in
the event of a failure. Since the pneumatic door system uses
compressed air to open and close, failure diagnosis has
generally been performed using compressed air pressure
sensors or acoustic sensors during open and close operations
[1][2]. However, pressure sensors have high-cost problems
that require modification of air cylinders for installation, and
difficulties when analyzing air pressure patterns due to the
nature of compressible fluids. In addition, sound sensors also Fig. 1. Pneumatic door structure and failure diagnosis simulator
have limitations due to the nature of railway vehicles where
a lot of ambient noise is mixed during vehicle operation.
Therefore, unlike previous studies, a displacement sensor was The pneumatic doors used in the test use a single cylinder
used to develop equipment that can perform machine learning double acting pneumatic type air cylinder. By the control
with displacement and speed data of the door panel during the power of the DC100V, the solenoid valve operates according
operation of the pneumatic door. In order to overcome the to the control command of the door switch to drive the belt,
challenges faced when using pressure sensors or acoustic and the door panel connected to it moves to perform the door
sensors, the travel distance of the door panel is measured so opening and closing function. The simulator was installed on
that failure diagnosis can be performed using distance and the ground using parts used in real vehicles and consisted of
speed data [3]. Universal sensors capable of measuring travel door engines, wire sensors, digital timers, data collectors and
distance include laser sensors or wire sensors. Laser sensors air compressors. When used as a fault diagnosis simulator,
have been widely used as high-precision measurement data for each fault and failure type can be collected and
classified and diagnosed by a separate machine learning
- 232 -
program by creating normal and abnormal conditions for each ln(1−0.95) 2
1
fault type and operating the door opening and closing switch. 𝑇4 = 193,990 × { }= 517,205 (3)
4×ln(1−0.1)
B. Displacement and Speed Data of Pneumatic Door B. Test Time of the Samples
In order to obtain data as a simulator for fault diagnosis, Table 1 shows the test time for failure by number of
the classification and definition of failure are important. samples performed. The non-failure test time was calculated
FMECA is a way to prevent failures or accidents by from 1 to 4 samples.
identifying potential failure modes that seriously affect the
product. The door engine was selected by performing fault TABLE I. NO FAILURE TEST TIME OF SAMPLE
data and the FMECA based on SAE-J1739 [6]. Figure 2 is an
No failure test time (cycle)
example of collecting and showing normal door operation Number of
sample 𝐵10life
data (normal) and an abnormal data with obstacles stuck in Parameters Test cycle
the upper part of the door engine. confidence level
𝐶𝐿 = 0.95,
4 193,990 517,205
unreliability 𝑝 =0.1,
shape parameter β = 2
IV. CONCULUSION
Instead of existing pressure and acoustic sensors, a fault
diagnosis simulator was developed using a wire sensor for
distance measurement. As a result of data acquisition with the
Fig. 2. Example of a normal(left) and abnormal(right) data of the simulator simulator, normal and abnormal data could be collected. In
addition, as a result of the durability evaluation, it was found
C. Durability Evaluation of Door Engine of the Simulator that the life of the device was guaranteed up to three years of
A no-failure test was performed to evaluate the durability the electric vehicle overhaul period. (1) By applying FMECA,
and ensure reliability of the developed pneumatic door the failure test time of the door engine with the highest risk
simulator. The no-failure test can prevent the cost of the life priority was calculated to be about 193,990 cycles. (2) As a
test from increasing by reducing the number of specimen. The result of the 4 sample tests, it was confirmed that there was no
test method is to continue the test with a predetermined fault for the 𝐵10 life calculated for the 517,205 no-failure test
number of samples until a set time of the target product, and time. After the test was completed, there were no
if a failure does not occur, the reliability of the product is abnormalities of the equipment. It was confirmed through this
determined to be satisfied. Assuming that the probability life study that it can be applied as a CBM sensor to pneumatic
distribution of mechanical parts generally follows a two- body side entrance doors of railway vehicles in the future.
parameter Weibull distribution, the no failure test time can be
calculated by the following equation [7]. ACKNOWLEDGMENT
1 This work was supported by the Korea Agency for
𝑙𝑛(1−𝐶𝐿 ) 𝛽 Infrastructure Technology Advancement (KAIA) by the
𝑇𝑛 = 𝐵𝑝 × { } (1)
𝑛×𝑙𝑛(1−𝑝) Ministry of Land, Infrastructure and Transport under Grant
RS-2019-KA152194.
where, 𝑇𝑛 : no-failure test, 𝐵𝑝 : the assurance life (life with a
product failure probability of p%), 𝐶𝐿 : the confidence level, REFERENCES
𝑛 : number of samples, 𝑝 : unreliability( 𝑝 = 0.1 for B10 [1] N. Lehrasab, H. P. B. Dassanayake, C. Roberts, S. Fararooy, and C. J.
lifetime), 𝛽 : the shape parameter. Goodman, “Industrial fault diagnosis: Pneumatic train door case
study,” Proceedings of the Institution of Mechanical Engineers, Part F:
For the durability test, an additional timer circuit Journal of Rail and Rapid Transit, vol. 216, no. 3. SAGE Publications,
manufactured separately was installed in the fault diagnosis pp. 175-183, May. 2002.
simulator to automatically perform the test. The test is [2] Y. Sun, G. Xie, Y. Cao, and T. Wen, “Strategy for Fault Diagnosis on
performed automatically by setting the operation mode of the Train Plug Doors Using Audio Sensors,” Sensors, vol. 19, no. 1. MDPI
door, setting the opening and closing time of the door switch. AG, Dec. 2018.
[3] G.H. Kang, H.C. Ki, S. H. An, J.H. Choi, and C.S. Kim, “A Genetic
III. RESULTS Algorithm and PCA-Based Feature Selection to Improve the Failure
Diagnosis Performance of Railway Vehicle Doors,” IEEE Access, vol.
A. No-failure Test Time 10, 2022, pp. 115787-115801.
[4] P. Turabimana, and C. Nkundineza, “Development of an On-Board
Based on the commuter electric vehicle maintenance Measurement System for Railway Vehicle Wheel Flange Wear,”
regulations of KORAIL, the 𝐵10 life is calculated to be Sensors, vol. 20(1), 303. 2020.
193,990 cycles, as shown in (2), considering 206 door [5] Y. Deng, Q. Li and Y. Lu, “A research on subway physical
opening and closing cycles per day, 365 days, and 86% per vulnerability based on network theory and FMECA,” Safety Science,
year. vol. 80, 2015, pp.127-134.
206 cycles 365day [6] SAE-J1739, “Potential Failure Mode and Effects Analysis in
𝐵10 𝑙𝑖𝑓𝑒 = × × 0.86 × 3 year = 193,990 (2) Manufacturing and Assembly Processes (Process FMEA), and
1day year
Potential Failure Mode and Effects Analysis for Machinery (Machinery
When the confidence level 𝐶𝐿 = 0.95, number of samples 𝑛 FMEA), SAE, June 2000.
= 4 , unreliability 𝑝 = 0.1 for 𝐵10 lifetime), the shape [7] K. W. Fertig and N. R. Mann, “Life-Test Sampling Plans for Two-
parameter 𝛽 = 2, the durability test time could be calculated Parameter Weibull Populations,” Technometrics, vol.22,1980, pp.165-
177.
by (3).
- 233 -
August 22(TUE) - 25(FRI), 2023
Oral Session 12
14:00~15:40 Thu, August 24, 2023
Room : A432
Chair: Seungkwon Shin (Korea Railroad Research Institute, Korea)
- 234 -
Mathematical model for determining the
orientation of a star sensor
Kuanysh Alipbayev 1 Kamila Saurova 2*
1
Almaty University of Energy and Communications named after G.Daukeyev, Kazakhstan
2
Almaty University of Energy and Communications named after G.Daukeyev, Kazakhstan
Email: [email protected]
Abstract— The star sensor (SS) is a device for determining However, the question of having adequate error models
the orientation of a spacecraft. The orientation of the star sensor and mathematical models for determining the orientation of
in outer space occurs through a lens and a detector that register the SS remains relevant.
images of the starry sky and compare them with stars from the
star catalog. Usually, a star sensor is based on a CCD sensor (a Now the use of the star sensor in the modern space industry
charge-coupled device), which consists of a set of photosensitive has become an integral part of the development of spacecraft.
photodiodes. Mathematical and software modeling is used for development
because it will be possible to test the system for faults and
This article discusses studies of improving the accuracy of failures, improve the system, and minimize failures on the real
spacecraft star sensors, as well as a mathematical model for vehicle.
determining the orientation of a star sensor.
The field of view of the lens can be calibrated from 10 to
Keywords— Star sensor. Mathematical modeling of SS. 25 degrees, and the mass of the sensor varies in the range of 2
Orientation of the star sensor. Orientation detection system. A way to 5 kilograms. The sensitivity of the detector also depends on
to increase accuracy. the model and customer requirements.
I. INTRODUCTION When using a star sensor, there is a lot of interference,
including image illumination and small particles hitting the
The star sensor (SS) can determine the orientation sensor lens. In order to practice such interference and failures,
parameters of a spacecraft with high accuracy, but under simulated conditions are produced for test models on the
favorable illumination conditions and within a limited range ground.
of its angular velocities [1].
All parts of the star sensor can be divided into:
Many works are devoted to the study of stellar sensor
errors. In [2] errors based on flight experiment data are optical components (lens, etc.);
considered. The errors caused by thermoelastic deformations photosensitive components (photodetector);
[3], electromagnetic compatibility [4], the use of SS to identify
the parameters of the inertial navigation system error model of electronic components (photodetector electronics,
the upper stage [5], etc. are studied. The influence of SS errors data processing electronics, memory).
on the accuracy of spacecraft orientation and platform
stabilization systems is discussed in [6-9]. The star sensor determines its location in the inertial
reference system by observing the starry sky in the visible
The paper [10] describes the main stages of developing the wavelength range. The sensor obtains an image of the visible
experimental model of the star tracker. In [11] the approach to starry sky by means of the lens built into the sensor. In the
the design and development of the optical head of the star received image, images of stars are detected, for which the
tracker and its lens for satellites is presented. Developed a coordinates are determined on the matrix radiation receiver.
technical model of the optical head of the star tracker, the basis The characteristics of the stars in the image are compared with
for the development of a prototype star tracker for use. [12] the catalog of stars stored in the star sensor's memory, for
An experimental model of the star tracker has been developed, which the celestial coordinates in the equatorial coordinate
with the following characteristics: field of view 20°, update system are known. The result of the comparison is the
frequency 2 Hz, deviation angle 40°, and the accuracy of identification of the celestial stars from the catalog with the
orientation determination along the optical axis/around the images of the stars in the image. Based on these stars, the
optical axis 15/50 angles. orientation of the star sensor is calculated with respect to the
inertial coordinate system associated with the stationary stars.
In [13] the work considers the introduction to the star This orientation determination action is periodically repeated.
sensors and procedures for recognizing the image of the starry The maximum frequency of repetition of the orientation
sky in several stages. The improvement of the accuracy of determination action is 10 Hz. Stellar sensor calibrations
spacecraft orientation parameters measurement by stellar should be performed as the photosensitive CCD degrades.
sensors is investigated. The calculation method using the Calibrations are repeated from the radiation environment.
algorithm of exact centroid estimation, which recovers the There is also a thermoelectric cooler in the sensor to increase
point spread function from the recorded astronomical images, the temperature of the photodetector.
is applied.
Minimization of the effects of atmospheric turbulence and
system noise on long-exposure images obtained with a
ground-based telescope has been proposed [14].
- 235 -
а23 = −cos 𝑏 sin 𝑦,
а31 = −sin 𝑎 cos 𝑏,
а32 = cos 𝑎 sin 𝑦 + sin 𝑎 sin 𝑏 cos 𝑦,
а33 = cos 𝑎 cos 𝑦 − sin 𝑎 sin 𝑏 sin 𝑦.
The equation of motion of the spacecraft:
𝐽𝜔 + 𝜔 × 𝐽𝜔 = 𝑀0 , (2)
where J is the tensor of inertia calculated on the principal
axes.
ω is the vector of absolute angular velocity.
Fig. 1. Basic elements of the star sensor 𝑀0 − The vector of the total momentum of external
forces.
II. MATH MODEL OF SYSTEM
First, the coordinate systems of motion of the star sensor Dynamic equations of kinematic Poisson relations:
should be considered. To do this, two right Cartesian 𝜏𝑖 = 𝜔 × 𝜏𝑖 , (3)
coordinate systems with the zero-point O in the center of the
celestial sphere is introduced.
where 𝜏𝑖 is the orth of the corresponding axis of the
moving coordinate system.
When integrating the equations of motion ((2), (3)), a zero
moment is used and the axisymmetric spacecraft, together
with it the sensor, performs a regular precession with known
parameters. For this purpose, mathematical modeling of
orbital and relative motion under the action of disturbing,
restoring, and controlling moments of different nature has
been implemented.
Here are some of them:
− gravitational moment:
3𝜇𝑔
𝑔𝑟
𝑀0 = 𝐸 × 𝐽𝐸,
𝑟3
Fig. 2. Relativity of inertial and moving coordinate systems
where µ is the gravitational parameter of the attracting
ОХ1 Х2 Х3 − integral (base) coordinate system. The first center;
axis lies on the plane of the celestial equator and stares at the r is a distance from the attracting center to the center of
point of the vernal equinox; the third axis lies on the Earth's mass of the spacecraft.
rotation axis; and the second axis is a complement of the right
triplet. E is local vertical component.
Ох1 х2 х3 − a movable but coupled coordinate system, − magnetic moment:
which axes are the axes of the satellite where the star sensor is 𝑚𝑎𝑔𝑛
𝑀0 = 𝑚 × 𝐵,
installed. It is assumed that the optical sensor coincides with
Ох1 , and the photo-editor plane is parallel to Ох2 х3 . where m is the vector of the total magnetic moment of the
carrier.
The orientation of the moving system can be written in
matrix form: B is the vector of induction of the Earth's magnetic field.
а11 а12 а13 − control torque that creates the flywheels:
Х = Ах = |а21 а22 а23 | ⋅ х, (1)
а31 а32 а33 𝑀0мах = −Н − 𝜔 × Н,
where H is the vector of kinetic momentum created by the
where flywheels.
а11 = cos 𝑎 cos 𝑏, During the numerical integration of motion, the vector
а12 = sin 𝑎 sin 𝑦 − cos 𝑎 sin 𝑏 cos 𝑦, equation (2) is projected onto the axes of the moving
coordinate system. Vectors m and H are naturally written in
а13 = sin 𝑎 cos 𝑦 + cos 𝑎 sin 𝑏 sin 𝑦, the moving system and require the transition to the coupled
а21 = sin 𝑏, axes. Let’s calculate these vectors using elements of the theory
of orbital motion.
а22 = cos 𝑏 cos 𝑦,
- 236 -
Let’s take the orbital coordinate system of the spacecraft E is the eccentric anomaly.
ОY1 Y2 Y3 . The first axis is directed to the radius vector of the
Expression (5) describes the relationship between time and
position point of the spacecraft on the orbit, the second axis is
the position of the spacecraft in orbit, expressed through the
on the plane of the orbit, and the third axis complements to the
eccentric anomaly, and expression (6) describes the
right orthogonal system.
relationship of eccentric and true anomalies.
Let’s use expression (4) to write vectors E and B in the
inertial coordinate system. The vector of the true vertical in
the system ОY1 Y2 Y3 can be written as
𝐸 𝑂𝑌 = (1 0 0 )𝑇 ,
and the geomagnetic field induction vector has the form:
−2 sin 𝑢 sin 𝑖
𝐵𝑂𝑌 = 𝐵0 ( cos 𝑢 sin 𝑖 )
cos 𝑖
The vector E in projections on the ОХ1 Х2 Х3 axes can be
Fig. 3. Orientation of the spacecraft coordinate system relative to the written as follows:
inertial
cos Ω cos 𝑢 − sin Ω sin 𝑢 cos 𝑖
The angles of longitude of the ascending node and 𝐸 𝑂𝑋 = (sin Ω cos 𝑢 + cos Ω sin 𝑢 cos 𝑖 ).
inclination determine the position of the orbital plane. The sin 𝑢 sin 𝑖
eccentricity of the orbit (e) determines its shape, and the orbit
parameter (p) its size. The angle of the pericenter argument
(the angle between the line of nodes and the direction to the Numerical integration of equations of motion (2) and (3)
pericenter of the orbit) sets the position of the orbit in the plane with the given initial conditions, system parameters, and
of the spacecraft. The moment in time when the spacecraft elements of the spacecraft orbit, allows to determine the
first passes through the pericenter is considered to be known. matrix of directional cosines (orientation at this moment of the
sensor relative to the inertial system) at each point in time.
Obtaining ОХ1 Х2 Х3 from ОY1 Y2 Y3 occurs through 3
consecutive rotations. The first rotation is performed by the
angle of longitude of the ascending node around the third
inertial axis, the second rotation by the angle of inclination of
the new position of the first inertial axis, the third rotation by
the angle of latitude argument around the new third inertial
axis. The relationship between the two coordinate systems can
be described in matrix form:
Х = СY, (4)
where
С
cos Ω cos 𝑢 − − cos Ω sin 𝑢 − sin Ω sin 𝑖
− sin Ω sin 𝑢 cos 𝑖 − sin Ω cos 𝑢 cos 𝑖
= sin Ω cos 𝑢 + − sin Ω sin 𝑢 + − cos Ω sin 𝑖
+ cos Ω sin 𝑢 cos 𝑖 + cos Ω cos 𝑢 cos 𝑖
( sin 𝑢 sin 𝑖 cos 𝑢 sin 𝑖 cos 𝑖 )
Fig. 4. Finding the vector direction of the star in the moving coordinate
system
The position of the spacecraft in orbit (the distance from
According to the user-defined field of view of the camera,
the attracting center to the center of mass of the spacecraft and
the “observable” part of space is defined. The spherical
the gravitational parameter of the attracting center) can be
coordinates of the stars (α, δ) are converted into the Cartesian
determined through the following equations:
coordinates of the inertial system ( ОХ1 Х2 Х3 ), which are
3⁄ translated into the moving system (Ох1 х2 х3 ) using the known
𝑎 2
𝑡 − 𝜏𝜋 = 1⁄ (𝐸 − 𝑒 sin 𝐸 ), (5) guide cosine matrix. The locations of stars in the moving
𝜇 2 coordinate system are translated into spherical coordinates (φ,
θ), which determines the direction of the vector to the star. The
𝜗 1+𝑒 𝐸 locations of stars in the moving coordinate system are
𝑡𝑔 =√ 𝑡𝑔 , (6)
2 1−𝑒 2 translated into spherical coordinates (φ, θ), which determines
the direction of the vector to the star.
𝑝
𝑟= Consider the HYDRA-CP model from the SODERN
1 + 𝑒 𝑐𝑜𝑠 𝜗
company as a test star sensor.
where а = р/(1 – е2) – the semi-major axis of the orbit;
- 237 -
The problem of star recognition is solved in an idealized
formulation. The simulated images of the starry sky do not
contain objects with complex configurations. In particular, the
tracks whose nature is significant angular velocities of the
camera and noises caused by the passage of protons through
the CCD at a large angle to its optical axis are not displayed.
As noted earlier, the images transmitted to the camera contain
only point objects (stars and interference) and noise imitating
the third-party inhomogeneous illumination.
The starry sky image recognition procedure is carried out
in several stages. First, the noise illumination is filtered by
dividing the image into sub-areas and subtracting the average
intensity values in the corresponding area from the current
intensity of each pixel. This partitioning is necessary to
Fig. 5. SODERN star sensor combat non-uniform illumination. After that, the filtered
image is scanned, and the signal maximums are determined
The tested star sensor has the following characteristics locally by sub-areas. The integral signal intensity in the
(Tab. 1) HYDRA-CP vicinity of the detected maxima according to the specified
detection criterion is compared with the current noise value. If
TABLE I. CHARACTERISTICS OF THE HYDRA-CP STAR SENSOR the criterion is fulfilled, the object is considered detected, and
its center of mass is calculated in the coordinate system
Sensor name HAS-2 (CMOS) associated with the detector plane:
Pixel size 18 mcm
∑ 𝜉𝑖 𝐼𝑖 ∑ 𝜂𝑖 𝐼𝑖
Matrix size 18.4 х 18.4 mm 𝜉𝑐 = , 𝜂𝑐 =
∑ 𝐼𝑖 ∑ 𝐼𝑖
Resolution 1024х1024 pixels
where (ξ, η) are coordinates of image pixels falling in the
Built-in lens 18.5 mm f2.6 H400 point signal localization area;
Maximum frame rate 30 fps.
I - corresponding intensities (Fig. 7)
Connection interface USB
Using the perspective projection, the observed stars from
the spherical system (φ, θ) move to the rectangular coordinate
system related to the model plane of the photodetector (Fig. 5)
and will be displayed on the monitor. The brightness and size
of the displayed stars are functions of the stellar magnitude;
the number of displayed stars is a configurable parameter.
It is also possible to implement the function of displaying
background objects: illuminating a part of space, finding and
imaging objects that are stationary (dust on the lens) and
moving (space debris) relative to the detector plane. Fig. 7. Calculating the star and determining the center of mass
The image transmitted by the simulator is “captured” by The procedure for recognizing the nodes of the calibration
the camera. Frame-by-frame analysis of the obtained video grids is slightly different from the recognition of simulated
images is carried out in a rectangular coordinate system, star images. In this case, it’s needed to find a large number of
associated with the detector's plane (Fig. 6). stars (from 100 to 400) without missing any. It is necessary to
establish the locations of all stars relative to the central point
of the grid to further establish an unambiguous
correspondence between the simulated and recognized stars.
Recognition begins with the central point, the size of
which is somewhat different from the rest of the stars and
serves as a good indication for its identification. After that, a
priori information about the grid structure is used to predict
the position of images of neighboring nodes from previously
detected ones. This method simplifies the search and detection
of stars by reducing it to a prediction-confirmation-correction
algorithm and allows to automatically obtain the desired
correspondence between the simulated and detected stars.
It is necessary to pay attention to the problem of point
object localization - the selected region of the image (groups
of pixels), which is considered when calculating the center of
Fig. 6. The coordinate system associated with the detector plane.
mass. In this paper, we used a rectangular area in which the
source image is inscribed. All the pixels of the image falling
- 238 -
into this area were considered when calculating its In the presence of a priori data about the sensor motion,
coordinates. But this method has significant drawbacks since obtained from third-party devices or pre-computed by LIS
it does not take into account the real shape of the image. algorithms, the position of stars in the detector system can be
Increased accuracy in determining the coordinates of the predicted with high accuracy. In this case, the recognition and
center of mass can be achieved by solving this problem of detection tasks are reduced to a tracking task performed
finding the contour of the image. The next level of accuracy according to the scheme, prediction-correction, which
can be obtained by using the distribution functions of the significantly increases the speed of the system.
observed signal of the point object. In other words, the
localized signal is interpolated on an analytical function (Fig.
8), estimates of the distribution parameters of which are
carried out at the stage of calibration of the device at different
angles of orientation of the source relative to the optical axis.
Fig. 9. Diagram of the angular distance table
In this work, an algorithm optimized for the search speed
was chosen. The size of the catalog is limited to the N brightest
Fig. 8. Three-dimensional distribution function of the obtained signal stars. Each star is assigned a unique number. A table of size N
x N is created, the cells of which contain the values of angular
In this case, various optical effects, such as distortion in distances between stars with the corresponding numbers (Fig.
the extreme regions of the image, can be taken into account 9). With the help of the built table a quick access to the
and, as a consequence, the accuracy of the instrument is distance by star numbers is carried out.
improved.
Then a list is created (Fig. 10), each row of which contains
Once the stars are identified and the actual angular information about the numbers of stars and the corresponding
distances between them are established using the prearranged angular distance between them. The list is limited according
calibration functions, they are ascertained by comparing the to the angle of view of the sensor used and is arranged in
acquired configurations with the configurations listed in the ascending order of distances.
stellar catalog.
The star detection problem along with the recognition task
is the main factor determining the quality of the instrument as
a whole. It is the most resource-intensive operation and, while
the correct solution of the identification problem in general
determines the accuracy, the speed of the system directly
depends on the methods of solving the detection problem.
One of the basic ideas, which is fundamental in the
implementation of many identification algorithms, is that the
stars in the catalog is represented in the form of vertices of
non-directional graph G1, the weight of its edges equal to the
angular distance, between the stars. The objects found after
analysis of the captured frame are also represented in the form
of a non-directional graph G2. The main task is to implement
fast methods for finding a subgraph from G, isomorphic to G2.
Fig. 10. List of distances between them with a comparison star catalog
Another group of identification algorithms includes the so-
called grid algorithms, the main idea of which is that a bit The algorithm for determining observable stars uses the 5
matrix (or pattern) is associated with each star. The value of brightest objects in the image. One object gets selected, the
the matrix field is 1 if a star adjacent to the selected one falls angular distances to the remaining four objects are calculated,
into it. Otherwise, the field value is 0. To determine the star and suitable pairs are found for each of them from the list of
configurations, a bitwise comparison of the matrices distances. Next, a list of 5 stars is generated for which the
constructed for the analyzed image with the corresponding distance from the first star to the remaining four is in the range
patterns from the catalog is used. δ ± ε, where ε is the accuracy of the sensor in determining the
All these algorithms are of the "Lost In Space" (LIS) type angular distance. For each list entry, all distances between the
and are characterized by the fact that they do not require any stars are determined (a distance table is used), and their
a priori information about the current orientation and angular correspondence to the distances in the image is checked. If at
velocity of the camera. least one distance does not match, we consider the next record.
- 239 -
The scheme of the identification algorithm is shown in Fig. Determination of the current orientation of the spacecraft
11. is reduced to the solution of the optimization problem of the
form:
If among the 5 selected stars in the image there are
“phantoms” (objects which are not present in the catalog), min ∑𝑛𝑖=1‖𝑋𝑖 − 𝐴𝑥𝑖 ‖,
{𝑎𝑗𝑘 }
there will be no correct configuration in the list. In this case,
the star identification procedure is repeated with another set of where n is the list of defined stars;
stars.
X – coordinates of the i-th star in the inertial coordinate
Using this algorithm, the detection was carried out while system;
displaying 1500 objects, while information was present in
only three hundred, that is, the proposed method successfully x – corresponding coordinates in the moving coordinate
bypasses the problem of the presence of dim stars and system. In this case, the elements of the matrix of directional
“phantom” objects in the images. cosines 𝑎𝑗𝑘 must satisfy the orthogonality conditions.
The problem of optimization of the detection rate in the In this case, let us consider a method for determining the
presence of “phantom” objects, as well as the tasks of detailed current matrix of the guide cosines, the essence of which is as
analysis of the algorithm in comparison with existing follows.
approaches, can be the subject of future research.
Let 𝑠𝑗𝑖 be the coordinates of three stars in the detector
coordinate system (j = 1...3 is the index of the corresponding
point, i = 1...3 is the coordinate in projection to the
corresponding axis), and 𝑆𝑗𝑖 be the corresponding coordinates
in the inertial system. The elements of the matrix of guiding
cosines are determined by solving the equations of the form:
𝑆𝑗1 𝑎11 𝑎12 𝑎13 𝑠𝑗1
(𝑆𝑗2 ) = ‖𝑎21 𝑎22 𝑎23 ‖ × (𝑠𝑗2 ).
𝑆𝑗3 𝑎31 𝑎32 𝑎33 𝑠𝑗3
The following equality is evident
3
𝑆𝑗𝑖 = ∑ 𝑎𝑗𝑘 𝑠𝑗𝑘 ,
𝑘=1
Fig. 11. Diagram of the star detection algorithm
which, after introducing an auxiliary matrix into
The object tracking algorithm can also be observed. In this consideration:
instance, advance information regarding the camera's current 𝑠11 𝑠12 𝑠13
orientation and angular velocity was conveyed from the
simulator to the frame processing module. The anticipated 𝐷 = |𝑠21 𝑠22 𝑠23 |
𝑠31 𝑠32 𝑠33
positions of the corresponding stars in the detector coordinate
system were projected, leading to the recognition of selected are grouped into the following 3 algebraic systems
image regions, verification of their existence, adjustment of
𝑆𝑗1 𝑎𝑖1
the centroids' positions, and the subsequent procedure of
comparing the detected configuration with the projected one. ( 𝑗2 ) = 𝐷 (𝑎𝑖2 ).
𝑆
Such an approach significantly amplified the speed of the 𝑆𝑗3 𝑎𝑖3
system.
From which it follows
Figure 12 shows an example of a recognized image and
𝑎𝑖1 𝑆1𝑖
identified objects. Among all detected point objects, nine
brightest objects (double boundary) were identified. The (𝑎𝑖2 ) = 𝐷 −1 (𝑆2𝑖 ).
detection was performed on 5 objects. 𝑎𝑖3 𝑆3𝑖
Then we perform several calculations (for different
triples), average the result, and orthogonalize the matrix of the
guide cosines.
Using the two known camera orientations A(k) and
A(k+1), separated by interval t, we can find the angular
velocity of the spacecraft by solving the Poisson equation:
0 −𝜔3 𝜔2
𝑑𝐴
= Ω𝐴, Ω = ( 𝜔3 0 −𝜔1 ),
𝑑𝑡
−𝜔2 𝜔1 0
where ω is the projection of the absolute angular velocity
on the axes of the inertial coordinate system.
Fig. 12. Determination of the stars image.
In this case
- 240 -
𝑑𝐴 −1 𝐴𝑘+1 − 𝐴𝑘 𝐴𝑘+1 + 𝐴𝑘 Russian Scientific and Technical Conference “Modern problems of
Ω= 𝐴 ≈ ( ) [2] orientation and navigation of spacecraft”, Russia, Tarusa, September
𝑑𝑡 ∆𝑡 2 10-13, 2018. – P.: 9 http://ofo.ikiweb.ru/publ/conf_2018_tez.pdf
[7] 7. G.A. Avanesov, R.V. Bessonov, A.N. Kurkina. Technology of
ground processing of star coordinates data to improve the accuracy of
georeferencing of Earth images from space // Modern Problems of
Earth Remote Sensing from Space. 2018. Т. 15. №6. С. 31-38
III. CONCLUSION http://jr.rse.cosmos.ru/article.aspx?id=2452
[8] 8. D.A. Slavgorodsky. Errors of gyrostabilized optical observation
Studies of accuracy enhancement by stellar sensors of system caused by peculiarities of optical element kinematics //
spacecrafts. A method for improving the accuracy of stellar Proceedings of XXII Conference of Young Scientists with international
sensors of spacecrafts and a mathematical model for participation. St. Petersburg, 17-20 March 2020. - P.: 16-18,
determining the orientation of a stellar sensor is presented. https://elibrary.ru/item.asp?id=44618751&selid=44618765
The experiment shows that the method evaluates orientation [9] 9. Fedorovich S.N. Increasing the accuracy of ball gyroscope rotors
accuracy, calculation time, required memory, size of the star by methods of precision spherical watering // Proceedings of XXII
Conference of Young Scientists “Navigation and Motion Control”, St.
catalog, sky coverage, etc. Petersburg, March 17-20, 2020. - P.: 74,
https://elibrary.ru/item.asp?id=44618751&selid=44618765
REFERENCES
[10] 10. M.M. Moldabekov, S.A. Yelubayev, K.A. Alipbayev, T.M.
[1] S.A. Dyatlov, R.V. Bessonov Review of stellar sensors of spacecraft Bopeyev, A.S. Sukhenko “Development of optical system of star
orientation // All-Russian Scientific-Technical Conference "Modern tracker for satellites" in the Proc. of international scientific conference
Problems of Spacecraft Orientation and Navigation Determination" "Reshetnevskie chteniya”, Russia, Krasnoyarsk, 2013, pp. 237-238.
Proceedings, Russia, 2009. – P.: 12 – 31, [11] 11. Features of design and development of the optical head of star
https://elibrary.ru/contents.asp?id=33699680 tracker. Moldabekov M. Proceedings of the SPIE/ Moldabekov M.,
[2] 2. G. Avanesov, R. Bessonov, A. Kurkina. Investigation of star Akhmedov, D.; Yelubayev, S.; Ten, V.; Bopeyev, T.; Alipbayev, K.;
sensor errors based on flight experiment data // Sixth All-Russian Sukhenko, A // The International Society for Optical Engineering,
Scientific and Technical Conference “Modern problems of spacecraft Amsterdam. - 2014, https://www.spiedigitallibrary.org/conference-
orientation and navigation”, Russia, Tarusa, September 10-13, 2018. – proceedings-of-spie/9241/924122/Features-of-design-and-
P.: 8 http://ofo.ikiweb.ru/publ/conf_2018_tez.pdf development-of-the-optical-head-of/10.1117/12.2067050.short.
[3] 3. G.A. Avanesov, R.V. Bessonov, A.S. Kvashnin, V.E. Shevelev. [12] 12. Software and mathematical support of Kazakhstani star tracker.
Investigation of stellar sensor errors caused by thermoelastic Akhmedov D. Proceedings of the SPIE/ Akhmedov D., Yelubayev S.,
deformations // Sixth All-Russian Scientific and Technical Conference Ten V., Bopeyev T., Alipbayev K., Sukhenko A. // The International
“Modern problems of spacecraft orientation and navigation”, Russia, Society for Optical Engineering, Software and mathematical support of
Tarusa, September 10-13, 2018. – P.: 29 Kazakhstani star tracker. - 2016.
http://ofo.ikiweb.ru/publ/conf_2018_tez.pdf [13] 13. Alipbayev K.1, Saurova K.2, Method of increasing the accuracy
[4] 4. N.N. Brysin, K.P. Lyubchenko, V.M. Muravyev, S.N. Romashin of the star sensor/MECHANICS OF GYROSCOPIC SYSTEMS. 2022.
et al. Investigation of stellar sensors for electromagnetic compatibility https://www.researchgate.net/publication/366718441_Method_of_inc
// Sixth All-Russian Scientific and Technical Conference “Modern reasing_the_accuracy_of_the_star_sensor
problems of spacecraft orientation and navigation,” Russia, Tarusa, [14] 14. Abubekerov M.K., Zakharov A.I., Prokhorov M.E.,
September 10-13, 2018. - P.: 283 Stekolshchikov O.Yu. Method of increasing the accuracy of star
http://ofo.ikiweb.ru/publ/conf_2018_tez.pdf orientation determination and long-term maintenance of increased
[5] 5. A.V. Malyshev, G.S. Sazanov. Identification of error model accuracy of orientation determination and a device for their
parameters of inertial navigation system of upper stage via star sensors implementation // Intelligent System of Thematic Study of
// Proceedings of XXII Conference of Young Scientists “Navigation Naukometric Data, Eurasian patent 026970, June 30, 2017.
and Motion Control”, St. Petersburg, March 17-20, 2020. – P.: 13-15,
https://elibrary.ru/item.asp?id=44618751&selid=44618765 .
[6] 6. A.A. Arefinkin, V.V. Vilensky, S.V. Voronkov, S.E. Zaitsev et
al. Methods to improve the accuracy of the orientation system with the
astro-correction loop of a small remote sensing spacecraft // Sixth All-
- 241 -
Target selection by a group of unmanned
aerial vehicles based on a swarm approach
Aigerim Mazakova1, Vadim Zhmud2, Sholpan Jomartova1, Тalgat Mazakov1
1
Al-Farabi Kazakh National University, Almaty, Kazakhstan
2
Altae-Sayan Branch of the Federal State Budgetary Institution of Science of the Geophysical Service of
the RAS, Novosibirsk, Russia
Email: [email protected]
Abstract-- The article deals with the problem of target One of the most important aspects of using swarm
selection by a group of unmanned aerial vehicles based on technology is the possibility of automatic control. This
a swarm approach using interval analysis and computer means that all devices are synchronized, communicating,
algebra. which allows them to accurately perform their tasks,
The UAV group is controlled from the mission control
center. The MCC sends to all drones the spatial
without operator intervention. Moreover, the use of
coordinates of the target and the required time to reach it. swarm systems reduces the number of errors and
Each of the UAVs independently solves the problem of accidents, which makes them more “reliable” in
attainability of the set goal, depending on its technical operation, especially in extreme situations.
characteristics and its own location. Thus, the process is
parallelized and carried out in real time. Overall, swarm drone systems are a step forward in
The answer about the possibility of achieving the goal robust automatic technology. The future lies in
by each drone is returned to the MCC, which, according continuing research on this technology and developing
to its other criteria, determines a separate drone (or a new strategies for its application.
subgroup of drones) that must solve the required task.
On the basis of interval mathematics, a criterion for II. LITERATURE REVIEW
the attainability of the goal by each UAV is obtained.
Based on the swarm approach, a list of unmanned aerial Drone swarm algorithms are an active area of
vehicles is defined, each of which is capable of performing research in autonomous systems. These algorithms
the task. This list may be empty if the selected target model the behavior of a swarm of drones based on
cannot be reached by any of the available UAVs. observations of natural swarms such as birds, fish, or
When developing the software, the library of interval insects. The goal of swarm algorithms is cooperation and
mathematics was used. The proposed approach can be coordination between individual drones to perform
applied to the study of various robotic systems.
complex tasks collectively.
Keywords—unmanned aerial vehicle (UAV), aerial Swarm algorithms are based on the principles of self-
target (AT), computing device (CD), mission control center
organization, communication and collective intelligence.
(MCC), reachability, swarm approach, interval
mathematics.
Unmanned vehicles in a swarm communicate, coordinate
their behavior, and adapt to a changing environment. This
I. INTRODUCTION allows them to efficiently perform tasks such as search
and rescue, environmental monitoring, or patrolling.
The swarm approach to the creation of unmanned
aerial vehicles is an innovative method that uses There are many studies in the literature on drone
innovative technologies to develop autonomous control swarm algorithms suggesting different approaches and
systems. It is based on a variety of unmanned aerial methods to solve the problem of swarm coordination.
vehicles working together with the help of data
processing algorithms and artificial intelligence. “Swarm Intelligence: From Natural to Artificial
Systems” (2003) by Eric Bonabeau [1]. This book
Thanks to this approach, UAVs can operate much provides a broad overview of swarm algorithms,
more efficiently and accurately than traditional systems including algorithms for unmanned systems. It describes
with a single vehicle. They are capable of performing the basic concepts and ideas behind swarm systems and
many different tasks, including transporting goods, provides numerous examples of how swarm algorithms
identifying enemies, detecting and preventing natural can be applied to various problems.
disasters, and many others.
“Bio-inspired Artificial Intelligence: Theories,
The advantage of such a group of devices is stability: Methods, and Technologies” (2008) by Floreano and
if one of the devices is lost, the rest will continue to work Mattiussi [2]. This book considers swarm algorithms as a
without changing performance. The use of swarm form of bioinspired artificial intelligence. It describes
systems also helps to avoid the need for a single large various swarm algorithms, including algorithms based on
drone, which in turn reduces the chances of malfunctions animal behavior, and gives examples of how these
and improves repairability. algorithms can be applied to various problems.
- 242 -
“A Survey on Swarm Robotic Modeling, Analysis In a group of drones, collective behavior can occur, in
and Hardware Architecture” (2018) by G. Seeja, A. which the devices act in concert and cooperate to achieve
Arockia Selvakumar and V. Berlin Hency [3]. This article a common goal. This may include forming formations,
reviews the state of research in swarm robotic systems, coordinated movement, avoidance of obstacles, and
including unmanned vehicles. It provides an overview of collective decision making. Collective behavior can be
various swarm coordination algorithms and methods and achieved through various algorithms and strategies such
provides examples of the implementation of swarm as swarm algorithms, cooperative control algorithms, and
systems in various applications. distributed scheduling algorithms.
“Cooperative Control of Multi-Agent Systems: A Group drones are used in various fields, including
Consensus Region Approach” (2015) by Zhongkui Li civil aviation, defense, reconnaissance and monitoring,
and Zhisheng Duan [4]. This book describes methods for search and rescue, agriculture, and more. They can be
coordinating and controlling multi-agent systems, used for complex missions such as patrolling large areas,
including unmanned swarm systems. It presents various detecting and tracking targets, multi-point surveying, or
algorithms and strategies to achieve consistent behavior coordinating in group operations.
and cooperation between unmanned vehicles.
A group of drones presents some challenges, such as
Swarm algorithms use various coordination methods ensuring reliable communication, managing collective
such as task allocation, motion synchronization, and security, and coordinating in a dynamic and
pattern formation. Some well-known algorithms include unpredictable environment. Future research aims to
swarm behavior modeling, marker-assisted swarm develop more intelligent and adaptive methods for
algorithms, and algorithms using various types of coordinating, sharing information, and distributing tasks,
communication and information exchange between as well as integrating group drones with other
drones. technologies such as artificial intelligence and
autonomous decision-making systems.
Drone swarm algorithms find applications in various
fields, including civil aviation, defense, geodesy, III. FORMULATION OF THE PROBLEM
environmental monitoring, etc. They can be used to
perform complex tasks that require collective action, An aviation group of UAVs is considered, consisting
such as patrolling large areas, detecting and tracking of K drones. The actions of the group are synchronized
targets, aerial photogrammetry, and even forming (controlled) from the Mission Control Center (MCC).
temporary communication networks.
Each of the UAVs has its own characteristics that
Drone swarm algorithms still present some determine its capabilities in terms of range and flight
challenges, such as ensuring communication reliability, speed. The drone can only exchange information with
efficient resource allocation, and adapting to a dynamic the MCC. In addition, it is equipped with a CD, which
environment. Future research is aimed at developing can determine its own coordinates in real time and, on
more intelligent and adaptive swarm algorithms, taking their basis, solve problems about the reachability of a
into account ethical and legal aspects, as well as creating certain target by a drone, based on the coordinates of the
integrated mixed reality systems for drones to interact AT and its own technical capabilities.
with people.
The MCC simultaneously transmits to all drones the
A drone group, also known as a multi-agent system or coordinates of the air target and the time limit for
a drone collective, is a collection of two or more drones reaching it. Each of the UAVs, having received the
working together to complete a task or achieve a common coordinates of the target, solves the problem of the
goal. Group drones have the potential to solve complex possibility of reaching the AT. Then he informs the
tasks that are impossible or difficult to perform with one MCC about the possibility and the minimum time to
unmanned vehicle. achieve the goal.
A drone group requires effective coordination and Based on the information received from the UAV
communication among its members. Drones must group, the MCC determines the priority of the UAV to
exchange information, plan their movement, share tasks complete the task.
and coordinate their actions. To do this, various
communication methods are used, including wireless The mathematical model of the UAV can be
networks, data transmission and sensory information. described by the following system of ordinary
differential equations:
In a group of drones, each vehicle can perform its own
unique task or work together with other vehicles to 𝑉̇ = 𝑔(𝑛𝑥𝑎 − 𝑠𝑖𝑛Θ)
complete a complex task. The distribution of tasks Θ̇ = 𝑔(𝑛𝑦𝑎 𝑐𝑜𝑠𝛾 − 𝑐𝑜𝑠Θ)/𝑉
includes the definition of the roles and functions of each Ψ̇ = −𝑔𝑛𝑦𝑎 𝑠𝑖𝑛𝛾/(𝑉𝑐𝑜𝑠Θ) (1)
device, the optimal allocation of resources and the
𝑥̇ = 𝑉𝑐𝑜𝑠Θ𝑐𝑜𝑠Ψ
effective use of the capabilities of each device in 𝑦̇ = 𝑉𝑠𝑖𝑛Θ
accordance with the requirements of the task. { 𝑧̇ = −𝑉𝑐𝑜𝑠Θ𝑠𝑖𝑛Ψ
- 243 -
𝑃𝑐𝑜𝑠𝛼−𝑋𝑎 𝑃𝑠𝑖𝑛𝛼+𝑌𝑎 𝑞𝑖,1 (𝑇) = 𝑥ц , 𝑞𝑖,2 (𝑇) = 𝑦ц , 𝑞𝑖,3 (𝑇) = 𝑧ц ,
𝑛𝑥𝑎 = , 𝑛𝑦𝑎 = (2)
𝑚𝑔 𝑚𝑔
𝑞𝑇 = (𝑥ц , 𝑦ц , 𝑧ц ). (7)
Here x, y, z are the coordinates of the center of mass of
the aircraft in the normal earth coordinate system, V is
Each of the UAVs returns to the MCC the answer
the flight speed, ϴ is the angle of inclination of the
in the form of two numbers. The first number means
trajectory, Ψ is the heading angle, α is the angle of
the possibility of achieving the goal and takes two
attack, γ is the bank angle, P is the engine thrust, 𝑋𝑎 is
values: 0 – the goal cannot be achieved and 1 – the
aerodynamic drag, 𝑌𝑎 is aerodynamic lift, m – aircraft
goal is in the area of its attainability. The second
mass, g – gravitational acceleration, 𝑛𝑥𝑎 – longitudinal
number in the case of attainability of the AT is equal
g-load and 𝑛𝑦𝑎 – lateral g-load (in flow axes). to the minimum value of the time for which the goal
can be achieved.
As control variables in (1), overloads 𝑛𝑥𝑎 , 𝑛𝑦𝑎 and bank
angle γ are taken [5-7]. 3) The MCC generates a list of responses from all
drones in the form of a 2*K matrix R. The values of
Let us introduce the notation: the elements of the matrix R are defined as follows:
𝑥0 𝑥1 R(1,i) = 0 (if the AT cannot be reached by the i-th
𝑥 y0 y1 UAV) and R(1,i) = 1 (the target is achievable); R(2,i)
y z0 z1 = 𝑇𝑖 , where 𝑇𝑖 ≤ 𝑇 is the minimum time for which the
𝑞 = z , 𝑞0 = 𝑉 , 𝑞1 = 𝑉 (3) i-th drone reaches the target.
0 1
𝑉
[Ψ] Θ0 Θ1
[Ψ0 ] [Ψ0 ] Next, the MCC selects from the list R the number
of the drone {s} with the minimum time value 𝑇𝑖 ,
For each UAV, based on equations (1)-(2), we rewrite provided that the goal is reached. Mathematically, this
the mathematical model in the form of the following can be written as follows:
systems of differential equations:
𝑇𝑠 = 𝑚𝑖𝑛{ 𝑅(2, 𝑖)|𝑅(1, 𝑖) = 1, 𝑖 = 1, 𝑘} (8)
𝑑𝑞𝑖
= 𝑓𝑖 (𝑞𝑖 , 𝑡) + 𝐵𝑖 (𝑡)𝑢𝑖 , 𝑖 = ̅̅̅̅̅
1, 𝑘 (4)
𝑑𝑡 The task of reaching the AT is assigned to the drone
with the number {s}. The rest of the UAVs go into
where 𝑢𝑖 – control-vector of dimension m.
standby mode for further instructions from the MCC.
There are certain restrictions on control for each UAV
IV. THE CRITERION OF ATTAINABILITY OF THE I-TH
𝑢𝑖 (𝑡) ∈ 𝑈𝑖 = UAV OF THE TARGET
𝑢𝑖 (𝑡): 𝑢𝑖,𝑗 (𝑡) ∈ 𝐶[[𝑡0 , 𝑡1 ]; −𝐿𝑖,𝑗 ≤ 𝑢𝑖,𝑗 (𝑡) ≤ 𝐿𝑖,𝑗 ,
{ }.(5) Due to the properties of the equations of the Cauchy
𝑖 = ̅̅̅̅̅
1, 𝑘, 𝑗 = ̅̅̅̅̅̅
1, 𝑚, 𝑡 ∈ [𝑡0 , 𝑡1 ] problem (4)-(6) imposed on the right side of the system
of equations, for a fixed control 𝑢𝑖 (𝑡) ∈ 𝑈𝑖 , the
The problem considered in the article is reduced to conditions of the existence and uniqueness theorem for
the study of several problems. the solution 𝑞𝑖 (𝑡), 𝑡 ∈ [0, 𝑇] [8] are satisfied.
1) The MCC transmits to all drones the time limit Let us rewrite the Cauchy problem (4)-(6) in the
and the coordinates of the air target in the following integral recurrent form:
form:
T – in time (0, T) the target must be reached by at 𝑡 𝑡
𝑞𝑖 (𝑡) = 𝑞𝑖,0 + ∫0 𝑓𝑖 (𝑞𝑖 (𝜏), 𝜏)𝑑𝜏 + ∫0 𝐵𝑖 (𝑡)𝑢𝑖 (𝜏)𝑑𝜏 . (9)
least one UAV;
(𝑥ц , 𝑦ц , 𝑧ц ) – air target coordinates. Equation (9) is a special case of the Volterra-Urysohn
integral equation of the second kind [9]
Note. If it is impossible to reach the AT by all drones,
the MCC recognizes this fact and switches to solving 𝑡
𝑦(𝑡) = 𝑔(𝑡) + ∫0 𝐾(𝑡, 𝜏, 𝑦(𝜏))𝑑𝜏, (10)
other tasks.
2) all UAVs simultaneously solve the following where
reachability problem, which we will formulate for 𝑦(𝑡) = 𝑞𝑖 (𝑡);
𝑡
the i-th UAV:
𝑔(𝑡) = 𝑞𝑖,0 + ∫ 𝐵𝑖 𝑢𝑖 (𝜏)𝑑𝜏 ; 𝐾(𝑡, 𝜏, 𝑦(𝜏))
can a drone whose dynamics is described by 0
equations (4) with technical characteristics (5) and = 𝑓𝑖 (𝑦(𝜏), 𝜏).
with initial conditions:
To solve the problem, we apply the results of interval
𝑞𝑖 (0) = 𝑞𝑖,0 (6) analysis [10-11]. Denote the interval vector function
𝑡
in time less than or equal to T, reach the air target 𝒈(𝑡) = 𝑞𝑖,0 + ∫0 𝐵𝑖 (𝜏)𝒗𝒊 𝑑𝜏, 𝑡 ∈ [0, 𝑇]. (11)
- 244 -
Here the sign 𝒗𝒊 = (𝒗𝒊,𝟏 , 𝒗𝒊,𝟐 , … , 𝒗𝒊,𝒎 ) denotes an Proof. Let us rewrite the approximate solution (14) at
interval m-vector with interval elements 𝒗𝒊,𝒋 = (0, 𝐿𝑖,𝑗 ). time T in the form of the following interval integral
expression
Theorem 1. Let equation (4) have a unique continuous
solution for a given continuous function 𝑢𝑖 (𝑡) ∈ 𝐶[0, 𝑇] 𝒒𝑖,𝑟 (𝑇) = ∑𝑟𝑘=0 𝝋𝒌 (𝑇), (15)
satisfying condition (5). This solution can be found by
simple iteration (successive approximations). 𝝋𝟎 (𝑇) = 𝒈(𝑇); 𝝋𝒌+𝟏 (𝑇) = 𝒈(𝑇) +
𝑇
∫0 𝒇𝒊 (𝑡, 𝜏, 𝝋𝒌 (𝜏))𝑑𝜏.
Proof. The method of simple iteration as applied to the
integral equation consists in obtaining a sequence of where the sign * denotes interval, 𝒒𝑖,𝑟 (𝑇) – the interval
functions (approximations) 𝑞𝑖,𝑘 (𝑡), 𝑘 = 0,1,2 …, by boundary vector at time T and r is selected from the
means of the recurrence relation: 𝑀𝑘 𝑇 𝑘
condition 𝑁 {exp(𝑀𝑇) − ∑𝑟𝑘=0 =} ≤ 𝜀 (𝜀–
𝑘!
𝑡 required accuracy).
𝑞𝑖,𝑘+1 (𝑡) = 𝑞𝑖,0 + ∫0 𝐾(𝑡, 𝜏, 𝑞𝑖,𝑘 (𝜏))𝑑𝜏 +
𝑡
∫0 𝐵𝑖 (𝑡)𝑢𝑖 (𝜏)𝑑𝜏, 𝑡 ∈ [0, 𝑇], k=1, 2 … Hence, from the fact that the boundary vector 𝑞𝑇
belongs to the interval vector 𝒒(𝑇), it follows that the
Denote target is reachable by the i-th drone.
𝑡
𝑔(𝑡) = 𝑞𝑖,0 + ∫0 𝐵𝑖 (𝜏)𝑢𝑖 (𝜏)𝑑𝜏 , 𝜑0 (𝑡) = 𝑔(𝑡); This implies the validity of the assertion of the
theorem.
𝑡
𝜑𝑘+1 (𝑡) = 𝑔(𝑡) + ∫0 𝐾(𝑡, 𝜏, 𝜑𝑘 (𝜏))𝑑𝜏.
When theorem 2 is satisfied, i.e., if the target is
Then the desired solution can be represented as a reachable by the i–th UAV, the following procedure is
series proposed to calculate the minimum reachability time
Ti =0; tay = 0.01;
𝑞(𝑡) = ∑∞
𝑘=0 𝜑𝑘 , (12)
while (Ti<=T) do begin
which converges if 𝑓(𝑡) is continuous on [0,T] and Ti=T+tay;
𝐾(𝑡, 𝜏) is continuous on 0 ≤ 𝑡, 𝜏 ≤ 𝑇. the interval vector 𝒒𝒊,𝒓 (𝑇𝑖) is calculated.
if the vector 𝒒𝑻 belongs to the first three elements of
Denote 𝑀 = sup |𝐾(𝑡, 𝑠)|, 𝑁 = ‖𝑞𝑖,0 ‖ ∗ 𝑇 ∗ 𝐿 ∗ the interval vector 𝒒𝒊,𝒓 (𝑇𝑖), then the loop exits
0≤𝑡,𝑠≤𝑇 break;
‖𝐵𝑖 ‖. It is possible to evaluate the k-th approximation to end; {while}
the desired solution by means of the expression:
The value of Ti corresponds to the minimum time for
|𝜑𝑘 (𝑡)| ≤ 𝑁𝑀𝑘 𝑇 𝑘 /𝑘! (13) the i-th UAV to reach the target.
From (13) follows the factorial convergence of the DISCUSSION
method of successive approximations.
The article develops a method for selecting a target
As an approximate solution we can take by a group of unmanned aerial vehicles based on a swarm
approach using interval analysis and computer algebra.
𝑞𝑖,𝑟 (𝑡) = ∑𝑟𝑘=0 𝜑𝑘 (𝑡), 𝑡 ∈ [0, 𝑇], (14)
For a group of UAVs controlled from the MCC,
whose accuracy is estimated by the following expression: spatial coordinates and the required time to reach the
∞ target are specified. To carry out numerical calculations
in real time, the process for each UAV is parallelized, i.e.
|𝑞𝑖 (𝑡) − 𝑞𝑖,𝑟 (𝑡)| = | ∑ 𝜑𝑘 (𝑡)|
each of the UAVs independently solves the problem of
𝑘=𝑟+1
∞ attainability of the set goal, depending on its technical
𝑁𝑀𝑘 𝑇 𝑘 characteristics and its own location.
≤ ∑ = 𝑁{exp(𝑀𝑇)
𝑘!
𝑘=𝑟+1
𝑟 On the basis of interval mathematics, a criterion for
𝑘
𝑀 𝑇𝑘 the attainability of the goal by each UAV is obtained.
− ∑ =}
𝑘! Based on the swarm approach, a list of unmanned aerial
𝑘=0
vehicles is defined, each of which is capable of
This implies the uniform and absolute convergence of performing the task. This list may be empty if the selected
series (14) for any fixed control 𝑢𝑖 (𝑡) ∈ 𝑈𝑖 , 𝑡 ∈ [0, 𝑇]. target cannot be reached by any of the available UAVs.
Theorem is proven.
When developing the software, the library of interval
Theorem 2. The AT is reachable by the i-th UAV if mathematics was used.
the 3-dimensional vector 𝒒𝑻 , composed of the coordinates
of the target (7), belongs to the first three elements of the
interval vector 𝒒𝒊,𝒓 (𝑇).
- 245 -
The developed library of interval functions is
universal. To use it, it is enough to set the numerical
values of the parameters in the form of intervals.
The results obtained are applicable to the study of any
dynamical system described by ordinary differential
equations. Therefore, the application of the results of the
article has great prospects for automating the solution of
many problems of mathematical modeling.
The work was done at the expense of the Research
Institute of Mathematics and Mechanics at Al-Farabi
Kazakh National University.
REFERENCES
[1] E. Bonabeau, M. Dorigo, G. Theraulaz. “Swarm Intelligence: From
Natural to Artificial Systems”, O’Reilly Emerging Technology
Conference, Santa Clara, CA., pp.22-25, April 2003.
[2] D. Floreano, C. Mattiussi. Bio-inspired Artificial Intelligence:
Theories, Methods, and Technologies. – The MIT Press,
Cambridge, 2008.
[3] G. Seeja, A. Arockia Selvakumar, V. Berlin Hency “A Survey on
Swarm Robotic Modeling, Analysis and Hardware Architecture”
Procedia Computer Science 133 (2018). pp. 478–485.
[4] Zhongkui Li, Zhisheng Duan. Cooperative Control of Multi-Agent
Systems: A Consensus Region Approach. – CRC Press, 2015.
[5] N. Gyagenda, J.V. Hatilima, H. Roth, V. A. Zhmud. “Review of
GNSS-independent UAV navigation techniques” Robotics and
Autonomous Systems. Volume 152. Issue C, June 2022. DOI:
https://doi.org/10.1016/j.robot.2022.104069.
[6] A.N. Kvitko. “Methods for solving boundary value problems for
controlled systems of ordinary differential equations and their
application in solving problems of controlling the motion of the
center of mass of an aircraft” Avtoref. dissertation doctor. physics
and mathematics sciences, spec. 01.01.11 – System analysis and
automatic control, St. Petersburg, 1992. – 30 p.
[7] A. Mazakova, S. Jomartova, T. Mazakov, T. Shormanov, B.
Amirkhanov “Controllability of an unmanned aerial vehicle” IEEE
7th International Energy Conference (ENERGYCON), 2022, pp.
1-5, doi: 10.1109/ENERGYCON53164.2022.9830244.
[8] L.S. Pontryagin. Ordinary differential equations. – M.: Nauka,
1974. – 332 p.
[9] A.F. Verlan, V.S. Sizikov. Integral equations: methods, algorithms,
programs. – Kyiv: Naukova Dumka, 1986. – 543 p.
[10] G. Bairbekova, T. Mazakov, Sh. Djomartova, S. Nugmanova
“Interval arithmetic in calculations” Open Engineering formerly
Central European Journal of Engineering Editor-in-Chief: Noor,
Ahmed, Open Engineering. Volume 6, Issue 1. – pp.259-263, DOI:
https://doi.org/10.1515/eng-2016-0036, October 2016.
[11] T. Mazakov, W. Wójcik, Sh. Jomartova, N. Karymsakova, G.
Ziyatbekova, and A. Tursynbai. “The stability interval of the set of
linear system” Journal of electronics and telecommunications,
2021. – Vol. 67. – No. 2. – pp. 155-161. DOI:
10.24425/ijet.2021.135958
- 246 -
DEVELOPMENT OF AN ADAPTIVE DRIVE
FOR AEROSPACE ENGINEERING
Ivanov K.,1 Alipbaev K.,2* Tulekenova D.,3
1,2
Gumarbek Daukeyev Almaty University of Energy and Communication, Almaty, Kazakhstan,
3
Al-Farabi Kazakh National University, Almaty, Kazakhstan
Email: [email protected]
Abstract— The adaptive drive is a gear mechanism with simplifications. The paper provides a concise formulation of the
constant gearing and a variable gear ratio that operates without fundamental principles of force adaptation theory for gear
a control system. It is based on the scientific discovery of the drives and explores the dynamics of the adaptive drive for the
"Force adaptation effect in mechanics" [5], which states that a spacecraft docking mechanism.
planetary kinematic chain with two degrees of freedom, formed
by gear wheels in a movable closed circuit, can adapt to a Keywords— adaptive drive, force adaptation, mechanism with
variable load.Although the force adaptation conditions two degrees of freedom, closed loop, dynamics, friction coupling,
theoretically enable self-adjustment of the gearbox, tests of the coupling mechanism.
drive have revealed limited determinability of motion. Analysis
of the interaction between the mechanism's parameters has I. INTRODUCTION
shown that using actual movements based on the principle of A docking device serves as a controlled electromechanical
possible works does not provide equilibrium and system designed to establish a connection between a moving
determinability of the mechanism.The adaptive drive addresses spacecraft (SC) and an orbital station (OS) [1]. The docking
these challenges by simplifying the design and improving
apparatus comprises a docking mechanism (DM) and a
operational reliability, meeting the main requirements of gear
controlled electromechanical actuator. The coupling
drives in aerospace technology: small size and weight, control
automation, fail-safe operation, and durability. Actuators in
mechanism functions as a mechanical system facilitating the
space technology are utilized for transferring motion to interaction between the spacecraft and the orbital station. It
actuating mechanisms, as well as for orientation and motion adjusts to the continuously changing conditions of their
stabilization.The weight of an actuator depends on the motor interaction. The controlled electromechanical drive includes
power, which must be matched to the maximum load. Studies of an electric motor, gear differential gear, controlling
manipulator drives have demonstrated that the maximum load electromechanical devices, and a control system.
is associated with overcoming inertial forces. To reduce the
In order to significantly simplify the controlled
resistance moment experienced by the motor, a drive with a
variable transmission ratio is necessary. This work proposes an
electromechanical drive of the coupling mechanism, a
adaptive manipulator drive with a variable load-dependent gear proposal has been made to replace the toothed differential
ratio, achieved through an adaptive gear variator drive. This transfer mechanism with a revolutionary transfer mechanism
drive not only enables smooth adjustment of the transmission called the adaptive toothed variator.
ratio but also adapts to the variable drag torque. The toothed
variator offers a simple design, small size, and light weight, II. MATERIALS AND METHODS
which are crucial for space devices.The adaptive drive The motion modes of the drive are determined through
comprises an electric motor, a variable ratio gear train, and a dynamic analysis, which takes into account the relationship
control system. With its two degrees of freedom, the adaptive between the parameters of movement, such as forces and
transmission mechanism can drive an executive actuator at a velocities. The acting forces dictate the speeds of motion for
speed inversely proportional to the external load while the links, as represented in the plan of linear velocities [2].
maintaining constant motor power.
This implies that the drive will not experience a breakdown
even if there is interference or a complete stop in the movement
of the working body. The paper focuses on the development of
the theory of force adaptation based on motion determinacy,
aiming to create a reliable gear drive for aerospace engineering.
The docking mechanism serves as a mechanical system for
the interaction between spacecraft and orbital stations, adapting
to the continuously changing conditions of their interaction. The
existing docking mechanism drive is a complex mechanical Fig. 1. a)-b) adaptive toothed variator with linear velocity plan, c) link 5
system comprising an electric motor, a toothed multistage withapplied forces, d) Satellite joint 5 with balancing friction torque
transmission mechanism, and a control system [7,8]. Previously,
a simplified adaptive drive was developed, incorporating an Problem Statement for Dynamic Analysis: Given the
adaptive toothed variator as the transmission mechanism. This constant input parameters of the variator, including angular
adaptive drive simplifies the design and enhances operational velocity 𝜔𝐻1 and driving momentum 𝑀𝐻1 , the objective is to
reliability. determine the change in output velocity 𝜔𝐻2 based on the
given values of the variable output moment of resistance 𝑀𝐻2 .
The mathematical model developed for the existing central-
type docking mechanism can be utilized for the docking On the plan of linear velocities (Fig. 1b), the line 𝐴𝑏
mechanism with the adaptive drive by applying the specified representing the angular velocity 𝜔𝐻1 remains constant, while
- 247 -
the line 𝐴𝑘 indicating the output angular velocity 𝜔𝐻2 is characteristic 𝜔𝐻2 = 𝑀𝐻1 𝜔𝐻1 /𝑀𝐻2 at 𝑀𝐻2 >
determined by the position of the velocity vector 𝑉𝐻2 at point 𝑀𝐻1 , 𝜔𝐻2 < 𝜔𝐻1 . There is force adaptation to the variable
𝑘. The interaction between these angular velocities occurs load with instantaneous adaptation of the speed of the working
through the angular velocity 𝜔5 of the line 𝑃𝑘𝑠 of link 5. body to the driving conditions at constant motor parameters.
Point P is the point of intersection of the 𝑔𝑒 line of angular
On the velocity diagram, the variable force 𝑅Н2 moves the
velocity 𝜔5 of link 5 with the vertical zero line 𝐴𝐵. The point
point 𝑘 of the line 𝑃𝑘𝑠, which rotates around the point 𝑠 and
𝑠 of the intersection of the angular velocity lines 𝜔𝐻1 and 𝜔5
can occupy positions in the gearbox area between the start
determines the end of the velocity vector 𝑉𝑆 , the same for the
position 𝑠𝑛 and the end position 𝑠𝑚. In the end position, when
links 5 and 𝐻1 . The point 𝑠 occupies a constant position, and
the critical load is reached, the point 𝑘 moves to the point 𝐾
the position of the point 𝑘, which depends on 𝜔𝐻2 , is variable.
that represents the stopping of the output driver. A change in
The interaction between the kinematic and force the load causes the line to rotate clockwise or vice versa
parameters of the mechanism, according to the force accordingly. The pole P can be above or below the point
adaptation condition (9), can be observed as the interaction of depending on the load.
forces 𝐹𝐻1 , 𝑅Н2 and 𝑅05 on link 5 (Fig. 1c). This interaction
The multiplier mode at 𝑀𝐻2 < 𝑀𝐻1 , 𝜔𝐻2 > 𝜔𝐻1 -
on the velocity plan is depicted as a change in the position of
section BD of the diagram. At point D, there is 𝑀𝐻2 =
the point under the influence of the variable force 𝑅Н2 . In
𝑀𝐻2𝑚𝑖𝑛 . The pole P is located above point A. The line
other words, it involves the rotation of the line 𝑃𝑘𝑠 around the
𝜔5 rotates around the point S in the area to the right of the
stationary point 𝑠 where the constant force 𝐹𝐻1 is applied. The
initial position 𝑠𝑛.
interaction between these parameters follows the condition
∑ 𝑀𝑠 = 0 or 𝑅05 ⋅ 𝑃𝑆 = 𝑅𝐻2 ⋅ 𝐾𝑆 , which can be further 3) Stop mode at 𝑀𝐻2 = ∞, 𝜔𝐻2 = 0 - point C' of the
simplified to the condition . diagram 𝑀𝐻2 = 𝑀𝐻2𝑚𝑎𝑥 . The line passes through point K on
the velocity diagram 𝜔5 . The motor continues to run and the
𝐹′𝐻1 ⋅ 𝑃𝑆 = 𝑅𝐻2 ⋅ 𝑃𝐾 (1)
actuator enters the single-motion state.
0.5𝑀𝐻1 𝑟3 𝑟
After substituting the value 𝐹′𝐻1 = ( + 1) = III. DRAWING THE TRACTION CHARACTERISTICS OF ADAPTIVE
𝑟𝐻1 𝑟6 𝑟4
𝑟 𝑟
0.5𝐹𝐻1 ( 3 + 1) is transformed into a condition and reduced DRIVE
𝑟6 𝑟4
to the form Utilizing streamlined force interaction parameters,
simplified force interaction parameters will be used to
𝐹𝐻1 ⋅ 𝐴𝐵 = 𝑅𝐻2 ⋅ 𝐴𝑆 (2) construct the traction characteristic [5].
Here 𝐹𝐻1 ⋅ 𝑃𝑆 = 𝑉𝑆 /𝜔5, 𝐾𝑆 = 𝑃𝑆 − 𝑃𝐾 = (𝑉𝑆 − 𝑉𝐾 )/
𝜔5 . After substituting in (1) we obtain 𝐹𝐻1 ⋅ 𝑃𝑆 = 𝑅𝐻2 ⋅ 𝐾𝑆
Thus, the position of the line 𝑃𝑘𝑠 determines the mode of
motion of the mechanism. The mechanism can have the
following modes of motion, represented on the diagram 𝑀 =
𝑀(𝜔) , which characterizes the relationship of force and
kinematic parameters of the drive (Fig. 1a).
1) The beginning of motion (start) at switching on the
motor - the section 𝐴𝐵 of the diagram corresponding to the
characteristics of the induction electric motor М = М0 −
𝑏𝜔0 . Here М0 - the starting motor torque, 𝜔0 - starting angular Fig. 2. 1- Adaptive drive; 2- motor; 3- power supply with voltage regulator;
velocity, 𝑏is the starting coefficient. The start occurs at the 4-6- input and output angular velocity meters; 5- power supply.
increased power of the motor. The drive allows the start of the
motor together with the toothed variator. The characteristic of
the motor at the start coincides with the characteristic of the
output shaft of the drive taking into account the gear ratio of
1. The acceleration takes place with an increase in the output
angular velocity and some decrease in the output torque of
resistance M.
The point B at the end of the start determines the constant
motor parameters in the operating mode of motion 𝑀𝑒 , 𝜔𝑒 .
The section 𝐵𝐸 of the full motor characteristic is not used.
The variator has the same parameters.
At point B, there is a single-motion at 𝑀𝐻2 =
𝑀𝐻1 , 𝜔𝐻2 = 𝜔𝐻1 . The angular velocity line 𝑃𝑘𝑠 of link 5 is
at its original position , which coincides with the line 𝐴𝑠𝑏. The
pole P (instantaneous center of velocity of link 5) coincides
with point A.
2) Gearbox operating mode with overcoming variable
(including sharply varying) loads - Load increasing - section
Fig. 3. 3D Model of the Adaptive Drive
BC of the diagram corresponding to the adaptive variator
- 248 -
Initial data: 1) 𝑃 = 𝑀𝜔 – electric motor power, here 𝜔 =
𝜔𝐻1 – angular velocity of electric motor.
𝑀 = 𝑀𝐻1 – electric motor torque М = 2.5 𝑁𝑚 , 𝑃 =
𝑀𝜔 = 2.5 ⋅ 40 = 100 Nm/s.
2) 𝑀𝑖 = 𝑀𝐻2𝑖 𝑖 = 𝐴, 𝐵, 𝐶, 𝐶′, 𝐶′′, 𝐷, 𝐸 – output torque of
resistance.
1. Start. Section AB: Initial data: 𝑀𝑖 = 𝑀𝐻2𝑖 𝑖 = 𝐴, 𝐵,
𝑀𝐴 = 3𝑀 = 3 ⋅ 2.5 = 7.5 𝑁𝑚, 𝑀𝐵 = 𝑀 = 2.5𝑁𝑚 .
𝜂 = 0,90 – Motor efficiency at minimum load.
Formula for calculating the angular velocity in rad/s
𝑃𝜂 100×0.9 𝑟𝑎𝑑 Fig. 4. Diagram of angular velocity versus drag torque
𝜔𝑖 = 𝑃𝜂/𝑀𝑖 , 𝜔𝐴 = 0, 𝜔𝐵 = = = 36 .
𝑀𝐵 2.5 𝑠
IV. ANALYSIS OF DRIVING REGIMES
Initial point of operating mode B: 𝜔В = 36 𝑟𝑎𝑑/𝑠, МВ =
2,5 𝑁𝑚 The analysis of the driving modes and a comparison
between the adaptive drive and the existing drive yield the
2. Operating mode. following main results:
2.1 Increase in load. Section 𝑩𝑪𝑪′ . During the start phase (AB section), both the adaptive
drive and the existing drive exhibit a motor characteristic that
Initial data 𝑀𝑖 𝑖 = 𝐶, 𝐶′ , 𝑀𝐶 = 15 𝑁𝑚 , М𝐶′ =
aligns with the output shaft characteristic, taking into account
25 𝑁𝑚,𝑀𝑓𝑖 = 𝑓𝑀𝑖 is the friction torque applied to the output a transmission ratio of 1. Acceleration occurs as the output
shaft, angular velocity and the starting torque of resistance M
𝑓 = 0,2 - the average reduction factor, 𝑀𝑓𝐶 = 𝑓𝑀𝐶 = increase.
0.2 × 15 = 3 𝑁𝑚. In the operational mode with variable resistance (section
𝑀𝑓𝐶 ′ = 𝑓𝑀𝐶 ′ = 0.2 × 25 = 5 𝑁𝑚. BC), the existing drive requires changes in motor parameters
and M through its control system. The control system,
Output angular velocity consisting of a clutch, electromagnetic components, and
100
friction brakes, transmits signals to adjust the motor
𝜔𝑖 = 𝑃/(𝑀𝑖 + 𝑀𝑓𝑖 ) , 𝜔𝐶 = = 5.55 𝑟𝑎𝑑/𝑠 , 𝜔𝐶′ = parameters based on the variable drag torque M. The coupling
15+3
100 mechanism and the drive mechanism interact through their
= 3.33 𝑟𝑎𝑑/𝑠.
25+5 respective complex control systems.
2.2. Load reduction from the starting position. Plot BD. On the other hand, the adaptive drive operates without a
Initial data: 𝑀𝐷 = 1 𝑁𝑚 - moment of resistance control system. In the BC section, the variable drag torque is
autonomously overcome by the adaptive variator through self-
𝑀𝑓𝐷 = 𝑓𝑀𝐷 = 0.2 ⋅ 1 = 0.2 𝑁𝑚 - the friction torque regulation. Additionally, the adaptive variator can
applied to the output shaft. independently switch to the multiplier mode (BD section)
100 when the drag torque decreases, greatly expanding the control
Output angular velocity 𝜔𝐷 = = 83.3 𝑟𝑎𝑑/𝑠. range.
1+0.2
3. Stop mode of motion 𝐶 ′ - the point shifts to the position Consequently, the adaptive self-regulating drive,
′′
𝐶 - emergency stop when the motor is running 𝜔𝐶′′ = 0, operating without a control system, achieves full and
𝑀С′′ = 30 𝑁𝑚, instantaneous adequacy to the variable drag torque, ensuring
trouble-free operation. In the event of an emergency overload
4. Stop of the electric motor after switching off - the (point C'), the actuator's output shaft will stop while the motor
section 𝐵𝐸 is a section of the electric motor characteristic. continues running, causing the mechanism to enter a one-
Point B shifts to position E. 𝜔𝐸 = 40 𝑟𝑎𝑑/𝑠, 𝑀𝐸 = 0 𝑁𝑚. degree-of-freedom state. Furthermore, a motor operating
After the motor is switched off, point B passes to the origin without a control system exhibits higher efficiency, reducing
of coordinates O. the actuator's weight due to improved efficiency and the
absence of braking and acceleration mechanisms.
TABLE I. PARAMETERS OF VARIATOR OUTPUT SHAFT
Comparison based on design parameters:
Point А В С C’ C’’ D E
1. The adaptive drive utilizes a single motor,
Momentum 7. 2. 15 25 30 1 0 eliminating the need for a separate motor
of resistance, 5 5 for electromagnetic braking.
Nm
Angular 0 36 5.55 3.33 0 83.3 4 2. It achieves full adaptation and
velocity, 0 instantaneous and adequate response to load
rad/s changes without requiring a control system
or adjustments, thereby preventing
emergencies.
Key results of the analysis:
3. The motor control system is eliminated.
- 249 -
4. Reliability and quality of drive operation
are increased.
5. The design is simplified.
6. The drive's size and weight are reduced.
V. CONCLUSION
The developed mathematical model of the central type
existing docking mechanism can be effectively applied to a
docking mechanism equipped with an adaptive actuator,
utilizing the provided simplifications.
Structural modifications:
1. A comprehensive dynamic analysis of the
actuator's motion modes has been
conducted.
2. The results of the dynamic analysis affirm
the significant advantages of the designed
drive.
ACKNOWLEDGMENT
The research was supported by grant no. AP09258712
"Development of self-adjusting electric drives for
spacecrafts".
REFERENCES
[1] Crockett Samuel J. Shiftless, continuously-aligning transmission. Patent
of USA 4,932,928, Cl. F16H 47/08, U.S. Cl. 475/51; 475/47.1990, 9 p.
[2] Harries John. Power transmission system comprising two sets of
epicyclic gears. Patent of Great Britain GB2238090 (A). 1991, 11 p.
[3] Levitskij N.I. Teoriya mekhaniziov i mashin. Moskva. Nauka. 1979. рр.
576.
[4] Markeev A.P. Teoreticheskaya mekhanika. Moskva. Nauka. 1990. рр.
416.
[5] Ivanov K.S. Prospects of Creation of Mechanisms with Two Degree of
Freedom. IFToMM World Congress on Mechanism and Machine
Science. Advances in Mechanism and Machine Science. Springer
Nature Switzerland AG 2019. https://doi.org/10.1007/978-3-030-
20131-9_93. PP 937-946.
[6] Ivanov K.S., Tulekenova D.T., M. Ceccarelli. An Adaptive Drive of
Spacecraft Docking Mechanism // Mechanisms and Machine Science,
Meder2021. Springer. Vol 3. ISSN 2211-0992. 2021. pp. 168-178.
[7] Syromyatnikov V.S. Stykovochnye ustrojstva kosmicheskih apparatov //
Mashinostroenie, Moskva. 1984. рр. 216.
[8] Yаskevich A. V. Kinematicheskaya skhema stykovochnogo mekhanizma
tipa «shtyr'–konus» dlya perspektivnyh kosmicheskih korablej. №
4(19)/2017 KOSMICHESKAYA TEKHNIKA I TEKHNOLOGII. RF.
Moskva. 2017. рр. 95 – 104.
[9] Yаskevich A. V., Pavlov V. N., Shepelkin N. A. Stykovochnyj
mekhanizm kosmicheskogo apparata. Patent RF № RU 2706639.
19.11.2019. pp. 12.
[10] Yaskevich A. V. Matematicheskie modeli gisterezisa, opisyvayushchie
deformacii mekhanizmov dlya stykovki kosmicheskih apparatov.
Elektronnyj zhurnal «Trudy MAI». Vyp. 83. Rezhim dostupa:
http://mai.ru/science/trudy/index.php. 2015. pp.– 23.
[11] Golubev YU. F., YAskevich A.V. Komp'yuternoe modelirovanie
dinamiki stykovochnyh mekhanizmov central'nogo tipa dlya
kosmicheskih apparatov // Preprinty IPM im. M.V.Keldysha.
doi:10.20948/prepr-2019-89. URL:
http://library.keldysh.ru/preprint.asp?id=2019-89. № 89. 2019. pp. 40
- 250 -
Underwater robot propulsion system
Ivanov K.,1 Alipbayev K.1*, Saurova K.1
1
Almaty University of Power Engineering and Telecommunications, Kazakhstan
Email: [email protected]
Abstract— Research and development of uninhabited represent a new class of managed objects with distinct tasks,
underwater robotic systems and complexes have been completed practical applications, technological features, and system
[1]. The history of underwater robotics formation and compositions. Simultaneously, the systems incorporated into
development spans over 40 years, during which significant AUUVs and ship equipment differ significantly in their
changes have occurred in the means and methods of organizing, purposes and physical principles of operation, imposing strict
creating, and applying these systems and complexes. There are and often contradictory requirements on design technology
three types of uninhabited underwater vehicles (UUV): towed, and internal system organization. Expanding the functionality
cable-controlled, and autonomous. of AUUVs is also associated with addressing several new
Towed UUVs (TUUVs) are utilized for surveying large areas
theoretical challenges, including control, navigation,
when searching for sunken objects, as well as for various types
of oceanographic measurements. They are equipped with search
communication, vehicle orientation, environmental data
equipment and meters to measure physical and chemical collection, and storage, as well as ensuring device safety in
parameters of the water environment. These devices are normal and exceptional circumstances.
connected to the towing vessel by a cable that transmits Noteworthy advancements in recent years have been made
electricity and information. As the depth of immersion increases possible through corporate collaboration and the involvement
and the tether cable becomes longer, operational difficulties for of specialized firms in equipping underwater vehicles with
the "ship unit" complex are significantly amplified, resulting in
industrial systems based on international standards,
increased mass and dimensions of the marine equipment.
Consequently, this leads to an increase in the displacement and
measurement instruments, and scientific equipment.
operational costs of the towing vessel. However, towed systems Currently, a considerable body of methodological and
are more efficient in obtaining essential information, which is practical experience has been accumulated worldwide in the
crucial for search operations. creation and utilization of underwater robotic systems to
address research and applied challenges across various
Keywords— remote-controlled tethered, underwater industries. Progress in this field lies in the development of
robotic systems, autonomous underwater vehicles,
more advanced systems, technologies, and multifunctional
multifunctional robotic complex, modular technology.
complexes that can tackle a wide range of tasks in complex
underwater environments. Consequently, UUVs have become
I. INTRODUCTION
specialized: search towed, remote-controlled working with
Remote-controlled tethered UUVs (RCTUUVs) are cables, and autonomous research vehicles. Therefore, it is
typically employed as operational devices designed for lifting widely believed that a complex comprising devices of these
sunken objects and performing various tasks underwater. three types is optimal for addressing a broad spectrum of tasks.
They can also be utilized for surveying objects and localized Moreover, many experts express the opinion that such devices
areas of the seabed. RCTUUVs are equipped with different should generally be multifunctional, relying on modular
actuators depending on the nature of the task. Manipulators, technology, interchangeable equipment, and the ability to
which mimic the motor functions of human hands, are the undergo functional restructuring.
most commonly used and versatile components.
Phototelevision installations enable object inspection, bottom The use of new technologies has significantly expanded
relief analysis, documentation of situations, and manipulator the scope of AUUVs. Currently, numerous projects and
control. Hydroacoustic stations are typically employed to existing developments focus not only on previously
guide RCUUVs to the designated work area. These devices established relatively affordable applications but also on
possess a high level of maneuverability but have limited entirely new endeavors in oceanography, servicing various
range. In recent years, the use of fiber-optic cables has industries, and military programs.
significantly expanded the operational capabilities of The development of priority areas in underwater robotics
RCTUUVs, particularly in inspection work and the involves the advancement of system architectures, models,
examination of various underwater structures. and methods for intelligent control and spatial orientation,
Initially, autonomous vehicles (AUUVs) were primarily ensuring the safety and survivability of autonomous
utilized for military purposes, such as mine search, underwater robots in uncertain and extreme environmental
identification, and destruction, simulation of underwater conditions. Achieving optimal navigation and control tasks
targets, homing weapons, and ice research. Subsequently, through integrated processing of all available information
AUUVs began to be predominantly used as information- about the environment and the underwater robot's condition is
gathering robots for research purposes in the exploration of of utmost importance. Such tasks encompass reaching a
the World Ocean. specific point or spatial area, real-time analysis of scenes, and
decision-making under complex and extreme environmental
The practice of creating and implementing UUVs conditions. An urgent task in the intellectualization of AUUV
demonstrates that AUUVs occupy a prominent position within control systems is the construction of comprehensive
the multifunctional robotic complex. Modern AUUVs modeling complexes capable of generating a virtual
- 251 -
environment, visualizing the device's missions, and robots, their systems, and complexes. Currently, the institute
supporting the operation of sensor devices in simulation is conducting research on the topic: "Research on priority
mode. Additionally, addressing the more complex problem of directions for creating underwater robotic systems for
3D reconstruction of the underwater environment, crucial for automated survey of the water environment and seabed
real-time spatial route planning using the vehicle's own topography, marine infrastructures, and physical fields of the
intelligence, is also intended. A problem-oriented ocean." This research is part of a scientific research program
geoformation system designed for interactive display of the funded by the budget, and additional research and
underwater environment and robot control is a necessary development work is carried out based on state tasks for
component of such tools. various departments.
In addressing the aforementioned general problems, there To propel underwater robots efficiently, optimal
is a notable trend towards integrating the functional properties propellers that minimize energy consumption should be
of autonomous and remote-controlled underwater robots to employed. The most common engine design is in the form of
create versatile groups of information-interacting devices that a propeller, which differs fundamentally from the natural
are relatively simple, reliable, and efficient. Some of these propulsion used by fish—tail propellers. Fish tails perform
problems are already being solved at the current level of reciprocating movements and can achieve high speeds (up to
understanding, while others will be resolved in the near future. 80 km/h for dolphins). Currently, underwater robot designs
For instance, the development of small autonomous vehicles have been created that simulate movement using a fish tail
with energy-intensive and renewable power sources enables (robot fish). However, these structures only provide forced
the implementation of an automated network for oscillatory motion for the main link, while the remaining links
oceanographic measurements and underwater situational have free movement.
awareness across the vast expanses of the world's oceans.
Similar advancements can be achieved in the creation of high- To create a water engine, a multi-link lever mechanism of
precision integrated underwater navigation systems based on a vibrating machine can be utilized [1]. This mechanism can
autonomous, sonar, and satellite technologies. Many modern have a spatial structure with forced movement of all links,
autonomous vehicles already possess navigation systems that ensuring high motor efficiency. This work focuses on
allow for location correction during tasks that require describing and researching the design of a three-dimensional
increased range and navigation accuracy. propulsion system for a water manipulator with forced
movement of all links. The report presents a fundamentally
The advancement of the robotic complex is directly new lever mechanism for the water engine and conducts a
dependent on the technological foundation that can meet the dynamic analysis of its operation.
ever-growing requirements for tactical, technical, and
functional characteristics of devices and their systems. In this
regard, the existing experience in modular technology is II. DESIGN OF A VOLUMETRIC WATER PROPULSION SYSTEM
invaluable, but it also necessitates proper analysis and
generalization, taking into account global trends. Several The volumetric water propulsion system is depicted in the
pressing problems are associated with the implementation of top view (Fig. 1). It consists of a base (rack) 0, an input crank
ongoing and innovative projects, including: 1 with a motor, and an output structural group 2-3-4-5-6-7-8-
9-10-11, forming a closed loop. The links 3, 5, 7, 9, and 11 are
− Introduction of new system technologies based on equipped with vertical rectangular blades, which constitute the
unified elements that enable the construction of different working surface of the engine interacting with the water
hardware and software configurations, allowing for structure environment. The kinematic chain of this system comprises
reconfiguration based on tasks and actual working conditions. n=11 moving links and p=16 rotational kinematic pairs,
− Establishment of a metrological experimental base resulting in a total number of degrees of freedom, denoted as
for the development, testing, and pilot operation of the created W.
equipment.
T=3n-2p=3x11-2x16=1 (1)
An important aspect of modern AUUV development is
related to the miniaturization of electronic systems and
devices in general. Microelectronic structures and This indicates that the kinematic chain is capable of
technologies are widely utilized in the architecture of compact motion and functions as a mechanism. Each component of the
devices and their systems. This issue is particularly relevant in mechanism executes well-defined predetermined movements.
the mass production of small devices intended for operation in When in operation, the input crank 1 imparts motion to the
shallow depths. While miniaturization of individual electronic connected structural group, resulting in reciprocating
devices (microprocessors, sensors, actuators) is relevant for all movements. In Fig. 1, the contour lines represent the initial
types of vehicles, there are limitations to miniaturization when position of the mechanism, while the dotted lines represent
creating deep-sea vehicles due to specific tactical and intermediate positions. The intermediate positions of the
technical requirements. However, significant progress can be mechanism exhibit symmetry relative to the centerline.
observed in the creation of specialized underwater robots
designed to fulfill a limited range of tasks. The autonomous
underwater robots developed in recent years can already serve
as prototypes for the next generation of "intelligent" devices.
The Institute of Marine Technology Problems (IMTP) at
FAB RAS, the leading organization in Russia in the field of
underwater robotics, has gained unique experience in the
creation and practical use of various types of underwater Fig. 1 - Volumetric propulsion of the underwater robot
- 252 -
𝐽10 = 0, 𝐽11 = 0, 𝐽9 ≠ 0,
It is indeed possible to analyze the kinematics of the
1 (−𝐽2 𝜔22 − 𝐽3 𝜔32 − 𝐽4 𝜔42 − 𝐽5 𝜔52 − 𝐽6 𝜔62 − 𝐽7 𝜔72 − 𝐽8 𝜔82 −
mechanism by creating velocity and acceleration profiles. 𝜔9 = − (√
−𝑚2 𝑣22 − 𝑚3 𝑣32 − 𝑚4 𝑣42 − 𝑚5 𝑣52 − 𝑚6 𝑣62 −𝑚7 𝑣72 − 𝑚8 𝑣82 −𝑚9 𝑣92 )
)
√𝐽9
Regarding the dynamics of the water propulsion
mechanism, a velocity profile can be constructed for an
extreme position. This profile can then be used to calculate Roots for the variable 𝑣2
the kinetic energy using the formula:
1 (−𝐽 𝜔2 − 𝐽3 𝜔32 − 𝐽4 𝜔42 − 𝐽5 𝜔52 − 𝐽6 𝜔62 − 𝐽7 𝜔72 − 𝐽8 𝜔82 − 𝐽9 𝜔92 − 𝐽10 𝜔10
2 2
− 𝐽11 𝜔11
𝑣2 = (√ 2 2 )
−𝑚3 𝑣32 − 𝑚4 𝑣42 − 𝑚5 𝑣52 − 𝑚6 𝑣62 −𝑚7 𝑣72 − 𝑚8 𝑣82 −𝑚9 𝑣92 )
√ 𝑚2
T = 0.5 J i i2 + 0.5 miVi 2 (2)
i = 2...11 i = 2,3,5,7,8,9
Derivative:
Here 𝐽- link moment of inertia, 𝜔 - link angular velocity, 11 9
𝜕 1 1
𝑚 - link mass, 𝑉- link center of mass velocity ( ∑ 𝐽𝑖 𝜔𝑖2 + ∑ 𝑚𝑖 𝑣𝑖2) =
𝜕𝑣2 2 2
𝑖=2 𝑖=2
1
𝜕( (𝐽2𝜔22 + 𝐽3𝜔32 + 𝐽4 𝜔42 + 𝐽5𝜔52 + 𝐽6𝜔62 + 𝐽7𝜔72 + 𝐽8𝜔82 + 𝐽9 𝜔92 + 𝐽10𝜔10
2 2 )
+ 𝐽11𝜔11
∑𝑖=2…11 𝐽𝑖 𝜔𝑖 + ∑𝑖=2…9 𝑚𝑖 𝑣𝑖 (3) 2
1 2
+ (𝑚2𝑣22 + 𝑚3 𝑣32 + 𝑚4 𝑣42 + 𝑚5𝑣5 + 𝑚6 ))𝑣62+𝑚7 𝑣72 + 𝑚8 𝑣82+𝑚9 𝑣92))
2
𝜕𝑣2
Input interpretation:
11 9 Initial kinetic energy:
1 1
𝑇 = ∑ 𝐽𝑖 𝜔𝑖2 + ∑ 𝑚𝑖 𝑣𝑖2
2 2 𝑇0 = 0; ∆𝑇 = 𝑇 − 𝑇0
𝑖=2 𝑖=2
Result: From here:
1 ∆𝑇
𝑇 = (𝐽2 𝜔22 + 𝐽3 𝜔32 + 𝐽4 𝜔42 + 𝐽5 𝜔52 + 𝐽6 𝜔62 + 𝐽7 𝜔72 = 𝑀1 𝜔1 − 𝑀𝑁 𝜔𝑁
2 𝑡
+ 𝐽8 𝜔82 + 𝐽9 𝜔92 + 𝐽10 𝜔10
2 2 )
+ 𝐽11 𝜔11 +
1 The formula of the forces of adaptation:
2 2 2 2 2 2
(𝑚 𝑣 + 𝑚3 𝑣3 + 𝑚4 𝑣4 + 𝑚5 𝑣5 + 𝑚6 𝑣6 +𝑚7 𝑣7
2 2 2
+ 𝑚8 𝑣82 +𝑚9 𝑣92 ) ∆𝑇
− + 𝑀1 𝜔1
𝜔𝑁 = 𝑡
Alternate form: 𝑀𝑁
1 1 1 1 1 1 Then
− 𝐽2 𝜔22 − 𝐽3 𝜔32 − 𝐽4 𝜔42 − 𝐽5 𝜔52 − 𝐽6 𝜔62 − 𝐽7 𝜔72 −
2 2 2 2 2
2
1 1 1 1 1 (I = 2 … 11) J
𝐽8 𝜔82 − 𝐽9 𝜔92 − 𝐽10 𝜔10 2 2
− 𝐽11 𝜔11 − 𝑚2 𝑣22 −
2 2 2 2 2 𝑀
1 1 1 1 1 𝑀 = 𝐽𝜀; J =
𝑚3 𝑣32 − 𝑚4 𝑣42 − 𝑚5 𝑣52 − 𝑚6 𝑣62 − 𝑚7 𝑣72 − 𝜀
2 2 2 2 2
1 1
𝑚8 𝑣82 − 𝑚9 𝑣92 + 𝑇 = 0 Result of the drive:
2 2
J (I={2,3,4,5,6,7,8,9,10,11})
Expanded form:
Using closest interpretation:
1 1 1 1 1
𝑇 = 𝐽2 𝜔22 + 𝐽3 𝜔32 + 𝐽4 𝜔42 + 𝐽5 𝜔52 + 𝐽6 𝜔62 + (I = 2 … 11) = M
2 2 2 2 2
1 1 1 1 1
𝐽7 𝜔72 + 𝐽8 𝜔82 + 𝐽9 𝜔92 + 𝐽10 𝜔10 2 2
+ 𝐽11 𝜔11 + More interpretations:
2 2 2 2 2
1 2 1 2 1 2 1 2
𝑚 𝑣 + 𝑚3 𝑣3 + 𝑚4 𝑣4 + 𝑚5 𝑣5 + (I = 2 … 11) J; (I = 2 … 9) m
2 2 2 2 2 2
1 1 1 1
𝑚6 𝑣62 + 𝑚7 𝑣72 + 𝑚8 𝑣82 + 𝑚9 𝑣92 From the diagram (Fig. 2), it is evident that each
2 2 2 2
component of the mechanism executes precisely determined
Roots: forced movements.
𝐽11 ≠ 0,
(−𝐽2 𝜔22 − 𝐽3 𝜔32 − 𝐽4 𝜔42 − 𝐽5 𝜔52 − 𝐽6 𝜔62 −
1
𝜔11 = − (√ 𝐽7 𝜔72 − 𝐽8 𝜔82 − 𝐽9 𝜔92 − 𝐽10 𝜔10
2
)
√𝐽11 −𝑚2 𝑣22 − 𝑚3 𝑣32 − 𝑚4 𝑣42 − 𝑚5 𝑣52 − 𝑚6 𝑣62 −𝑚7 𝑣72 − 𝑚8 𝑣82 −𝑚9 𝑣92 )
𝐽11 = 0, 𝐽10 ≠ 0,
(−𝐽2 𝜔22 − 𝐽3 𝜔32 − 𝐽4 𝜔42 − 𝐽5 𝜔52 − 𝐽6 𝜔62 −
1
𝜔10 = − (√ 𝐽7 𝜔72 − 𝐽8 𝜔82 − 𝐽9 𝜔92 − )
√𝐽10 −𝑚2 𝑣22 − 𝑚3 𝑣32 − 𝑚4 𝑣42 − 𝑚5 𝑣52 − 𝑚6 𝑣62 −𝑚7 𝑣72 − 𝑚8 𝑣82 −𝑚9 𝑣92 )
Fig. 2 - The percentage ratio of the mechanism of each link
- 253 -
Additionally, the diagram (Fig. 3) illustrates the operating
time of the input crank 1, which undergoes reciprocating
movements. The graph demonstrates the movement of the
mechanism in response to the applied forces.
Fig. 3 - The graph of the movement of the mechanism under
the action of specified forces
The kinetic energy value obtained can be utilized for a
dynamic analysis of the mechanism using the Lagrange
equation. This equation enables the calculation of forces and
accelerations within the system.
Alternatively, to achieve the desired uneven motion, the
moment of inertia of a flywheel can be determined. The
flywheel's inertia can be adjusted to introduce variations in
the motion pattern as required.
CONCLUSION
In conclusion, the developed water robot propeller
mechanism has the potential to enhance the efficiency of its
movement. By incorporating this mechanism, the water robot
can achieve improved propulsion capabilities, leading to
increased efficiency in its overall performance.
ACKNOWLEDGMENT
The research was supported by grant no. AP09258712
"Development of self-adjusting electric drives
for spacecrafts".
REFERENCES
[1] Inzartsev A.V., Kiselev L.V., Kostenko V.V. etc. Underwater robotic
systems and complexes: systems, technologies, application. IMTP
FAB RAS. Vladivostok. 2018.368 p.
[2] Ivanov K.S. Vibrating machine. USSR author's certificate No. 713607.
02/05/1980.
- 254 -
Research and design of an efficient adaptive
drive with balancing friction coupling
Konstantin Ivanov, Kazakhstan ([email protected]), Aidos Sultan, Kazakhstan ([email protected])*, Kuanysh Alipbayev,
Kazakhstan ([email protected]), Komekbayev Arman, Kazakhstan ([email protected])
Almaty University of Power Engineering and Telecommunications named after Gumarbek Daukeyev, Almaty, 050013,
Kazakhstan
Corresponding author: [email protected] (Aidos Sultan)
Abstract— Contemporary research efforts in the scientific Research has demonstrated that multi-speed
and technical domain prioritize the automation of processes, transmissions can decrease a vehicle's total energy
systems, and devices, with notable applications in fields such as consumption, leading to extended travel range with smaller
aerospace, robotics, and electric transportation technology. battery packs, heightened dynamic performance, and
Adaptive systems provide practical solutions to the technical enhanced climbing capabilities. However, the efficacy of
and design complexities inherent in automating technological traditional multi-stage gearboxes typically falls short
tools and processes. The development of an efficient and compared to their single-stage counterparts [6].
dependable transmission system for electric autonomous and
mobile vehicles is possible through the application of an Recent studies have shown that multi-ratio systems can
adaptive mechanical drive with two degrees of freedom. The reduce energy consumption by 2 to 20% for various driving
proposed innovative transmission system, a continuous multi- cycles [5], [6], [7] and increase drive wheel torque by 35%
speed drive featuring an auxiliary friction coupling, facilitates [8], although this results in to a decrease in efficiency.
the adaptation of this two-mobile system. This unique
adaptation ability of the two-mobile self-regulating mechanical For heavy-duty vehicles that, due to their mass, need
drive allows it to independently adjust to external load via a very high torque for acceleration, multi-speed gearboxes
balancing friction coupling that connects the frictional moment may be more necessary so that a machine with lower torque
to the relative angular velocity. and less tractive effort can be used [9].
This paper explores the interplay of parameters within a The results of «Development of a Multi-Speed Dual
two-mobile system to enhance the efficiency of the self- Clutch CVT for Electric Vehicles» show that electric
adjusting adaptive drive across a broad range of regulation, vehicles can gain significant range advantages from multi-
utilizing friction coupling. speed transmissions. CVT shows the greatest potential and
is one of the most promising transmissions for electric
Keywords— Self-adjusting adaptive drive; additional vehicles [10]. However, the use of a double clutch or other
constraint; friction coupling; friction moment; regulatory auxiliary control devices complicates the design of the drive
efficiency. and reduces efficiency.
I. INTRODUCTION It was noted in [11] that adaptive transmission control
can be used to further improve the adaptability to working
Modern automatically controlled drives are constructed
conditions and the economy of the vehicle. However, this
through the refinement of multi-stage Continuously Variable
also leads to a more complex design.
Transmission (CVT) designs. As mechatronics technology
advances, a variety of novel drives are being incorporated The paper [12] proposes a dual motor configuration in
into drive systems to boost transmission efficiency. [1]. which two smaller motors are connected via a planetary
There is a growing trend towards decreasing the number of gear, as opposed to the standard configuration which uses a
stages in systems while preserving an extensive control single larger motor directly connected to the drive wheels
range. In efforts to enhance the efficiency of drive systems with a fixed ratio reduction gear. The main disadvantage of
for mobile robots and electric vehicles, the application of this configuration is the problem of additional unsprung
multi-speed transmissions is being extensively examined. mass due to the twin-motor design, which degrades the
[2]. dynamic performance of the electric vehicle. Also, the
investigated multi-speed twin-engine transmission is not
The utilization of multi-speed gearboxes in the
suitable for driving in unprepared terrain, since this type of
transmissions of battery electric vehicles is under active
configuration was only effective at low speeds and high
research. Their multi-ratio configuration helps keep the
torques. The dual motor drive system requires the use of a
electric motor functioning within a higher efficiency region.
control, which also complicates the design.
This results in decreased energy usage by the vehicle and
improved dynamic performance. [3]. The paper [13] proposes a two-speed automatic manual
transmission with a selectable one-way clutch. The
The use of a multi-stage gearbox increases the cost of
developed drive consists of a high-speed motor and a
producing an electric vehicle and requires the introduction
mechanical switching drive, and can carry out switching
of a gearbox control [4] and the development of a gear shift
without interrupting the energy flow. The main disadvantage
algorithm [5].
of the studies performed is that the automatically controlled
- 255 -
drive is built on the basis of the use of single-moving
mechanical elements connected by a control system. As a
result, a most complex controlled electromechanical system
is created, which has a low adequacy to a changing external
load and low efficiency.
On the above grounds, to create an efficient, compact
system and an automatically controlled drive, it is proposed
to use a previously developed adaptive mechanical system
with two degrees of freedom without a control system [14].
The self-adjusting adaptive drive [15] is fundamentally
different from all the drives considered above in that it Fig. 1. Adaptive drive
provides effective overcoming of the resistance torque in a
given wide range due to the use of two degrees of freedom.
The adaptive gear variator's kinematic chain, as depicted
The controllability of a bi-mobile system's motion with in Fig. 2a, comprises a rack 0, an input carrier H_1, a
a single input is accomplished by the introduction of cyclical gear assembly of 1-2-3-6-5-4, and an output carrier
an additional constraint that facilitates the replacement H_2. The gears 1, 2, 3, 6, 5, 4 shape a flexible closed loop
of the arising unbalanced force with a computed with a dynamic linkage. Additional gears, 8 and 7, constitute
resistance moment. This supplementary constraint is an extra parallel transmission originating from carrier H_2
realized by employing a friction joint which provides a to satellite 5. This closed loop encompasses an input satellite
compensatory resistance moment. Nevertheless, the 2, an array of sun wheels labeled 1-4, a cluster of ring
mechanical converter, characterized as a friction joint wheels denoted as 3-6, and an output satellite 5.Drive works
in the investigated scheme, does not serve as a as follows: the input shaft with the input carrier transmits
structural element due to its substantial geometric the movement to the input satellite 2. Input satellite 2
transmits motion to gear blocks 1-4 and 3-6. Wheel blocks
parameters (trunnion radius) which are inconsistent 1-4 and 3-6 transmit the movement to the output satellite 5
with the scheme. and the output carrier . At the same time, a kinematic
This article explores the interaction of parameters within chain with an additional parallel transmission of gears 8-7
a two-mobile system with the objective of enhancing the transmits the movement from the input carrier to the
efficiency of a self-regulating adaptive drive, given a broad output satellite 5 and the carrier . Parallel gear 8-7 has a
range of regulation, based on the utilization of a balancing gear ratio equal to the gear ratio of the planetary kinematic
friction coupling. chain.
The subsequent structure of this article is arranged as
follows: Section 2 details the kinematics of the adaptive
drive. Section 3 introduces an optimization algorithm for
determining the optimal kinematic and power parameters of
an adaptive drive. In section 4, a numerical computation of
the kinematic and power parameters of an adaptive drive is
executed, culminating in the determination of efficiency.
Section 5 reviews the findings obtained. Section 6 offers the
principal conclusions. Lastly, Section 7 discloses the
funding sources.
II. DESCRIPTION OF THE KINEMATIC CHAIN OF THE ADAPTIVE
DRIVE
An adaptive drive, also referred to as AD, is a
mechanism that self-modifies as depicted in Fig. 1. This Fig. 2. Adaptive drive (a), plan of linear velocities of
drive embodies the principle of power adaptation, where the drive (b), determination of relative angular velocity (c)
speed of the output link correlates inversely with the
fluctuating output load. Though it possesses two degrees of Parallel gear 8-7 can create an effective force-velocity
freedom and a single input, it seemingly conflicts with the constraint. In the present work, the creation of force-
prerequisites for a mechanism's existence and its motion's velocity constraint by means of an adjustable friction
definability. The underlying assumption is that the precise coupling 10 connecting wheels 5 and 7 using an adjustable
definability of motion provides the essential and adequate tension screw 9 is given. The tensioning screw 9 creates a
conditions for power adaptation. [15]. A necessary condition normal reaction and frictional moment in the coupling 10
is provided by a movable closed loop made up of gears. The required for equilibration.
sufficient condition provides an additional constraint
between force and velocity. Figure 2c shows a simplified plan of linear velocities for
easier visual perception.
Figure 3 shows a structural description of a 3D model of
an adaptive drive.
- 256 -
– radius of the joint.
– the relative angular velocity of links 7 and 5.
The sufficient condition for power adaptation is
represented by the power balance equation with the addition
of power consumed by friction.
The power balance for a kinematic chain withadditional
frictional parallel constraint is as follows:
(4)
Force adaptation Eq.:
Fig. 3. Structural description of a 3D model of an
adaptive drive (5)
III. KINEMATIC ANALYSIS OF ADAPTIVE DRIVE
The interaction of kinematic and force parameters will The required friction torque can be pre-determined using
be represented in the form of a power balance and the Eq. with further refinement.
performed according to the principle of virtual works using
actual displacements. Eq. (5) eventually determines the mathematical model of
the adaptive drive when the necessary and sufficient
The interaction parameters are: external forces and condition for force adaptation is met.
(corresponding to moments and ) acting on the
closed loop at points and , angular velocities The values of angular velocities and are
of the initial links , and linear velocities of points determined as follows:
and (Fig. 2b, 2c). 1) For planetary gear:
The previously developed equations of interaction
between the kinematic and power parameters [15] define a (6)
mathematical model of the adaptive drive without additional
constraint. 2) For an additional gear:
The necessary condition for adaptation is provided by a
movable closed loop composed of gear wheels. The closed (7)
loop equilibrium condition (without gear 8-7) is:
The relative angular velocity is determined by the
(1) Eq.:
Eq. (1) allows us to obtain the necessary power (8)
adaptation condition, taking into account the negative sign
of the output power: The efficiency of the mechanism is determined by the
Eq.:
(2)
% (9)
The effect of power adaptation is as follows: for a given
constant input power and a given output drag IV. NUMERICAL CALCULATION OF THE MATHEMATICAL
torque , the output angular velocity is in inverse MODEL OF THE ADAPTIVE DRIVE
proportion to the variable output drag torque .
The necessary condition for adaptation expresses the Computational evaluation of the drive's mathematical
model allows us to effortlessly and promptly validate the
theoretical possibility of creating an adaptation. The real
proposed theoretical hypotheses. Let's examine the drive's
adaptation can only take place if an additional sufficient
performance through the numerical investigation of the
adaptation condition is met. created mathematical models:
A sufficient condition provides an additional constraint
of force to velocity. The additional constraint is expressed 1) AD-1 model, which aligns with the essential
by the Eq. of power consumption for friction in the friction and sufficient condition for power adaptation [15].
coupling 10: 2) AD-2 model corresponding to the
approximate-sufficient condition of power adaptation,
. (3) where .
3) AD-3 model corresponding to the real-
sufficient condition of power adaptation, where
Here – is the required frictional torque , .
– the normal reaction created by the tension screw 9,
– coefficient of friction,
- 257 -
As an example, let's define the parameter values for one .
of the calculation options.
Initial data: Let’s determine the value of the efficiency of the
The adaptive drive has the given constant motor power mechanism according to the following Eq.:
parameters on the input link and the given
value of variable output torque resistance on the output %= .
link (figure 2a).
It is required to determine the power and kinematic According to this solution algorithm, we determine the
parameters of the drive. remaining parameters that determine the sufficient condition
Model №1. of power adaptation at the corresponding values of and
from Table 1. The obtained data are summarized in
Given: = 100 s-1, = 10 Nm, Table 2.
= [10, 15, 20, 25, 30, 35, 40, 45, 50] Nm,
TABLE II. RESULTS OF CALCULATIONS FOR AD-1 WITH A
= 4 Nm, SUFFICIENT CONDITION OF POWER ADAPTATION
= 40; = 16; = 72; = 16; = 40; = 96. Avera
% ge
Determine:
%
and . 10 100, 100 0 0 100 10
Solution: 0 0
15 66,7 13,4 - 213, 52,5 78,
Using Eq. (2), calculate the output angular velocity .
53,3 2 7
The obtained data are summarized in Table 1. 20 50,0 - - 319, 34,0 68,
29,9 79,9 6 0
TABLE I. VALUES OF OUTPUT DRAG TORQUE AND OUTPUT
ANGULAR VELOCITY FOR AD-1 WITH NECESSARY CONDITION OF 25 40,0 - -96 384, 24,6 61,
POWER ADAPTATION 56,0 0 6
10 15 20 25 30 35 40 45 50 30 33,3 - - 427, 19,1 57,
73,5 106, 2 3
100, 66, 50, 40, 33, 28, 25, 22, 20, 8
0 7 0 0 3 6 0 2 0 63,4
35 28,6 - - 457, 15,5 54,
Table 1 values correspond to the necessary condition of
power adaptation. To determine a sufficient condition of
85,6 114, 0 3
power adaptation, it is necessary to solve the following task. 2
40 25,0 - -120 480, 13,0 52,
Next, we define the values and
95,0 0 0
.
45 22,2 - - 498, 11,1 50,
As an example, let's calculate the required parameters 102, 124, 4 2
for = 10Nm and corresponding = 100 . 4 6
Power balance of kinematic chain with friction joint K: 50 20,0 - - 511, 9,8 48,
107, 127, 6 8
, 9 9
where Based on the data in Table 2, plot the traction
. characteristic for AD-1 (without additional frictional
Here ω5=ωH2+(ω3-ωH2)u(H2/56)=100+(100-100), constraint) as a dependence of the output resistance moment
on the output shaft speed (Fig. 4a).
where = Model №2.
, Given: = 100 s-1, = 10 Nm,
Let’s determine the value of the frictional power = [10, 15, 20, 25, 30, 35, 40, 45, 50] Nm,
using the following Eq.:
- average value for preliminary
. approximate calculation,
= 40; = 16; = 72; = 16; = 40; = 96; =
Let’s determine the value of the output angular velocity 20; = 36.
using the following Eq.: Determine:
- 258 -
and .
Solution: s
-1
.
First, the decision is made for AD-2 with a preliminary
relative angular velocity . 2) For an additional gear at = 72 s-1:
Using Eq. (5) and the output torque data from Table 1,
we determine the values of the output angular velocity . .
As an example, determine the value of at = 10
Nm using the following constants:
s-1.
W.
Determine the variable value of the relative angular
velocity by the Eq. (8):
s-1.
W.
Determine the value of the power for friction using
the Eq. (3):
Calculation of the output angular velocity at =
10 Nm:
.
= 10 Nm:
Determine the value of the output angular velocity
using the Eq. (5):
s-1.
Using this solution algorithm, determine the remaining
parameters , which determine the approximate Determine the value of efficiency according to the Eq.
sufficient condition of power adaptation at the (9):
corresponding values of . The obtained data are
summarized in Table 3. %= .
TABLE III. Values of output torque and output
angular velocity for AD-2 with an approximately sufficient After performing calculations according to Eqs. (5)...(9),
condition of power adaptation respectively, for each value of and from Table 3,
build Table 4 with the results of calculations
10 15 20 25 30 35 40 45 50 ( and ) for AD-3 with a real-
sufficient condition of power adaptation (there is an
72, 48, 36, 28, 24, 20, 18, 16, 14, additional frictional constraint with a real relative angular
0 0 0 8 0 6 0 0 4 velocity).
Next, consider the actual operating conditions of the TABLE IV. VALUES OF THE OUTPUT TORQUE AND OUTPUT
drive. ANGULAR VELOCITY FOR AD-3 WITH A REAL-SUFFICIENT CONDITION OF
POWER ADAPTATION
Model №3.
AD-3 with a real adaptation condition at .
Avera
s-1 s -1
s -1 s-1 W s -1
% ge
Using Eqs. (5)…(9), calculate the values of the output %
angular velocity . 10 72, 27, 21, - 30, 96, 96,
As an example, below is a numerical calculation for 0 1 6 5,5 8 9 9
= 72 s-1. 15 48, - - - 57, 62, 94,
Let's determine and for all values of (Table 3) 0 35, 45, 10, 7 8 2
using Eqs. (6) and (7) respectively: 3 6 3
1) For planetary gear at = 72 s-1:
20 36, - - - 71, 46, 92,
92,3
0 66, 79, 12, 1 4 9
5 2 7
= 25 28, - - - 79, 36, 92,
-1
s . 8 85, 99, 14, 5 8 1
2 4 2
30 24, - - - 84, 30, 91,
- 259 -
0 97, 112 15, 6 5 5 range for output angular velocity;
7 ,8 1
4)The AD-3 model's efficiency stands at an
35 20, - - - 89, 26, 91,
average of 92%, a significant improvement
6 106 122 15, 0 0 1
,4 ,3 9
compared to the AD-1 model, which averages
40 18, - - - 91, 22, 90,
63% efficiency;
0 113 129 16, 3 7 9 5) Collating findings from points 3 and 4, it is
,3 ,6 3 evident that the control range of the AD-3 model
45 16, - - - 94, 20, 90, decreased slightly compared to the AD-1 model,
0 118 135 16, 1 1 6 but this was accompanied by a substantial
,4 ,2 8 increase in efficiency.
50 14, - - - 95, 18, 90,
CONCLUSION
4 122 139 17, 8 1 4
,6 ,7 1 The analysis performed on the mechanism with two
degrees of freedom and an additional constraint
demonstrates a reliable adaptation of the output
Based on the data from Table 4, build a graph of the
traction characteristic of AD-3 with additional friction working component to a varying load. The model
coupling in the form of a dependence of the output chosen for this investigation, featuring an additional
resistance moment on the output shaft speed (Fig. balancing constraint in the form of a friction coupling,
4b). guarantees the system's reliable functioning and high
Figure 4 shows comparative graphs of the characteristics efficiency.
of AD-1 and AD-3. Although a slight increase in power losses due
to the inclusion of a friction coupling mildly
impacts efficiency and the control range, these
remain within a fairly broad spectrum. This
outcome can be attributed to the strategic
positioning of the friction coupling, which allows
for a low relative velocity of the interacting links.
The friction coupling establishes an effective
force-velocity limitation.
Figure 4 - Graphs of the dependence of the output This friction coupling forms an efficient force-
moment of resistance velocity restriction that substitutes the unbalanced
4a - Graph of the dependence of the output moment of reaction, ensuring the determinability of motion
resistance on the output angular velocity for within a specified extensive regulatory range.
AD-1 with a sufficient condition of power adaptation,
VI. SOURCE OF FINANCING
4b – Graph of the dependence of the output moment of
resistance on the output angular velocity for AD- This research has been funded by the Science
3 with a real-sufficient condition of power adaptation. Committee of the Ministry of Education and Science
of the Republic of Kazakhstan (Grant No.
AP09258712 «Development of self-adjusting electric
V. ANALYSIS OF THE RESULTS drives for spacecrafts»).
Examination of the generated results leads to
REFERENCES
the following deductions:
[1] Kim, D.H.; Kim, J.W.; Choi, S.B. Design and modeling of energy
1) The theoretical patterns of the mechanism's efficient dual clutch transmission with ball-ramp self-energizing
mechanism. IEEE Trans. Veh. Technol. 2019, 69, 2525–2536.
parameters interaction that we obtained are indeed [2] Ahssan, M.R.; Ektesabi, M.M.; Gorji, S.A. Electric vehicle with
valid. Graph 4b substantiates the consistent multi-speed transmission: A review on performances and
complexities. SAE Int. J. Altern. Powertrains 2018, 7, 169–182.
functionality of the evolved drive and the [3] Eduardo Louback, Fabricio Machado, Lucas Bruck, Phillip J.
manifestation of the power adaptation effect; Kollmeyer, Ali Emadi, «Real-Time Performance and Driveability
Analysis of a Clutchless Multi-Speed Gearbox for Battery Electric
2) The engineered mechanism boasts similar Vehicle Applications», 2022 IEEE Transportation Electrification
Conference & Expo (ITEC), pp.1041-1046, 2022.
order power adaptation and high operational [4] H. Laitinen, A. Lajunen, and K. Tammi, «Improving electric vehicle
reliability, shown by the definability of energy efficiency with two-speed gearbox», in Proc. IEEE Veh.
Power Propulsion Conf., Apr. 2017, pp. 1–5.
movement; [5] T. Hofman and C. H. Dai, «Energy efficiency analysis and
comparison of transmission technologies for an electric vehicle», in
3) The AP-3 model demonstrates a dependable Proc. IEEE Veh. Power Propulsion Conf., Mar. 2011, pp. 23–28.
- 260 -
[6] A. Sorniotti, S. Subramanyan, A. Turner, C. Cavallino, F. Viotto, and
S. Bertolotto, «Selection of the optimal gearbox layout for an electric
vehicle», SAE Int. J. Engines, vol. 4, no. 1, pp. 1267–1280, 2011.
[7] Q. Ren, D. A. Crolla, and A. Morris, «Effect of transmission design
on electric vehicle (EV) performance», in Proc. IEEE Veh. Power
Propulsion Conf., Oct. 2009, pp. 1260–1265.
[8] B. Eberleh and T. Hartkopf, «A high speed induction machine with
two speed transmission as drive for electric vehicles», in Proc. Int.
Symp. Power Electron., Elect. Drives, Automat. Motion, May 2006,
pp. 249–254.
[9] F. A. Machado, P. J. Kollmeyer, D. G. Barroso and A. Emadi, «Multi-
speed gearboxes for battery electric vehicles: Current status and future
trends», IEEE Open Journal of Vehicular Technology, vol. 2, pp. 419-
435, 2021.
[10] J. Ruan, P. D. Walker, J. Wu, N. Zhang, and B. Zhang, «Development
of continuously variable transmission and multi-speed dual-clutch
transmission for pure electric vehicle», Adv. Mech. Eng., vol. 10, no.
2, pp. 1–15, Feb. 2018.
[11] Jianjun Hu, Feng Xiao, Hang Peng, Wei Zhao. CVT discrete speed
ratio optimizations based on energy efficiency for PHEV. Alexandria
Engineering Journal Volume 61, Issue 5, May 2022, Pages 4095-
4105.
[12] Mantriota G, Reina G. Dual-Motor Planetary Transmission to
Improve Efficiency in Electric Vehicles. Machines 2021, 9, 58.
[13] Meng, D.; Wang, F.; Wang, Y.; Gao, B. In-Wheel Two-Speed AMT
with Selectable One-Way Clutch for Electric Vehicles. Actuators
2021, 10, 220.
[14] Ivanov K.S. Almaty, KZ. Gebrauchsmusterinhaber. Bezeichnung –
Einrichtung zur automatischen und kontinuierichlen Drehmoment –
und Drehzahlveranderung einer Abtriebswelle je nach
Fahrwiderstand. Urkunde uber die Eintragung des Gebrauchsmusters
Nr. 20 2012 101 273.1. Tag der Eintragung 02.05.2012. Deutsches
Patent und Markenamts. Bundesrepublik Deutschland. 2012.
[15] Ivanov K.S. Theory of adaptive mechanical drive. Theoretical and
Applied Mechanics Letters. Elsevier. Volume 11, Issue 4, 2021. P. 1 –
5.
- 261 -
August 22(TUE) - 25(FRI), 2023
Oral Session 13
14:00~15:40 Thu, August 24, 2023
Room : A434
Chair: Ainur Begimbetova (AUPET, Kazakhstan)
- 262 -
Method of predicting the service life of a linear
adjustment transformer based on the analysis of
operational reliability data
I.T. Utepbergenov1*, A.A. Tazhibayev1,3, Ye.U. Zhumagaliyev2, D.A. Yesengaliyev2, S.A. Yermetov3
1
Almaty University of Power Engineering and Telecommunications named after Gumarbek Daukeyev, Kazakhstan,
2
ZhubanovAktobe Regional University, Kazakhstan,
3
TrencoR&D LLP, Kazakhstan,
Email: [email protected]
Abstract— The main idea of the study is to study the The increase in the life cycle of the transformer resource is
possibility of identifying the service life reserves of transformers carried out by means of carrying out complex work, which is
on the basis of improving the quality of assessment and carried out on the basis of study operations and diagnostics in
prediction of reliability and calculation of operational reliability accordance with industry regulatory and technical documents.
by processing the entire volume of stored data of transformers When planning and carrying out such work, data are
obtained during the operation period. The goal of the work is to accumulated on the entire “fleet” of aging transformers
develop a methodology for assessing and predicting the throughout the entire service life. The application of an
reliability and calculating the operational reliability of a line improved power transformer maintenance plan using
regulating transformer using the available accumulated data on
reliability-oriented asset management can extend the service
its condition at the time of maintenance and repair. For
analytical studies, static data on the time between failures of the
life of transformers up to 25–30 years [6]. However, this is
LTOL-16000/10 type line regulating transformer for 32 years possible if there are reserves for the service life of
were used. The R software environment was used to calculate transformers.
standard statistical characteristics, construct distribution The main idea of the study is to study the possibility of
histograms, a leveling curve, and check the law of distribution identifying the reserves of the service life of transformers
of the time between failures of the transformers under study. based on improving the quality of assessment and prediction
The objective values of quantitative indicators of the operational
of reliability and calculation of operational reliability by
reliability of transformers are determined on the basis of
processing the entire volume of stored data of transformers
empirical operation data necessary for the design and
improvement of transformer units. It is shown that when
obtained during its operation. At the same time, the results of
applying the developed methodology, taking into account the data processing are the basis for designing and improving the
rate of thermal aging, the calculated maximum service life of system for maintenance and repair of the transformer.
power transformers of the studied class can increase to 30.97 The goal of the work is to develop an improved
years, subject to the operating conditions of power transformers methodology for assessing, predicting and identifying the
that ensure the failure rate in the degradation interval is not reliability reserve and calculating the operational reliability of
more than the failure rate of the running-in period. The results
a line regulating transformer using the available accumulated
obtained are necessary for making a decision on planning the
scope of maintenance and repair work when working with
data on its condition at the time of maintenance and repair.
transformers that have exhausted their standard resource. II. STUDY METHODS
Keywords— linear adjustment transformer; operational The methodological basis of the proposed methodology
reliability; statistical analysis; data set; predicting the service life for assessing and predicting reliability and calculating the
operational reliability of the line control transformer is the
I. INTRODUCTION static processing of data [7] on their wear and failures.
Improving the operational reliability of transformers is The work uses the data of a service organization obtained
becoming increasingly important due to the aging of the fleet during the operation of LTOL-16000/10 type line control
of transformers and the development of a certain standard transformers up to 16,000 kVA for 32 years in the process of
average life (25 years) of their service [1-3]. The operational their inspection and planning of maintenance and repair.
reliability and, accordingly, the service life of a line regulating
transformer are assessed through the following parameters: The R software environment [8] was used to calculate
reliability, serviceability and life cycle. The operational standard statistical characteristics, construct distribution
reliability and service life of transformers are affected by: the histograms, a leveling curve, and check the law of distribution
frequency of shutdowns in case of accidents, defects and of the time between failures of the transformers under study.
damages, the quality of active and structural materials used in Also, this environment is used to calculate the reliability
the manufacture of the transformer; environmental conditions indicators of transformers.
during transformer operation; lack of proper care for
transformer devices and poor quality of their repair [4, 5].
- 263 -
III. DEVELOPMENT AND IMPLEMENTATION OF THE ∆t – interval boundaries.
METHODOLOGY Output data:
A. Selection of input and output data 1) Failure rate - λ;
2) Average time between failures - Тav;
The following input data were used to assess and predict
3) Probability of failure-free operation - Р(t);
the reliability and calculate the operational reliability of the
transformer: 4) Average time of failure-free operation - Т.
t – time between failures, h; B. Calculation of statistical characteristics and
n – total number of observed values; construction of a histogram and a leveling curve for the
ni – number of values in each interval; distribution of time between failures
k – number of static distribution intervals;
TABLE I. STATIC DATA ON THE TIME BETWEEN FAILURES OF THE LTOL -16000/10 TYPE LINE CONTROL TRANSFORMER FOR 32 YEARS FOR
ANALYTICAL STUDIES
t, time between failures, (hours)
350 700 5,300 1,430 7,600 250 10,400 500
5,000 3,800 1,000 6,200 3,020 2,400 11,550 14,550
19,500 2,750 310 1,960 2,200 5,890 5,400 8,500
470 1,800 2,500 7,500 21,070 17,550 2,520 9,000
The minimum series term is 250 and the maximum is The value of the interval ∆t is determined by the formula:
21,070 hours. !!"# "!!$% %+,%+
∆t =
#$%.'()*
= #$%.'()'% = 4,666.2 ℎ (1)
The range of the series is tmax - tmin = 21,070–250 =
20,820 hours For each interval, calculations were made, and the results
The entire range of static data on the time between failures are presented in Table 2.
ti (n = 32) is divided into equal intervals, (i.e. this is the
number of failures of the transformer at a certain interval).
TABLE II. STATIC DATA PROCESSING
No. interval ∆t, hours ni Pi = ni/n ti interval average P i ti Σ ni/n ni/n∆t10-6 Pi[ti-M(t)]2
value
1 0-2,000 2,000 9 0.28 1,000 281 0.28 141 6,684,082
2 2,000-4,000 2,000 7 0.22 3,000 656 0.50 109 1,808,105
3 4,000-6,000 2,000 5 0.16 5,000 781 0.66 78 119,629
4 6,000-8,000 2,000 3 0.09 7,000 656 0.75 47 118,652
5 8,000-
10,000 2,000 2 0.06 9,000 563 0.81 31 610,352
6 10,000-
12,000 2,000 2 0.06 11,000 688 0.88 31 1,641,602
7 12,000-
14,000 2,000 0 0.00 13,000 0 0.88 0 0
8 14,000-
16,000 2,000 1 0.03 15,000 469 0.91 16 2,602,051
9 16,000-
18,000 2,000 1 0.03 17,000 531 0.94 16 3,867,676
10 18,000-
20,000 2,000 1 0.03 19,000 594 0.97 16 5,383,301
11 20,000-
22,000 2,000 1 0.03 21,000 656 1.00 16 7,148,926
Σ 32 M(t) = 5,875 h D(t) = 29,984,375
Static dispersion D (t) will be: Often in statistical studies, the following characteristics
are used, the standard error in determining the arithmetic
D(t) = ∑{Pi[ti м- M(t)]2} =29,984,375 h (2) average (mathematical expectation):
Standard deviation: ∆М(t) =
-(!)
=
2,452.,
= 967.99 ℎ (5)
√* √'%
σ(t) =)𝐷(𝑡) = )29,984,375 = 5,475.8 ℎ (3)
The mean-square error of determining the mean-square
deviation:
Coefficient of variation:
-(!) 2,452.,
-(!) 2,452., σ(t) = = = 484.0 ℎ (6)
V(t) = 1(!) = 2,,52
= 0.93 (4) √%* √74
- 264 -
Then, rounding the values of the deviations, we obtain: As a result of the obtained calculated data, we build a
histogram of the time between failures, which is shown in Fig.
𝑀(𝑡) = 5,875 ∓ 967,99(ℎ) (7) 1.
Fig. 1. Distribution histogram and leveling curve
In Figure 1, a leveling distribution curve is constructed, related to a limited number of experimental data, or whether
*
which is a graph of the function a(t) = *∆!∙#+&' which, while they are significant and related to the fact that the fitted curve
maintaining the essential features of the statistical distribution, poorly aligns the distribution data.
is free from random irregularities in the histogram. Pearson’s goodness-of-fit test χ2:
C. Selection of the distribution law
Based on the nature of the histogram, it can be assumed !"! #"$ " (&! )()
χ2= ∑*+,- (9)
that the random variable under study is distributed according "$ " (&! )
where:
to an exponential law. This also proves the almost complete
coincidence of the value of the mathematical expectation and k is the number of statistical distribution intervals, k = 11;
the mean-square deviation of the random variable t. ni is the number of values of a random variable in each
interval;
Taking as the mathematical expectation of the time n is the total number of observed values of the random
between failures its assessment (static average) M(t) = 5,875 variable, n = 32;
h, we can write: f(t)=λe-λt. Since under the exponential P’(ti) is the theoretical probability of a random variable
#
distribution law λ= = 𝑐𝑜𝑛𝑠𝑡, then falling into the i-th interval.
М(!)
The probabilities of a random variable falling into the i-th
f (t)= λ exp (-λt) = 0.00017e -0.00017t (8) interval are given in Table 3. They are numerically equal to
When choosing a theoretical distribution curve, some the increment of the distribution function on the interval:
discrepancies between it and the statistical distribution are P’(ti<t< ti+1)= F(ti+1)-F(ti
inevitable. At the same time, it is necessary to know whether
these discrepancies are due only to random circumstances
TABLE III. THEORETICAL VALUE OF PROBABILITIES
𝒌
[𝒏𝒊 − 𝒏𝑷" (𝒕𝒊 )]𝟐
No. interval λti ei-0.00017t P'(ti) .
𝒏𝑷" (𝒕𝒊 )
𝒊%𝟏
1 0-2,000 0 1 0.29 0.008
2 2,000-4,000 0.3 0.71 0.21 0.012
3 4,000-6,000 0.7 0.51 0.15 0.008
4 6,000-8,000 1.0 0.36 0.1 0.013
5 8,000-10,000 1.4 0.26 0.074 0.057
6 10,000-12,000 1.7 0.18 0.053 0.054
- 265 -
7 12,000-14,000 2.0 0.13 0.04 1.28
8 14,000-16,000 2.4 0.09 0.03 0.002
9 16,000-18,000 2.7 0.07 0.019 0.25
10 18,000-20,000 3.1 0.047 0.014 0.68
11 20,000-22,000 3.4 0.033 0.010 1.45
3.7 0.024 Σ= 3.812
The parameter R - the number of degrees of freedom. For P (t) = e- λt; f (t)= λe- λt (10)
the exponential distribution law of a random variable, the
bond number is S=2. We determine R=11-2=9. From the # #+(
Tav = ; = = 4,166 ℎ.
reference literature, the value χ2 = 4.168 was determined with %.4
the degree of freedom R=9 and the significance level α=0.9 ,,+++
[9]. Р(8,000) = 𝑒𝑥𝑝 >−
4,#77
@ = 𝑒 "#.<% = 0.147
The calculated value χ2=3.812<4.168, therefore, the #4,+++
Р(14,000) = 𝑒𝑥𝑝 >− @ = 𝑒 "'.'7 = 0.03
hypothesis of exponential distribution and failures is accepted. 4,#77
%+,+++
D. Calculation of reliability indicators of the LTOL- Р(20,000) = 𝑒𝑥𝑝 >− 4,#77
@ = 𝑒 "4., = 0.008
16000/10 type transformers
On the basis of the calculated data obtained, we plotted the
It is necessary to determine the probabilities of no-failure graph of dependence of the probability of failure-free
operation and the average time between failures of the LTOL- operation on the time of the first order P(t) = e-λt of the LTOL-
16000/10 type transformer, for time ti = 8,000, 14,000, 20,000 16000/10 type transformer, shown in Fig. 2
h, if the failure rate λ = 2.4·10-4 h-1. With an exponential
distribution law, the following relations are true:
Fig. 2. Probability of failure-free operation of transformers
Determine the probability of failure: E. Assessment of the transformer service life, taking
into account the rate of thermal aging.
Q8,000 = 1 – Р8,000 = 0.853;
Experience confirms that the actual service life of the
Q14,000 =1 – Р14,000 =0.965; transformer is on average no more than 20 years.
Transformers with the same operating characteristics may
Q20,000 =1 – Р20,000 =0.992. have different service life. Thermal factors such as
insufficient cooling, excessive harmonics, and improper
operating conditions can reduce the life of a transformer.
- 266 -
According to IEC 60076–12:2008 [10], in terms of According to formula (12), the thermodynamic
service life, the transformer of the considered LTOL- temperature of the most heated point was calculated: Т =
16000/10 type belongs to the “A” class of the insulating 273+30+100 = 403 K
system temperature. The nominal service life of the Thermal aging rate:
transformer L, at a constant thermodynamic temperature T of
the hottest point, ˚C is determined by the following formula: К = 180,000 ∙ (3.1 · 10"4 )"# ∙ 𝑒𝑥𝑝 > @=
"#2,<++
4+'
L = а · exp b/T (h) (11) 42.58 ℎ – service life consumption per hour.
The thermodynamic temperature of the hottest point has With a load time of 240 hours, the estimated service life
the form: for ten days at this temperature will be 42.58∙240=10,219
hours of the entire service life of the transformer.
Т = 273 + То + ∆Т, (12) The expected service life at a nominal hottest point
temperature of 100°C is 18,000 hours. Then in ten days at a
#+,%#<
where: То is ambient temperature, ˚С (30˚С); temperature of 130˚С it will be 100 ∙ = 5.68 % of the
#,+,+++
∆Т is the temperature difference between the hottest point nominal service life of the transformer.
of the winding (130˚С) and the ambient temperature for a To study the failure rate for the period of running-in and
given load, ˚С. aging of the transformer, as well as to determine the average
(∆Т = 130–30 = 100 ˚С). time of failure-free operation of the transformer, the uniform
Calculation of the thermal aging rate K is determined as distribution law and the Weibull distribution law were used
the time consumption of the entire service life per hour of [11].
operating time at temperature T, ˚С Figure 3 shows a histogram approximating the probability
density (7) of the distribution for failure of transformers with
"=
𝐾 = 18,000 ∙ а"# ∙ 𝑒𝑥𝑝 ( > ), (13) a power of up to 16,000 kVA
),++, )
where “a” and “b” are constants for the equation for 𝑓? (𝑡) = 0,069 ∙ 𝑡 "+.++% ∙ 𝑒 "(+,#2<,∙! , at 0≤t≤12, years
calculating the service life of the transformer (Arrhenius
equation constants). In our case, a = 3.1×10-4 (h); b = 15,900 fII (t)= 0.0099, at 12≤t≤23, years
(K).
&+ ∙! ',-(,- )
𝑓??? (𝑡) = 0,165 ∙ 10"< ∙ 𝑡 2,24,2 ∙ 𝑒 "(+,+%2∙#+ ,
23≤t≤40, years
Fig. 3. Histogram and probability density of distribution for failure of transformers up to 16,000 kVA
It is obtained that for three resource intervals (I - interval from 0 to 12 years, in the interval 23-40 years - also
running-in, II - normal operation, III - degradation), the time by Weibull distribution, with parameters α3 = 0.025 10-9 and
between failures of transformers is described by the Weibull δ3 = 6.5485, and on the intermediate interval 12-23 - uniform
distribution with parameters α1 = 0.1598 and δ1 = 0.998 in the distribution law with coefficient α3 = 0.0099.
- 267 -
>. @#% >1 @%' >0 @4+
"+.++% "A+.+#2<,∙! ).++, B
𝑇 = I 𝑡 ∙ 0.069 ∙ 𝑡 ∙𝑒 𝑑𝑡 + I 𝑡 ∙ 0.0099𝑑𝑡 + I 𝑡 ∙ 0.0165 ∙ 10", ∙ 𝑡 2.24,2
>) >. @#% >1 @%'
&+ ∙! '.-(,-)
∙ 𝑒 "(+.+%2∙#+ 𝑑𝑡 = 21.34 𝑦𝑒𝑎𝑟𝑠
IV. ISCUSSION OF THE RESULTS working with transformers that have exhausted their standard
resource.
The results of calculations of operational reliability
according to the proposed method of the line regulating ACKNOWLEDGMENT
transformer during operation up to 16,000 kVA using the The the research was carried out within the framework of
available accumulated data makes it possible to identify the grant funding of Young scientists of the Committee of Science
presence of a reliability reserve and, accordingly, the service of the Ministry of Science and Higher Education of the
life. Namely, the calculated maximum service life of Republic of Kazakhstan (IRN No. AR09058053
transformers was 30.97 years, while the average time of "Development of a linear adjustment transformer type LTOL-
failure-free operation of transformers is 21.34 years. 16000/10).
From the constructed graph of the dependence of the
probability of failure-free operation on the time of the first REFERENCES
order P(t) = e- λt of the LTOL-16000/10 type transformer, the [1] Molebe P.P., Makhanya B.S., Pretorius J.H.C. The reliability and
maximum time of failure-free operation is determined to be maintainability analysis of transmission transformers in South Africa.
2018 IEEE International conference on environment and electrical
18,000 hours. engineering 2018 IEEE Industrial and commercial power systems
Europe/ Univ Palermo, Palermo, Italy.
In addition, according to the results of the calculation, it is
[2] Kiesewetter D.V., Zhuravleva N. M., Reznik A.S., Tukacheva
possible to recommend limiting the operation of power A.V., Smirnova E.G., Khripunov A.K. On possibility of power
transformers in the degradation interval to ensure the transformer operational reliability increase / 10th International
conditions when the failure rate does not exceed the failure conference 2016 Electric power quality and supply reliability
rate of the running-in period. conference (PQ), Tallinn, Estonia, 2016. pp. 192 - 197.
[3] Arjan van Schijndel, J.M. Wetzer J.M. Wetzer, P.A.A.F. Wouters.
For the hottest point of the winding Т=130˚С with a load Forecasting transformer reliability / IEEE conference on electrical
time of 240 hours, the service life of the LTOL-16000/10 type insulation and dielectric phenomena 2006. pp. 577-582.
transformer is 5.68% of the nominal service life. [4] L.M. Suleimanova. Increasing the operational resources of power
transformers while ensuring electromagnetic compatibility for
V. CONCLUSIONS overvoltages: Abstract of the dissertation of the candidate of technical
sciences. - Samara, 2006 - 23 p.
In the course of this study, a methodology was developed
[5] Chojnacki A.L. Reliability parameters and properties of medium
and tested for assessing, predicting and identifying the voltage low voltage transformers / Przeglad elektrotechniczny. №4,
reliability reserve and calculating the operational reliability 2008. р.74-77.
of the line regulating transformer using the available [6] Aysun Koksal, Aydogan Ozdemir. Improved transformer maintenance
accumulated data on its condition at the time of maintenance plan for reliability centred asset management of power transmission
system / IET generation, transmission & distribution, 2016, Vol. 10,
and repair. Iss. 8, pp. 1976-1983.
The calculated average time of failure-free operation of [7] Rexphepi V. An analysis of power transformer outages and reliability
line regulating transformers of a given type from the studied monitoring / 4th International conference on power and energy systems
data set for 32 years was 21.34 years. engineering (CPESE 2017), Berlin, Germany, p. 418 - 422.
It is shown that the calculated maximum service life of [8] S.E. Mastitsky, V.K. Shitikov. Statistical analysis and data
power transformers of the studied class can increase up to visualization with R. - E-book, access address, 2014: http://r-
analytics.blogspot.com
30.97 years, subject to the operating conditions of power
[9] Edwin B. Wilson, Margaret M. Hilferty. The distribution of chi-square.
transformers that ensure the failure rate in the degradation Proc. Natl. Acad. Sci. USA. 17 (12), pp. 684–688.
interval is not more than the failure rate of the running-in [10] IEC 60076-12:2008
period. [11] Weibull W. A statistical distribution function wide applicability /
The results obtained are necessary for making a decision Journal of applied mechanics (ASME), 1951, Т. 18, pp. 293-297.
on planning the scope of maintenance and repair work when
- 268 -
Sensorless operation of six-phase PMSM based
on high-frequency signal injection with torque
ripple reduction method
Seong-hoon Kim, Kwan-yuhl Cho* Hag-wone Kim and Sang-min Lee
Korea National University of Transportation, Republic of Korea
Email: [email protected]
Abstract— Sensorless control of PM synchronous motor II. VSD CONTROL OF SIX-PHASE PMSM
based on signal injection is widely used in low-speed operation
range because it can estimate the rotor position at low speeds as
The VSD(Vector Space Decomposition) method[1] is
well as at standstill. However, this method causes torque ripples adopted to control of six-phase PMSM. The advantages of
due to injected voltage, which makes the acoustic noise and VSD method is high torque dynamics performance because it
mechanical vibration. Six-phase PMSM consists of two three- can be eliminated Coupled component. Fig.2 shows that VSD
phase PMSMs with parallel connection so that the two harmonic reference frame.
dq voltages for the rotor position estimation should be injected
to each three-phase PWM inverters. In this paper, the torque
ripple cancellation method is proposed. The torque ripples can
be canceled since the torque ripples generated in each three
phase PMSMs have complementary phases. 1kW six-phase
PMSM verifies the effectiveness of the proposed method.
Keywords—PMSM, six-phase, dual three-phase, torque ripple
reduction, sensorless control, signal injection Fig. 2. VSD(Vector Space Decomposition) reference frame.
I. INTRODUCTION According to the VSD transform, the voltage equation can
be written as (1)-(2) and torque equation can be obtained as
The rotor position information is necessary to PM motor (3). Only the first reference frame(DQ) generate torque.
control. Generally, position sensor such as encoder or resolver
used to measurement rotor position information. But, the 𝑣 𝑟 + 𝑝𝐿 −𝜔 𝐿 0
𝑖
position sensor increases systems cost. In order to solve this = +𝜔
𝑣 𝜔 𝐿 𝑟 + 𝑝𝐿 𝑖 √3ϕ
problem, the sensorless control is adopted in many application.
This sensorless control can eliminate the position sensor and
uses the estimated rotor position to control of motor. 𝑣 𝑟 + 𝑝𝐿 −𝜔 𝐿 𝑖
𝑣 = 𝑖
Sensorless control of PM synchronous motor based on signal 𝜔𝐿 𝑟 + 𝑝𝐿
injection is widely used in low speed operation range because
it can estimate the rotor position at low speeds as well as at 𝑇 = {√3ϕ 𝑖 + (𝐿 − 𝐿 )𝑖 𝑖 }
standstill. But the signal injection methods cause torque ripple
of injected signal frequency, which makes the acoustic noise
Where LDQ and Ldqz is inductance of reference frame, Rs
and vibration. Six-phase PMSM consists of two-three-phase
PMSMs with parallel connection so that the complementary is stator winding resistance, ϕ is flux linkage of Permanent
signals for the rotor position estimation can be injected to each magnet, 𝑝 is differential operator and P is number of poles.
three-phase inverter. In this paper, the torque ripple
III. SINGAL INJECTION SENSORLESS CONTROL
cancellation method using complementary signal injection for
six-phase PMSM is proposed. The rotor position is estimated To estimate rotor position, the high-frequency square
from the information of six-phase currents induced by the wave signal injection method [2] is adopted. In Fig. 3 is shown
complem-entary voltages injected into each three-phase the Injected voltage.
windings of PMSM. The torque ripples can be canceled since
the torque ripples generated in each three-phase PMSMs gave
complementary phases. In Fig.1 the Two inverters for six-
phase PMSM is shown.
Fig. 3. Injected SWT(Square Wave Type) voltage.
Where 𝑉 is magnitude of injected voltage and Ts is
sampling time. Frequency of injected voltage is quarter of
switching frequency. The relationship between the induced
Fig. 1. Two inverters for Six-phase PMSM.
- 269 -
high-frequency current and rotor position information can be
written as follow:
A. Experiment result
𝑖 cos𝜃 Fig.5 shows the estimated rotor position based on the
𝑖 = demodulated current. Motor drive 300rpm at 20% load. Inject
sin𝜃
voltage magnitude is 20V that is 10% of DC link voltage. The
∆𝑖 =𝑖 [𝑘] − 𝑖 [𝑘 − 1] error of the estimated position is 1.8 degrees, confirming that
the rotor position and speed are estimated well.
Where ℎ is high frequency component which injected.
The rotor position information within high frequency induced
current. The estimated rotor position can be obtained as follow:
𝜃 = 𝑡𝑎𝑛 | |
∆𝑖 𝑖𝑓 𝑉 >0
𝑤ℎ𝑒𝑟𝑒 |𝑖 |=
−∆𝑖 𝑜𝑡ℎ𝑒𝑟𝑤𝑖𝑠𝑒
Where 𝜃 is estimated rotor position and |𝑖 | is
demodulated signal of high-frequency induced current .
Fig. 5. Estimated rotor position and current 300rpm, 20% load
IV. PROPOSED TORQUE RIPPLE CANCELLATION
Fig.6 shows the phase currents, DQ and dqz axis currents
The relationship between the first VSD D-reference frame with and without harmonic voltage injection. It can be noted
and each three phases conventional 𝑑 -reference frames as that the a1-axis and a2-axis harmonic currents have similar
follow: waveforms with phase difference of 180 degrees, so that The
current ripples are appeared only in dz axis current.
𝑖 𝑖 +𝑖
=
𝑖 𝑖 −𝑖
Where d1 is 𝑎𝑏𝑐 three-phases d-reference frame and d2
is 𝑎𝑏𝑐 three-phases d-reference frame. D is sum of each
three-phase and dz is difference of each three-phase. The Fig.3
as shown proposed torque ripple cancellation algorithm. The
torque ripples can be canceled since the torque ripples
generated in each three-phase PMSMs gave complementary
phases.
Fig. 6. Phase and DQ and dqz currents with and without voltage injection
VI. SUMMARY
In this paper, the torque ripple reduction algorithm for
Fig. 4. Proposed torque ripple cancellation algorithm. sensorless control based on signal injection is proposed.
Proposed method can be estimated rotor position and can be
V. EXPERIMENT canceled torque ripple. Experimental result with 1kW six-
To verify the effectiveness of proposed algorithm, 1kW phase PMSM verifies that the proposed method estimates the
six-phase PMSM was used. the motor parameters used in rotor speed and position well and reduces the torque ripple due
experiment is shown as follow: to injected harmonic voltages.
TABLE I. MOTOR PARAMETERS USED IN EXPERIMENT REFERENCES
Motor parameters
Symbol [1] Y. Hu, Z. Q. Zhu and M. Odavic, "Comparison of two-individual
Name of parameters Value Unit
current control and vector space decomposition control for dual three-
P Number of poles 20 phase PMSM," IEEE Trans. on Industry Applications, vol. 53, no. 5,
Rs Stator winding resistance 1.42 Ohm pp. 4483-4492, 2017.
[2] Y. Yoon, S. Sul, S. Morimoto and K. Ide, "High-bandwidth sensorless
LDQ DQ-axis Inductance 9.0 /10.3 mH algorithm for ac machines based on square-wave-type voltage
Ldqz dqz-axis Inductance 2.3 /2.6 mH injection," in IEEE Trans. on Industry Applications, vol. 47, no. 3, pp.
1361-1370, 2011.
Psif Back EMF constant 0.068 Wb
- 270 -
Study on the Operation Algorithm of TVPP
Considering Distribution Line Constraints in
Jeju Power System
Ji-Won Lee 1, Dong-Il Cho 2, Seong-Jun Park 1, Tae-Yang Nam 1, Byung-Ki Kim 3, Jae-Sun Huh 4,
Nam-Gil Paik 4, Jae-Chul Kim 1, Won-Sik Moon 1*
1
Soongsil University, South Korea
2
Korea University, South Korea
3
Korea Institute of Energy Reasearch, South Korea
4
KEPCO, South Korea
Email: [email protected]
Abstract— Recently, there has been increasing interest in demand resources such as solar power, ESS(Energy Storage
renewable energy due to the issue of global warming. Through System), and DR(Demand Response) are collected and
the use of renewable energy, efforts are being made to reduce integrated through distributed resource owners and
carbon emissions from conventional power generation and intermediaries to operate as a virtual power plant.
achieve carbon neutrality. However, renewable energy sources
have greater variability compared to traditional generators,
which can have an impact on grid operations and result in
output limitations. In this paper, we propose an algorithm that
introduces a Virtual Power Plant (VPP) to minimize output
limitations by utilizing the resources within the VPP in
situations where output limitations are increasing.
Keywords—Virtual Power Plant, Distribution Line
Constraints, Technical Virtual Power Plant
I. INTRODUCTION
Due to the issue of global warming, there is a growing
global interest in renewable energy. European countries are
leading the way in expanding the acceptance rate of renewable
energy and are planning to achieve carbon neutrality through
this. In Korea, the proportion of volatile renewable energy Fig. 1. Virtual Power Plant Architecture
centered on wind power and solar power is being expanded to B. Classification based on operational objectives
achieve energy policy. It is expected to expand the capacity of
renewable energy to 58GW by 2030[1]. However, the VPP can be divided into CVPP and TVPP depending on
increase in renewable energy generation facilities leads to the operation objectives[5].
problems such as curtailment due to output volatility and Commercial VPP is a VPP operated by an intermediary for
degradation of power system stability[2]. To respond to this, the purpose of market participation, with no geographical
the concept of a Virtual Power Plant (VPP), which integrates restrictions on the recruitment of distributed resources, and
and manages multiple distributed resources as one, is being contributes to the balance of supply and demand across the
introduced in Korea. VPP operates like a generator, and system. The operation of VPP for profit-making purposes,
through this, it can respond more efficiently to curtailment such as small-scale intermediary trading markets, can be
than before[3]. In this paper, we simulated the overvoltage understood as CVPP. By gathering and optimizing various
situation that can occur in the distribution system when using distributed resources, intermediaries can create various
VPP. In the simulated situation, we proposed an algorithm to business models based on the profits generated, such as power
relieve overvoltage using VPP resources. trading agency.
II. VIRTUAL POWER PLANT Technical VPP is a VPP operated by a network operator
mainly for the purpose of stabilizing the distribution system,
A. Characteristics of Virtual Power Plant recruiting distributed resources in the region and using them
Virtual Power Plant(VPP) is a system designed to operate for auxiliary purposes such as voltage and frequency control.
various distributed energy resources as if they were a single Through the operation of TVPP, distributed resource owners
power plant, using information and communication provide necessary distributed resources for stabilizing the
technology and automatic control technology[4]. The concept distribution network when needed and receive compensation
of a VPP is illustrated in Figure 1. While it is not a power plant for the use of distributed resources by the distribution network
that physically exists in a specific location, it has the same and
- 271 -
operator. VPP operators can recruit distributed resources in B. Operation Algorithm
nearby areas to delay the investment in new distribution The proposed algorithm is illustrated in Figure 2. Once the
networks by utilizing them for peak power reduction and output capacity to be delivered is allocated, the situation
distribution network voltage adjustment. If flexible resources within the power distribution system is checked for
are needed, they can also participate in the ancillary service overvoltage. The upper limit of the ultra-high voltage
market using TVPP resources[]. maintenance range according to the performance standard of
III. CURRENT STATUS OF JEJU POWER SYSTEM distribution electrical equipment is 23,000V (1.039p.u), and
the lower limit is 20,800V (0.908p.u)[7]. If overvoltage occurs,
A. Current status of Jeju Power System reactive power control through ESS is first implemented. The
As of 2022, in Jeju Island, the capacity of renewable power reactive power is absorbed to lower the voltage. If overvoltage
generation facilities is 295MW for wind power and 368MW still occurs after this, a command is issued to increase the
for photovoltaic, accounting for 18% and 23% of the total charging amount of the EV charger. If the voltage does not
power generation facilities, respectively. According to the decrease even after these measures, active power control of
10th Basic Plan for Electricity Supply and Demand, the the ESS is implemented, and if the voltage still does not
photovoltaic is expected to increase to 1,299MW and wind decrease, curtailment of the PV is finally carried out.
power generation to 2,345MW by 2030[6]. The increase in
renewable energy generation facilities leads to issues such as
curtailment and degradation of system stability due to output
variation.
B. Current Status of Curtailment
Table 1 shows the status of wind power curtailment in the
Jeju region. With the increase in renewable power generation,
the number of curtailments for wind power has increased to
104 times in 2022 since it was first implemented in 2015. By
estimating the loss cost through the average SMP of each
respective year, the loss cost due to output control in 2022 is
6.4 billion KRW[7]. From 2021, output control, which had
only targeted wind power, has been expanded to solar power,
occurring once in 2021, 28 times in 2022, and 12 times until
March 2023.
TABLE I. CURRENT STATUS OF WIND POWER CURTAILMENT IN JEJU
ISLAND
Loss
Curtailment Average
Number of Cost
Year Volume SMP
Curtailments (Million
(MWh) (KRW/kWh)
KRW)
2015 3 152 126 19
2016 6 252 92 23
2017 14 1,300 120 156
2018 15 1,366 147 200
2019 46 9,224 153 1,411
2020 77 19,450 101 1,964
2021 64 12,016 128 1,538
2022 104 25,634 251 6,434
Fig. 2. Proposed TVPP Algorithm
IV. TVPP OPERATION ALGORITHM
V. CASE STUDY
A. Assumption of Constraint Scenario
A VPP aggregator in the form of CVPP participates in the A. Modeling of Distribution System
power market and is allocated capacity. However, problems The parameters of the simulated power distribution system
such as voltage rise can occur within the distribution system used for the case study are shown in Table 2. We assumed a
to which the actual VPP is connected. In such situations, an situation where 13MW of output had to be produced through
algorithm is needed that can solve the voltage rise problem photovoltaic power generation. Figure 3 is a graph showing
through other VPP resources connected. In this paper, an overvoltage occurring when 13MW of output is produced in
algorithm was proposed that includes PV, ESS, and EV as the simulated distribution system. In the situation without
VPP resources. control, a voltage of 1.045pu appeared.
- 272 -
TABLE II. PARAMERTERS OF THE SIMULATED POWER DISTRIBUTION
SYSTEM
Classification Parameters
Line length 22 [km]
AWOC-58 [mm²], AWOC-160 [mm²],
Wire Type
ALOC-95 [mm²]
Photovoltaics Capacity 13.5 [MW]
ESS Capacity 1 [MW]
EV Charger Capacity 3 [MW]
Fig. 5. Voltage after reactive power control of ESS and EV charging control
VI. CONCLUSION
In order to achieve carbon neutrality, the proportion of
renewable energy generation is increasing. However,
renewable energy has the characteristic of inconsistent output,
leading to problems such as curtailment and degradation of
grid stability. In particular, Jeju Island has a higher proportion
of renewable energy than other regions in Korea, making the
curtailment problem more severe. To solve this, the concept
of a Virtual Power Plant (VPP), which operates various
resources like a single power plant, has been proposed. In this
paper, we defined the problems that can occur when operating
a grid in the Jeju area through a VPP. We then proposed an
Fig. 3. End-node voltage of the simulation system diagram algorithm to solve these potential problems and were able to
B. Results of Case Study confirm that the algorithm works well through a case study. In
the future, it is expected that the composition of the algorithm
The parameters of the simulated power distribution system can be changed by adding various resources such as wind
used for the case study are shown in Table 2. We assumed a power and Power to Gas (P2G).
situation where 13MW of output had to be produced through
photovoltaic power generation. Figure 3 is a graph showing ACKNOWLEDGMENT
overvoltage occurring when 13MW of output is produced in
This work was conducted under the framework of the
the simulated distribution system. In the situation without
research and development program of the Korea Institute of
control, a voltage of 1.045pu appeared.
Energy Research(C3-2416)
REFERENCES
[1] Jin-Yong Jung, Yoon-Sung Cho, "A Study on the Evaluation of System
Inertia Impact and Stability Improvement Plan According to
Renewable Energy Expansion," The transactions of The Korean
Institute of Electrical Engineers, Vol.70, No. 9, pp.1263-1273,
September 2021.
[2] Gi-Hoon Kim, Yoon-Cheul Jeung, Kyoung-Soo Kang, Yeong-Jun
Choi, Hee-Sang Ko, “A Novel Energy System Integration Scheme for
Virtual Power Plants based on Sector Coupling.” Journal of Energy &
Climate Change, Vol. 17, No.2, pp. 228-237, December 2022.
[3] Seonghyeon Kim, Sung-chan Park, Myung-Hwan Lee, Sungyun Choi,
"Sector-Coupling based microgrid for responding to cutailment
renewable energy," The Korean Institute of Electrical Engineers
Conference, pp. 95-96, October 2021.
[4] Sun-Hwa Lee, "The System and Status of Virtual Power Plants in
Korea," KDB survey monthly report, No. 767, pp. 25-34, October 2019.
[5] KEPCO Economy & Management Research Institute, "KEMRI power
economy review," Vol. 4, April 2021.
[6] Ministry of Trade, Industry and Energy, “The 10th Basic Plan for
Fig. 4. Voltage during reactive power control of ESS Electricity Supply and Demand,” January 2023.
[7] Korea Power Exchange, "Annual power system operation performance
in Jeju", March 2023.
If the EV charging amount is increased according to the [8] Korea Electrice Power Corporation(KEPCO), "Regulation on the Use
algorithm, it can be confirmed that the voltage decreases to of Electrical Equipment for Transmission and Distribution; Appendix
1.031pu at 13:00, one hour later, resolving the issue. 4; Performance Standard for Distribution Electrical Equipment.
- 273 -
Study of the direct and inverse problem of
the thermal conductivity of a rod
Sholpan Jomartova1, Тalgat Mazakov1, Nurdaulet Isimov1, Aigerim Mazakova1,
Kalamkas Begaliyeva1
1
Al-Farabi Kazakh National University, Almaty, Kazakhstan
Email: [email protected]
Abstract-- In this paper, we study the thermophysical conditions, including the initial temperature distribution
state of a rod of limited length with a constant cross section in the body and boundary conditions on the body surface,
in the form of a square. which can be specified in one of three ways: surface
The study of thermal conductivity in such rods is of temperature, heat flux, and heat transfer coefficient [2].
practical importance for various technical systems, such as
jet and hydrogen engines, power plants and spacecraft. The
Temperature is a parameter that characterizes the
solution of this problem is important for determining the
steady thermophysical state of the rods under the influence energy of the thermal motion of particles of a substance.
of various heat sources. Consequently, the process of heat propagation and its
A three-dimensional body is considered, where the left direction are inextricably linked with the temperature
end of the rod coincides with the origin of the coordinates. distribution inside the body. In general, the temperature
A constant heat transfer coefficient over the entire surface is not the same at different points of the body and depends
of the rod is taken into account, as well as the effects of point on time: T = T(x, y, z, t).
temperature and surface heat transfer. The problem is
solved by reducing it to a system of linear integral After a considerable period of time, the temperature
equations. of all parts of the body equalizes and becomes equal to
The article developed a program that finds the control
the temperature of the medium (this is true for the case
(given temperature) at the left end of the rod, providing the
required temperature at the right end for a given fixed time. when the volume of the medium is much larger than the
The results of numerical calculations are output to files and volume of the body and its temperature practically does
presented in the form of tables and one-dimensional graphs not change with time) [3].
demonstrating the dynamics of temperature changes over
time. For the inverse boundary value problem of heat
Keywords -- linear integral equations, temperature, transfer, an approximate solution was constructed by the
thermal conductivity, thermal insulation, thermal physics quasi-inversion method and an order-exact estimate of
process, energy. the error of the constructed approximate solution was
obtained for one of the well-posedness classes of the
I. INTRODUCTION inverse boundary value problem [4].
Limited length rods are widely used in various For an ill-posed problem with inverse time of a
technical systems, including jet and hydrogen engines, semilinear differential-operator equation, a stable
gas generators, nuclear and thermal power plants, approximate solution is constructed and an estimate of its
processing industrial lines and spacecraft power plants. error is given [5].
These load-bearing elements operate under the
simultaneous influence of various heat sources. The method of spectral analysis established a
Therefore, an urgent problem is the development of criterion for the uniqueness of the solution of the inverse
special methods, computational algorithms and software problem of finding the initial condition. For these
packages that will allow us to investigate the steady problems, uniqueness, existence, and stability theorems
thermophysical state of rods of limited length under the for solutions were proved [6].
influence of heterogeneous heat sources.
II. STATEMENT OF THE PROBLEM
There are several methods for solving heat
conduction problems, including analytical, analog, In this paper, we consider a three-dimensional body,
numerical, graphical, and experimental. Each of these the constant cross section of which has the shape of a
methods is based on different forms of equations. The square. It is assumed that the left end of the rod coincides
experimental method is used when other methods fail, with the origin of coordinates and the heat transfer
and it allows you to determine the thermophysical coefficient is assumed to be constant over the entire
properties of the material, such as thermal conductivity surface of the rod. It is also assumed that heat of a fixed
and specific heat [1]. temperature is constantly supplied to the left side of the
rod.
To solve problems of heat conduction in solids of
complex shape, analytical and numerical methods are
used. Solutions are possible under known boundary
- 274 -
The distribution of heat in a homogeneous rod in the Let us cover the domain D with a uniform grid with
absence of heat sources is described by the three- steps ∆x, ∆y, and ∆z along the x, y, and z axes,
dimensional heat equation: respectively. We write the following difference-
differential approximation of equation (1):
𝜕𝑇 𝜕2 𝑇 𝜕2 𝑇 𝜕2 𝑇
= 𝑎2 ( + + ), (1)
𝜕𝑡 𝜕𝑥 2 𝜕𝑦 2 𝜕𝑧 2 𝑑𝑇𝑖,𝑗,𝑘 𝑇𝑖+1,𝑗,𝑘 −2𝑇𝑖,𝑗,𝑘 +𝑇𝑖−1,𝑗,𝑘 𝑇𝑖,𝑗+1,𝑘 −2𝑇𝑖,𝑗,𝑘 +𝑇𝑖,𝑗−1,𝑘
2
where 𝑎 – thermal diffusivity, = 𝑎2 ( + +
𝑑𝑡 ∆𝑥 2 ∆𝑦 2
𝑡0 ≤ 𝑡 ≤ 𝑡1 – period of time during which the
𝑇𝑖,𝑗,𝑘+1 −2𝑇𝑖,𝑗,𝑘 +𝑇𝑖,𝑗,𝑘−1
process of thermal conductivity of the rod is studied, + ), (5)
∆𝑧 2
x, y, z – spatial variables 0 ≤ 𝑦, 𝑧 ≤ 𝑙1 , 0 ≤ 𝑥 ≤ 𝑙2 ,
𝑙 𝑙 𝑙
𝑥ц , 𝑦ц , 𝑧ц – rod center: 𝑥ц = 2 , 𝑦ц = 1 , 𝑧ц = 1 ,
2 2 2 ̅̅̅̅̅̅̅̅̅̅̅̅
𝑡 ∈ [0, 𝑡1 ], 𝑖 = 2, 𝑛𝑥 − 1, 𝑗 = ̅̅̅̅̅̅̅̅̅̅̅̅
2, 𝑛𝑦 − 1,𝑘 = ̅̅̅̅̅̅̅̅̅̅̅
2, 𝑛𝑧 − 1,
𝑙1 – rod width and height,
∆x = 𝑙1 /𝑛𝑥, ∆y = 𝑙1 /𝑛𝑦, ∆x = 𝑙2 /𝑛𝑧, nxyz=nx*ny*nz.
𝑙2 – rod length,
D – parallelepiped {0≤ 𝑦, 𝑧 ≤ 𝑙1 , 0 ≤ 𝑥 ≤ 𝑙2 }, Г –
Here ∆x is the step and nx is the number of split points
border D, along the Ox axis, ∆y is the step and ny is the number of
𝑄 = {𝑥, 𝑦, 𝑧, 𝑡|(𝑥, 𝑦, 𝑧) ∈ 𝐷, 𝑡 ∈ [𝑡0 , 𝑡1 ]}. split points along the Oy axis, ∆z is the step and nz is the
The initial and boundary conditions are given as: number of split points along the Oz axis, the indices i, j,
𝜕𝑇
k are in x, y, and z coordinates, respectively [7].
| = 0, (2)
𝜕𝑛 Г
Initial conditions (2) take the form
𝑇(0, 𝑦, 𝑧, 𝑡) = 𝑢, 𝑢 ∈ 𝑈 (3)
𝑇1,𝑛𝑦ц,𝑛𝑧ц (0) = 𝑢,
𝑇(0, 𝑦ц , 𝑧ц , 𝑡1 ) = 𝑞. (4)
𝑛𝑥ц = [𝑛𝑥/2], 𝑛𝑦ц = [𝑛𝑦/2], 𝑛𝑧ц = [𝑛𝑧/2]. (6)
The set of allowable initial temperatures is denoted
by 𝑈 = {𝑢: 0 ≤ 𝑢 ≤ 𝑢𝑚𝑎𝑥 , 𝑢 ∈ 𝑅1 }, where 𝑢𝑚𝑎𝑥 is the The given final state will take the form
maximum allowable initial temperature, depending on
𝑇𝑛𝑥,𝑛𝑦ц ,𝑛𝑧ц (𝑡1 ) = 𝑞. (7)
the operating conditions and technical capabilities of the
device as a whole. This makes it possible to take into
account various operating scenarios and variations in Boundary conditions (2) are approximated by the
system parameters. following differential equations
𝑑𝑇1,𝑗,𝑘 𝑑𝑇𝑛𝑥,𝑗,𝑘 𝑑𝑇𝑖,1,𝑘 𝑑𝑇𝑖,𝑛𝑦,𝑘 𝑑𝑇𝑖,𝑗,1
However, the analytical solution of partial differential = 0, = 0, = 0, = 0, =
𝑑𝑡 𝑑𝑡 𝑑𝑡 𝑑𝑡 𝑑𝑡
equations, such as the heat equation, using modern
𝑑𝑇𝑖,𝑗,𝑛𝑧
mathematical methods is often difficult or even 0, = 0, (8)
𝑑𝑡
impossible. The heat conduction equation, presented as a
differential energy conservation equation for an isochoric 𝑖 = ̅̅̅̅̅̅
1, 𝑛𝑥 , 𝑗 = ̅̅̅̅̅̅
1, 𝑛𝑦,𝑘 = ̅̅̅̅̅̅
1, 𝑛𝑧.
process of heat transfer, relates temporal and spatial
temperature changes in a solid. We introduce a vector g of dimension nxyz and a
matrix A of dimension nxyz*nxyz.
To solve the problem of heat conduction, it is
necessary to take into account the initial and boundary The elements of the vector g are defined as follows:
conditions in combination with the geometric shape of
the body. The complete solution of the problem allows us g 𝑝 (t) = 𝑇𝑖,𝑗,𝑘 (t),𝑝 = (𝑖 − 1) ∗ 𝑛𝑦 ∗ 𝑛𝑧 + (𝑗 − 1) ∗
to find the temperature field in the body at any time t.
𝑛𝑧 + 𝑛𝑧
However, often there is a need to solve an inverse
problem, where it is required to find the control u so that 𝑖 = ̅̅̅̅̅̅
1, 𝑛𝑥 , 𝑗 = ̅̅̅̅̅̅
1, 𝑛𝑦,𝑘 = ̅̅̅̅̅̅
1, 𝑛𝑧.
the solution satisfies the given conditions at a specific
time 𝑡1 at the right end of the body. The elements of the matrix A are determined through
the coefficients of equation (5) and the given steps ∆x,
The analytical solution of partial differential ∆y, ∆z along the corresponding axes.
equations is a difficult task. However, there are many
numerical methods that allow one to obtain an Then the original problem (1)-(4) is reduced to the
approximate solution and apply them in practice. following Cauchy problem for a system of linear ordinary
differential equations:
In this paper, it is proposed to reduce the heat equation
to a system of integral equations and find the control u 𝑑g
= Ag, g(t 0 ) = g 0 , 𝑡 ∈ [0, 𝑡1 ] (9)
from the solution of an algebraic equation obtained from 𝑑𝑡
boundary conditions. This approach makes it possible to
obtain results suitable for practical use. Since the conditions of the Cauchy theorem are
satisfied for the system of differential equations (9), then
III. DEVELOPMENT OF A COMPUTATIONAL for any initial states 𝑔0 there is a unique solution 𝑥(𝑡) ∈
ALGORITHM 𝐶 1 [0, 𝑡1 ].
- 275 -
The solution of equation (8) can be represented as: Theorem 2. For any 𝜀 ≥ 0 there is a number 𝑟 such
1
𝑡
that for all 𝑘 > 𝑟 we have ‖𝐴𝑘 𝑡1𝑘 ‖ ≤ 𝜀.
𝑘!
g(𝑡) = g(0) + ∫0 𝐴g(𝜏)𝑑𝜏, 𝑡 ∈ [0, 𝑡1 ] (10)
Proof. We introduce the notation 𝛾 = ‖𝐴‖ 𝑡1 , 𝛾𝑘 =
Equation (10) is a particular case of the Volterra 1 𝑘 𝛾𝑘
𝛾 . Consider the series 1 + ∑∞
𝑘=1 . Let us calculate
integral equation of the second kind [8]: 𝑘!
𝛾𝑘+1 𝛾
𝑘!
𝜌 = lim = lim = 0. Because 𝜌 ≤ 1, then the
𝑡 𝑘→∞ 𝛾𝑘 𝑘→∞ 𝑘+1
𝑦(𝑡) = 𝑓(𝑡) + 𝜇 ∫0 𝐾(𝑡, 𝜏)𝑦(𝜏)𝑑𝜏, (11) constructed series is convergent. For any 𝜀 ≥ 0 there
𝛾
exists a number r such that for all 𝑘 > 𝑟 we have 𝑘+1 ≤
𝛾𝑘
where
𝜌 + 𝜀.
𝑦(𝑡) = g(𝑡); 𝑓(𝑡) = g(0); 𝐾(𝑡, 𝜏) = 𝐴; 𝜇 = 1. 𝛾 𝛾
Then ≤ 𝜀 or 𝑟 = [ ]. (Integer part of division).
𝑟+1 𝜀
Theorem 1. Equation (11) has a unique continuous This implies the validity of the assertion of the theorem.
solution for any initial states g 0 . This solution can be
1
found by the formula y(𝑡) = (𝐸 + 𝐴𝑡 + 𝐴2 ∗ 𝑡 2 … + Let 𝑝ц denote the index of the point corresponding to
2
1 the center of the rod section (1, 𝑛𝑦ц , 𝑛𝑧ц ).
𝐴𝑘 ∗ 𝑡 𝑘 + ⋯ ) ∗ 𝑓(𝑡) .
𝑘!
Then the initial condition takes the form
Proof. Since the matrix A and the vector of initial
conditions g 0 are constant, then all the conditions of g 𝑝ц (0) = 𝑢. (15)
Theorem 7.1 [8] are satisfied.
The given final state (6) takes the form
For equation (11), we denote the Volterra operator by
𝑡 g 𝑝ц (𝑡1 ) = 𝑞. (16)
𝑄𝑦 = ∫0 𝐾(𝑡, 𝜏)𝑦(𝜏)𝑑𝜏 .
We introduce the notation
Let us define the iterated kernel of the Volterra operator:
1
𝑛 𝑡 𝐶 = (𝐸 + ∑𝑟𝑘=1 𝐴𝑘 𝑡1𝑘 ) (17)
𝑄 𝑦= ∫0 𝐾𝑛 (𝑡, 𝜏)𝑦(𝜏)𝑑𝜏 , 𝑘!
𝑡 Then equation (14) can be written as
𝐾𝑛 (𝑡, 𝑠) = ∫𝑠 𝐾(𝑡, 𝜏)𝐾𝑛−1 (𝜏, 𝑠)𝑑𝜏,
g(𝑡1 ) = 𝐶 ∗ 𝑓(𝑡1 ) (18)
∞
𝑅(𝑡, 𝑠) = ∑ 𝐾𝑛 (𝜏, 𝑠). Substituting equation (18) into (15) and (16), we obtain
𝑛=1
g 𝑝ц (0) = 𝑢,
Let 𝑀 = sup |𝐾(𝑡, 𝑠)| ≤ ‖𝐴‖ . Then for repeated
0≤𝑡,𝑠≤𝑇
g 𝑝ц (𝑡1 ) = 𝑞. (19)
nuclei it is true
𝑀𝑛 ∗ (𝑡 − 𝑠)𝑛 From (19) we can express 𝑓(𝑡1 ) in terms of q:
| 𝐾𝑛 (𝑡, 𝑠)| ≤ 𝑓(𝑡1 ) = 𝐶 −1 𝑞.
(𝑛 − 1)!
Then the solution of equation (11) takes the form
Since 𝑓(𝑡) = g(0); 𝑡 ∈ [𝑡0 , 𝑡1 ], then the first equation
𝑡 (19) implies
у(𝑡) = 𝑓(𝑡) + ∫0 𝑅(𝑡, 𝜏)𝑓(𝜏)𝑑𝜏 . (12)
𝑢 = [𝐶 −1 𝑞]𝑝ц . (20)
Instead of у(𝜏) in (12) we substitute g(𝜏).
Theorem 3. If there is such a 𝑡1 > 0 that the matrix C
From here, due to the independence of the matrix A
defined by formula (17) is invertible, then control (20)
from time, we obtain
provides heating of the end central point of the rod to a
1 1 given temperature at time 𝑡1 .
g(𝑡) = (𝐸 + 𝐴𝑡 + 𝐴2 𝑡 2 … + 𝐴𝑘 𝑡 𝑘 + ⋯ ) ∗ 𝑓(𝑡) (13)
2 𝑘!
Proof. It follows from the previous discussions.
This implies the validity of the assertion of the theorem.
The following iterative solution algorithm is
1
For sufficiently large k, the quantity ‖𝐴𝑘 𝑡 𝑘 ‖ becomes proposed:
𝑘!
small and (13) can be rewritten as:
1. The desired time 𝑡1 and the temperature at the end
1 of the rod q are set.
g(𝑡) = (𝐸 + ∑𝑟𝑘=1 𝐴𝑘 𝑡 𝑘 ) ∗ 𝑓(𝑡), 𝑡 ∈ [0, 𝑡1 ] (14)
𝑘!
2. Based on the data of the original problem (1)-(3),
matrix A and matrix C are constructed.
- 276 -
3. Calculate the inverse matrix 𝐶 −1 .
4. According to formula (20), the temperature at the
initial moment of time at the left end of the rod (control)
is calculated, which is necessary for the rod to warm up
to the temperature at the right end in time 𝑡1 .
5. The resulting solution is output to Rezult.txt and to
the GrafX.txt file for subsequent visualization using the
MatLab software [9-10].
IV. NUMERICAL SOLUTION OF PROBLEMS FOR SPECIFIC
INITIAL DATA
An innovative program was developed that can
effectively solve the inverse control problem for the
equation of thermal conductivity of a square bar. It is
based on the application of a system of linear integral
equations, and an appropriate algorithm has been Fig 1. Graph of temperature distribution along the center of the rod in
developed for its solution. the X direction from the origin [11]
A unique feature of this program is its ability to Due to the large temperature difference from 0 to 200
present the results of numerical calculations in a degrees, Figure 2 shows a 2 graph of temperature
convenient format. The results obtained over time can be distribution along the center of the rod in the X direction
visualized in the form of one-dimensional graphs. This with a one-step offset from the origin in dynamics.
allows researchers and engineers to analyze data more
deeply and make informed decisions based on the results.
In addition, the program saves the results of
numerical calculations in several files, which ensures
their safety and the possibility of subsequent use. This is
important for further research, as well as for documenting
the results obtained and sharing information with other
specialists in the field of thermal conductivity.
In general, the developed program is a powerful tool
for solving the inverse control problem in the equation of
thermal conductivity of a square bar. It simplifies the
process of numerical simulation, analysis and data
visualization, which contributes to a more accurate
understanding of thermal conductivity and helps to
develop effective control strategies in various
applications. The calculations were carried out with the
following initial data:
𝑙1 = 1.0; 𝑙2 = 10.0; ∆𝑡 = 0.01; Fig 2. Graph of temperature distribution along the center of the rod in
nx =10; ny= 6; nz= 6; 𝑞 = 200. the X direction with a one-step offset from the origin [11]
Below are the results of experimental calculations in CONCLUSION
the form of figures 1-2.
The article explores the following two problems:
Figure 1 shows a graph of the distribution of
temperature along the center of the rod in the X direction 1. Direct problem. For a given fixed temperature 𝑢 ∈
from the origin in dynamics. 𝑈, find the time 𝑡1∗ ≤ 𝑡1 during which the center of the
right end of the rod heats up to a temperature q.
2. Inverse problem. Determine the required
temperature value 𝑢 ∈ 𝑈 so that the center of the right
end of the rod is heated to a temperature q for a given
time 𝑡1∗ ≤ 𝑡1 .
An iterative algorithm was developed for solving the
direct problem, which makes it possible to estimate the
time required to heat the right end of the rod to a given
temperature q. This algorithm is an effective tool for
- 277 -
modeling the process of heat conduction and allows you [5] G.R. Jaffe, V.W. Brar, M.G. Lagally, M.A. Eriksson “A simple
numerical method for evaluating heat dissipation from curved
to get reliable predictions of the heating time. wires with periodic applied heating” Applied Physics Letters,
2021. – 119(16). – DOI: 10.1063/5.0059648.
To solve the inverse problem, an iterative algorithm [6] J.Amrit, K.Nemchenko, T. Vikhtinskaya “Effect of diffuse
based on the dichotomy method was proposed. This phonon boundary scattering on heat flow “ Journal of Applied
algorithm allows you to determine the parameters Physics, 2021. – 129(8). – DOI: 10.1063/5.0036935.
necessary to achieve a given temperature at the right end [7] D. F. Sikovsky. Methods of Computational Thermal Physics.
Novosibirsk: Novosib. state un-ty, 2013. – 98 p.
of the rod. It provides an opportunity to effectively
[8] A. B. Vasil’ev and N. A. Tikhonov, Integral Equations. – M.:
control the process of heat conduction based on the MSU, 1989. – 156 p.
obtained data and estimates. [9] I.E. Anufriev, A.B. Smirnov, E.N. Smirnova. Matlab 7. – St.
Petersburg: BHV-Petersburg, 2005. – 1104 p.
As a result of the development of two programs for [10] V.P. Dyakonov. Matlab 6.0/6.1/6.5/6.5+SP1+Simulink 5/5.
modeling the temperature propagation along the rod, Processing of signals and images. – M.: SOLON-Press, 2005. –
accurate numerical calculations were obtained. The 592 p.
results of numerical calculations over time were [11] K.B. Begaliyeva “Development of an automated control system
presented in the form of tables and graphs, which makes for the thermophysical state of the rod during operation”
Dissertation. PhD in spec. 6D070200, Almaty, 2023. – 96 p.
it possible to visualize and analyze the data in a more
[12] T. Mazakov, W.Wójcik, Sh.Jomartova, N.Karymsakova,
convenient way. G.Ziyatbekova, A.Tursynbai “The Stability Interval of the Set of
Linear System” INTL Journal of Electronics and
Two programs have been developed for finding the Telecommunications. – 2021. – Vol. 67, N. 2. – P.155-161. DOI:
temperature distribution along the rod, which puts the 10.24425/ijet.2021.135958.
results of numerical calculations into several files. The [13] N. Isimov, T. Mazakov, O. Mamytbekov, G. Ziyatbekova
“Application of fuzzy and interval analysis to the study of the
results of numerical calculations in dynamics (over time) prediction and control model of the epidemiologic situation”
are displayed in the form of a table and displayed in the Journal of Theoretical and Applied Information Technology,
form of graphs. Pakistan, 2018. – Vol. 96, – Issue 14, – рр. 4358-4368.
The obtained results of numerical calculations are
consistent with the experimental data, which confirms the
reliability of the developed programs. For ease of use and
exchange of information, the results are also saved to text
files that can be processed using the MatLab system. The
corresponding program was developed for plotting one-
dimensional graphs of temperature dynamics.
One of the promising directions for further research
in this area is the use of interval mathematics for a more
detailed study of the heat equation [12-13]. Interval
mathematics allows you to take into account the
uncertainty and errors of the input data and model
parameters, which can lead to more accurate results and
improve the reliability of modeling.
All work was carried out with the support of the funds
of the Research Institute of Mathematics and Mechanics
at the Al-Farabi Kazakhstan National University. This
provides additional assurance of the quality and
reliability of the results and highlights the significance of
this study.
REFERENCES
[1] D.S. Karpovich, O.N. Susha, N.P. Korovkina, V.P.Kobrinets
“Analytical and numerical methods for solving the heat equation”
Proceedings of BSTU. Physical and Mathematical Sciences and
Informatics, 2015, No. 6. pp.122-127.
[2] B.A.Voronenko, A.G.Krysin, V.V. Pelenko, O.A. Tsuranov
Analytical description of the process of non-stationary heat
conduction. – St. Petersburg: NRU ITMO; IKiBT, 2014. – 48 p.
[3] E.M. Dede, Z. Yu, P. Schmalenberg, H. Iizuka “Thermal
metamaterials for radiative plus conductive heat flow control”
Applied Physics Letters, 2020. – 116(19). – DOI:
10.1063/5.0007574.
[4] D.Visaria, A.Jain “Machine-learning-assisted space-
transformation accelerates discovery of high thermal conductivity
alloys” Applied Physics Letters, 2020. – 117(20). – DOI:
10.1063/5.0028241.
- 278 -
Switching Loss Analysis by Interleaving
Method of 3-LV Boost Converter with Coupled
Inductor
Hae In Kim1 , Hag Wone Kim1*, Kwan Yuhl Cho1 and Jeong Won Kang1
1
Department of Control and Instrumentation Engineering, Korea University of Transportation, Chungju,Korea
Email: [email protected]
Abstract— Parallel configuration of power conversion
devices has been widely adopted as a relatively simple method
for achieving high power. In addition to enabling large-scale
integration, parallel configuration offers benefits such as
reducing output ripple and minimizing passive component sizes
through interleaving control between modules. However,
caution must be exercised in its application due to potential
issues such as circulating currents and unbalances in current or
voltage. This paper analyzes and compares two different
interleaving methods in terms of semiconductor switching
device losses.
Keywords—Multi-Level Converter, Interleaving, Parallel
Operation, Circulation Current, Power Loss
I. INTRODUCTION
As global carbon regulations gradually spread, interest in
solar power generation for large-scale renewable energy
generation is increasing. Solar PV is currently the world's Fig. 1: Parallel 3-LV Boost Converter with Coupled Inductor
fastest-growing energy technology, with 50 GW of new
capacity being added annually [1]. In order to increase the A. Comparison of interleaving methods
efficiency according to the gradually increasing power
generation, research on multi-level converters is being Earlier, it was mentioned that TLBC generally operates
actively conducted. Among them, TLBC (3-LV Boost the upper and lower switches with a phase difference of 180
Converter) has the simplest form among non-isolated multi- degrees. However, in the two-phase parallel type TLBC, a
level converters. TLBC has two boost converters connected in total of four switches, the upper and lower switches of one
series, and generally operates with a phase difference of 180 module and the upper and lower switches of two modules, are
degrees between the upper and lower switches. The switching operated with a phase difference of 90 degrees. At this time,
method can reduce the current and voltage ripple of the it can be divided into an N-type Interleaving method and a Z-
converter. However, there is one option for the switching type Interleaving method according to the order in which the
method of parallel TLBC. It can be divided into N-type switches are operated. The N-type Interleaving method is a
Interleaving and Z-type Interleaving according to the method method in which the switch in one module is operated and
of configuring the phase difference between the modules and then the switch in the next module is operated, and the Z-type
the phase difference between the upper and lower switches. Interleaving method is a method in which the upper switch of
The two interleaving methods are compared in various aspects, each module is operated and the lower switch is operated in
such as inductor current ripple, output current ripple, and this order. Fig. 2 shows the waveform of the switching signal
output filter capacitor current. Therefore, in this paper, the two
according to the interleaving method.
interleaving methods are compared in terms of the switching
Since only the timing at which each switch operates is
loss of the semiconductor switching device and verified
through the Psim simulation program. different according to the interleaving method, since the
II. PARALLEL TYPE 3-LV BOOST CONVERTER WITH
COUPLED INDUCTOR
Multi-level converters, which are mainly applied to high-
power systems, prefer parallel configuration of power
conversion modules due to the advantage of convenient
capacity increase. In addition, it is an effective configuration
for reducing output current and voltage ripple through
interleaving control that operates by giving a phase difference (a) N-type Interleaving (b) Z-type Interleaving
between modules. Fig. 1 shows the circuit diagram of the
parallel type TLBC. However, there are several caveats to the Fig. 2: Switching method by interleaving method
parallel configuration and control of TLBC.
- 279 -
average characteristics are the same, the steady-state
characteristics by the state averaging technique are the same,
but affect the circulating current and the output filter
capacitor current.
B. Neutral point connection of output filter capacitor.
When configuring TLBC as a parallel module, the neutral
point of the output filter capacitor between each module can
be connected or separated, and the current of the output
capacitor is different accordingly. The configuration of
TLBC with the neutral point connected is advantageous in (a) N-type Interleaving
terms of the lifespan of the output filter capacitor and is easy
to control.[2] Therefore, this paper discusses the configuration
in which the neutral point of the output filter capacitor of the
parallel module TLBC is connected. This affects regardless
of the interleaving scheme.
C. Necessity of Bottom Inductor and Coupling Inductor
In parallel type TLBC, the lower inductor must be applied to
balance the impedance of the upper module and the lower
module and to prevent the short circuit of the output capacitor.
A coupled inductor refers to an inductor in which magnetic (b) Z-type Interleaving
fluxes of two or more inductors are combined into one
Fig. 4: Inductor Current Waveform by Interleaving Method
common core in a similar form to a transformer. The coupled
inductor can reduce the volume and weight of the inductor
through magnetic flux offset. In this paper, the upper and
lower ends of the TLBC are coupled to each other in the form Equation (1) represents the switching loss of the
of an inductor, and the coupling inductor of the TLBC has a semiconductor switching device. P_SW is the switching loss,
coupling coefficient limit, so a coupling inductor with a E_on is the switch-on loss, E_off is the switch-off loss, and
coupling coefficient of about 0.6 was applied. f_sw is the switching frequency
III. COMPARISON OF SWITCHING LOSS ACCORDING 𝑃 = 𝐸 +𝐸 ∗𝑓 (1)
TO INTERLEAVING METHOD
It has been mentioned above that interleaving control giving
a phase difference between switches can be applied to reduce Table 1 Simulation Parameter
output current and voltage ripple.
Parameter Mark Value Unit
Power 𝑃 16*2 kW
Input Voltage 𝑉 800 V
Input Current 𝐼 20*2 V
Switching
𝑓𝑠𝑤 18.6 kHz
Frequency
Self Inductance 𝐿 470 uH
Mutual Inductance 𝐿 470*0.6 uH
Inductor’s ESR 𝑅 0.1 Ω
Output Filter
𝐶 1200 uF
Capacitor
Output Voltage 𝑉 1200 V
Duty 𝐷 0.386 -
Fig. 3: Parallel TLBC's PSIM Schematic
- 280 -
Fig.5 shows the loss of each switching device according to IV. CONCLUSION
the interleaving method through simulation and In this paper, the switching loss of each switch of the parallel
calculation.According to Fig.5, it can be seen that the overall type TLBC for each interleaving method was
loss of the semiconductor switching device does not change analyzed. The interleaving method is divided into an N-type
significantly depending on the interleaving method. However, interleaving method and a Z-type interleaving method
in the case of Z-type interleaving, it can be seen that the loss according to the phase delay order of each switch. Although
is biased to a specific switch, which has a bad effect on the the sum of switching losses of all switches for each
stability of the overall system. interleaving method is almost the same, it was found that Z-
type interleaving has different losses between switches,
which is disadvantageous in terms of the lifespan of the
overall power conversion module. The effect of the intiliving
method has been compared in various aspects. In the future,
it is necessary to compare the overall power
conversion efficiency and input/output characteristics.
REFERENCES
[1] Breyer, Christian, et al. "On the role of solar photovoltaics in global
energy transition scenarios." Progress in Photovoltaics: Research and
Applications 25.8 (2017): 727-745.
[2] Lee, Kang-Mun, et al. "3-level boost converter having lower inductor
for interleaving operation." The Transactions of the Korean Institute of
Power Electronics 26.2 (2021): 96-105.
(a) N-type Interleaving (b) Z-type Interleaving
Fig. 5: Switching loss per switch according to interleaving
method
- 281 -
August 22(TUE) - 25(FRI), 2023
Oral Session 14
14:00~15:40 Thu, August 24, 2023
Room : A435
Chair: Kuanysh Alipbaev (AUPET, Kazakhstan)
- 282 -
Finding trends in the development of narrow
beam inter-satellite communication antennas for
Cubesat satellites using patent analysis
*Note: Sub-titles are not captured in Xplore and should not be used
Zh.Suimenbayeva1,2, B.Aitchanov1, K.Alipbayev*1, A.Komekbayev1, A. Aden1, A.Mukasheva1
1
Almaty University of Power Engineering and Telecommunication named G.Daukeev, Kazakhstan
2
Satpayev University, Kazakhstan
Email: [email protected]
Abstract— Currently, the interest of the scientific and between satellites. Antennas can be of various types, including
commercial community in the space industry is increasingly mono-field, dipole, patch antennas, and others, depending on
turning to CubeSat-type nanosatellites. One of the key aspects communication requirements and other mission parameters.
in the design of such nanosatellites is the choice of antenna One of the main problems in the design of antennas for
parameters that provide communication between the satellite Cubesat is the need to minimize their size and weight, as well
and the Earth or between satellites. Antennas for CubeSat can as to ensure their reliability and efficiency in outer space. Over
be of various types, such as monopole, dipole, patch antennas, time, with the development of technologies and materials,
and others, depending on communication requirements and antennas for Cubesat are becoming more advanced, which
mission characteristics.
makes it possible to improve the quality of communication
Of particular interest is the use of narrow-beam (narrow- and expand the possibilities of using these satellites. Of
beam) antennas in nanosatellites, which provide a more particular interest is the use of narrow-beam (narrow-
accurate and efficient use of energy, which allows to increase the directional) antennas in nanosatellites, which provide more
communication range and improve the quality of data accurate and efficient use of energy, which makes it possible
transmission. This article discusses the process of conducting a to increase the communication range and improve the quality
patent search on the WIPO, FIPS and Google Patent databases of data transmission. As CubeSat technology continues to
for antenna systems used in CubeSat-type nanosatellites. evolve, protecting intellectual property rights through patents
is becoming increasingly important. In this article, we will
The analysis of patent bases allows to identify trends in the look at the process of conducting a patent search for narrow-
development of antenna system technologies for CubeSat, as angle antenna systems for nanosatellites such as CubeSats. It
well as to identify key innovations and innovative approaches should be noted that conducting a patent search for narrow-
used in this area. The study is being conducted to promote the angle antenna systems for CubeSats is an important step in the
development of more efficient and advanced antenna systems patent application process. This can help identify previously
for small spacecraft such as the CubeSat, which improves patented inventions that are similar to a new invention and
communications and ensures optimal performance of provide valuable information about the state of the art in the
nanosatellites in outer space.
field. By conducting a thorough patent search, inventors can
Keywords—patent search, inter-satellite communication,
increase the chances of successfully obtaining a patent for
cubesat, antenna sytems their invention..
II. TYPES OF ANTENNAS FOR CUBESAT
I. INTRODUCTION
Currently, the eyes of the scientific and commercial Despite the limited size and mass of Cubesats, there are many
community of the space industry are increasingly turned types of highly directional antennas that can be used in these
towards the use of nanosatellites such as Cubesat. Cubesat is nanosatellites to provide inter-satellite communications. Let's
the standard for the design and manufacture of small satellites. consider the main ones
The standard was developed in the early 2000s at California A. Patch Antenna
Polytechnic University and Stanford University. Cubesat
belongs to the class of so-called nanosatellites and usually has These antennas are flat metal plates that can be easily
a Cubesat, a standard for the design and manufacture of small integrated into the Cubesat chassis. They provide a narrow
satellites based on a cubic module known as a "Unit". One beam, making them ideal for inter-satellite communications.
Cubesat unit has standard dimensions of 10x10x10 For example, authors from School of Electronics and
centimeters (or 1 liter of volume) and a weight of no more than Telecommunications [1] introduced a metasurface-based
1.33 kg. It is important to note that the advantages of Cubesats, antenna designed for CubeSat applications in the S-band. The
including their low cost, small size and rapid deployment, antenna consists of a single-feed slotted patch and a
make them an ideal solution for many tasks in the field of metasurface made of periodic metallic plates. It achieves a
space activities. One of the key aspects of designing a Cubesat wide operating bandwidth and circularly polarized radiation.
nanosatellite is the choice of antenna parameters, since it The prototype was successfully tested, showing good
provides communication between the satellite and the Earth or radiation efficiency, symmetric pattern, and a high front-to-
- 283 -
back ratio. The design is compatible with standard CubeSat missions, demonstrating their successful use for
structures, making it suitable for various CubeSat applications communication, Earth observation, and combined GNSS-R
in the S-band. and microwave radiometry payloads. The deployability and
compactness of helix antennas make them a suitable choice
for CubeSats, addressing the challenges posed by the limited
space and stowage requirements of these small satellites.
Figure 2 -Extruded view of the Nadir Antenna and
Deployment Subsystem [11]
2. Dipole Antennas: Dipole antennas are simple, half-
wavelength wire antennas that can be folded for launch and
then deployed to their full length in space. They are
commonly used in CubeSats operating at lower frequencies
(e.g., S-band, L-band).
3. Yagi-Uda Antennas: Yagi-Uda antennas are
directional antennas with multiple elements. They can be
designed with collapsible elements for compact stowage and
then expanded in space to achieve their operational
configuration.
4. Planar Phased Arrays: Planar phased arrays are flat
antenna arrays with electronically steerable beams. They can
be made deployable by using flexible materials and folding
Figure 1- A fabricated sample of the antenna [1] techniques.
5. Origami Antennas: Origami antennas for CubeSats
New design of a dual aperture-coupled microstripline-fed refer to innovative antenna designs that utilize the principles
antenna array with circular polarization for CubeSat of origami, the art of paper folding, to create compact and
applications in the C-band proposed by a group of Portuguese efficient antennas suitable for small satellite missions. These
scientists [2] . The antenna array is designed to be mounted on antennas can operate in various frequency bands, including
CubeSat deployable wings along with solar panels, Terahertz (THz), L-band, S-band, C-band, and more,
maximizing space utilization and improving the link budget. depending on the mission requirements. They can be
The antenna exhibits good isolation between Tx/Rx signals designed for different types of polarization, such as circular
and offers a lower profile approach. It offers an innovative polarization or linear polarization, and offer different
solution for high-performance antenna systems in CubeSats, radiation patterns to suit specific applications like Earth
overcoming limitations of single-element approaches. An observation, communication, and remote sensing.[18-21]
interesting solution was proposed by the authors [3] for
wideband microstrip patch antenna for CubeSat applications
in the Ka-band. The antenna design includes a parasitic
element and slots to enhance impedance bandwidth and gain
performance. It demonstrates a wide -10dB impedance
bandwidth and excellent cross-polarization discrimination in
the E-plane, making it suitable for CubeSat applications in the
Ka-band.
Articles [1-10] cover various innovative patch antenna
designs and solutions tailored for CubeSat applications. They
address challenges related to size, weight, performance, and
compatibility, providing valuable contributions to the
advancement of small satellite communication systems.
B. Deployable antennas
For CubeSats are a crucial component in space missions
where compactness and efficient deployment are essential.
They allow the antennas to be stowed in a compact form
during launch and then deploy in space to achieve the desired
Figure 3 - Fabricated antenna on rigid FR4 board: (a) Planar, (b) folded
operational configuration. Here are some common types of [20]
deployable antennas used in CubeSats:
1. Helical Antennas: Helical antennas are wire
In modern scientific research, they are actively engaged in the
antennas wound in a helix shape. They are lightweight,
design and development of new models of communication
compact, and offer circular polarization. During deployment,
antennas for small spacecraft such as Cubesat. Such antennas
the helical structure uncoils to form the antenna. Helical
require innovative approaches, given the limited size and
antennas are often used for communication in UHF and VHF
weight of the cubesat, and are aimed at providing efficient
frequency bands. A review of articles [11-17] showed the
inter-satellite communications.
versatility and effectiveness of helix antennas in CubeSat
- 284 -
III. PATENT SEARCH ANALYSIS In [22], a patent is given that describes the method involves
An analysis of data from the World Intellectual Property identifying satellites to form a temporary communication
Organization (WIPO) database for the keywords "antenna group, determining available radio resources for their
cubesat" and "cubesat inter-satellite communications" for the communications, and then allocating a subset of these
period from 2011 to 2023 shows a clear increase in the number resources for specific use within a defined timeframe.
of publications in this area. The period from 2015 to 2022
stands out in particular, where there was a noticeable jump in
the number of patents (Figure 4).
The largest number of patents were registered in the US,
China and India (Figure 5), with Chinese inventors holding
more patents related to the mechanical component of cubesat
antennas, including opening methods and new designs. On the
other hand, American scientists have more patents related to
new principles and methods of organizing inter-satellite
communications.
This can be attributed to the successful experience of
launching the Starlink fleet of satellites, which has stimulated
interest and research in the field of inter-satellite
communications. It should also be noted that scientific
research in this area actively interacts with industry and Figure 6 - Illustration of ways in which satellites may need and utilize
practical applications, such as space missions and radio resources fРor inter-satellite communications
communication systems. This contributes to the dynamic
development of technologies and innovations in the field of The analysis of the FIPS Russian patent database resulted
cubesat antenna systems and ensures their more efficient in the identification of a number of patents dated from 2015 to
operation in the space environment.Keep your text and 2022. Based on the results of the patent search, several key
graphic files separate until after the text has been formatted trends in inter-satellite communications can be identified: an
and styled. Do not use hard tabs, and limit use of hard returns increase in interest in communication technologies between
to only one return at the end of a paragraph. Do not add any space vehicles, which is reflected in an increase in the number
kind of pagination anywhere in the paper. Do not number text of patent documents for the specified period. There is also an
heads-the template will do that for you. improvement in the efficiency and speed of data transmission
between satellites, which is likely due to the development of
new modulation methods and improvements in antenna
systems. Engineers and scientists are striving to create
compact and lightweight communication systems, as
evidenced by the weight limit of up to 50 kg in some patents.
Finally, improving the reliability of communications is likely
to become relevant in connection with the expansion of the
use of inter-satellite communications in various space
missions, and relevant patent documents are devoted to this.
Patent [23] is of the greatest interest. The proposed method is
based on the global transmission of information between near-
Earth correspondents through the spacecraft cluster of
information transmission "Kosmoset" using inter-satellite
radio and laser communication channels with simultaneous
Figure 4 - Patent search results analysis by years (WIPO)
radio monitoring of the Earth's surface, airspace, ionosphere
and surrounding space, radar and optical monitoring of the
underlying surface of the Earth, as well as monitoring the
Earth's gravitational field with the help of onboard measuring
equipment of the corresponding spacecraft clusters included
in the system.
Figure 5. Patent search results analysis by countries (WIPO)
- 285 -
Figure 6 - principle of the prototype Capitalize only the first word in a paper title, except for proper
nouns and element symbols.
For papers published in translation journals, please give
the English citation first, followed by the original foreign-
IV. CONCLUSION language citation [6].
After conducting a patent analysis using the databases of [1] Ta S.X., Nguyen V.C., Nguyen K.K., Dao-Ngoc C.Single-feed slotted-
WIPO (World Intellectual Property Organization), FIPS (US patch antenna loaded with metasurface for CubeSat applications (2021)
Federal Patent and Trademark Office) and Google Patents, we International Journal of RF and Microwave Computer-Aided
can conclude that over the past 5 years there has been a Engineering, 31 (4), DOI: 10.1002/mmce.22560
noticeable increase in new inventions in the field of satellite [2] Kaur R., Mayank, Tandon T., Schumacher J., Sundaramoorthy G.,
Sahoo U.K.S-Band Patch Antenna System for CubeSat Applications
antenna systems for satellites CubeSat type. (2021) Advances in Intelligent Systems and Computing, 1363 AISC,
An important trend is the development and pp. 102 – 112 DOI: 10.1007/978-3-030-73100-7_8
implementation of new design solutions for antenna systems. [3] Hammoumi M.E., Idrissi N.E.A.E., Raad R., Theoharis P.I., Tubbal
Inventors are actively working to improve the shapes and F.A Wideband Compact Patch Antenna for Ka-band and CubeSat
structures of the antennas to improve their performance, Applications (2021) 2021 15th International Conference on Signal
compactness and efficiency in the limited space of the Processing and Communication Systems, ICSPCS 2021 – Proceedings
DOI: 10.1109/ICSPCS53099.2021.9660306
CubeSat.
Another important area of research is the development of [4] Priscila G., Nur L.O., Edwar A Design of ADS-B Receiver Antenna
using Truncated Circular Microstrip Patch in Cubesat Form Factor
new materials for antenna systems. The use of new materials (2022) 2022 IEEE International Conference on Aerospace Electronics
can significantly improve the characteristics of antennas, such and Remote Sensing Technology, ICARES 2022 – Proceedings DOI:
as resistance to extreme space conditions, reduction in mass 10.1109/ICARES56907.2022.9993600
and increase in radiation efficiency. [5] Abulgasem S., Tubbal F., Raad R., Theoharis P.I., Liu S., Khan M.U.A.
The integration of modern antenna systems into the A wideband metal-only patch antenna for cubesat (2021) Electronics
cubesat includes the development of specialized control (Switzerland), 10 (1), art. no. 50, pp. 1 – 13 DOI:
10.3390/electronics10010050
systems for electronic components, as well as appropriate
[6] Mubasshir Hossain M., Latif S.I., Spencer E.A.Hybrid Perturbations in
algorithms and software to ensure fine tuning and orientation Stacked Patch-Ring Circularly Polarized Microstrip Antennas for
of antennas in space. CubeSat Applications (2022) IEEE Aerospace and Electronic Systems
In general, modern cubesat inter-satellite communication Magazine, 37 (3), pp. 24 – 31 DOI: 10.1109/MAES.2022.3141436
antenna systems are complex technical systems that combine [7] Abulgasem S., Tubbal F., Raad R., Theoharis P.I.Dual Band All-Metal
innovative design solutions, new materials and advanced T-Shaped Patch Antenna for X/Ku-band CubeSat Applications (2022)
communication technologies. These systems are aimed at 2022 5th International Conference on Communications, Signal
Processing, and their Applications, ICCSPA 2022 DOI:
providing efficient, reliable and high-speed communication 10.1109/ICCSPA55860.2022.10019252
between cubesat satellites, which is essential for successful [8] Muntoni G., Montisci G., Melis A., Lodi M.B., Curreli N., Simone M.,
space missions and scientific research. Tedeschi G., Fanti A., Pisanu T., Kriegel I., Athanassiou A.,
Mazzarella G.A Curved 3D-Printed S-Band Patch Antenna for Plastic
FUNDING CubeSat (2022) IEEE Open Journal of Antennas and Propagation, 3,
This research has been funded by the Science Committee of pp. 1351 – 1363 DOI: 10.1109/OJAP.2022.3222454
the Ministry of Science and Higher Education of the Republic [9] Hammoumi M.E., El Amrani El Idrissi N., Raad R., Theoharis P.I.,
Tubbal F., Abulgasem S. Ultra Wideband Dual Circularly Polarized
of Kazakhstan AP14869120 "Design and creation of highly Patch Antenna for 5G and CubeSat Applications (2022) Proceedings -
directional satellite-to-satellite antenna systems with the best 2022 9th International Conference on Wireless Networks and Mobile
and optimal characteristics for the Cubesat form factor." Communications, WINCOM 2022 DOI:
10.1109/WINCOM55661.2022.9966460
[10] Ramezanzadehyazdi S., Shafai C., Isleifson D., Shafai L., Ferguson P.
ACKNOWLEDGMENT (Heading 5) An Optically Transparent Meshed Patch Antenna With Enhanced
Bandwidth for CubeSat Applications (2022) 2022 IEEE International
The authors would like to thank the organizing committee of Conference on Wireless for Space and Extreme Environments, WiSEE
2022, pp. 44 – 48 DOI: 10.1109/WiSEE49342.2022.9926940
ICEF2023 for the giving opportunity to publish the paper.
[11] Drobczyk M., Jetzschmann M., Nohka F., Budroweit J. A low-cost
Nanosatellite Approach: AISat-1 after 5 Years in Orbit (2020) 2020
REFERENCES IEEE Space Hardware and Radio Conference, SHaRC 2020, art. no.
The template will number citations consecutively within 9034063, pp. 5 – 8 DOI: 10.1109/SHaRC47220.2020.9034063
brackets [1]. The sentence punctuation follows the bracket [2]. [12] Fernandez L., Sobrino M., Ruiz-De-azua J.A., Calveras A., Camps
Refer simply to the reference number, as in [3]—do not use A.Design of a Deployable Helix Antenna at L-Band for a 1-Unit
CubeSat: From Theoretical Analysis to Flight Model Results(2022)
“Ref. [3]” or “reference [3]” except at the beginning of a Sensors, 22 (10), art. no. 3633 DOI: 10.3390/s22103633
sentence: “Reference [3] was the first ...” [13] Salem N.A.-D.M., Hussein A.I., Mabrook M.M.A proposed Array of
Number footnotes separately in superscripts. Place the Quadrifilar Helix Antenna for CubeSat applications (2023) 20th
International Learning and Technology Conference, L and T 2023, pp.
actual footnote at the bottom of the column in which it was 171 – 175 DOI: 10.1109/LT58159.2023.10092351
cited. Do not put footnotes in the abstract or reference list. Use [14] Meirambekuly N., Karibayev B.A., Namazbayev T.A., Ibrayev G.-
letters for table footnotes. G.A.E., Orynbassar S.O., Ivanovich S.A., Temirbayev A.A. A High
Gain Deployable L/S Band Conical Helix Antenna Integrated With
Unless there are six authors or more give all authors’ Optical System for Earth Observation CubeSats (2023) IEEE Access,
names; do not use “et al.”. Papers that have not been 11, pp. 23097 – 23106 DOI: 10.1109/ACCESS.2023.3253556
published, even if they have been submitted for publication, [15] Sureda M., Sobrino M., Millan O., Aguilella A., Solanellas A., Badia
should be cited as “unpublished” [4]. Papers that have been M., Munoz-Martin J.F., Fernandez L., Ruiz-De-Azua J.A., Camps A.
accepted for publication should be cited as “in press” [5]. Design and Testing of a Helix Antenna Deployment System for a 1U
- 286 -
CubeSat (2021) IEEE Access, 9, art. no. 9416495, pp. 66103 – 66114
DOI: 10.1109/ACCESS.2021.3075660
[16] Fernandez L., Sobrino M., Milian O., Aguilella A., Solanellas A.,
Badia M., Munoz-Martin J.F., Ruiz-de-Azua J.A., Sureda M., Camps
A.Deployment mechanism for a L-band helix antenna in 1-Unit
Cubesat (2022) Acta Astronautica, 196, pp. 394 – 399 DOI:
10.1016/j.actaastro.2020.09.005
[17] Alqaraghuli A.J., Singh A., Jornet J.M. Case Study on Deployable
Origami Antennas for Terahertz CubeSat Networks(2023) 2023
International Applied Computational Electromagnetics Society
Symposium, ACES-Monterey 2023 DOI:
10.23919/ACES57841.2023.10114788
[18] Velez C.A., Kaddour A.-S., Georgakopoulos S.V., Ynchausti C.,
Magleby S., Howell L.L. A Novel Miura-Ori Origami Reflectarray
Antenna for CubeSat Applications (2021) 2021 IEEE International
Symposium on Antennas and Propagation and North American Radio
Science Meeting, APS/URSI 2021 - Proceedings, pp. 1759 – 1760
DOI: 10.1109/APS/URSI47566.2021.9704247
[19] Hwang M., Kim G., Kim S., Jeong N.S. Origami-Inspired Radiation
Pattern and Shape Reconfigurable Dipole Array Antenna at C-Band for
CubeSat Applications
[20] (2021) IEEE Transactions on Antennas and Propagation, 69 (5), art. no.
9234044, pp. 2697 – 2705 DOI: 10.1109/TAP.2020.3030908
[21] Alqaraghuli A.J., Singh A., Jornet J.M. Compact Two-Stage Origami
Horn Antenna for Terahertz CubeSat Networks (2022) 2022 IEEE
International Symposium on Antennas and Propagation and USNC-
URSI Radio Science Meeting, AP-S/URSI 2022 - Proceedings, pp.
1102 – 1103 DOI: 10.1109/AP-S/USNC-URSI47032.2022.9886236
[22] Patent WO2021058111 - Method, arrangement, and computer
program product for inter-satellite links to reuse the resources of feeder
and service links. 2019,
https://patentscope.wipo.int/search/en/detail.jsf?docId=WO20210581
11
[23] Galkevich A. I., Fateev V. F. A method for creating a global
information environment in near-Earth space and a multifunctional
space information system "Paradigm" based on a network of low-orbit
spacecraft for its implementation. Patent No. 2701415,2019, https://
www.fips.ru/iiss/document.xhtml?faces-
redirect=true&id=77f3e8413b979b22aa33fe23068ae52cJ.
- 287 -
A method for improving the accuracy of navigation receivers using GPS as an
example.
Khussainov Ansar ([email protected]), Ibraim Merey ([email protected])*.
Gumarbek Daukeev Almaty University of Energy and Telecommunications, Almaty, 050013, Kazakhstan
Correspondent author: Ibraim Merey ([email protected])*.
Abstract. In this paper, the analysis of the development of a program to increase the accuracy coefficient of
raw data is carried out. GPS modules allow errors of 1-7 m on average. Based on these data, more accurate devices
were obtained. Measurements and comparative analyses of the data obtained with the developed method were carried
out. The main objective of the study is to improve the accuracy of GPS modules, to find a more profitable option for
filtering data to reduce the measurement error.
Keywords. GPS, Kalman filter, algorithm, ecology, navigation.
Introduction
The data used comes from the GPS receiver in NMEA-0183 format per hour of Grand VTG
messages. The speed measured by the equipment and transmitted in the VTG message indicates the
presence of jumps and other errors, which greatly complicates the task of controlling the GPS receiver.
Therefore, filtering is necessary to eliminate GPS noise. The reasons for interference in GPS data are
different. The main hindrances are:
- Atmospheric interference.
- Obstacles in the way of the signal.
- GPS position in orbit.
For example, the low inclination of GPS orbits (approximately 55°) seriously reduces accuracy in
the circumpolar regions of the Earth.
All this together leads to jumps in position, a change of course and other troubles. In this work, first
of all, it was necessary to get a filtered speed.
Therefore, it was decided to build a filter model in MATLAB, where, having obtained the speed as
a derivative of the GPS data presented by the GGA message, compare it with the original speed data from
the VTG message [1, 2].
Separation of navigation data
Navigation data filtering can be divided into hardware, performed by the GLONASS/GPS module,
and software, performed by the terminal software. Hardware filtering consists in the analysis of noise and
distortion of satellite signals and their subsequent processing using digital mathematical filters unique to
each receiver model. Software filtering is performed at the terminal or dispatch center and consists in the
final screening of navigation data obtained from the navigation module using statistical smoothing
algorithms, taking into account the readings of sensors installed on the mobile object, built-in factors of
loss of positioning accuracy and other analytical methods. In the case of vehicles, such sensors may be:
- Ignition relay, engine operation sensor, wheel rotation sensor and others. Thanks to their use, the
mobile terminal processes navigation information from the GLONASS/GPS module only after the sensors
are triggered, thereby eliminating rough emissions.
- Software filtering is a more universal way of screening out erroneous data and outliers, since in
this case there is no need to be tied to a specific manufacturer and type of navigation receiver.
It is advisable to carry out software filtering on the terminal of a mobile object, since in this case
the load on the data transmission channel between the mobile object and the dispatch center is significantly
reduced, and the workload of the information storage server is also reduced. Also in this case, it is possible
to transfer data in stack mode (priority for new data). Such an implementation does not require significant
resources for organizing calculations and is feasible on terminal equipment.
The existing data filtering methods used in satellite monitoring systems are based on the values of
the built-in positioning accuracy factors (HDOP, VDOP, PDOP), as well as the operation of sensors
installed on a mobile object.
To solve the problem of filtering navigation data, statistical filtering methods are known, including
mathematical smoothing algorithms (Kalman filter, least squares method, 3 sigma rule, median filter), as
well as criteria for discarding data outliers (criterion Grabs, criterion Chauvenet, criterion Pearce, Dixon's
- 288 -
Q-test).
In the case when the number of consecutive outliers is small, the use of various smoothing
algorithms makes it possible to achieve high accuracy of navigation data on the location of the object. In a
situation where successive outliers form a whole "bundle", smoothing algorithms do not cope (low
accuracy), therefore it is necessary to use methods that can discard erroneous information.
Thus, the use of various smoothing algorithms does not completely avoid errors in the data, since
such algorithms can achieve high accuracy only when analyzing data for a certain period (formed sample),
which do not cope well with smoothing newly incoming data, especially in cases where the latter are
erroneous [3, 4].
Definitions of the Kalman filter
The Kalman filter is an effective recursive filter that evaluates the state vector of a dynamic system
using a series of incomplete and noisy measurements. Named after Rudolf Kalman.
The Kalman filter is widely used in engineering and econometric applications: from radars and
vision systems to estimates of parameters of macroeconomic models. Kalman filtering is an important part
of control theory and plays an important role in the creation of control systems. Together with the linear-
quadratic regulator, the Kalman filter allows solving the problem of linear-quadratic Gaussian control. The
Kalman filter and the linear-quadratic regulator are a possible solution to most fundamental problems in
control theory.
In most applications, the dimension of the object state vector exceeds the dimension of the
observation data vector. At the same time, the Kalman filter allows you to evaluate the complete internal
state of the object [5,6,7].
The Kalman filter is designed to recursively evaluate the state vector of a priori known dynamic
system, that is, to calculate the current state of the system, it is necessary to know the current measurement,
as well as the previous state of the filter itself. Thus, the Kalman filter, like other recursive filters, is
implemented in a temporal, not in a frequency representation, but, unlike other similar filters, the Kalman
filter operates not only with state estimates, but also with estimates of the uncertainty (distribution density)
of the state vector, based on the Bayes formula of conditional probability.
The algorithm works in two stages. At the prediction stage, the Kalman filter extrapolates the values
of state variables, as well as their uncertainties. At the second stage, according to the measurement data
(obtained with some error), the extrapolation result is refined. Due to the step-by-step nature of the
algorithm, it can monitor the state of an object in real time (without looking ahead, using only current
measurements and information about the previous state and its uncertainty).
A good example of the filter's capabilities is obtaining optimal, continuously updated estimates of
the position and velocity of some object based on the results of a time series of inaccurate measurements
of its location. For example, in radar there is a task of tracking a target, determining its location, speed and
acceleration, while the measurement results are received gradually and are very noisy. The Kalman filter
uses a probabilistic model of target dynamics that specifies the type of probable movement of the object,
which reduces the impact of noise and provides good estimates of the position of the object at the present,
future or past time [8,9,10].
The Kalman filter operates with the concept of a system state vector (a set of parameters describing
the state of the system at some point in time) and its statistical description. In general, the dynamics of a
certain state vector is described by the probability densities of the distribution of its components at each
moment of time. In the presence of a certain mathematical model of observations of the system, as well as
a model of a priori change in the parameters of the state vector (namely, as a Markov forming process), it
is possible to write an equation for the a posteriori probability density of the state vector at any time. This
differential equation is called the Stratonovich equation. The Stratonovich equation is not solved in general.
An analytical solution can be obtained only in the case of a number of constraints (assumptions):
● Gaussian a priori and a posteriori probability densities of the state vector at any time (including
the initial one)
● Gaussian generating noises
● Gaussian noise observations
● White noise observations
● Linearity of the observation model
● Linearity of the model of the forming process (which, recall, should be a Markov process)
The classical Kalman filter is an equation for calculating the first and second moments of a
posteriori probability density (in the sense of a vector of mathematical expectations and a matrix of
variances, including mutual ones) under these constraints. Due to the fact that for a normal probability
- 289 -
density, the mathematical expectation and the variance matrix completely set the probability density, we
can say that the Kalman filter calculates the a posteriori probability density of the state vector at each time,
which means it fully describes the state vector as a random vector quantity.
The calculated values of mathematical expectations at the same time are optimal estimates
according to the standard error criterion, which causes its wide application.
There are several varieties of the Kalman filter, differing in approximations and tricks that have to
be applied to reduce the filter to the described form and reduce its dimension:
● Advanced Kalman filter. Reduction of nonlinear models of observations and the forming process
by means of linearization by means of Taylor series expansion;
● Sigma-point Kalman filter. It is used in tasks in which simple linearization leads to the destruction
of useful connections between the components of the state vector. In this case, the "linearization" is based
on a sigma-point transformation. Used to reduce the dimension of the problem;
● Options with a non-linear additional filter are possible, allowing to bring non-Gaussian
observations to normal;
● There are options with a "whitewashing" filter that allows you to work with "colored" noise, etc.
In addition, there are analogues of the Kalman filter that use the continuous time model in whole
or in part:
● Kalman — Bucy filter, in which both the evolution of the system and the measurements have the
form of functions of continuous time;
● Hybrid Kalman filter that uses continuous time to describe the evolution of the system and
discrete moments of time for measurements.
Kalman filters are based on time-discretized linear dynamical systems. Such systems are modeled
by Markov chains using linear operators and terms with normal distribution. The state of the system is
described by a vector of finite dimension — the state vector. At each clock cycle, the linear operator acts
on the state vector and translates it into another state vector (deterministic state change), some vector of
normal noise (random factors) is added and, in general, a control vector simulating the impact of the control
system.
The Kalman filter can be considered as an analogue of hidden Markov models, with the difference
that the variables describing the state of the system are elements of an infinite set of real numbers (unlike
the finite set of the state space in hidden Markov models). In addition, hidden Markov models can use
arbitrary distributions for subsequent values of the state vector, unlike the Kalman filter, which uses a
normally distributed noise model. There is a strict relationship between the equations of the Kalman filter
and the hidden Markov model.
When using the Kalman filter to obtain estimates of the process state vector from a series of noisy
measurements, it is necessary to present a model of this process in accordance with the filter structure —
in the form of a matrix equation of a certain type.
Mathematical model of the system
The receiver has two inputs, from which it is necessary to calculate the useful signal x. However,
since it is difficult to obtain accurate data, an approximate or estimated value (1) is calculated.
{𝑧1 = 𝑥 + 𝑛1 𝑧2 = 𝑥 + 𝑛2 (1)
𝑧1 , 𝑧2 – Input data of the device
n_1 – data from GPS.
n_2 – data from GLONASS.
1
𝜔(𝑛1 ) = ∙ 𝑒𝑥𝑝{−𝑛1 2𝜎12
𝜎1 √2𝜋
1
𝜔(𝑛2 ) = 𝜎2 √2𝜋
∙ 𝑒𝑥𝑝{−𝑛1 2𝜎22 (2)
If there is a mathematical algorithm that evaluates the output of the device, let's also evaluate the
performance of this algorithm, since the algorithm itself depends on the performance criteria. Whichever
algorithm is used, an epsilon error will inevitably occur, which is an error between the actual value of the
coordinate and the resulting estimate at the output of the resulting algorithm, so the optimal algorithm will
generate an error with a mathematical expectation equal to zero (2) [11].
Since we get both positive and negative data, it is not easy to take the estimate itself as an estimate,
so we take the square of the error we have
𝜀 = 𝑥 − 𝑥̂ = 𝑥 − 𝑢 (𝑧1 , 𝑧2 ) (3)
- 290 -
𝑅(𝜀) = 𝜀 2 = (𝑥 − 𝑥̂)2 (4)
Since the value of R can be small for one dimension and large for many dimensions, it is necessary
to construct an equation that minimizes the square of the non-specific error (3,4). On average,
mathematically speaking, this is the expectation that the error should be as small as possible. The
application of the above operations gives equation (5).
𝑅(𝜀) = 𝑀{𝜀 2 } = 𝑀{(𝑥 − 𝑥̂)2 } = 𝜎𝜀2 (5)
In the case of Gaussian noise, it was shown that the optimal, i.e. the best, algorithm can be
constructed using a linear combination in (6), so the next task is to build an algorithm that minimizes the
error criterion.
𝑢(𝑧1 , 𝑧2 ) = 𝛼1 𝑧1 + 𝛼2 𝑧2 (6)
Multiply the values of z_1 and z_2 by some unknown gravity coefficients α_1 and α_2.Equation
(7) can be obtained.
𝑅(𝛼1 , 𝛼2 ) = 𝑀{(𝑥 − 𝑥̂)2 } = 𝑀{(𝑥 − [𝛼1 𝑧1 + 𝛼2 𝑧2 ])2 } (7)
These expressions of two unknowns can be optimally reduced to finding just one coefficient if these
coefficients meet this condition. (8)
𝛼1 + 𝛼2 = 1 ⟹ 𝛼1 = 1 − 𝛼2 (8)
This can also be illustrated according to the graph in Figure 1.
Figure 1 – is a graph of compliance with the condition.
𝑥̂ = (1 − 𝛼)𝑧1 + 𝛼𝑧2 (9)
Now this equation depends on the parameter a (9)
𝑅(𝛼1 , 𝛼2 ) = 𝑀{(𝑥 − 𝑥̂)2 } = 𝑀{(𝑥 − [𝛼1 𝑧1 + 𝛼2 𝑧2 ])2 } (10)
Substituting the values of observations z_1 and z_2 into the corresponding formula, the quality
criterion takes the form (11).
𝑃 = 𝑀{((1 − 𝛼)𝑛1 + 𝛼𝑛2 )2 } (11)
According to probability theory, the observed voices n_1 and n_2 can be considered independent
of each other. Then it is decomposed into the sum of squares of mathematical expectations, which is the
following equation (12).
- 291 -
𝑃 = 𝑀{((1 − 𝛼)𝑛1 ))2 } + 𝑀{(𝛼𝑛2 )2 } = (1 − 𝛼)2 𝜎12 + 𝛼 2 𝜎22 (12)
The optimal parameter α of (12) is a parabola depending on α. The following equation (13) is
differentiated.
1 𝜎2
𝛼 = 𝜎2 +𝜎 2 (13)
1 2
At the same time, the observation processing algorithm itself takes this form (14).
𝜎12 𝜎12
𝑥 = ∙ 𝑧1 + ∙ 𝑧2 (14)
𝜎12 +𝜎22 𝜎12 +𝜎22
The result is obtained from two observations. If we consider the evaluation algorithm as a special
case of the implementation of the Kalman filter, then we can generally understand the essence of its work.
(15,16)
To obtain results from the two observations obtained. The evaluation algorithm is perceived as a
special case of the implementation of the Kalman filter and in general it reflects the essence of his work.
(15,16)
𝑃𝑘−1 ∙𝜎𝑘2
𝑃𝑘 = 𝑃𝑘−1
(15)
𝑃
𝑥̂𝑘 = 𝑥̂𝑘 + 𝜎𝑘2 ∙ (𝑧𝑘 − 𝑥̂𝑘−1 ) (16)
𝑘
This expression is a Kalman filter in discrete time, its block diagram can be presented in the
following block diagram (Figure 2).
Figure 2 – Block diagram of the Kalman filter in floppy time
Another observation comes to the input, you need to subtract the previous estimate from it, that is,
we form a formula, then multiply the multiplier, then add it all up again with the previous estimates and get
X at the output. As the number of observations increases, more and more accurate estimates of x_k will be
obtained. The key feature, due to the recurrent principle of the filter, with its computational complexity, the
operability remains constant for any number of observations [12, 13, 14].
Experimental part and method of execution
To begin with, an algorithm was created that is designed to improve data filtering. The
implementation of this algorithm was done in the programming language “Python” with the help of two
main libraries::
1. Numpy is an open source library for the Python programming language.
2. Matplotlib.pyplot is a library in the Python programming language for visualizing data with
two–dimensional and three-dimensional graphics. The resulting images can be used as illustrations in
publications (Figure 3).
- 292 -
Figure 3 – Algorithm of the GPS accuracy improvement method.
The signals were filtered using an algorithm, and the error variance was calculated and the
estimation errors were analyzed. 100 observations were selected, sufficient to evaluate the performance of
the algorithm. The average variance of noise and signal was taken in order to simulate real conditions as
much as possible [15].
At the end of the verification of the algorithm's operability, we proceed to the assembly and testing
of the equipment. For the first experiment, a simulation model based on the “Tinkercad" software was used.
This program allowed to simulate the present conditions and the creation of the device (Figure 4).
Figure 4 – The created model based on the site.
After full testing, you need to proceed to the assembly in order to test the algorithm in real
conditions (Figure 5).
- 293 -
Figure 5 – Connection of the board according to the scheme.
An Arduino board was used to operate the algorithm, and a signal was received as data, received
and filtered from a vibration sensor. It was decided that in order to test the algorithm, a constant sensor
operation time will be used (Figure 6).
With the help of the developed algorithm, it is possible to improve the quality of the data obtained,
which will lead to better data. When testing the algorithm, problems were identified due to the fact that the
data received from the satellite differs from the data received in this way, This problem was solved by
optimizing the code for RAW and GDA format data. So the algorithm was able to produce very good
indicators compared to its competitors like GLONASS and Galileo [16,17].
Figure 6 – The data obtained when using the Arduino board.
After checking, you need to start assembling this model using operated devices, such as (GPS
module (EVK–6T–0–001 series), GPS antenna (M827B series),) (Figure 7).
Figure 7 – GPS module (EVK–6T–0–001 series)
- 294 -
This GPS module is used to receive data from Satellites after connecting to the network, using an
algorithm to get raw data from it and optimize them. All the data came to us in RAW format after an
algorithm was used that helped to get more accurate data (Figure 8).
Figure 8 – GPS antenna (M827B series)
This antenna was needed in order to receive data and contact the satellite, and then correct the
received data. During the experiment, data were obtained from different satellites (Figure 9).
Figure 9 – Module for connecting and transmitting RT data.
When connecting two modules, data was transmitted, but there was a problem that the amount of
interference increased due to additional data transmission. At this stage of the assembly, the algorithm was
again optimized taking into account this problem (Figure 10).
- 295 -
Figure 10 - Analysis of the obtained data for the entire time of the experiment.
After successful execution of the algorithm, it was decided to compare all the data obtained, from
the first model to the last. At the final stage of the design, it was revealed that the quality of the algorithm
itself had been increased over the entire time of operation, and it also turned out to achieve high-quality
data processing. If we compare competitors such as GLONASS and Galileo, then this algorithm will be
able to use one of the effective ways of filtering data to increase accuracy, which was demonstrated above
in experiments [18].
Results and Comparisons.
After the complete assembly of the finished model, changes in navigation data are displayed. First
of all, data is received without filtering, and after using the filter, you can observe changes in coordinates.
Using the program “OpenStreetMap” – you can track changes in coordinates. By this method, it is possible
to analyze and compare the quality of the algorithm. Since different paths and connections to different
satellites were previously used, in order to achieve the best result, it is necessary to enter data for the most
successful result. So for the correctness of the experiment, the GoogleMaps map application will be used
(Figure 11).
Figure 11 – Testing the algorithm using OpenStreetMap.
Now, as the zone and vehicle were determined and selected, it is possible to test correctly and apply
raw data to begin with, and then use the algorithm and see how effective the algorithm is (Figure 12).
- 296 -
Figure 12 – Received data. Blue color - raw data, Red color - data after processing.
The efficiency of the algorithm and the accuracy of data processing have been verified by proper
tests. In Figure 12, the initial data is indicated in blue, that is, data without filtering, and the already filtered
and processed data is indicated in red. This can be especially noted in turns, where the motion vector
changes and the raw data has a large error.
Conclusion
Based on the results of this study, it is possible to identify the optimal method of filtering navigation
data for recognizing a more accurate location using a GPS receiver. The method implies a reduction in the
error of navigation data using Kalman filtering. The work of the GPS receiver using the Kalman filter was
done on MatLab/Simulink software. Figure 13 shows graphs of distance, speed and acceleration relative to
the GPS receiver.
Figure 13 – Output of the GPS receiver results using the Kalman filter.
- 297 -
Two values were considered - actual and estimated. The data obtained contribute to
understanding the evaluation of the accuracy of the algorithm using the unit of measurement. At the same
time, it is possible to determine the influence of distance, speed and acceleration on the dependence of
data acquisition and their truth.
BIBLIOGRAPHY
[1] F. Babich, M. Comisso, A. Cuttin, M. Marchese, and F. Patrone,
[2] ‘‘Nanosatellite-5G integration in the millimeter wave domain: A full topdown approach,’’ IEEE
Trans. Mobile Comput. , vol. 19, no. 2, pp. 390–404,
[3] Feb. 2020.
[4] H. Al-Hraishawi, H. Chougrani, S. Kisseleff, E. Lagunas, and S. Chatzinotas, ‘‘A survey on non-
geostationary satellite systems: The communication perspective,’’ 2021, arXiv:2107.05312.
[5] Zhou Haitao, Wang Lei, Fu Wenju, Han Yi, Li Tao, Li Wei, Chen Ruizhi, “Real-time single-
frequency precise point positioning using BDS-3 PPP-B2b corrections”, Dec 2022.
[6] Sun Jin, Chen Zhengyu, Wang Fu, “A Novel ML-Aided Methodology for SINS/GPS Integrated
Navigation Systems during GPS Outages”, vol 23, Dec 2022.
[7] Technical Specification Group Radio Access Network; Solutionsfor NR to Support Non-Terrestrial
Networks (NTN) (Release 16),document 3GPP TR 38.821, Version 16.0.0, 3rd Generation Partnership
Project, 2019.
[8] Leyva-Mayorga, B. Soret, and P. Popovski, ‘‘Inter-plane inter-satellite connectivity in dense LEO
constellations,’’ IEEE Trans. Wireless Commun. , vol. 20, no. 6, pp. 3430–3443, Jun. 2021.
[9] I. D. Portillo, B. G. Cameron, and E. F. Crawley, ‘‘A technical comparison of three low earth orbit
satellite constellation systems to provide global broadband,’’ Acta Astronautica, vol. 159, pp. 123–135,
Jun. 2019.
[10] AWS Ground Station. Accessed: Sep. 1, 2021. [Online]. Available:
https://aws.amazon.com/ground-station
[11] Nelson Acosta, Juan Toloza, “Techniques to improve the GPS precision”, vol 3, pp 2-4
[12] Union of Concerned Scientists. UCS Satellite Database 2021. Accessed: Oct. 1, 2021. [Online].
Available: https://www.ucsusa.org/resources/satellite-database
[13] R. G. W. Latio, ‘‘Social and cultural issues: The impact of digital divide on development and how
satellite addresses this problem,’’ Online J. Space Commun. , vol. 2, no. 5, pp. 1–17, 2021.
[14] H. Al-Hraishawi, E. Lagunas, and S. Chatzinotas, ‘‘Traffic simulator formulti beam satellite
communication systems,’’ in Proc. 10th Adv. Satell. Multimedia Syst. Conf., 16th Signal Process. Space
Commun. Workshop(ASMS/SPSC), Oct. 2020, pp. 1–8.
[15] Y. Turk and E. Zeydan, ‘‘Satellite backhauling for next generation cellular networks: Challenges
and opportunities,’’ IEEE Commun. Mag. , vol. 57,no. 12, pp. 52–57, Dec. 2019.
[16] BegümErçevik Sönmez and DenizErinsel Önder, “The influence of GPS-based navigation systems
on perception and image formation: A case study in urban environments” vol. 86, March 2019, pp. 102-
112.
[17] H. Al-Hraishawi, S. Chatzinotas, and B. Ottersten, ‘‘Broadband nongeostationary satellite
communication systems: Research challenges andkey opportunities,’’ in Proc. IEEE Int. Conf. Commun.
Workshops (ICCWorkshops), Jun. 2021, pp. 1–6.
[18] Kanouj M.M., Klovok A.V., “Adaptive Unscented Kalman Filter for Tracking GPS signals in the
Case of an Unknown and Time-Varying Noise Covariance” vol 3, Jule 2021, pp 224-235
[19] Zeng Zhanjing, Chen Po-Ju, Xiao Xiao, Liu Peixue, Zhang Jie, “The mediating and moderating
effects on the intention to use navigation apps”, vol 5, Nov 2022, pp 972-991
[20] European Space Agency. European Data Relay Satellite System (EDRS). Accessed: Aug. 15, 2021.
[Online]. Available: https://artes.esa.int/edrs/overview
- 298 -
August 22(TUE) - 25(FRI), 2023
Poster Session 1
12:00~12:40 Thu, August 24, 2023
Room : Building A 4th Floor
Chair: Prof. Jeamoon Kim
(Korea National University of Transportation, Korea)
Prof. Jaesang Cha (Chiba University, Japan)
Prof. Ainur Begimbetova (AUPET, Kazakhsatn)
Prof. Changjun Ann (Chiba University, Japan)
- 299 -
A Study on the Partial Discharge
Characteristics of VTs and CTs by Applying
Repeated Impulses
Ji Man Park 1 Ja Yoon Kang 1 Young Chae Mun 1 and Ga Ram Han 1*
1
Korea Electrical Safety Corporation Research Institute, Republic of Korea
Email: [email protected]
Abstract—As the use of epoxy-insulated electrical equipment This paper presents the results of partial discharge
continues to increase, VTs(Voltage Transformers) and measurements on VTs and CTs by applying repetitive
CTs(Current Transformers) have also been replaced by epoxy impulses. For measuring PD, six VTs and CTs from three
insulation rather than oil-filled insulation. However, the partial manufacturers were used, and the maximum voltage was
discharge measurement technique for determining the applied up to 180 kV. In order to acquire PD data,
deterioration state of epoxy insulation facilities, such as VTs and TECHIMP’s AQUILA model with HFCT sensor and a
CTs, is slow to develop. In this paper, partial discharge contact-type electromagnetic wave partial discharge
measurement was performed for PTs and CTs by applying measurement equipment (CALVUS INSTRUMENTS) were
repeated extra-high voltage impulses, and the results were
adopted for comparison of two different measurement
analyzed. An 800kV impulse generator(HAEFLY) was used for
applying impulses to VTs and CTs. Partial discharge (PD)
techniques. The experimental system was set up consists of
measurements were carried out using CALVUS two transformers(one for step-up and the other is step-down),
INSTRUMENTS (C3.5), a contact-type electromagnetic wave a VT (CT), a power cable and a load, and experiments were
partial discharge measurement equipment, and carried out by replacing VT and CT samples. As a result of the
AQUILA(TECHIMP) which measures PD by an HFCT(High- experiment, it was confirmed that the facility deteriorated and
Frequency Current Transformer) sensor. Impulse was applied partial discharge occurred only when a considerable number
to six VTs and CTs of three manufacturers and partial of impulses were applied. From the PD measurement results,
discharge measurement was performed and the results were by comparing HFCT and contact-type measurement
analyzed in this paper. equipment, PD characteristics of epoxy insulation with
repetitive impulse stress was analyzed.
Keywords—partial discharge, voltage Transformer, current
Transformer, Degradation, Impulse II. EXPERIMENTAL SETUP
I. INTRODUCTION A. Configuraton of Simple Test System
Due to the excellent insulation performance of epoxy, If the PD measurements is executed for individual
existing insulation materials are steadily replaced by epoxy. facilities or artificial defects in insulation, it is difficult to
These electrical facilities with epoxy insulation, such as cast- obtain data usable for on-site. Therefore, in this paper, a
resin transformers, VTs & CTs and line post insulators, have simple experimental system was constructed and the PD was
played an important role in operation of electrical network. measured by replacing VT and CT samples. The simple
Therefore, it is extremely important to prevent unexpected experimental system consists of two cast-resin transformers,
failures by applying various diagnosis techniques. VT/CT sample, an XLPE power cable and adjustable load.
One cast-resin transformer was used for step-up (380V to
Measurements on insulation resistance and dielectric 22.9kV) and the other for step-down (22.9kV to 380V). Two
strength are widely used techniques for estimating condition cast-resin transformers and VT(CT) were connected with
of epoxy insulation. However, because these techniques 22.9kV CNCV XLPE power cables. Fig. 1 shows the
involve inevitable power outage, economic loses and configured simplified experimental system.
inconvenience will be incurred to network owners. In addition,
owing to the pre-works that separating from individual facility
to electrical network, the entire process of inspection takes
more than an hour.
In order to solve these problems, inspection techniques
without power outages is actively being adopted. Partial
discharge measurement is one of the effective method for
inspecting the deterioration status of epoxy insulation in live-
wire state, and widely applied PD measurement sensor is an
high frequency current transformer (HFCT) that is installed to
ground wire. However, PD measurement using HFCT sensor
has obvious limitations that it is difficult to identify defected
facility why various type of signals flow into the ground wire.
The fact that limitation of an HFCT and lack of alternative
options make it difficult to access of study on the partial
discharge characteristics of epoxy insulation [1-4]. Fig. 1. Configration of simple test system
- 300 -
B. Partial Discharge Measurement Equipment
In order to measure PD from deteriorated VTs and CTs by
repetitive impulses, two type of PD measurement equipment
were used. One is AQUILA model manufactured by
TECHIMP, which is using HFCT sensor for PD measurement.
The other is C3.5 model that is contact-type electromagnetic
wave detection sensor manufactured by CALVUS
INSTRUMENTS. C3.5 model consists of modularized sensor,
connection bar and a tablet with analyzing software. Fig. 2
shows equipment for PD measurement.
III. PARTIAL DISCHARGE MESUREMENT OF VTS AND CTS
The performance of the epoxy insulation is degraded due Fig. 3. 800kV impulse generator(HAEFLY), picture of impulset
to the influence of overvoltage or overcurrent. In this paper, application and PD measurement
the insulation performance was artificially degraded through
repeated lightening impulse application to VTs and CTs. IV. MEASUREMENT RESULTS
When the manufacturing process is completed, VTs and Table 1 shows the experimental results. In the case of CT
CTs must conduct lightening impulse tests in accordance with by company K, PD was detected after 2,090 impulse
the Korea Electro-Technical Code (KEC) regulations. The test applications. Based on the electromagnetic partial discharge
method is to apply a lightening impulse voltage of 95 kV (125 method, 2,086pC was measured as well as 68pC was detected
kV) 15 times in positive polarity and 15 times in negative in VT of Company D and 142pC was detected in VT of
polarity for epoxy insulated CT, and there shall be no company K. During the lightening impulse experiment, in the
abnormality even after applying the lightening impulse. In this waveform shape of the impulse generator, it is observed that
study, lightening impulses were applied 30 times to positive pulse distortion according to the PD occurrence.
and negative polarities and PD measurements were executed
every end of the application of lightening impulses. Applying TABLE I. TABLE TYPE STYLES
30 times of lightening impulses and PD measurements were PD
repeated until the PD was occurred in target facilities. The Manufac Number of
Maximum PD magni
turer / applied
application of impulses proceeded differently for each facilities impulses
voltage occurrence tude
manufacturer and maximum impulse voltage was 180 kV. [pC]
Company
2,090 175kV O 2,086
There are various VT and CT manufacturers in Korea, and K / CT
it is impossible to perform PD measurement tests for all Company
6,000 150kV X -
manufacturers. Therefore, in this study, three representative W / CT
Company
manufacturers (K, D, and W companies) were selected and D / CT
4,600 180kV X -
experiments were executed on a total of six units, one each of Company
VT and CT. In the case of CT of Company K, the initial 2,437 150kV X -
K / VT
lightening impulse voltage started at 95 kV, and after the Company
218 140kV Δ 68
impulse voltage was applied 1,110 times, the test voltage was W / VT
Company
gradually increased to a maximum of 175 kV. VT and CT of
D / VT
80 140kV Δ 142
other companies were also experimented in this way, and in
the case of Company D, no partial discharge occurred even
after repeated impulse application, so an impulse voltage was
gradually increase up to 180 kV.
The application of lightening impulses were performed
total 15,424 times. However, a CT that manufactured by W
company, PD was not measured even though 6,000 times of
lightening impulses were applied. Details on measurement
results will be covered in chapter 4.
Fig. 2. Equipment for PD measurement, AQUILA by TECHIMP(left) and
electromagnetic wave measurement module(right)
Fig. 4. Comparison for the pulse shape of the first applied impulse and
Identify applicable funding agency here. If none, delete this text box. 2,088th applied impulse
- 301 -
A. PD Measurement results TABLE II. TABLE TYPE STYLES
The contact-type electromagnetic (EM) wave Manufactu Number of Dissipation factor
measurement module has two different PD detection mode. rer / applied Note
early middle late
facilities impulses
One is [high] mode, which is detecting all of the PD signals Company
from target facility (scanning mode) and the other is [low] 2,090 0.33 0.44 0.30 ↑ 0.11
K / CT
mode that detecting PD signals over 10MHz (defect detecting Company
6,000 0.39 0.60 0.53 ↑ 0.21
mode). Among VTs and CTs, only company K’s CT shows W / CT
the meaningful results of PD measurement. Table 1 shows the Company
4,600 0.25 0.62 0.35 ↑ 0.37
measurement result of HFCT and EM wave measurement D / CT
Company
module. 2,437 0.66 0.98 1.02 ↑ 0.36
K / VT
In case of PD measurement using EM wave measurement Company
218 0.64 0.60 0.62 ↓ 0.04
W / VT
module, measurement point was set as three points, ground Company
wire, left and right of surface, respectively. On the ground D / VT 80 0.48 14.1 0.59 ↑ 13.6
wire, maximum PD magnitude was measured 113 pC in gain
[high] mode. On the left side of surface, 104 pC was measured
and on the right side 55 pC was measured in same gain mode.
In [low] mode, PD measurement results of ground wire, left
and right surface were 2,086 pC, 150 pC and 86 pC,
respectively. In the body of target CT, due to the thick epoxy
insulation, PD measurement was impossible. PD magnitude
was high in the order of ground wire, left side and right side.
When PD measurement was performed using HFCT sensor,
Maximum PD magnitude was 1,689 pC.
B. Dissipation factor measurement
As a result of measuring the dissipation tests (tan δ) after
the lightening impulse application for VT and CT, there was a
tendency that measurement value was repeated to rise and
decrease irregularly. It is assumed that environmental factors
such as temperature and humidity affect the value of
dissipation factor rather than value of PD measurement. In
case of company K's VT, it rose sharply from 0.48% to 14.1%, Fig. 6. PD measurement result of HFCT sensor installed on ground wire
but there was no direct correlation with the occurrence of PD.
In case of company D's VT that the change in tan δ was not V. CONCLUSION
large and there was no change in tan δ of company C’s CT In this paper, PD measurement was performed on VTs and
after 2,000th impulse applied. CTs by applying repetitive lightening impulses. Total three
VTs and three CTs were used to investigate PD characteristics
Comparing the results of the lightening impulse
of epoxy insulation. Lightening impulses were applied using
experiment with the dissipation factor measurement, it is
800kV impulse generator manufactured by HAEFLY, and PD
analyzed that the change in the tan δ measurement is greatly
measurement was executed by HFCT sensor and EM wave
affected by environmental factors, and it is considered
measurement module. Among these six VTs and CTs, valid
unreasonable to judge the insulation degradation of the facility
PD data was obtained from the company K’s CT.
only by dissipation factor measurement.
Epoxy insulated facilities, such as cast-resin transformers,
line post insulator and VTs & CTs, have performed in an
electrical network. Accordingly, it is very important to ensure
reliable operation of these facilities for preventing unexpected
outage of electrical network. In the near future, PD
measurement on various type of defective epoxy insulation
specimen will be performed.
REFERENCES
[1] Soon-Yong Kim, “Partial Discharge Diagnosis for a Generator
Potential Transformer Based on Epoxy Insulators Including Voids,”
Journal of International Council on Electrical Engineering, Vol. 3, No.
2, pp.174∼178, 2013.
[2] V. S. Deolankar1, “Partial Discharge Analysis in High Voltage Current
Transformers,” International Refereed Journal of Engineering and
Science, Vol 6, No. 5, PP. 24∼30, 2017.
[3] Nascimento, “Evaluation of Partial Discharge Signatures Using
Inductive Coupling at On-Site Measuring for Instrument Transformers,”
International Journal of Emerging Electric Power Systems, 2018.
[4] Wen-Yeau Chang, “Partial Discharge Pattern Recognition of Cast
Fig. 5. PD measurement screen shot of company K’s CT on the ground wire
Resin Current Transformers Using Radial Basis Function Neural
using EM wave measurement module
Network,” J Electr Eng Technol, Vol. 9, No. 1, pp. 293∼300, 2014.
- 302 -
An Application of Terahertz Wave for Detecting
Insulation Defects in 22.9kV XLPE Power Cable
Ji-Man Park 1, Young-Chae Mun 1 and Ja-Yoon Kang 1*
1
Electrical Safety Research Institute of Korea Electrical Safety Corporation, Korea
Email: [email protected], [email protected]
Abstract— Recently, various on-line diagnostic XLPE using electrodes. Terahertz was applied to these
techniques for detecting insulation defects in electrical samples, and the results were analyzed.
facilities are being studied, and partial discharge measurement
is the most commonly used for detecting insulation defects. II. PHYSICAL FAULT DETECTION AND ANALYSIS
However, since the partial discharge diagnosis technique A. Specimen
relies only on signals, it is not possible to estimate a clear
source of defects, and it is also very difficult to determine the The physical defect sample is a simulation of void defects
type of defects. In this paper, we tried to detect images by in XLPE. Figure 1 is a photograph for describing a physical
applying terahertz waves for each defect type of XLPE. For defect sample. A hole was made using a 2mm drill for void
the test, 22.9Kv (Korean electrical system nominal voltage) defects. As shown in Fig. 1(a), six holes were made with the
CNCV-w XLPE (Concentric Neural Cross-linked drill 1 mm away from the surface. The 1 mm distance is H2
polyethylene insulated Vinyl Sheathed Cable) was used. and H6, the 2 mm distance is H3 and H5, and the 3 mm
Experiments were conducted on samples of physically and distance is H1 and H4. The 180︒ to 360︒ sections included
electrically generated defects, and it was analyzed whether a semiconductor layer actually used on the XLPE surface.
terahertz waves could be detected for each type of defect in This was included to confirm the effect of transmission
the XLPE cable. characteristics on the semiconductor layer because terahertz
waves are reflected against metal objects. The scan direction
Keywords— Terahertz wave, Internal defect detection, was counterclockwise at 0︒ as shown in the picture. (b) in
XLPE cable, Scanning image, Electrical tree Figure 1 is a side photograph after the hole is made.
I. INTRODUCTION
The development of inspection and diagnosis technology
for high-voltage power facilities is very important for ensuring
safety and efficient maintenance. The inspection of electrical
facilities has been mainly conducted in the form of measuring
insulation resistance and insulation resistance after shutting
down the entire facility, but in industrial facilities such as
factories, a single power outage causes huge losses, and in
residential areas, home appliances fail or inconvenience.
Therefore, the development of uninterruptible inspection (a)
technology is actively underway at home and abroad, and in
particular, the development of sensors and constant
monitoring technologies for partial discharge measurement is
actively underway. However, since these techniques must be
judged by the detected signal, the know-how of experienced
experts is required, and the biggest problem is that the
deterioration state must be determined only by the
measurement signal, making quantitative judgment difficult.
Terahertz technology is a technology that can see the (b)
inside of objects using terahertz waves, and has recently been
Fig. 1. XLPE Cable Physical Defects
actively applied to airport checkpoints and food inspections,
and has the advantage of being harmless to the human body.
B. The result of an experiment
If terahertz waves can be applied to insulators to image
internal defects, it can be a good alternative as it can be used Figure 2 summarizes the results of applying terahertz
by non-experts as well as intuitive judgment, but little research waves to physical defect samples. Figure 2 shows the result of
is currently being conducted to apply terahertz technology to detecting the internal image with (a) time domain and (b)
electrical facilities.[1] frequency domain, respectively. As a result of image detection,
a void defect estimated from H1 to H6 was estimated.
In this paper, terahertz was applied to detect internal Through these results, it was confirmed that the
images by defect type of XLPE. The defect type was tested on semiconductor layer used in XLPE is capable of internal
samples generated physically and electrically. As for the transmission and does not affect securing defects in XLPE. In
physical type, holes were artificially created in the XLPE, and addition, it was confirmed that defects in mm could be
as for the electrical type, an electric tree was created in the detected.
- 303 -
B. The result of an experiment
Figure 4 summarizes the results of applying terahertz
waves to electrical defect samples. The red box indicates
where the electric tree image should be secured. However,
images presumed to be electrodes in two types were detected,
but no images were objectively judged to be electric trees.
Unlike physical defects, it is presumed to be due to limitations
in resolution to detect defects in units of um.
(a)
(b) (a) Busy type
Fig. 2. Internal Image Detection Results of Physical Defects in XLPE Cable
III. INTERNAL IMAGE DETECTION RESULTS OF PHYSICAL
DEFECTS IN XLPE CABLE
A. Specimen
The electrical defect was created by simulating the electric
tree generated during use in the cable. A needle electrode was
inserted into the XLPE to transfer the high voltage. The
electric tree was prepared in a bush type and a branch type.
Figure 3 is an enlarged photograph of the XLPE electrical
defect sample.
(b) Branch type
Fig. 4. Results of internal image detection of xlpe cable electrical fault
IV. CONCLUSION
In this paper, we experimented on whether it is possible to
detect internal defects in power facilities as images using
terahertz waves. The experiment was conducted by simulating
physical and electrical defects on the XLPE of the XLPE cable.
As a result of the experiment, it was confirmed that image
detection can be performed inside XLPE. However, physical
(a) Busy type defects in XLPE were secured, but electrical defects were not
secured. As a result, the inside of the cable could not be
completely detected due to limitations such as transmission
depth and resolution. In the future, we will study scan modules
related to resolution and scan methods of XLPE cables to
detect accurate internal images.
REFERENCES
[1] C. Guo, W. Xu, M. Cai, S. Duan, J. Fu and X. Zhang, "A Review:
(b) Branch type Application of Terahertz Nondestructive Testing Technology in
Electrical Insulation Materials," in IEEE Access, vol. 10, pp. 121547-
Fig. 3. Electrical fault with XLPE cable 121560, 2022
- 304 -
A Study on the On/off-line Framework for
Asset Management of 154 kV Electrical Facilities
Young Kyu Mo1*, Min Hwang1, Pil-sung Woo1, Joon-Ho Son1, Youngseok Kim1
1
Electrical Safety Research Institute, Korea Electrical Safety Corporation, Republic of Korea
Email: [email protected]
Abstract—The safety of electrical facilities that supply II. ON/OFF-LINE FRAMEWORK FOR ASSET MANAGEMENT
electrical energy requires operation and management based on
the balance of efficiency and reliability. To ensure the safety of A. Asset management of electrical facilities
electrical facilities, rational management is required to maintain For asset management of electrical facilities, life cycle
the performance of the facilities that make up the facilities above management is required to record the entire process from
a certain standard at all times. To this end, an important
design to disposal. Figure 1 shows the life cycle management
indicator is the degree to which the original performance and
process of a typical electrical facilities. It can be divided into
function are maintained within the period of use based on the
reliability and safety of electrical facilities. Continuous diagnosis
four stages, such as design, construction, operation, and
and history management are required to maintain the degree of disposal, based on the state of electrical facilities, and repair
facility's health similar to its initial performance from a or replacement is performed through safety management,
maintenance perspective. This paper proposes the concept of an regular inspection, and precise diagnosis during facility
asset management framework for 154 kV electric facilities to operation. When repair is not possible due to aging or when
ensure the safety of electric facilities operating electric facilities replacement is economically advantageous, electrical
for large-capacity customers. The proposed on/off-line asset facilities is disposed. The asset management of 154 kV power
management framework consists of a conceptual model of the equipment presented in this paper is a combination of online
system to provide life-cycle electrical safety services for 154kV real-time monitoring through sensors and IoT modules
large-capacity electrical facilities using platform-based on/off- optimized for abnormal condition detection, such as partial
line data. discharge, voltage, current, insulation condition, and
temperature, and offline data such as pre-use inspection,
Keywords— Asset Management, Safety Management, Safety regular inspection, and precision diagnosis of electrical
diagnosis, Life prediction, Life assessment facilities.
I. INTRODUCTION
Electrical facilities are characterized by long-term
operation, and each elements that makes up the facility
degrades due to electrical, thermal, and environmental stress
over time[1]. Moreover, the long-term operation of power
facilities in this aging deterioration causes the operation of
electrical facilities to be suspended due to unexpected failures,
resulting in serious electrical accidents such as defects in
production products due to power outages and, in the worst
case, electric fires/electric shock[2]. In particular, if such a
Fig. 1. Life cycle management process of a typical electrical facilities.
sudden failure of power facilities occurs in large-capacity
customers, which are important national industrial facilities
B. On/off-line data definition
such as semiconductor production plants, it could cause
enormous damage that threatens the survival of companies, On/off-line data is essential data for asset management of
people's lives, and even the national economy. To ensure the 154kV electrical facilities. Online collected data is converted
safety of electrical facilities, rational management is required into data based on time series, and the status and deterioration
to maintain the performance of the facilities that make up the information data of electrical facilities such as partial
facilities above a certain standard at all times[3]. To this end, discharge, thermal image, and power quality sensor can be
an important indicator is the degree to which the original acquired in real time. Offline collected data can be acquired
performance and function are maintained within the period of through visual and facility measurement inspections by field
use based on the reliability and safety of electrical facilities. experts at regular intervals. By utilizing the acquired on/off-
Continuous diagnosis and history management are required to line data, it is possible to perform effective asset management
maintain the degree of facility's health similar to its initial by providing services such as abnormal symptom monitoring
performance from a maintenance perspective. The safety of electrical facilities, safety rating, life assessment, and life
management of existing electrical facilities is managed by prediction. Korea Electrical Safety Corporation selected
regular inspections according to a certain cycle, safety inspection items and sensors applied to 154k oil-immersed
diagnosis by customer requests, and safety management transformers and cables as they conducted “digital system
agencies. Accordingly, in order to efficiently manage facilities, inspection”. Figure 2 shows the digital system inspection
it is necessary to understand the status of the facility's history items and sensors.
management base and to provide information on the
repair/replacement decision-making of facility owners based
on the use of information during the life cycle.
- 305 -
healthiness score can be calculated by classifying grades
according to the diagnostic results of parameters such as
facility condition, life, and degree of deterioration, and
multiplying the score for each grade by a weight. The sum of
the calculated scores is the health index, which was developed
by applying the diagnostic work of the Korea Electrical Safety
Corporation based on the IEC White paper health index
algorithm.
Fig. 2. Digital system inspection item and sensor for each electrical
facilities.
C. On/off-line data structure design for asset management
Common data for asset management of electrical facilities
can be divided into three categories: customer data, facility
data, and environmental data. Customer data consists of 16
components, including customer number, business name,
address, kind of facility, facility classification, type of facility,
electric power distribution method, power generation purpose,
etc. Facility data consists of a total of 166 components, and
the number of configuration data differs depending on the type Fig. 4. IEC ealth index algorithm.
of electrical facility. In common, it is individually defined
according to the characteristics of the facility, such as
installation date, facility name, facility capacity, facility
location, manufacturing date, management history, and life III. CONCLUSION
criteria. Environmental data can be divided into inspection In this paper, the concept of an asset management
data and diagnosis data, and is defined as the load factor of framework for electrical facilities is presented for securing
electrical facilities, period of use, facility usage environment, safety and efficient management of electrical facilities for
and fulfillment the duties of electrical safety management data. large-capacity customers who operate 154kV class electrical
Figure 3 shows a conceptual diagram of on/offline data facilities. The proposed on/off-line asset management
structure design for electrical facility asset management. framework can be used to prevent safety accidents, improve
efficiency, and reduce maintenance costs of 154kV large-
capacity electrical facilities by utilizing on-site measurement
data and on/offline data acquired through online sensors. In
addition, the inspection items and sensors applied to the digital
system inspection of 154kV class electrical facilities were
selected and the research direction of future electrical facility
asset management was presented. Based on the digital system
inspection items presented in this paper, we plan to conduct
advanced research and empirical research on condition
assessment, life assessment, and life prediction required for
asset management of 154kV electrical facilities.
Fig. 3. Conceptual diagram of on/off-line data structure design for REFERENCES
electrical facility asset management.
[1] Jongwook Jung, Sun koo Kim, “A Consideration of the Expected
D. Electrical facilities assessment system Service Life Cycles of Electrical Equipments in Buildings,” Journal of
The evaluation of electrical facilities is divided into two the Korean Institute of Illuminating and Electrical Installation
Engineers, vol. 24, no. 12, pp. 131–137, 2010.
categories: safety level and health index. Safety level is
[2] Hong-Seok Lee, “ Development status of Power facility asset
composed of electrical factors and environmental factors and management system,” The Proceedings of the Korean Institute of
the scores of electrical factors and environmental factors can Illuminating and Electrical Installation Engineers, vol. 35, no. 2, pp.
be calculated by multiplying the scores and weights of each 23-28, 2021.
assessment item. The weight was set in consideration of the [3] Min Hwang, Yeong-Seok Kim, Gi-Ryong Jo, Do-Yun Kim, Pil-Sung
expert survey, analytic hierarchy process (AHP), and accident Woo, Jun-Ho Son, Young-Kyu Mo, "Study on the Conception of
history. After deriving the score for each electrical facility, it Framework for Life Cycle-based Safety Management System", Korean
Electrical Society Conference Proceedings, pp. 2109-2110, 2022.
can be evaluated by dividing it into 5 grades from A (excellent)
[4] Min Hwang, Do-Yun Kim, Gi-Ryong Jo, Young-Seok Kim."A Study
to E (dangerous) according to the electrical safety rating on the Establishment Model for Life Cycle-based Safety Management
system of the Korea Electrical Safety Corporation[4]. System of Extra-high Voltage Facilities"The transactions of The
Healthiness is an index that expresses the state of electrical Korean Institute of Electrical Engineers, vol.71, no.12, pp. 1909-1916,
facility and can be evaluated based on diagnostic data. The 2022.
- 306 -
Design of Real-time Monitoring/Diagnosis
Algorithm for Electrical Facilities
Hoon Jung, Young-Taek Lim, Jong-Chan Lee and Joonho Ahn1*
1
Future Innovation Institution of Seoul National University, Korea
Email: [email protected]
Abstract— With the development of IoT technology, facility In addition, when analyzing accidents by cause of
monitoring/diagnosis technology through real-time monitoring is electrical facilities, natural deterioration (250 cases, 7.8%),
required for the operation of various and safe electrical facilities insulation failure (241 cases, 7.5%), protection malfunction
using AI. This study aims to design an algorithm to perform failure (118 cases, 3.7%), and overload (117 cases, 3.6%).
determination and prediction by learning abnormal signals of
electrical facilities with artificial intelligence through real-time Natural degradation (7.8%) means that there are the most
monitoring of electrical facilities using current, infrared, high accidents caused by aging of facilities, and insulation defects
frequency, vibration, gas sensors, etc. To design the algorithm, it (7.5%) are also believed to be the fundamental cause of
will analyze the failure mechanism of electrical facilities, learn insulation defects due to deterioration in facilities that have
logical consistency with sensors through failure case analysis, progressed slowly during the life cycle.
quickly determine the presence or absence of failure of electrical
facilities through failure judgment and prediction. Protective device malfunction (3.7%) and overload (3.6%)
refer to stress from the outside of the facility, and this is due
Keywords—electrical facilities, real-time monitoring, to malfunction or strain on the facility due to external stress
algorithm such as overvoltage and overcurrent.
In the case of overload (3.6%), it is judged to be an
unexpected increase in load during design.
I. INTRODUCTION
As a result of examining the disaster status, it is difficult
Looking at the current status of electrical disasters in to determine the cause of many accidents caused by electricity,
Korea in 2021, others account for the largest number of fires and the electrical safety inspection currently in operation is
in electrical facilities. This shows that it is not easy to clearly determined based on the standard range of measurements
identify the cause of the accident for the facility because it has measured during inspection.
not been identified exactly what the cause of the fire caused
by the electrical facility is, or if it is a minor accident to be It is necessary to continuously monitor/diagnose the
identified. condition of the facility through a real-time diagnosis and risk
prediction system of electrical facilities, and to clarify the
Looking at the number of electrical fires by facility, cause of failure through data through artificial intelligence
wiring/wiring devices accounted for the largest number of learning through data.
cases with 1,887 (22.9%), followed by electrical facilities with
991 (12.0%) and seasonal devices with 928 (11.3%). However, A safety management system for electrical facilities is
2,537 cases (30.7%) are unknown, and others are high at 394 required to build big data through data analysis by reviewing
cases (4.8%). the characteristics, type of failure, and cause of failure of
sensors for monitoring/diagnosis.
When analyzing accidents by electrical facilities, 1,883
cases (58.5%) occurred in low-pressure electrical facilities, 38
cases (1.2%) in high-pressure electrical facilities, and 1,299
cases (40.3%) in special high-pressure electrical facilities. II. FAILURE MECHANISM FOR ELECTRICAL FACILITIES
Low-pressure electrical equipment accidents were followed There are three types of failure mechanisms of electrical
by 176 wiring breakers (9.3%), 171 (9.1%), 164 short circuit equipment: initial failure, accidental failure, and wear failure.
breakers (8.7%), and 163 indoor wiring (8.7%), and 146
special high-pressure electrical equipment accidents (15.3%) Early failure is caused by initial stress after installing
were transformers (11.2%), 115 (8%). electrical equipment, and the failure rate tends to decrease. It
is known to be mainly due to design, manufacturing errors,
In the case of low-pressure electrical facilities, many inexperience in use, damage during transport, and insufficient
accidents occurred in load devices such as breakers, indoor debugging.
wiring, and electric heat, and in the case of special high-
pressure electrical facilities, accidents occurred mainly in the In Random Failure, electrical facilities are stabilized and
switchboard such as transformers, transformers, switches, and have a constant and low failure rate, and this is due to
so on. exceeding design limits, continuous vibration and shock,
higher stress than design, overwork and user error, failures not
found in debugging, and natural disasters.
- 307 -
Wear-out failure refers to a time when the failure rate Therefore, in order to predict failures through real-time
increases due to long-term use, and it is analyzed that aging, monitoring/diagnosis, it is necessary to set a failure mode to
contraction or cracking, insufficient maintenance, and be reflected in the artificial intelligence algorithm and apply
improper overhaul such as wear, aging, corrosion, fatigue and different criteria reflecting the characteristics of statistical
insulation deterioration. failure sections during demonstration and operation.
III. REAL-TIME MONITORING AND DIAGNOSTIC MECHANISM
ANALYSIS
Sensors with various functions are required for electrical
facility diagnosis, and sensors acquire information on
phenomena caused by degradation in electrical facilities such
as temperature, vibration, and sound according to their
characteristics and deliver data to real-time
Fig. 1. Bathtub Curve for Electrical Facilities monitoring/diagnosis platforms.
TABLE II. EXAMPLES OF VARIOUS SENSORS FOR ELECGTRICAL
EQUIPMENT DIAGNOSIS
TABLE I. CAUSE OF OCCURRENCE AND IMPROVEMENTS measurement sensor Examples
Mode Cause of Occurrence Improvements - Vibration sensor
Vibration - a bearing analysis
Early Failure Caused by initial stre - Thorough quality control - Waveform proce
Measurement - adjust imbalance
ss, failure rate decrea Aging, Burn-in, HASS, ES ssing
sing trend S - detection of abnormalities in
- Stabilization process such - acoustic sensor bearing parts
as debugging, burn-in test, acoustic
- unltrasonic senso - partial discharge sound detec
etc measurement
r tion
- Conservation Prevention - ultrasonic flaw detection
Activities
- Winding temperature monito
- Manufacturer's - thermocouple the
ring
dependence on quality rmometer
- Bearing temperature monitor
assurance - resistance thermo
ing
Random Constant failure rate - a fundamental change in d meter
Thermometry - Rotating part temperature m
Failure and low safety factor e s i g n - a radiation therm
easurement
ometer
- Normal operation – appro - Detection of abnormal losses
- an infrared therm
p r i a t e u s e by temperature distributio
ometer
- Prevention Maintenance n
- Corrective Maintenance Pressure - Diagnosis of clogging of the
- Pressure sensor
Measurement flow path
Were-out Increased failure rate- - Prevention and Break-down
Failure due to prolonged us Maintenance - cooling quantity
flow
- flux sensor - measurement of cooling flow
e measurement
rate
displacement - displacement sen
- stress measurement
According to the bathtub curve, the failure mechanism of detection sor
electrical equipment shows the characteristics of three - Measurement of the stroke c
sections during the life cycle. Accordingly, it is important to haracteristics of the breake
optical
understand the history of the facility (manufacturing and measurement
- optical sensor r
installation date, failure history, environmental history, etc.) - Gas insulated internal abnor
malities detection device
for real-time monitoring/diagnosis of electrical facilities.
- Measurement of piping thick
The failure of an electrical facility does not occur as a ness
single failure, but as a cause of a failure, several similar Radiation - radiation sensor - Scale attachment status meas
phenomena occur at the same time, and several phenomena detection - X-ray film urement
occur sequentially depending on the degradation mechanism. - Inspection of equipment for
damage
In addition, the initial failure is caused by instantaneous
errors or signals inside and outside due to the initial stress of - Monitoring of Pressure Vess
Acoustic
- Ultrasonic sensor els in Operation of Large S
the operation of the installed electrical equipment, and the Emission
tructures
accidental failure is mainly caused by instant errors or signals
caused by external effects (lakes, short circuits, power outages, - Gas Analysis in Transformer
etc.). Chemical
- Gas sensor
flow
Measurement - Core Monitor
However, wear and tear is a period of failure caused by - Lubricating oil diagnosis
deterioration depending on the life cycle, and accidents such
as power outages caused by facility failures can be prevented
by reinforcement learning based on big data accumulated This requires processing into valid data to preprocess and
rather than instantaneous parts such as initial failures and learn sensor data, and real-time monitoring/diagnosis of
accidental failures. electrical facilities using various sensors is required, but active
- 308 -
use of sensors leads to cost and it is important to ensure the However, in this case, artificial intelligence must learn
safety of electrical facilities. through big data to distinguish between normal and failure
signals, but facility failures are not common, so artificial
Sensors such as vibration, sound, temperature, gas, etc. intelligence identifies failure signals by itself and predicts
can identify abnormal signs of electrical facilities, and data failures by comparing the frequency and intensity of failure
such as PD (partial discharge), power quality, etc. can be signals.
secured by measuring voltage/current, which is an electrical
characteristic. The advantage of this method is that multiple types of
facilities are monitored under the same conditions through the
Electrical facilities installed and operated in buildings same sensor, so consistency in data can be secured, and
have their own characteristics, and the facility's normal artificial intelligence can improve precision by learning
operation period is different, and the facility's soundness patterns that identify failure signals on its own through
should be determined by grasping details. unsupervised learning of abnormal signals.
The disadvantage is that it is difficult to determine the
cause of the failure of the facility's abnormalities, and the
TABLE III. EXAMPLES OF VARIOUS SENSORS FOR ELECGTRICAL
EQUIPMENT DIAGNOSIS result value may vary depending on the resolution and
precision of the sensor.
FACILITIES History management contents
Weight, body insulation flow rate, reference impac
t voltage insulation strength, transformer type, iron
core type, neutral point arrester, neutral point grou
Transformer
nding, OT protection and cooling method, rated vo
ltage(1,2,3rd), rated current, reference capacity, im
pedance, connection
Rated voltage, rated current, operating method, dri
Disconnector
ve motor, type
Rated voltage, rated current, reference impact volt
age insulation strength, cut-off capacity, cut-off sp
eed, maximum cut-off current, open time, no load i
Circuit breaker
nput time, trip power voltage, input power voltage,
operation type, weight, OCB flow rate, opening/cl
osing method, breaker type Fig. 2. Diagnostic Mechanism of Monitoring Type
Current Rated voltage, raated current, insulated method, C
transformer ore ratio
Rated voltage, nominal discharge current, residual
The diagnostic mechanism can perform a real-time
Lighting arrester voltage, power frequency discharge voltage, impul diagnosis process through a sensor installed in the facility's
se sparkover voltage, weight measurement element, and performs monitoring and diagnosis
Rated voltage, rated current, type, BCT core ratio,
based on the standard values that allow the facility to operate
Bushing stably.
BCT core burden
Primary/Secondary voltage, insulation flow rate, w The measurement-type diagnostic system has a reference
Discharge coil
eight value for normal operation of the facility, and monitors in real
time whether the value measured by the sensor exceeds the
Electric power
condenser
Circuit voltage, bank current, bank capacity, cell reference value. To this end, it is necessary to learn normal
signals that artificial intelligence can learn, and it has the
Charging Input voltage, input current, output voltage, droppe advantage of analyzing abnormal signals to determine the
equipment r operation method cause of the facility's failure.
Potential However, in order to determine the cause of the failure of
Type, potential transformer voltage
transformer the facility, it is necessary to understand the phenomenon that
causes the failure, so artificial intelligence learning through
big data is necessary through "supervised machine learning."
In addition, artificial intelligence can analyze and upgrade
Real-time monitoring/diagnosis mechanisms can be failure signals according to the failure mechanism to improve
largely divided into two methods: monitoring and diagnosis. the diagnosis mechanism on its own for failure precursor
In the case of surveillance-type mechanisms, the goal is to symptoms.
preferentially grasp the presence or absence of abnormalities
in real-time electrical facilities based on predetermined The advantage of this method is that it is easier to identify
general-purpose sensors. the cause of the failure of the facility because sensors are
selected and measured according to the facility, and the
The abnormal signal learned through the sensor is learning difficulty of the failure pattern increases according to
determined, and the failure type of the facility is clustered and the reinforcement learning conditions.
notified to the safety manager. This requires an algorithm that
allows artificial intelligence to isolate abnormal signals from Detailed data on facilities can be secured as a disadvantage,
normal signals by itself through unsupervised machine but life prediction due to deterioration is expected to be
learning. difficult to grasp due to the large number of cases because it
- 309 -
is necessary to learn based on information from various The figure below shows a new relationship diagram using
sensors. FTAs and FMEA for short circuit breakers. In the figure, it is
necessary to compare and analyze abnormal signals from the
measurement sensor with the cause of failure in the real-time
monitoring/diagnosis flow from left to right around the
measurement sensor to find out the failure signal by itself.
ACKNOWLEDGMENT
This work was supported by the Korea Institute of Energy
Technology Evaluation and Planning(20215910100080).
Fig. 3. Diagnostic Mechanism of Measurement Type
REFERENCES
IV. DESIGN OF REAL-TIME MONITORING AND DIAGNOSTIC
ALGORITHM FOR ELECTRICAL FACILITIESD
[1] Min-Hee Kim, Kyo-Hong Jin, “Development of a Deep Learning
For real-time monitoring and diagnosis of electrical Algorithm for Anomaly Detection of Manufacturing Facility” Journal
facilities, it is first necessary to understand the mechanism for of the Korea Institute of Information and Communication Engineering,
facility failure. Basic data on the safety management level, vol.26, No.2, pp. 199–206, Feb 2022.
such as the year of manufacture and installation of installed [2] Ji-Young Jung, Failure Diagnosis Using Text Mining and Deep
Learning: Development of Prediction Algorithm for Responsible
electrical facilities, should be obtained to identify which type Department of the Failure, Thesis, 2019.
of failure mechanism falls under. [3] Jong-Min Park, Seung-Hyun Ahn, and Tae-Jin Ha, “Development
In addition, a real-time monitoring/diagnosis algorithm Design for Multiple Agent Linked ICT Based Electricity Safety
Reinforcement Operation System”, Journal of Information Technology
can be designed through a relationship diagram by connecting and Applied Engineering, vol.11, no.2, pp.17-20, 2021.
the analysis of the cause of failure with the analysis of the [4] Jeong-Yul Seo, “Analysis of Failure Causes and Life Time Prediction
cause of failure using FTA and FMEA methods. of Low Voltage Circuit Breaker using Data Mining”, Thesis, 2010.
Fig. 4. Example of a figure caption.
- 310 -
Leakage Current Analysis of PCS
considering the Ground System
Dongmin Kim 1 and Hanju Cha 1*
1
Chungnam National University, Republic of Korea
Abstract—In a distributed power system using a PCS, a
parasitic capacitance exists between input stage modules and
ground due to the structural characteristics. Bad effects of the
parasitic capacitances are mitigated by PWM technique or
isolation transformer. However, leakage current paths are
created through a parasitic capacitance across the transformer
and cable connected to the grid, and thus leakage current flows
even in the IT ground system. In this paper, the leakage currents
in the IT ground system with or without a cable are discussed in
the 3-phase 3-level PCS system, and the leakage current effect
by the cable is verified through simulation and experiment.
Keywords— leakage current, cable impedance, transformer
impedance, ground system
Fig. 1. 3-phase 3-level PCS system structure
I. INTRODUCTION
In the distributed power PCS system, parasitic capacitance When a common mode voltage is applied to the CP, a
occurs between the input module and the ground due to leakage current can flow, and the generated leakage current
structural characteristics, and the parasitic capacitance ranges is proportional to the magnitude of parasitic capacitance,
from several nF to several uF depending on conditions such common mode voltage, switching frequency and dv/dt of
as frame structure, climatic conditions, and distance between PCS switching pulses.
cells [1]. When the common-mode voltage caused by PWM To reduce the leakage current generated in the PCS system,
switching is applied across the parasitic capacitance, a a transformer is installed between the PCS and grid. In the
leakage current may flow. Magnitude of the leakage current case of an ideal transformer, there is no parasitic capacitance
depends on the size of the parasitic capacitance, slope of the CT inside the transformer. Thus a leakage current path is
common-mode voltage, and the switching frequency, and it blocked and leakage current does not flow. However, a stray
may increase EMI and safety issues [2]. The leakage current capacitance exists between the primary and secondary
is reduced by using the isolation of the transformer [3]. winding in the real transformer, and then leakage current is
However, the Y-Y transformer contains a parasitic able to flow through the transformer. A leakage current path
capacitance between the transformer windings which makes is created due to the parasitic capacitance existing inside the
a leakage current path allowing a leakage current to flow. In transformer, and the leakage current is a zero sequence
addition to the transformer leakage path, significant length of component. Fig. 2 shows a PCS system connected to IT
cable between the PCS and grid is able to create a new ground system. In order to measure a stray capacitance of the
leakage current path. In this paper, a common-mode voltage transformer, the primary and secondary windings are short-
that generates the leakage current in a 3-phase 3-level circuited, and impedance between neutral points of the Y-Y
transformer (the primary neutral terminal P and secondary
converter is discussed, and leakage current before and after
neutral terminal S) are measured. In Fig. 2, there is a cable on
adding a cable in the PCS system connected as IT system is
two paths between the PCS and the transformer or between
also analyzed and verified by experiment. the CP and the transformer. Since an actual cable has an
II. LEAKAGE CURRENT GENERATED IN PCS impedance which modeled as a resistance, an inductance and
a capacitance among the 3- phase line and the neutral line,
Fig.1 shows the 3-phase, 3-level PCS configuration, and it
consist of the input DC source such as PV and battery, IGBT
power stack, LCL filter, grid and stray capacitance CP. The
space vector PWM(SVPWM) method is applied to the 3-level
converter, and the magnitude of common mode voltage at the
3-level outputs has discrete values of ±VDC/3, ±VDC/6 and 0.
Formula of the common mode voltage generated by PCS
switching is expressed at equation (1) as sum of the output
phase voltages referenced from the DC link neutral point O.
𝑉 = (1)
Fig. 2. PCS connected to IT-type ground system
- 311 -
a new leakage current path is represented by the capacitances influenced by the cable impedance, internal parasitic
CC. Therefore, understanding of the cable and its equivalent
circuit for the 3 phase 4 wire cable are required. With the use
of transmission line model, the propagation constant of a
cable can be derived as follows.
Z R jL (2)
Y G jC (3)
R jL G jC (4)
If the cable is considered as a lossless system, then the
characteristics impedance of the cable can be simplified as
R j L R j L L (5) (a)
Zo
G jC C
III. EXPERIMENTAL RESLTS
PCS system connected to the IT ground system is
assembled to analyze the path of leakage current. Fig. 3 shows
the experimental set-up, (a) shows the 10 kW 3-phase PCS,
and (b) shows the 3-phase 4-wire cable. A Y-Y transformer is
used for the transformer, and the length of 16m (or 50m) of
IEC 60227-3 450/750V 25mm2 is used for the cable. System
parameters are : 220V, 60 Hz of AC grid, 500V of VDC ,
switching frequency of 10kHz, and parasitic capacitance CP of
1.5uF.
(b)
Fig. 4 Experimental results (a) leakage current (b) expended waveform of
common mode voltage and leakage current
Fig. 3 Experiment set-up (a) PCS and LCL filter (b) 3 phase 4 wire cable
Fig. 4 (a) shows the waveform of the leakage current
flowing in the IT ground system. In the IT ground-type PCS
configuration, the leakage current is a lot reduced by using Fig. 5 Experimental results of leakage current after adding 16m cable
the isolation characteristics of the transformer compare to the
TN ground system. However, due to a parasitic capacitance capacitance of the transformer and effect of stray
inside the actual transformer, a leakage current path is formed, capacitances of CP CT and CC have been discussed. The
and it enables a leakage current to flow through the leakage current increased from 53.7mA to 63.8mA after
transformer. Therefore, about 53.7mA of leakage current adding a cable.
flows in the IT ground-type PCS. Fig. 4 (b) shows expanded
.
waveforms of common mode voltage and leakage current
flowing through the transformer primary and secondary REFERENCES
windings. Since the leakage current is generated by the
[1] J. Wang, Y. Xun, X. Liu and S. Yu, "An on-line extraction method for
common mode voltage with the period of 100usec, leakage the parasitic capacitance of the photovoltaic panel," in CPSS
current has the same period of 100usec, and illustrates a shape Transactions on Power Electronics and Applications, vol. 4, no. 4, pp.
of surge current at the moment of the common mode voltage 320-327, Dec. 2019
transition. Fig. 5 shows the leakage current after adding 16 m [2] J. Wang, X. Liu, Y. Xun and S. Yu, "Common Mode Noise Reduction
length of cable between the PCS and transformer. The of Three-Level Active Neutral Point Clamped Inverters With Uncertain
leakage current increases to 63.8mA after adding the cable Parasitic Capacitance of Photovoltaic Panels," in IEEE Transactions on
Power Electronics, vol. 35, no. 7, pp. 6974-6988, July 2020
because capacitances between 3-phase lines and the neutral
[3] V.Grishanov, Evgeniy and S. Brovanov. “Theoretical Aspects of the
line of the cable builds up a new leakage current path. Common-Mode Leakage Current Suppression in a Photovoltaic
Power Generation System Based on Multi-Level H-Bridge Type
IV. CONCLUSION Converters”, 2018 XIV International Scientific-Technical Conference
on Actual Problems of Electronics Instrument Engineering (APEIE),
This paper studied the leakage current of PCS system pp. 32-37, 2018.
connected to the IT ground system. Leakage current
- 312 -
Estimation of the probability of
disconnection of urban electric networks in
emergency situations based on the use of
GIS maps
Y.Umbetkulov*1, K.Bakenov1, T.Аgimov1, M.Khazimov1, L.Uteshkalieva1
1
Almaty University of Power Engineering and Telecommunication named G.Daukeev, Kazakhstan
Email: [email protected]
networks with the dispatching control of the power system;
Abstract - The article discusses the use of geoinformation - information about the availability or supply of materials
systems (GIS) in assessing the probability of disconnection of for temporary power lines.
urban electrical networks in emergency situations (strong In recent years, GIS maps have appeared in almost many
earthquakes). It is noted that one of the effective tools for cities around the world. The use of GIS in the energy system has
managing and monitoring the general condition of distribution
electric networks is the use of GIS maps reflecting the spatial
significantly increased efficiency in the energy sector. Their
position of transformer substations and power transmission lines. main advantage: visibility and informativeness.
The probabilities of mechanical damage or disconnection of The electrical network in GIS can be visualized. Just by
elements of consumer transformer substations and 10 kV power looking at the map and clicking on a certain function (say, a
transmission lines were assessed. Based on the use of GIS maps, a transformer), all the information (both location and technical
method for quantifying damage to power transmission lines and information) will be displayed. This provides a more flexible
consumer transformer substations at different earthquake understanding of the electrical network and, consequently, a
intensities has been developed. faster approach to solving production problems.
The methodology for analyzing and assessing the probability GIS is not only computer systems designed to create maps,
of disconnection is recommended for specific or a group of
substations in busy neighborhoods or districts of the city. At the
but also powerful tools for geographical analysis. GIS can be
same time, the collection of statistical data can be recorded on GIS defined as integrated sets of hardware, software, databases and
maps of such substations of urban distribution networks. The processes designed to collect, pre-process, analyze and visualize
research results are a useful reference material for planning and data that can be spatially localized and linked.
organizing emergency power supply in natural and man-made
emergencies. II. RESEARCH METHODS
Keywords - emergencies, electrical networks, transformer Currently, large cities are interested in creating a
substation, probability of disconnection, GIS map comprehensive GIS map, taking into account the application of
all important communications. This will make it possible to
I. INTRODUCTION effectively solve the issues of urban planning, construction of
At the present stage of the development of innovation and metro stations, underpasses, new facilities, transformer
digitalization of spatial images, the use of maps developed on substations and power transmission lines.
the basis of geoinformation technology or geoinformation In this article, fragments of GIS maps developed by
systems (GIS) seems to be a rather interesting tool for scientists of the Almaty University of Energy and
operational management and monitoring of the operational Communications named after G.Daukeev were used to assess
state of urban distribution electric networks (DEN). the probability of damage to elements of urban electrical
GIS maps allow you to create spatial models of real networks [4].
electrical networks with fixing the locations of supply Figure 1 shows a section of 6-10 kV distribution electric
transformer substations (TS) and consumer transformer networks in one of the districts of Almaty. Cables of 6-10 kV
substations (TS) on objects, as well as to designate the real are marked with red lines, and consumer transformer substations
trajectories of overhead and cable power lines. In addition, the TS-6-10/0.4 kV are marked with squares.
use of GIS maps provides optimization of the design of power Note that in the GIS map applications, the characteristics
plants, analyze and control electrical loads and timely assess of transformer substations are plotted, their coordinates, types
the technical condition of electrical equipment and timely carry of transformers, their power and other parameters. In the part of
out the necessary measures for their repair and maintenance or cables, their type, brand, length and other characteristics are
expansion of the power facility, its modernization, maintenance indicated. GIS data, as a rule, is constantly updated.
and repair, expansion and optimization of the power system, etc.
[1-3].
The option of centralized power supply in emergency
situations depends on the operability of existing electrical
networks and information from GIS. The logistics flow in this
case consists of the following main elements:
- flow of information about the state of the object;
- flow of information about the state of electrical networks
(from the power system to TS- 10/0.4 kV);
- coordination of operational switching in electrical
- 313 -
taking into account the opinions of specialists in this industry.
The "failure tree" method is the most suitable method for
assessing the risk of accidents at power plants, since the
graphical part of the failure tree is very similar to the wiring
diagram. Conventions and logical operators for these events are
used to create a "failure tree" [6,7].
Statistical data or quantitative values of the probability of
a major failure allow us to obtain a quantitative estimate of the
average annual frequency of accidents at a particular power
plant.
As an example, an assessment of the probability of
electricity supply to TS 1, located near a multi-storey residential
building, with possible strong earthquakes of 7 and 8 points was
carried out. On the GIS map, the TS1 in question is powered by
separate cable lines from TS 2 and TS3, which, in turn, receive
electricity from different PS with a voltage of 35/10 kV.
The design scheme of this section of the electrical network
is shown in Figure 3.
Figure 1- GIS map of the city
In case of strong earthquakes, there is a high probability
of disconnecting the centralized power supply for two main
reasons:
- damage (in whole or in part) to electrical networks and
systems;
- dispatching shutdown of electrical networks due to the
Figure 3 - Design scheme of the electric network section
prevention of possible fires or gas explosions from short
circuits in the electrical wiring of buildings.
Using a fragment of such a GIS map shown in Figure 2, Using well-known logistics operators AND and OR,
Figure 4 shows a "failure tree" for the example under
this article estimates the probability of a power outage to a
consideration, that is, a block diagram for assessing the risk of
10/0.4 kV consumer substation based on known methods of
power outages (or failure) at TS1 [7].
logistic calculation.
Figure 4 - «Failure tree» of the risk assessment of blackouts
Figure 4 shows that the disconnection of TS1 (marked A1)
depends on damage to the cable line (B 1) or mechanical damage
to TS1 (B2) itself or simultaneous disconnection of TS2 and
TS3.
A similar pattern is observed in other chains of the scheme.
Well-known recommendations on logistics operators AND
Figure 2 - Cable lines and consumer TS-10/0.4 kV and OR were used to build mathematical models. The analytical
formulas for calculating the probabilities of TS1 (PA1)
If it is necessary to quantify and qualitatively assess disconnection in an earthquake of magnitude 7 and 8 are given
probable accidents at DEN, well-known methods are used [5,6]: below.
- graphoanalytic, consisting in the use of logical methods
for analyzing "failure tree" and "event tree" or design models РА1 = 1 – (1 – РВ1)( 1 – РВ2)( 1 – РВ3) (1)
of the structure;
- statistical, consisting in the fullest possible use of РВ3 = РС1·РС2 (2)
accident and malfunction statistics, as well as data on the
reliability of analog objects; РС1 = 1 – (1 – РD1)( 1 – РD2)( 1 – РD3) (3)
- expert, which consists in developing an assessment by
- 314 -
РС2 = 1 – (1 – РD4)( 1 – РD5)( 1 – РD6) (4) probability of TS1 shutdown decreases, earthquake with
magnitude 7 points from A1 = 0.42 to A1 = 0.14 and with
РD3 = 1 – (1 – РE1)( 1 – РE3) (5) magnitude 8 points - from A1 = 0.75 to A1 = 0.24. This
increases the chance of supplying electricity to TS1 and
РD4 = 1 – (1 – РE2)( 1 – РE3) (6) ensuring successful rescue operations even in case of strong
earthquakes.
The quantitative values of the expected probable failures It should be noted that the dispatching shutdown (in order
of the elements of the "tree of failures" were determined as to avoid gas explosions and fires from electricity) can be
follows: reduced by turning off switchgear built-in or attached to
РD1, РD4 and РB1 – from literary sources [8,9]; residential multi-storey buildings that are powered by TS-10/0.4
РE3 – forecast of the Department of Emergency Situations kV. These simple operations can be carried out by electricians
of the city of Almaty; on duty or operational services of electrical networks.
РD2, РD5, РB2, РE1, РE2 – forecasts of specialists of energy If it is impossible to supply electricity through electric
companies and scientists of Almaty University of Energy and networks, various MIE will have to be used for rescue
Communications named after G.Daukeev; operations [10,11], in the future – an expanding fleet of electric
РA1, РB3, РC1, РC2, РD3 and РD6 are the results of cars or electric buses.
calculations.
Here is a detailed calculation РА1 (for example, when РE3 III. CONCLUSION
= 0.5):
РД6 = 1 – (1 – 0,1) (1 – 0,5) = 0,55 It is shown that the main task of GIS is to model "real
electrical networks", conduct spatial analysis and ensure high
РД3 = 1 – (1 – 0,1) (1 – 0,5) = 0,55 accuracy of the optimization procedure. This helps to carry out
various analyses of the electrical load consumed, the location
РС1 = 1 – (1 –0)( 1 –0,1)( 1 – 0,55) = 0,6 and power of transformer substations and the optimal trajectory
of laying cable and overhead power lines. In addition, GIS maps
РС2 = 1 – (1 – 0)( 1 – 0,1)( 1 – 0,55) = 0,6 of DEN can be used to navigate the architecture of the city
during the construction of bridges, pedestrian crossings and
РВ3 = РС1·РС2 = 0,6 ·0,6 = 0,36 various buildings and structures.
The possibility of using GIS maps to assess the probability
РА1 = 1 – (1 –0)( 1 –0,1)( 1 – 0,36) = 0,42 of disconnection of TS-10/0.4 kV consumer substations of
electric networks in case of strong earthquakes is shown.
Calculations were carried out for probable earthquakes of Methodological bases for estimating the probability of
magnitude 7 and 8 points. The results are shown in Table 1. disconnection of individual or a group of TS 10/0.4 kV urban
electric networks using GIS maps are proposed. A "failure tree"
Table 1 – Results of accident risk assessment
has been developed to quantify the probability of disconnecting
Elements Name magnitude magnitude
consumer substations at different earthquake intensities.
of the of the " failure 7 points 8 points
Calculations showed that the probabilities of a power
"failure tree" element РЕ3=0,5 РЕ3=0 РЕ3=0,75 РЕ3=0
outage on TS1 during an earthquake of magnitude 7 points were:
tree"
A1 = 0.42 (with a dispatcher shutdown E3 = 0.5) and A1 = 0.14
РА1 Power outage on 0,42 0,14 0,75 0,24 (with E3 = 0), respectively, with magnitude 8 points - A1 = 0.75
TS1 (with a dispatcher shutdown E3 = 0.5) and A1 = 0.24 (when E3
РВ1, РD1, Cable lines 0 0 0 0 = 0).
РD4 Currently, GIS technologies and GIS maps are successfully
РВ2, РD2, Mechanical 0,1 0,1 0,15 0,15 used in emergency situations of natural (earthquake, flood) and
РD5 damage to the man-made nature (industrial accidents and catastrophes). In
TS 1, TS2, TS3 particular, for the management and routing of mobile power
of the outdoor sources and the organization of emergency and restoration work
installation on bridges and other activities [10,11,12].
РD3, РD6 Power outage on 0,55 0,1 0,8 0,15 The results obtained can be used in emergency services.
PS 35/10 kV
FUNDING
РС1, РС2 Power outage on 0,6 0,19 0,83 0,3
TS 2 and TS3
This research has been funded by the Science Committee
РЕ1, РЕ2 Mechanical 0,1 0,1 0,15 0,15 of the Ministry of Science and Higher Education of the Republic of
damage to the Kazakhstan AP19679469 "Development of a wide-range valve
PS of the generator and stable output voltage for small wind power plants"
outdoor
installation ACKNOWLEDGMENT
РЕ3 Dispatcher 0,5 0 0,75 0
shutdown The authors would like to thank the organizing committee
of ICEF2023 for the giving opportunity to publish the paper.
Table 1 shows that the probability of a power outage from
TS1 at an earthquake intensity of magnitude 7 points was PA1 REFERENCES
= 0.42. At the same time, the probability of disconnecting the
electrical network by dispatchers is assumed to be E3 = 0.5. [1] S. Mathankumar and P. Loganathan. GIS Based Electrical System
Planning and Network Analysis // World Engineering & Applied
In the absence of a dispatcher shutdown (E3 = 0), the Sciences Journal 6 (4): 215-225, 2015. ISSN 2079-2204
- 315 -
[2] N. Rezaee, M. Nayeripour, A. Roosta, T. Niknam. Role of GIS in
Distribution Power Systems //World Academy of Science.
Engineering and Technology / International Journal of Electrical and
Computer Engineering. Vol:3,№:12, 2009.
[3] Umbetkulov Y., Sakhariyev B., Amitov Y. On the question of
providing electricity for rescue and other emergency operations in
earthquakes / AUPET Bulletin, Series of Technical Science. №2(49),
2020 – p.54-61.
[4] Sakhariev B.S. Practical application of geoinformation technologies
in the analysis of electrical networks // VIT-2004, Almaty, KazNU,
2004. pp. 23-27.
[5] Guidelines for conducting risk analysis of hazardous production
facilities (RD-03-418-01 ). Approved by the resolution of
Gosgortehnadzor of Russia dated 10.07.2001 No. 30.
[6] Мethodological guidelines for conducting an analysis of the risk of
accidents of hydraulic structures. STP VNIIG 230.2.001-00 / Edited
by E.N. Bellendir, S.V. Solsky, N.Ya. Nikitina. SPb.: JSC "VNIIG
named after B.E. Vedeneev". 2000.
[7] Тechnogenic risk: textbook / N.N.Chura; edited by V.A.Devisilov. –
M.: KNORUS, 2011.- 280s.
[8] Operational forecasting of the engineering situation in emergency
situations / under the general editorship of S.K. Shoigu, G.P. Sakov,
M.P. Tsivilev, I.S. Polyakov, etc. – M.: CJSC "PAPYRUS", 1998. –
166c.
[9] Kurdyumov V.I., Zotov B.I. Design and calculation of security means.
– M.: KolosS, 2005.- 216 p.
[10] Umbetkulov E.K., Sydykov S.K. The use of electric generating sets
based on driving gear of a tractor in emergency situations in the city
of Almaty // Scientific journal "Research, results" KazNAU, Almaty,
№ 3 (055) 2012-S.150-153
[11] Tsai, M.-K., & Yau, N.-J. Using mobile disaster response system in
bridge management //Journal of Civil Engineering and Management,
20(5),2014. pp 737-745.
[12] P.G. Mikhailov, A.O. Kassimov, Y. Umbetkulov. Systems of
Monitoring and Control of Emergency Situations in Facilities and
Territories of the Republic of Kazakhstan//International Multi-
Conference on Industrial Engineering and Modern Technologies
(FarEastCon). 6-9 Oct. 2020.
- 316 -
A Study on Common Mode Effects
According to Battery Rack
Configurations in Energy Storage Systems
Yong-Eun Choi 1, Hyo-Seok Oh 1, Byung-Min Moon 1, Chin-Young Chang 1 and Jae-Moon Kim 1*
1
Korea National University of Transportation, Korea
Email: [email protected]
Abstract— Energy storage system is known to increase noise This is a method of measuring CMV by installing a
and increase the possibility of common mode voltage generation measuring instrument behind the Li-ion battery rack and using
depending on the installation method of transformers and the DC voltage.
number of battery racks, and may cause malfunctions of
facilities. Therefore, in this paper, the common mode voltage
effect according to the increase in the number of battery racks
B. CMV measurement method
was examined using the ungrounded transformer of Energy
storage system. ESS CMV is measured through the CMV instrument
installation method shown in Figure 2. To measure CMV in
Keywords— Energy Storage System, Common Mode Voltage, the configured circuit, connect the (+) and (-) poles of the Li-
Power Conditioning System, Li-Ion Battery, Isolation ion battery Rack to the Voltage Probe respectively[3]. Then,
Transformer connect the ground to the enclosure of the battery rack and
measure it through the CMV instrument of the ESS.
I. INTRODUCTION
An ESS(Energy Storage System) is a technology that has
recently attracted worldwide attention due to the need for
depletion of fossil fuels and reduction of carbon emissions.
However, depending on the installation method of the ESS
transformer, it is known that noise increases and the
possibility of CMV(Common Mode Voltage) increases,
which also causes malfunction of the facility[1]. Therefore, in
this paper, a power conversion system was constructed using
power conversion software, and the CMV change due to the
increase in battery rack according to the grounding method of
the ESS transformer was analyzed.
Fig. 2. ESS Li-ion battery rack CMV meter installation method
II. CIRCUIT CONFIGURATION FOR ESS CMV MEASUREMENT
C. ESS transformer and battery rack configuration
A. ESS configuration circuit The transformer of ESS is grounded as shown in Figure 3,
The composition of ESS can be largely divided into AC and it can be composed of Figure 3(a) ungrounded type and
grid, PCS(Power Conditioning System), and battery, and Figure 3(b) grounded type.
Figure 1 shows the composition of ESS. AC power consists of
a linking transformer and protection device, and PCS consists
of IGBT, DC capacitor, and filter. The battery part consists of
a Li-ion battery and a protective device[2].
Power Conditioning System(PCS) Li-Ion Battery
P Voltage
Probe
AC Grid
TR (+)
Instrument
Measuring
CMV
(-)
N
(a) Ungrounded (b) Grounded
Ground
(Recommended) (Not recommended)
Fig. 1. ESS CMV test circuit configuration Fig. 3. ESS transformer grounding and battery configuration
- 317 -
In the AC power side of the ESS, the linkage transformer Simulation CMV showed 530[V]. Although within the
plays an important role. However, depending on the normal range below the standard M ax 1000[V], it was
grounding method of the transformer, it may cause CMV to confirmed that the peak to peak of CMV was about 70[V].
occur. Therefore, grounding type is not recommended in
Korea, and non-grounding type is recommended.
AC-side transformers of lithium-ion batteries currently B. CMV measurement of 10 battery racks
used in most ESSs can cause an increase in CMV depending ESS 10 battery racks were connected in parallel and the
on the characteristics of the grounding method. In addition, CMV measurement result waveforms are shown in Figure 6.
CMV tends to increase as the number of racks of batteries
decreases. The standard in Korea follows the battery
manufacturer's standard, and the standard is limited to Max
1000[V] when the rack's highest voltage is less than 999.6[V].
III. ESS CIRCUIT MODELLING
In this paper, the ESS simulation modeling used the
ungrounded transformer in Figure 3(a) of the recommended
transformer grounding method in Figure 3. In addition, to
analyze the occurrence of CMV according to the change in the Fig. 6. 10 battery racks CMV waveform
number of ESS battery racks, the circuit as shown in Figure 4
was configured. The circuit is divided into AC power, tie-in It was confirmed that the CMV of 10 battery racks
transformer(ungrounded), PCS and battery. The battery part connected in parallel was 500[V] within the standard value,
was additionally connected in parallel so that several racks and the peak to peak of the CMV was about 5[V]. Therefore,
could be connected. it was confirmed that the CMV value decreased as the number
of battery racks increased.
V. CONCLUSION
In this paper, the results of the simulation to analyze the
CMV effect when the battery rack increases according to the
ESS transformer grounding method are as follows.
Comparing CMVs connected in parallel with 1 battery
rack and 10 battery racks of ESS, it can be seen that there is a
Fig. 4. ESS circuit modelling
difference in CMV value of about 30[V]. In addition, as the
The transformer was applied considering the number of battery racks increases, checking CMV Peak to
transformation ratio at 380[V]/380[V], and the switching Peak confirms that 65[V] decreases. Therefore, when
frequency of the PCS switching element was set at 5[kHz]. In considering the safety of the ESS, the grounding method
addition, the battery rack configuration was configured by characteristics of the AC-side transformer and the number of
connecting in series assuming two battery modules of 499[V] battery racks must be carefully considered.
each. In addition, assuming that several racks exist, battery
racks were configured in parallel. The maximum voltage of
the battery rack was simulated to approximate 999.6[Vdc]. ACKNOWLEDGMENT
This work was supported by Korea Institute of Energy
Technology Evaluation and Planning(KETEP) grant funded
IV. CMV ANALYSIS SIMULATION ACCORDING TO ESS by the Korea government(MOTIE)(20220610100010, R&D
BATTERY QUANTITY CHANGE Project of Establishment for Renewable Energy-linked ESS
Safety Evaluation Center: based on PV and FC)
A. CMV measurement of one battery rack
Figure 5 shows the waveform of one battery rack using a
ungrounded ∆-∆ transformer. REFERENCES
[1] Y. E. Choi, H. S. Oh, B. M. Moon, J. M. Kim, “A Study on the Effect
of Common Mode Voltage(CMV) by Switching Device Short Circuit
During ESS Operation,” The Korean Institute of Electrical Engineers
(KIEE) conference proceedings, pp. 237-238, 2022.
[2] S. M. Choi, S. H. Kim, M. Y. Kim, D. S. Rho, “A Study on
Characteristics and Modeling of CMV by Grounding Methods of
Transformer for ESS,” Journal of the Korea Academia-Industrial
cooperation Society, vol. 22, no. 4, pp. 587-593, 2021.
[3] S. H. Kim, H. S. Choi, J. Y. Kim, A. Han, “A Study on the Cause of
CMV and Leakage Current, and Mitigation Strategy on the Lithium-
ion Battery of ESS,” The transactions of The Korean Institute of
Electrical Engineers, vol. 70, no. 1, pp. 61-71, 2021.
Fig. 5. One battery rack CMV waveform
- 318 -
Development of 48V PbC Hybrid Battery for Electric
Forklift Using High Quality Activated Carbon
Hoon Seo 1, In Seok Seo 1, Hyon-ok Lim 1and Kyung-Wan Koo 2*
1
DongYoung Ind. Co. Ltd., Republic of KOREA
2
HoSeo University, Republic of KOREA
Email: [email protected]
Abstract—Lead-acid batteries are used as power sources in activated carbon. The growth of lead sulfate crystals by an
industrial sites because of their low manufacturing costs and electrode reaction in the Pb cell increases electrode resistance
excellent price competitiveness compared to other secondary and eventually reduces the cycle life of the cell. By attempting
batteries. Unlike conventional Pb batteries, hybrid lead-carbon to introduce carbon with excellent electrical conductivity to
(PbC) batteries have been developed that take advantage of both reduce resistance, it is necessary to facilitate the transfer of
general lead storage batteries and capacitors, and PbC batteries electrons in the electrode and induce the reduction of lead
are attracting technical attention by overcoming shortcomings sulfate to metal lead to suppress the increase in resistance. To
of existing Pb batteries.In the battery market for forklifts and
this end, in a PbC hybrid battery, activated carbon physically
construction equipment, there is an urgent need to develop PbC
attached with lead is mixed with an electrode binder and then
hybrid battery technology with high power/long life
characteristics that may handle strict environments such as high
coated on an anode lead plate to manufacture a carbon
efficiency charge/discharge and discharge depth of 50%. electrode, thereby facilitating electron transfer and inducing a
reduction of lead sulfate to metal lead.
Keywords—Lead-carbon (Pb/C) battery, Secondary Battery ,
Bio-Cellulose (BC) membrane, Forklift
I. INTRODUCTION
Compared to conventional forklifts using internal
combustion engines using diesel or PLPG, electric forklifts
have low maintenance costs and are difficult to revitalize the
market due to their low life (2~3 years) and low output power
characteristics. PbC hybrid battery technology is being
demonstrated and commercialized in various application
fields (start-up, industrial, ESS), but in Korea, it is weak to
localize components and develop system completion
technology. Therefore, in order to localize PbC hybrid battery
parts with long life and high power characteristics and develop
system completion technology, it is urgently required to
develop activated carbon-based active material core part
technology with low performance due to carbon
corrosion/electrolysis. Battery requirements for electric
forklifts should preferentially ensure a long life of 10 years or
more, high safety without low cost and explosion risk, and
high reliability of technology that operates stably under Fig. 1. Edge regulation strategy with activation
extreme environmental conditions such as high temperature and physical/chemical post-processing
and low temperature. Accordingly, the hybrid PbC battery
technology to be developed in this research is a new battery
technology with complementary characteristics as a fusion Adjustment of properties through material surface
technology of a general lead storage battery and a capacitor. treatment
In other words, PbC batteries are next-generation battery
technologies that realize the advantages of capacitors and Improvements in the electrochemical performance of
secondary batteries, such as long life, high power, fast charge, carbon materials have been reported by doping carbon
and partial charge discharge characteristics. materials with B, F, N, etc. or controlling the interaction
between carbon/carbon materials. Based on this research, it is
II. EXPERIMENTS intended to improve the electrochemical performance of PbC
A Study on the Change of Surface Properties and the nano-composite active materials by inducing carbon edge site
Improvement of Electrochemical Performance through activity. Edgeless activated carbon was synthesized by
Carbon Edge Control removing the edge present on the surface of activated carbon,
and lead-carbon batteries were manufactured by including
To suppress reactivity by edges on the surface of activated them in the anode in the PbC battery manufactured using
carbon, research is conducted on long-life, high-performance edgeless activated carbon. Verified that the presence of an
PCB cells that induce reduction of metal lead by improving edge at a certain rate in edgeless activated carbon suppresses
charging importability, including surface-modified activated the reactivity of the edge, thereby suppressing various
carbon and anode manufactured using the surface-modified additional reactions.
- 319 -
Fig. 2. Typical surface control technologies (high
temperature heat treatment, nitrogen plasma, and
boron doping)
Fig.4. Comparison of battery prices for forklifts
III. RESULT AND DISCUSSION
In the case of the Pb battery composition, the action
material of the (+) electrode is lead peroxide (PbO2), and the
action material of the (-) electrode is lead (Pb), and it is
manufactured in this electrode plate to freely dissolve, diffuse,
move, and penetrate an electrolyte. There is also a separator,
a nonconductive thin plate, between the two plates. Dilute
sulfuric acid (H2SO4) is used as the electrolyte, and the rated
voltage per cell is 2v. In particular, the PbC hybrid battery
forms a conductive network by using lead (Pb) and activated
carbon composite materials as the (-) electrode action material
to improve reactivity of the electrode active material and
suppresses the growth of lead sulfate (PbSO4) generated
during charging/discharging.
Fig.3. Characteristics of Electrodes and Electrodes with
Typical Cathode Materials
It was confirmed that the crystallinity of the activated
carbon was increased and the specific surface area was
adjusted by adding C3N4, and the physical properties of the
PbC battery were improved when the PbC anode was prepared
using the C3N4-activated carbon. It was verified that, by
developing a PbC battery using a carbon active material
coating layer, a charging speed was improved by 64% Fig.5. EUCAR Profile Cycle Test Results for Prototypes
compared to the existing battery, and that the cycle life was
improved by 4 times or more compared to the existing battery
system under low rate operation (DOD 17.5%).
IV. CONCLUSIONS
The mainstream of electric forklift batteries are lead storage
batteries and lithium-ion batteries, but due to the high cost of
lithium-ion batteries, most of them are lead storage batteries The Pb-C composite battery contributes greatly to improving
in Korea. The use of lead batteries in material handling the life of the lead storage battery by well suppressing the
equipment is cost effective and provides high resilience in growth of lead sulfate crystals on the electrode surface caused
poor environments associated with chemicals, dust and by the battery load caused by the fork operation of the forklift.
vibration. Lead batteries require less maintenance than other Pb-C composite batteries are advantageous for frequent
types of batteries, and their use and replacement plans are charging and discharging due to capacitor characteristics, so a
consistent, simple, and predictable. high life can be expected in an incomplete charging state
- 320 -
(PSoC) due to frequent charging and discharging of batteries [4] Sanna Lander, Mikhail Vagin, Viktor Gueskine, Johan Erlandsson,
Yselaure Boissard, Leena Korhonen, Magnus Berggren, Lars Wågberg,
such as ISG (Idle Stop & Go) function of a vehicle. and Xavier Crispin Adv. Energy Sustainability Res. 2022, 3, 2200016.
The Pb-C composite battery does not require a spare battery [5] V. Hospodarova, E. Singovszka, N. Stevulova, Characterization of
and a battery replacement operation through fast charging and Cellulosic Fibers by FTIR Spectroscopy for Their Further
frequent charging for idle time because the battery remaining Implementation to Building Materials, Am. J. Anal. Chem. 9 (2018)
capacity needs to be charged below 20%. Electric forklifts 303-310. doi:10.4236/ajac.2018.96023.
using conventional lead batteries require at least 8 hours of [6] G. Palanisamy, T. Sadhasivam, W.S. Park, S.T. Bae, S.H. Roh, H.Y.
Jung, Tuning the Ion Selectivity and Chemical Stability of a
charging to use for 3-4 hours and charge when the battery life Biocellulose Membrane by PFSA Ionomer Reinforcement for
is 20% or less. Therefore, forklifts cannot be used during Vanadium Redox Flow Battery Applications, ACS Sustain. Chem. Eng.
charging time, requiring extra forklifts for smooth operation 8 (2020) 2040−2051. doi:10.1021/acssuschemeng.9b06631.
Electric forklift operation can be divided into two types: [7] Z. Zhao, J. Zhou, T. Fan, L. Li, Z. Liu, Y. Liu, M. Lu, An effetive
surface modification of polyester fabrics for improving the interfacial
running and fork raising, load is applied constantly, while fork deposition of polypyrrole layer, Mater. Chem. Phys. 203 (2018) 89-96.
raising requires instantaneous high power. The Pb-C doi:10.1016/j.matchemphys.2017.09.062.
composite battery can respond to high power demand. [8] X. Wen, Y. Zheng, J. Wu, L. Yue, C. Wang, J. Luan, Z. Wu, K. Wang,
In vitro and in vivo investigation of bacterial cellulose dressing
containing uniform silver sulfadiazine nanoparticles for burn wound
healing, Prog. Nat. Sci. Mater. Int. 25 (2015) 197-203.
[References] doi:10.1016/j.pnsc.2015.05.004.
[9] R. Du, F. Zhao, Q. Peng, Z. Zhou, Y. Han, Production and
[1] Sung-Hee Roh, Gowthami Palanisamy, T. Sadhasivam, Jae-Eun Jin, characterization of bacterial cellulose produced by Gluconacetobacter
Jin-Yong Shim, and Ho-Young Jung, ACS Sustainable Chem. Eng. 7 xylinus isolated from Chinese persimmon vinegar, Carbohydr. Polym.
(2019) 8789-8797. 194 (2018) 200-207. doi:10.1016/j.carbpol.2018.04.041.
[2] X. Song, U. Cvelbar, P. Strazar, L. Vossebein, A. Zille, Chemical, [10] G. Pacheco, C.R. Nogueira, A.B. Meneguin, E. Trovatti, M.C.C. Silva,
thermo-mechanical and antimicrobial properties of DBD plasma R.T.A. Machado, S.J.L. Ribeiro, E.C. da Silva Filho, H. da S. Barud,
treated disinfectant-impregnated wipes during storage, Polymers Development and characterization of bacterial cellulose produced by
(Basel). 11 (2019) 1769. doi:10.3390/polym11111769. cashew tree residues as alternative carbon source, Ind. Crops Prod. 107
[3] L.I. Fockaert, S. Pletincx, D. Ganzinga-Jurg, B. Boelen, T. Hauffman, (2017) 13-19 doi:10.1016/j.indcrop.2017.05.026.
H. Terryn, J.M.C. Mol, Chemisorption of polyester coatings on [11] M.W. Ullah, M. Ul-Islam, S. Khan, Y. Kim, J.K. Park, Structural and
zirconium-based conversion coated multi-metal substrates and their physico-mechanical characterization of bio-cellulose produced by a
stability in aqueous environment, Appl. Surf. Sci. 508 (2020) 144771. cell-free system, Carbohydr. Polym. 136 (2016) 908-916.
doi:10.1016/j.apsusc.2019.144771. doi:10.1016/j.carbpol.2015.10.010.
p
- 321 -
Experimental studies of a model of a dusted flow air heater for the development of a new hot-water boiler
Otynchiyeva, M. T.1 Bakhtyar B.T.2* Korobkov M.S.3 Tursunbayeva G.U.4
1
Narxoz University, Almaty
2
S.Seifullin Kazakh Agro Technical Research University, Astana
3
Almaty University of Power Engineering and Telecommunications named after Gumarbek Daukeyev, Almaty
3
S.Seifullin Kazakh Agro Technical Research University, Astana
Email: [email protected]
Abstract— The article presents the results of the determined through the Nusselt number in finned heat
development and creation of a solid fuel hot-water boiler, exchangers largely depends on the mass flow rate, realizing
in the creation of which a model of an air heater was used that the mass flow rate at a constant cross section increases
in which the processes of flow of a gas flow dusty with ash velocity, and, accordingly, the Reynolds number, it is obvious
were studied. Presented are the results of thermal, that an increase in the Reynolds number leads to an increase
aerodynamic and structural calculations for a new type of in the efficiency of heat transfer. In [8], a vacuum-type air
hot water boiler with a behind-the-screen gas duct, as well heater with helical inserts was investigated. Screw inserts are
as with a new solid fuel air heater with a capacity of 30 good flow turbulizers, which in theory should increase the
Gcal/h. The calculation results were compared with the efficiency of heat transfer. Studies have shown that the
KV-TK-100-150 hot-water boiler. The results of the installation of screw inserts leads to an increase in hydraulic
economic effect from the introduction of a hot water resistance, but also increases the efficiency of heat transfer,
boiler with a capacity of 30 Gcal/h are presented. which is shown through thermal efficiency. In [9], the
influence of a change in the relative roughness on the
Keywords— hot-water boiler, biogas, biogas plants, thermohydraulic characteristics of an air heater was
waste, biogas technology, heating system investigated. Studies of the effect with transverse ribs [10]
showed a significant increase in the coefficient Nu depending
Introduction on Re. At Re = 13000, the Nu coefficient was equal to 120.
The overwhelming majority of modern heating boilers run The study of hydraulic resistances showed that the use of ribs
on fuel oil or gas. This fuel, as it is known, is extremely scarce, significantly increases the resistance of the air heater, for
therefore, it is necessary to develop measures for efficient example, at a flow rate of 0.039 kg/s, the difference between
combustion of solid fuel in heating boilers to reduce the smooth and ribbed pipes was 20 Pa. The use of fins made of
harmful effect on the environment. This is reflected in the porous aluminum led to an increase in the efficiency of heat
program for the development of the fuel and energy complex transfer in the air heater [11]. Investigation of the effect of the
of Kazakhstan [1]. step and angle of the arc of discrete artificial roughness on heat
Boiler efficiency increase due to heat recovery in boiler air transfer in an air heater [12] showed that with a relative
heaters is an urgent topic [2-5]. For example, the work [2] roughness equal to 8.33, the Nu number increased
investigated the installation of a second air heater for the boiler significantly depending on the Re number. In [14], a broad
efficiency. The study showed that the installation of an air analysis of air heaters was carried out. The analysis showed
heater and a flue gas dryer can increase the boiler efficiency that the most effective Reynolds number for curved air heaters
by 1.5%, as well as reduce the flue gas temperature by 20℃ is in the range of 11000-16000.
when using lignite as fuel. Analysis of the air heater of a solid The issues of increasing the efficiency of boilers through
fuel boiler with a capacity of 300 MW for the spread of ash on various methods are widely considered in the literature [15-
the surface showed that ash deposits depend on temperature 21]. For example, work [15] presents the results of increasing
irregularities that appear due to unsteady operation of the air the efficiency of a boiler with an angular fuel supply, by
heater [3]. Authors in [4] investigated selective catalytic increasing the efficiency of fuel combustion processes and the
reduction in a regenerative air heater. Studies have shown that secondary air supply system. The work [16] presents the
the use of catalysis in an air heater can reduce the results of numerical simulation of solid fuel boilers. In [17],
concentration of nitrogen oxides by 11%. In [5], a new numerical studies were carried out using oxygen additives.
combustion air heating system is investigated using flue gases. The paper [18] presents the results of the analysis of the
Studies have shown that it is possible to reduce the surface of an air heater of a fluidized bed boiler. Analysis
temperature of flue gases to 30℃ in winter. Taking into showed that the surface of the air heater is largely influenced
account the rather small material presented on the study of air by the composition of the solid fuel. Known works [19] on the
heaters of boiler plants, the authors analyzed the air heaters study of a regenerative air heater showed that corrosion is
presented in general for energy. In particular, the results of formed on the cold parts of air heaters. In work [20], the study
studies of air heaters in solar collectors were analyzed, in view of the use of a regenerative air heater for the efficiency of a
of the sufficient similarity of heating processes. boiler plant was carried out. Studies have shown that efficient
The work [6] investigates the hydraulic resistance of a heat exchange in the air heater reduces the energy
solar heater in the form of a bell mouth. Studies of hydraulic consumption of the boiler.
resistance have shown that a decrease in resistance due to an Currently, there are few proven in operation and reliable
increase in the bell leads to an increase in the Nusselt number, in operation hot-water boilers for solid fuels in Kazakhstan
i.e. to increase of heat transfer. Mathematical modeling of a [21]. Moreover, boilers operating on fuel oil also have
solar heater [7] showed that the heat transfer process extremely low reliability due to the frequent failure of the
- 322 -
furnace walls and the first rows of tubes of convective from the flow remain in the cyclone, and the gases are sucked
bundles. Firewalls of modern boilers, both hot-water and through the gas path 18 by the induced fan into the house gas
steam, have a very low utilization factor. Screen pipes, as it is collector 26. The cyclone works satisfactorily at the maximum
known, perceive heat only by their projection onto the lining. gas flow rate through the experimental section and the flow
Even taking into account the back radiation of the lining on rate in the outer cylinder ω=3÷3.8 m/s. When the cyclone is
the pipes, the degree of use of the heating surfaces of wall operating below the maximum gas flow, an external air intake
screens does not exceed 35-38% [22]. Therefore, the pipe 15 is provided.
development of a new profile of a heating boiler with the aim It should be noted that the cyclone works satisfactorily
of using the reverse side of the furnace walls and new air (with lower gas consumption) when a sufficient draft of the
heater for convective heat exchange can lead to a significant smoke pump along the path 18 is discolored.
reduction in metal consumption. Certain difficulties arise when the ash is fed into the dust
The purpose of this work is to obtain initial data for the collector or the stabilizing entrance section of the
development of a technical design for a high-power hot-water experimental channel. There are two options for feeding ash.
boiler for solid fuel with dry ash removal using a special type Both options are shown in Figure 1.
of air heater for heating air. According to the first option, for the transportation of ash,
The carried out literary analysis shows that little attention the compressor air is supplied with a pressure of P=1.5-2.0
has been paid to the research of air heaters. In this regard, kg·s/cm2 ash-air ejector 7, where the necessary vacuum is
research on heat transfer and resistance was carried out on a created by the supplied air for the transportation of ash. The
model of an air heater for clean and dusty flows. These results air from the compressor enters the electric heater 10. Here it
will make it possible to identify the optimal design of the is heated to the temperature necessary to feed it to the
behind-the-screen gas duct, as well as to obtain formulas for experimental part (200 ÷ 250℃). Pre-calibration of the ash
calculating the heat transfer and resistance coefficients in it. feeder (6) is performed, the ash used in the experiments is
Experimental studies carried out on a model of an air sifted through a coarse sieve mesh to separate lumps, etc. The
heater channel on a dusty flow made it possible to select inner surface of the hopper is heated by spirals of an electric
specific dimensions for the air heater. heater, which is wound around its outer wall. The mixture
passes through the ejector 7 through the protarized flow meter
I. Description of the experimental stand for the study section 9 and from there to the experimental section, the ash
of heat transfer and resistance in the model of the air is discharged into the air path at a distance of 1500 mm above
heater channel for a dusty flow the test area, which ensures a satisfactory distribution over the
cross-section of the channel.
The proposed new type of hot water boiler with a screen According to the second variant, the ash-air ejector is
flue has a special design air heater in the convective flue, installed inside the dust collector, as shown in Figure 1,
designed to heat the air up to 340 ℃. This air heater is made section II-II. The conditions for transporting the sol depend on
of standard wavy sheets that form flat-wavy channels. the pressure of the air flow created by the ejector. The location
Particularly important is the study of heat transfer in a dusty of the ash-air ejector for this option when air is supplied from
flue, which is a fairly accurate approximation to the the air duct collector I must be checked experimentally.
combustion processes in real boilers burning solid fuel. In this
regard, an experimental stand and a technique for studying
heat transfer and resistance in the channel of an air heater on
wavy sheets under conditions of a dusty flow have been
developed.
Figure 1 shows a diagram of an experimental installation
for the study of heat transfer and resistance in the model of the
air heater channel for a dusty flow. Here is 1 air duct collector;
3-flow meter diaphragm; 4-10-electric heaters; 6-ash feeder;
8-ash-air path; 12-working central channel; 16-cyclone for
catching ash particles; 29-cooling medium channels; 21-
fittings for measuring the temperature of the walls of the gas-
air flow channel, pressures: 26-flue basin collector:
The air from the blower with a temperature of 25-35℃
enters the electric preheating furnace 4 through the adjusting
gate 19 and the flow washer 3. The stability of the power
consumed in the electric furnace is monitored using an
ammeter and a voltmeter. The heated air enters the ash
collector II or the input stabilizing section, where it is mixed
with the ash. The poppet ash feeder 6 provides a uniform
dosed supply of ash to the air stream. The stabilizing section
II (which is the ash collector) ensures the uniformity of the
flow and the alignment of the ash concentration along the
cross-section of the experimental section 12. From the
experimental section, the ignited flow enters the output section Fig. 1. Schematic of the experimental setup
12, and then into the precipitation cyclone 16. The particles
- 323 -
inlet, outlet and along the cross-section of the channels with
nichrome-constantan or chromel-copel thermocouples. A PP-
63 potentiometer or a millivoltmeter is used as a secondary
device for measuring the EMF of a thermocouple. The air flow
rate is determined by the diaphragms, the flow temperature at
all diaphragms is measured with mercury thermometers. In all
experiments, barometric pressure and ambient temperature are
measured. After isothermal blowdowns and the dependence of
the drag coefficient on the Reynolds number is obtained,
studies of the non-isothermal flow in the channels are carried
out. Resistance measurement in non-isothermal flow is
performed for all three channels while taking into account the
cooling and heating of the flow. To measure the temperature
of the channel walls, thermocouples were stamped in two
sections along the height of the channel and in five sections
along the length of the channel in the right and left walls to a
depth of 3 mm. The wall temperature is measured by
nichrome-constantan thermocouples with a diameter of 0.4
mm.
To ensure the reliability of the experimental data, the
balance experiments are carried out on the pilot plant on a pure
stream. In the balance experiment, all flow measurements
begin after reaching steady state. The stationary mode is
characterized by a constant temperature of the walls and air at
the entrance and exit from the experimental section. To
exclude the influence of radiant heat transfer between the
heated wall and the junction of the thermocouple,
measurements in the flow will be carried out using the method
of two thermocouples.
Fig. 2. Experimental section for the study of heat transfer
Before the experiment: in a dusty flow, the ash-air path is
and resistance in the model of the air heater channel
switched on (without ash supply), as shown in Figure 1. The
Nomenclature flow temperature in the ash-air path must correspond to the air
temperature in front of the stabilizing inlet section of the
R – radius of curvature, ˚;
experimental channel.
ψ – sector angle, ˚;
The flow temperature in the ash-air duct is controlled by
a – channel width, m;
chromel-copel thermocouples, which are installed after the
h – channel height, m;
compressor air electric furnace and in front of the segmented
Re – Reynolds number, - caliber – length to diameter ratio;
EMF – electormotive force diaphragm. The vacuum in the ash-air ejector is measured
through a hole drilled in the branch pipe in front of the
receiving chamber. The flow rate of the dusty flow is
Cooling medium air with a temperature of 25-35℃ is
measured by a segmental diaphragm; the calibration curve is
supplied from the manifold of the air duct I through the flow
shown in Figure 3. The insulation temperature is measured in
meter washers 3 along the path 20 to the experimental section.
several sections along the length of the experimental section
The diagram of the investigated element of the air heater
with chromel-copel thermocouples. When ash is supplied
is shown in Figure 2. Here R is the radius of curvature of the
directly to the inlet stabilizing section, where the ash-air
channel; t is the step of the wave; ψ is the angle of the sector;
ejector is installed (Figure 1.9, option II), there is no need to
a - channel width; h is the channel height; channel length -
supply compressor air. With this scheme, the supply of ash
1694 mm, average equivalent diameter - 0.04025 m. Relative
from the feeder to the experimental section is carried out
drip curvature - 6.92.
without suction of outside air. The ash feeder is installed near
To go to real conditions, it is necessary to obtain these
the receiving chamber of the ash-air ejector, while the ash-air
dependencies for a dusty flow. In this case, the range of
duct is replaced by a short section. Experiments on a dusty
Reynolds numbers varies from 6*103 to 50*103, which
flow are carried out with a steady dynamic equilibrium of gas-
corresponds to an average flow velocity from 3 to 25 m/s. To
suspended particles inside the experimental channel. Table 1
measure static pressures on the walls of the channel, and at the
gives a list of all measured values. Table 2 shows the amount
entrance and exit from the experimental section, holes with a
diameter of 0.5 mm were made and special fittings were of ash consumption required for one experiment, with a
calculated ash concentration in flue gases μ = 0.0544 kg/kg for
welded. Static pressure is measured on straight sections at a
Ekibastuz coal.
distance of 10 calibers from the entrance to the channel and 20
calibers from the exit from it. For isothermal flow, it is
sufficient to measure the resistance in the central channel. Air
pressure and flow rate are measured with U - samples and
micromanometers. The flow temperature is measured at the
- 324 -
Ambient temperature tam. -
Insulation temperature tin -
Wall temperature at the entrance and -
𝑡𝑤′ , 𝑡𝑤′′
exit from the experimental section
Ash-air flow temperature in front of the -
𝑡𝑎сд ,
segmental orifice plate
Compressor air temperature after -
𝑡𝑐′′
electric oven.
TABLE 2. Ash consumption
ω, m/s V, m3/hour μ, kg/kg Ga, kg/hour
20 331,2 0,044 9,06
15 248,4 0,044 6,81
10 116 0,044 4,55
5 83 0,044 2,27
The amount of heat is determined by the flow rate of the
main air and the transporting dusty flow, the temperature
Fig.3. Segment diaphragm graduation graph difference at the inlet and outlet from the experimental
section.
II. Technique for processing experimental data The amount of heat supplied to heat the flow is
determined by the formula:
Resistance coefficients are determined from the formula: on the central channel
𝑑𝑃 𝜉 ̅2
𝜌𝑢 ′
𝜋𝑅𝑑𝜑𝑍
=
𝑑𝑒𝑞
∗
2
𝑄𝑓𝑢 = 𝑉𝑐′ (𝐶𝑓′ 𝑡𝑓′ − 𝐶𝑓′′ 𝑡𝑓′′ ) + 𝐺𝑎 (С𝑎 𝑡𝑓′ − С𝑎 𝑡𝑓′′ ) (2)
(1)
where, 𝑉𝑐′ - air consumption reduced to normal
𝑑𝑃
where, is the static pressure gradient along the conditions, nm3/hour;
𝜋𝑅𝑑𝜑𝑍
expanded length of the experimental section: 𝐶𝑓′ , 𝐶𝑓′′ - heat capacity of the flow at the entrance and exit
𝑘𝑐𝑎𝑙
ξ - resistance coefficient; from the experimental section, 3 ;
𝑛𝑚 ℃
Z is the number of steps in the experimental section; 𝑡𝑓′ , 𝑡𝑓′′ - average flow temperature at the beginning and at
ρ is the flux density at its average temperature and the end of the experimental section. We assume that the
pressure; temperature of the flow of the semi-air path at the beginning
𝜑
𝑑𝜑 = - angle of the wave sector; and at the end of the experimental section is equal to the
180
𝑢̅ - average flow rate; temperature of the main flow.
𝑑𝑒𝑞. - average equivalent diameter. 𝐺𝑎 - ash consumption, kg/h;
𝑘𝑐𝑎𝑙
In a non-isothermal flow, in contrast to an isothermal С𝑎 - ash heat capacity,
𝑘𝑔 𝑑𝑒𝑔𝑟𝑒𝑒
flow, a correction is introduced to the measured pressure drop
It is accepted that the heat capacity of ash is equal to the
due to the pressure loss due to acceleration.
heat capacity of dust or solid fuel [3].
2 𝑡′′в −𝑡′в Ash consumption is determined from the expression
∆𝑃н = 𝜌𝑎𝑣𝑔 ∗ 𝑢𝑎𝑣𝑔 в then
𝑇ср
(kg/h):
∆𝑃 = ∆𝑃𝑚 ± ∆𝑃н (+ flow cooling) 𝑇н𝑃ст.ср
𝐺𝑎 = 𝑉𝑒𝑥𝑝.𝑠. 𝛾н 𝑎𝑣𝑔 𝜇
TABLE 1. Table of measured values 𝑃н𝑇𝑓
Measured values Designation Dimension (3)
Differential pressure across the mm Hg
diaphragms of the main and cooling ∆ℎ0𝑑 , ∆ℎ1𝑑 , ∆ℎ2𝑑
streams where Vexp.s. is the volume of gases through the experimental
Segment diaphragm pressure drop ∆ℎ𝑑 -
Static pressure in front of the - section where it is determined as (m3/hour):
diaphragms of the main and cooling ∆𝑃𝑑0 , ∆𝑃𝑑1 , ∆𝑃𝑑2
streams
Static pressure in front of the segmental - 𝑉𝑒𝑥𝑝.𝑠. = 𝑓 ∗ 𝜔 ∗ 3600
𝑃𝑑𝑠𝑡
diaphragm (4)
Static pressure at the inlet and outlet of ′𝑎𝑣𝑔 ′′𝑎𝑣𝑔 -
𝑃𝑤 , 𝑃𝑤 ,
the experimental section
Static pressure drop in the experimental - 𝑓 - section of the experimental channel, m2;
∆𝑃𝑒𝑥𝑝.𝑠.
section 𝜔 - flow rate, m/s;
𝑇 𝑃
Vacuum in the ash-air ejector Sa.i. -
𝛾н н ст.ср
ср - correction factor;
Vacuum in the gas-air duct at the - 𝑃н𝑇𝑓
Sc′′ 𝑎𝑣𝑔 𝑎𝑣𝑔
cyclone outlet 𝑇𝑓 и 𝑃ст - average temperature and flow pressure;
Barometric pressure B mm Hg
A𝑤 𝛼 𝑓.𝑎.
Air temperature at diaphragms 𝑡0𝑑 , 𝑡1𝑑 , 𝑡2𝑑 ℃ 𝜇= dimensionless ash concentration;
Air temperature before and after the - 100𝐺𝑔
𝑤
experimental section in the case of a 𝑡𝑓′ , 𝑡𝑓′′ A - working ash content, %;
clean flow 𝛼𝑓.𝑎. - the proportion of fly ash is determined in, %;
Ash-air flow temperature before and -
𝑡𝑎′ , 𝑡𝑎′′
after the experimental section
- 325 -
A𝑤 between the cubes of the air heater. A schematic diagram of
𝐺𝑔 = 1 − + 1.306𝛼 ∗ 𝑉 0 - flue gas consumption, kg /
100 ash extraction in this section of the convective gas duct is
kg;
shown in Figure 4. With this method, the extraction time is
𝛼 - the excess air ratio in the air heater zone is determined
according to the operational parameters of the boiler; long, which makes it difficult to extract a large amount of ash.
𝑉 0 - theoretical air volume, m3/kg.
Losses to the environment are determined by the average
temperature of the insulation and its outer surface, kcal/h:
𝑄𝑒𝑛𝑣 = 𝛼(𝑡𝑖𝑛𝑠 − 𝑡𝑒𝑛𝑣 ) ∗ 𝐹𝑖𝑛𝑠
Average logarithmic temperature difference is calculated
by the formula:
∆𝑡𝛿 −∆𝑡𝜇
∆𝑡𝑙𝑜𝑔 = ∆𝑡 (5)
2.3𝑙𝑔 𝛿
∆𝑡𝜇
where ∆𝑡𝛿 and ∆𝑡𝜇 - are temperature differences in the initial
and final sections of the experimental section.
The heat transfer coefficient α1 along the central channel from
air (or dusty flow) to the wall is determined by the formula:
′
𝑄𝑓𝑐𝑡𝑟 (𝑄𝑓𝑐𝑡𝑟 ) Fig. 4. Scheme of ash extraction from the boiler BKZ-
𝛼1 = (6) 160-100
∆𝑡𝑙𝑜𝑔𝐹𝑚
According to the third method, the ash of the Ekibastuz
here 𝐹𝑚 is the heat-perceiving surface of the channel coal was taken from the bunkers of the prechamber and fields
𝛼 𝛼
Then the Nusselt criteria are determined 𝑁𝑢3 = 1 9 and 1, 2, 4, 5, 6 of the electrostatic precipitators on the operating
𝜆
a graph of the dependence 𝑁𝑢3 = 𝑓(𝑅𝑒3 ) boiler PK-39 of the Pavlodar CHP at points 2 and 3 (Figure
Pollution coefficients are determined according to the 5). For all sampling methods, the ash concentration in the flue
well-known method [8] as the difference in thermal gases was determined.
resistance:
1 1
𝜀= 𝑎𝑣𝑔 − 𝑎𝑣𝑔
𝐾𝑑 𝐾𝑐
𝑚2 ℎ𝑜𝑢𝑟℃
𝑘𝑐𝑎𝑙
𝑎𝑣𝑔 𝑎𝑣𝑔
where 𝐾𝑑 , 𝐾𝑐 - average values of heat transfer
𝑘𝑐𝑎𝑙
coefficients for dusty and clean flows, 2 .
𝑚 ℎ𝑜𝑢𝑟℃
Heat transfer coefficient for net flow according to [3].
1
а) 𝐾𝑐 = 1
⁄𝛼1 +1⁄𝛼2
in this case, we assume that 𝛼1 = 𝛼2 since the channels
are identical;
1
for a dusty flow 𝐾𝑑 = 1 1 𝜀+ ⁄𝛼1 + ⁄𝛼2
Thermal efficiency coefficient
𝐾𝑑 1⁄ +1⁄
𝛼1 𝛼2
𝜓= 𝑜𝑟 𝜓 = (7)
𝐾𝑐 𝜀+1⁄𝛼1 +1⁄𝛼2
III. Ash sampling method and sampling results for Fig. 5. Scheme of ash extraction from the PK-39 boiler
experiments on a model of an air heater channel.
The results are shown in Table 3. Here Р stat is the statistical
3.1 Selection method pressure of gases in the gas duct, measured by a
To carry out experimental studies on the model of the air pneumometric tube; Р1, Р2 - gas pressure before and after the
heater channel, ash, the fractional composition of which is diaphragm.
unique to nature, should be used. For this, ash was taken from ∆Рd - pressure drop across the diaphragm;
the gas ducts of operating boilers in several ways. Tg d, td - temperature of gases in the gas duct and
According to the first method, coal ash from the diaphragm;
Karaganda field was taken from under the gas ducts of the Ρd - the density of gases is determined by the formula
convection shaft of the BKZ-160-100 boiler of the Almaty Р
273(𝐵+ 𝑠𝑡𝑎𝑡 )
State District Power Plant. This ash differed from fly ash in a ρd = 𝛾𝑁 13.6
(8)
(273+t𝑑 )760
significant size.
According to the second method, coal ash from the where 𝛾𝑁 – specific gravity of the measured gas referred to
Karaganda deposit was sampled in the section located reference conditions (0℃ - 760 mm Hg)
- 326 -
Vd – the gas flow rate through the diaphragm is determined № Ash ρср t2 Р1 Р2
by the formula collection mm ℃ mm mm H2O
∆𝑃𝑑
date H2O H2O
𝑉𝑑 = √ 𝑉 (9) 1 11.06.20 75 249 410 395
𝜌𝑑
2 13.06.20 87 256 415 390
where V is determined by the calibration curve of the flow 3 14.06.20 77 247 455 420
washer (d=12 mm) (Figure 6.). 4 20.06.20 79 259 420 400
5 21.06.20 106.2 258 375 350
f – inlet section of the sampling tube (for sampling, two 6 22.06.20 90 254 440 420
tubes with f1=0.0006 m2 and f2=0.0007 m2); 7 21.09.20 109 242 310 295
B – barometric pressure; 8 24.09.20 110.5 235 310 290
Ω – the velocity of the suctioned gases in the diaphragm
9 26.09.20 109 258 460 440
is determined by the formula
10 27.09.20 106 238 480 450
𝑉𝑑 11 02.10.20 118.2 241 480 440
𝜔= (10)
𝐹𝑤 ∗3600
TABLE 3. Continued table 3
where 𝐹𝑤 – washer cross section; № ∆Рд tд ρ 𝑽д f В
Ggas – the suction gas flow rate was determined by the mm ℃ kg/m3 m3/h m2 mm
formula H2O H2O
Р𝑃𝑠𝑡𝑎𝑡 1 15 37 1.21 4.5 0.0006 716
𝑇н (В+ 13.6 )
𝐺 𝑔𝑎𝑠 = 𝑉𝑑 𝛾𝑁 (11) 2 25 40 1.2 5.3 - 705
𝑃н (273+t𝑑 )
3 35 43 1.2 5.9 - 702
4 20 42 1.2 5.3 - 705
5 25 40 1.2 5.9 - 704
6 20 44 1.2 5.3 - 704
7 15 47 - 4.8 0.0007 709
8 20 50 - 5.5 - 709
9 20 45 - 5.5 - 707
10 30 53 - 6.7 - 707
11 40 39 - 7.9 - 706
TABLE 3. Continued table 3
№ ω Ggas Gdust μ
m/s kg/h kg/h kg/kg
1 0.46 2.899 0.150 0.0517
2 2.73 3.31 0.153 0.046
3 2.73 3.73 0.135 0.036
4 2.45 3.2 0.266 0.083
5 2.73 3.67 0.154 0.042
6 2.45 3.32 0.190 0.0572
Fig. 6. Calibration characteristic of the diaphragm. dsh = 7 1.9 3.096 0.051 0.016
12.0 mm 8 2.18 3.6 0.107 0.029
9 2.18 3.62 0.154 0.043
3.2 Ash sampling results 10 2.66 4.34 0.091 0.021
11 3.13 5.09 0.144 0.028
From the data in Table 3, the average concentration value
μ_ср=0.0444 kg / kg was determined, which was used to The results of the fractional composition of the ash
conduct experiments on the model. In addition, the analysis of sampled at point 1 (Figure 5) before the RAH, in the bunkers
the dispersed composition of ash was carried out for particles of the prechamber 2 and the fields of the electrostatic
from 0 to 50 microns by the method of centrifugal separation precipitators 3 are shown in Tables 4 and 5, here ρavg is the
on the "Bako" apparatus, and for particles from 50 to 280 mm true density of ash particles (without pores) was determined
- by sieve analysis. The defining characteristic in the sieve by the pycnometric method.
analysis of the selected ash is the residue on the R50 sieve, TABLE 4. Results of the analysis of ash on the centrifugal
according to which the conditional ash separation is carried separator "Bako"
out: with R50 = 27 ÷ 36% - fine ash; with R50 = 50 ÷ 62% - Ash density
the average ash taken from the bunkers of fields 1,3,4,5,6 and Ρavg=2023 kg/m3
the pre-chamber was characterized by the following residues Before RAH
on the R50 μm sieve: Particle Content Content Particle Content Conten
Field 1 - R50 = 35 ÷ 36% (fine ash); diamete of of diamete of t of
Field 3 - R50 = 27÷ 29% (fine ash); r in fraction smaller r in fraction smaller
Field 3,4 - R50 = 50 ÷ 62% (medium ash); microns s by particle microns s by particle
Field 4,5 - R50 = 50% (medium ash); δs mass of s; % δs δs mass of s; % δs
Field 6 - R50 = 50% (medium ash); particle particles
prechamber (gas distribution chamber) R50 = 77 ÷ s% %
79% (coarse ash). 1.56 2.2 2.2 1.56 3.1 3.1
TABLE 3. Results of selection of natural ash
- 327 -
3.73 4.0 6.2 3.73 4.0 7.1 taken from the fields of electric filters meets these
7.1 13.6 19.8 7.1 12.1 19.2 requirements. This conclusion is also obtained when
10.5 136. 33.4 10.5 11.7 31.9 comparing the mineral composition of ash (table 6). The graph
16.6 21.7 55.1 16.6 21.2 53.1 in Figure 9 clearly shows the convergence of two methods for
29.4 25 80.1 29.4 26.8 79.9 determining the fractional composition of ash (for small and
37.5 9.5 89.6 37.5 9.4 89.3 large particles).
43.9 4.3 92.9 43.9 4.4 93.7 TABLE 6. Composition of ash taken at the station
0.61 6.1 100 0.63 6.3 100 Name Ash before Ash after
TABLE 4. Continued table 4 RAH, % RAH, %
Particle density Calcination losses 4.06 3.32
Ρavg=1860 kg/m SiO2 55.5 54.55
Electrostatic precipitator Electrostatic precipitator R2O3 38.8 42.55
field 1 field 3 Al2O3 23.14 28.25
Particle Conten Content Particle Content Content Fe2O3 8.42 6.01
diamete t of of diamete of of TiO2 1.17 1.06
r in fractio smaller r in fractions smaller CaO 1.55 1.27
microns ns by particles microns by mass particle MgO 0.40 0.40
δs mass of ; δs % δs of s; δs % SO3 0.41 0.47
particle particles
s% %
2.31 4.1 4.1 2.31 2.1 2.1
3.9 4.8 8.9 3.9 3.5 5.6
7.41 12.0 20.9 7.41 11.5 17.1
10.9 14.1 35.0 10.9 12.2 29.3
17.3 19.9 54.9 17.3 20.9 50.2
30.7 24.2 79.1 30.7 27.0 77.2
39.1 9.0 88.1 39.1 9.2 86.4
45.8 1.2 89.3 45.8 4.5 90.9
0.78 7.8 97.1 0.91 9.1 100
TABLE 5. Results of sieve analysis of ash from Ekibastuz coal
№ Sieve, Electric filter
micron Field 1, % Field 3, %
1 R280 0.31 0.06 0.196 0.14
2 R140 1.6 1.49 1.51 1.57 Fig. 7. Diagram of distribution of ash fractions
3 R100 4.69 4.81 6.46 7.33
4 R90 15.94 15.4 16.13 16.99
5 R71 21.67 21.85 18.62 18.97
6 R50 35.5 35.43 27.69 28.96
7 Rbottom 64.5 64.57 72.31 71.04
TABLE 5. Continued table 5
№ Sieve, Electric filter
micron Field Field Field 3 and 4, % Field
4, % 5, % 6, %
1 R280 0.009 0.08 0.18 0.09 0.08
2 R140 0.24 1.03 1.74 2.09 0.34
3 R100 5.47 6.21 6.89 8.011 2.31
4 R90 22.5 23.64 24.32 25.81 19.13
5 R71 31.4 41.4 44.63 46.41 37.73
6 R50 79.4 49.5 53.52 57.21 48.99
7 Rbottom 20.6 50.5 46.48 42.79 51.01
TABLE 5. Continued table 5
Sieve, Before After RAH, % Prechamber, %
micron RAH, %
R280 2.18 1.35 1.65 1.0 1.14 1.63 1.62
R140 4.07 7.43 7.44 6.36 7.54 16.6 19.04
R100 6.79 27.11 11.57 8.24 11.98 35.74 38.72 Fig. 8. Disperse composition of ash. (cumulative
R90 16.4 51.6 21.6 19.2 22.83 55.09 59.35 distribution)
R71 23.4 69.81 24.2 22.65 30.47 64.66 66.51
R50 27.4 78.0 30.3 29.16 33.36 76.99 78.42
Rbottom 72.6 22.0 69.7 70.84 66.64 23.01 21.58
According to these results, in Figures 7 and 8, graphs of
the dispersed composition of ash are given for different
sampling points in the gas duct of the PK-39 boiler of the
Pavlodar CHP, and in Figure 9, the integral curve of the
fractional composition. These data make it possible to analyze
the fractional composition of the ash sampled at different
points of the boiler gas duct and to select the ash for carrying
out experiments on the model that is closest in fractional
composition to the original, i.e. in front of the air heater. Ash
- 328 -
Fig. 9. Curve of ash dispersion
Fig. 10. Distribution of static pressure along the length of
IV. Results of the study of hydraulic resistance and the channel
heat transfer in the air heater model
In Figure 10 shows the distribution of static pressure
along the length of a complex cross-section channel. A sharp
change in the static pressure at the initial section of the
channel is explained by the presence of a converging section.
At the outlet, a diffuser section is obtained, which leads to an
increase in the pressure at the entrance to the experimental
section, the wetted perimeter increases by 60%, and at the
outlet it decreases by 60%.
To align the velocity profile at the entrance to the channel
and at the exit from it, double metal meshes with a mesh size
of 1 * 1 mm are installed.
In the setup experiments, a coincidence of the static
pressure values measured with a Pitot tube (diameter of the
Figure 11. Dependence of hydraulic resistance on the Re
tube nozzle is 3 mm) and with bushings with a hole diameter
number
of 0.5 mm was obtained. The analysis shows that the pressure
change according to the linear law, for the investigated
regimes, takes place in the section of the channel between the
points x/l=0.36 and x/l=0.72.
When calculating the hydraulic resistance coefficient, the
static pressure drop was taken at a distance of 0.8 mm. In
Figure 11 shows the dependence of the drag coefficient on
Reynolds in coordinates lg100𝜉 on lgRe.
The processing of the data obtained by the least squares
method made it possible to obtain the generalized formula
𝜉 = 𝐵 ∗ 𝑅𝑒 𝑚 (12)
where B = 0.205; m=0.155.
The standard deviation of the generalization was 3.6%,
the arithmetic mean was 0.1%. Comparison of the hydraulic
resistance coefficient of the investigated channel shows that
in the entire investigated range of Reynolds numbers it is Fig. 12. Dependence of the Nu number on Re in the
higher than in a straight pipe by an average of 66.9. investigated channel
Such an increase in the drag coefficient should be
explained by the presence of laminar zones in the acute
corners of the complex channel and, thus, by a general
decrease in the flow velocity.
Data on the first commissioning non-isothermal
experiments are shown in Figure 12. They show that heat
transfer in a channel with a complex cross-section is on
average 20% higher than heat transfer in a straight pipe.
- 329 -
Having processed the obtained data on heat transfer by the turn (festoon), the flow rates can be reduced by increasing the
least squares method, we obtain a generalized formula. cross-section of the gas duct by a local increase in its width.
With this temperature, the combustion products enter the air
𝑁𝑢 = 𝐴𝑅𝑒 𝑘 = 0.22𝑅𝑒 0.819 (13) heater 6 of a special type, where the air is heated to a
temperature of 340℃.
This air heater (Fig.14) is made of standard corrugated
The standard deviation of the generalization was 1.5%, the
sheets, which form flat-volost channels with an equivalent
arithmetic mean - 0.0%. 𝑒𝑞 𝑒𝑞
diameter on the gas side 𝑑 𝑔 = 37 mm and on the air side 𝑑 𝑎
V. Computational and design studies of a boiler with a = 35 mm. The average radius of curvature of these channels
is 40 mm, and the angle of curvature is 𝜑 = 109°. These data
behind-the-screen solid fuel flue
correspond to the optimal geometry of the channel, for which
The study of the intensification of heat exchange in
the calculation formulas for heat transfer are experimentally
convective heating surfaces, carried out in the laboratory of
determined
boiler units, served as the basis for the development of
schematic diagrams of high-power hot-water boilers with a 𝑁𝑢 = 0,985 × 𝑅𝑒 0,5 𝑃𝑟 0,26 (14)
behind-the-screen gas duct and an air heater. To develop
structural diagrams of a hot-water boiler with a behind-the- and the reduced coefficient of resistance
screen gas duct, as well as to compare it with the existing КВ-
ТК-100-150 /2/, thermal, aerodynamic, hydraulic and 𝜉 = 8,8 × 𝑅𝑒 −0,43 (15)
structural calculations were carried out for a new type of hot The manufacture of this air heater for a pilot industrial
water boiler with a behind-the-screen gas duct. boiler will make it possible to obtain the necessary
experimental material for solving the problem of using such
air heaters in a dusty flow.
Fig. 14. Scheme of the air heater of a new boiler
Thus, this boiler profile excludes the use of convective
tube bundles, which in known boilers pose the greatest
Fig. 13. Scheme of a new hot water boiler
operational difficulties. The longitudinal flow in the behind-
the-shield gas duct allows for increased gas velocities, which
The heating boiler of the new type has a cylindrical or
will lead to self-blowing of the heating surfaces of the behind-
multifaceted furnace (1 with pinching (Figure 13). Products
the-shield gas duct. All this taken together will lead to a
of incomplete combustion from the furnace enter a vertical
reduction in the specific consumption of metal by more than
cylindrical or multifaceted afterburning and cooling chamber
two times, both on the heating surface, operating under
2. The latter is formed by pipes d = 50 x 4mm, welded
pressure, and on the air heater.
together by bridges. Thus, the screens of the cooling chamber
Since all boiler heating surfaces are gas-tight, all-welded,
are gas-tight, all-welded. In the upper part of the cooling
it becomes possible to organize the operation of the boiler
chamber, the pipes are divorced in two rows and form a under pressurization, which will lead to energy savings for
scallop 3 along the entire perimeter of the cooling chamber, own needs.
through which combustion products are discharged, washing
The water circulation scheme in the boiler pipes with a
the reverse side of the screens. a polyhedron concentrically
capacity of 30 Gcal/h has been developed in two versions
enclosing the cooling chamber. The pipes of this cylinder or
polyhedron, as well as in the cooling chamber, are welded to (Figure 15). The first option is designed as a six-way one.
each other by the bridges. Increased gas velocities in the Water is supplied to the lower collector, which has a partition
longitudinal flow are not dangerous from the point of view of that provides water supply to one third of the screen pipes (56
the nose, as ash particles slide along the pipes. In places of pcs.). In the upper part of the furnace, through the bypass
- 330 -
vertical collector, water enters the pipes of the behind-the- of the second version of the circulation scheme is a slightly
screen gas duct (56 pcs.) And is directed to the middle lower temperature head in the behind-the-screen gas duct,
collector, also equipped with a partition. From the middle which leads to an increase in the boiler heating surface.
collector, water enters the pipes of the second third of the Calculation of heat transfer in the behind-the-screen gas duct
screened gas duct and through the bypass vertical collector is at a water temperature at the boiler inlet of 120℃ shows that
directed to 56 pipes of the furnace screens. And further to the in this case the heating surface of the behind-the-screen gas
lower collector, from which it is directed to the remaining 56 duct increases by 39 m2, and the metal consumption by 700
pipes of the screens. Having passed the last third of the pipes, kg. This disadvantage of the circuit can be considered
insignificant in comparison with the bathroom advantages of
the water after the third bypass collector is directed to the
the second version of the circulation circuit. This scheme
middle collector, from which it enters the network. Inlet water
essentially provides a peak boiler operation and is accepted as
temperature 70℃ and outlet 150℃, water flow 375 t/h, water the main, most advantageous water circulation scheme for a
speed 1.4 m/s. The hydraulic resistance of the boiler is 9.66 30 Gcal/h boiler. For a 100 Gcal/h boiler, it is possible to use
mm H2O, and the power consumption for overcoming the a two-way water circulation scheme at an inlet and outlet
resistance of the boiler is N_k= 13 kW. The second variant of temperature of 70/150℃.
water circulation has two strokes. Water from the lower The boiler can be repaired by pulling the outer screens up.
collector enters simultaneously all pipes of the furnace walls In a low-power boiler (30 Gcal/h), the outer pipes of the
(168 pcs.). Getting simultaneously into all three vertical behind-the-screen gas duct are raised entirely, and for the I00
throughput collectors in the upper part of the furnace, water Gcal/h boiler, the outer pipes can be split into 3 or 6 sectors
from them is directed to all the other pipes of the behind-the- and diverted to the side or up to access the back side of the
screen gas duct and is collected in the middle collector, from furnace walls.
which it is sent to the network. To ensure a water velocity of The results of typical, aerodynamic calculations of a pilot
about 1.3 m/s, its consumption should be about 1000 t/h. At industrial boiler with a heating capacity of 30 and 100 Gcal/h
the same time, due to the bypass of bitter water at the inlet to are presented in Tables 7, and the weight characteristics of the
the boiler, its temperature rises to 120℃, and at the outlet - boiler are in Table 8. It can be seen from these tables that the
150℃. specific metal consumption per boiler is 30 Gcal/h by 2.38
times less than on the KV-TK-100 boiler. At the same time,
the temperature of the flue gases in the pilot industrial boiler
of the proposed type is expected to be 68 ℃ lower than that of
the comparable KV-TK-100 boiler and the efficiency is 3%
higher. The share of pipes in the total mass of the boiler metal
of 30 Gcal/h is 40%. The overall economic effect is $ 26
thousand / year. Calculations for a 100 Gcal/h boiler with a
behind-the-screen gas duct with new air heater show that its
performance is better than that of a 30 Gcal/h boiler. So, the
specific consumption of metal is 2 t/Gcal (instead of 2.2 for
the boiler 30 Gcal/h and 5.43 g/Gcal for the KV-TK-100
boiler). The share of pipes in the total mass of metal has been
reduced to 32.5%. The economic effect per boiler 100 Gcal/h
is expected to be about 100 thousand $/year, while the cost of
the boiler pays off within 1.26 years. The design
characteristics for a new type of hot water boiler are given in
table 9.
TABLE 7. Calculation results of boilers
№ Name of calculated values unit of
measurement
1 Boiler performance Gcal/h
2 𝑘𝑐𝑎𝑙 kg/h
Fig. 15. Schemes of water recirculation in a new boiler Fuel consumption 𝑄 𝐻 𝑃 = 6240 𝑘𝑔
3 Specific consumption of equivalent fuel 𝑘𝑔𝑜𝑒.
In this variant of water storage, the hydraulic resistance of 𝐺𝑐𝑎𝑙
the boiler is only ∆P = 2.6 mm.H2O., and the power 4 Heat voltage of the furnace kcal/m3h
consumption is only N_k= 9.2 kW. 5 Inlet water temperature ℃
Comparing these options with each other, it can be noted 6 Leaving water temperature ℃
that the second option is significantly more profitable than the 7 Boiler pressure kg/cm2
first. The high temperature of the water at the boiler inlet 8 Gas temperature at the outlet from the furnace ℃
9 Gas temperature in front of the air heater ℃
ensures a more durable operation of the heating surfaces, since 10 Flue gas temperature ℃
it eliminates sulfuric acid corrosion. The power consumption 11 Hot air temperature ℃
for overcoming the resistance within the boiler is also 12 Excess air ratio -
noticeably lower in the second version. All pipes of the 13 Heat loss with flue gases %
furnace walls, the most stressed, are washed by an upward 14 Heat loss from fur. underburning %
flow of water, which also increases the reliability of the boiler 15 Heat loss from chem. underburning %
16 Heat loss to the environment %
and allows shading lower water velocities. The disadvantage 17 Boiler efficiency %
- 331 -
18 Gas velocity in the behind-the-screen gas duct m/s Name of parts and their Quantity Length Weight, kg
19 Water velocity in the boiler pipes '' size for 1 or area 30 100
20 Shielded gas duct resistance kg/m2 boiler Gcal/h Gcal/h
21 Resistance of the air heater on the gas side '' Total metal 66200 200
22 Total gas resistance of the boiler and air heater '' kg ton
23 Air path resistance to burners ''
24 Boiler hydraulic resistance kg/cm2 Including metal under pressure (pipes) for a 30 Gcal/h
25 Boiler metal weight ton
26 Specific metal consumption t/Gcal boiler - 26.56 t or 40%, for a 100 Gcal/h boiler - 65.0 t or
TABLE 7. Continued table 7 32.5%. Specific metal consumption for a 30 Gcal/h boiler is
№ Calculated values for boilers 2.21 t/Gcal and for a 100 Gcal/h boiler - 2 t/Gcal. Specific pipe
KazNII-30 (developed KazNII-100 (developed KV-TK-100 consumption for a 30 Gcal/h boiler is 0.88 t/Gcal and for a 100
boiler) boiler) Gcal/h boiler - 0.65 t/Gcal. The KV-TK-100-150 boiler has a
1 30 100 100
2 5119 17200 17737 specific metal consumption of 5.25 t/Gcal. (Power
3 153 153 153 engineering 5, 1980, 1-30-02).
TABLE 9. Design data of a hot water boiler 30 Gcal/h and 100
4 179000 136000 117000
5 70/120 70 70
Gcal/h.
Name of parts and their size Unit of Size
6 150 150 150
measurement 30 100
7 25 25 25
Furnace volume m3 178,5 790
8 1045 1070 1130
The radiation-perceiving m2 200,7 550
9 400 400 428
surface of the firebox screens
10 140 140 209
The radiation-perceiving m2 39,67 1200
11 343 343 377
surface of festoon
12 1,2 1,2 1,2
Лучевоспринимающая m2 11,3 -
13 5,2 5,2 8,0 поверхность фестона
14 1,0 1,0 1,0 Cross-section for the passage m 2
6,13 22,2
15 - - - of gases in the festoon
16 0,5 0,5 0,65 Diameter and wall thickness mm 50/4 50/4
17 93,3 93,3 90,35 of festoon pipes and screens
18 20,4 26,4 - Heating surface of the m 2
597,3 1930
19 1,4/1,25 1,1 - behind-the-screen gas duct
20 48,4 68,0 - Cross-section of the off- m2 2,1 5,45
21 103 110 - screen flue (gas)
22 158 178 - Hydraulic diameter of the m 0,095 0,15
23 100 100 160 channel of the off-screen flue
24 0,966/0,26 0,2 2,6 Heating surface of the air m2 1492,7 4688
25 66,2/66,9 200 525 heater (heat-sensitive)
26 2,21/2,23 2,0 5,25 2
Cross-section for the passage m 2,2 7,32
of gases in the air heater
The numerator is the circulation diagram of the 1st option, Cross section for the passage m2 2,11 7,04
of air in the air heater
the denominator is the circulation diagram of the 2nd option. Hydraulic diameter of the m 0,0352 -
channel on the air side
TABLE 8. Weight characteristics of the boiler 30 Gcal/h and Hydraulic diameter of the m 0,037 -
100 Gcal/h. channel on the gas side
2
Cross-section of pipes on the m 0,233 0,341
Name of parts and their Quantity Length Weight, kg
water side (2nd version of
size for 1 or area 30 100 the circulation)
boiler Gcal/h Gcal/h
Pipes of firebox screens, 168/ 10080/
off-screen flue, spacers of 246 14260 28420 72280 VI. Simple calculation of the economic effect from the
screen pipes and off- m introduction of a new hot water boiler with a capacity of
screen flue
Collectors, branch pipes 1 61,6 m 3340 7740
30 Gcal/h
Ribs-turbulators in the 336 2320 m 4550 30000 From the constructive calculation of the 30 Gcal/h boiler and
behind-the-screen gas duct the table of weight characteristics, it can be seen that the total
Gas outlet box from the 1 50 m2 1570 4000 mass of the boiler metal is 66.9 tons. The well-known KV-
boiler to the air heater
sheet TK-100-150 boiler has a metal mass of 525 tons, and its cost
Boiler lining with 1 270 m2 1000 2500 is $ 264 thousand. Installation of the boiler, according to the
aluminum sheet price list of installation operations, is $ 25 thousand. Thus,
Air heater, plate made of 1 1493 17500 55000 the total cost of the boiler with installation is $ 300 thousand.
wavy sheets m2
a) supply and discharge 1 65 m2 1500 3500
Hence, the specific cost of recycled metal is 300,000 / 525 =
lines 572 $/t, and the specific consumption of metal is 5.25 t/Gcal.
b) air preheater end plates 398 45 m2 530 1100 Taking this specific metal consumption, it is possible to
Welding seams leg - 14000 880 2200 determine the mass of a boiler of 30 Gcal/h of a conventional
m type.
Total base metal 59090 178300
kg kg
Elements of the frame, 1 - 7110 21700
М𝑏 = 5,25 ∗ 30 = 157,5 𝑡.
platform, stairs kg Thus, the saving of metal for one boiler of 30 Gcal/h will be:
approximately-12.0% of
the weight of the base
metal 𝐸𝑚 = 157,5 − 66,9 = 90,6 𝑡.
- 332 -
and the reduction in capital costs is K = 90.6 * 572 = $ 52 Constructive diagram of new type hot water boiler for
thousand. In addition to a significant reduction in capital solid fuel with a productivity of 30 have been developed.
costs, the new type of boiler consumes less fuel by 5 kg/Gcal Thermal, aerodynamic, hydraulic and structural calculations
(Table 7). of a water-heating boiler with a behind-the-screen gas duct
The total fuel economy with an annual number of boiler and an air heater for a capacity of 30 Gcal/h have been carried
5∗30∗5000 out. At the same time, two variants of water circulation in the
operation hours of ≈ 5000 will be 𝐸 = =
1000 boiler pipes are considered: six strokes and two strokes. A
750 t.f.e./year or natural fuel (𝑄 𝐻
𝑃 = 6240 kcal/kg).
comparative analysis of a water-heating boiler of the
The average cost of Kuznetsk coal with transportation costs
proposed type with a well-known boiler of the KV-TK-100-
is about 20 $/t. Hence the annual savings due to reduced fuel
150 type for Kuznetsk coal is carried out.
consumption will be
The calculation of the economic effect from the
𝑆𝑓 = 842 ∗ 20 = 16800 $ 𝑎 𝑦𝑒𝑎𝑟. introduction of a hot water boiler of a new profile with a
capacity of 30 Gcal/h is given.
The hydraulic resistance of the new boiler reaches 2.6 mm A hot water boiler of a new type allows using the reverse
H2O, instead of 15 m for known boilers of similar side of the furnace walls for convective heat exchange, which
performance. This leads to a reduction in the power of the significantly increases its efficiency, and the possibility of
circulation pump by 11 kW. Hence, the energy savings on the eliminating convective bundles from their schemes of a new
circulation pump will be boiler significantly increases the reliability of the new boiler.
The brief conclusions drawn from this study are as
11 * 5000 = 55000 kWh/year. follows:
The cost of electricity in the utility system is 0.05 $/kWh. 2. Comparison of the new type boiler with the well-known
Hence the savings by reducing energy consumption will be: KV-TK-100-150 boiler shows an improvement in the
technical and economic indicators of the new boiler: the
55000 ∗ 0,05 boiler efficiency increases by 3% due to a decrease in the flue
𝑆= = 27,5 $ 𝑎 𝑦𝑒𝑎𝑟.
100 gas temperature; fuel consumption is reduced by 5 kg/Gcal.
The total savings in operating costs from the introduction 3. Saving metal consumption per boiler with a capacity of
of a 30 Gcal/h boiler will be: 30 Gcal/h is 90.6 tons. Specific metal consumption for a 30
Gcal/h boiler is 2.38 times less than for a KV-TK-100-150
𝑆𝑖 = 16840 + 27.5 = 18868 $ 𝑎 𝑦𝑒𝑎𝑟. boiler.
4. Comparison of two options for water circulation in the
Saving the reduced costs, considering the reduction in boiler (six-way and two-way) shows the advantages of the
capital investments, will be: two-way option: increased reliability of the furnace walls,
𝐸𝑟 = 18868 + 0,15 ∗ 52000 = 26668 $ 𝑎 𝑦𝑒𝑎𝑟. reduced power consumption to overcome the hydraulic
resistance of the boiler (1.4 times).
The approximate cost of a boiler of 30 Gcal/h can be 5. Simple calculations of the cost of a new type of boiler
determined by multiplying the specific cost of the metal by 30 Gcal/h have shown that the costs of its manufacture due to
its weight. fuel and metal savings are paid off in 1.5 years.
С𝑏 = 572 ∗ 66,9 = 38300 $. ACKNOWLEDGMENT
Consequently, the cost of the boiler pays off by saving the
reduced costs during: This research has been funded by the Science Committee
of the Ministry of Science and Higher Education of the
38300 Republic of Kazakhstan Grant № AP13068541 project
= 1.5 𝑦𝑒𝑎𝑟𝑠
26015 «Development of an experimental energy complex based on
a modernized boiler plant using biofuel»,
The design of the new boiler differs from the known ones
in many welding works. Welding of fins and heat-dissipating REFERENCES
turbulizing fins of the boiler 30 Gcal/h requires additional [1] Concept of development of the fuel and energy complex of the
welding work in the amount of I4 thousand m. The cost of Republic of Kazakhstan until 2030. Approved by the decree of the
Government of the Republic of Kazakhstan dated June 28, 2014 No.
these works for one boiler is about $ 4 thousand. These 724. https://policy.asiapacificenergy.org/node/369; 2021 [accessed
additional capital expenditures, as calculations show, pay off 09.03.2021].
in 5 years only due to energy savings due to the complete [2] Somchart Chantasiriwan, Optimum installation of flue gas dryer and
additional air heater to increase the efficiency of coal-fired utility
absence of leaks in the gas ducts. Thus, all-welded hermetic boiler, Energy, Volume 221, 2021, 119769, ISSN 0360-5442,
gas ducts, despite the increase in welding work, pay for https://doi.org/10.1016/j.energy.2021.119769.
themselves, which has become a well-known fact that ensures [3] Hari B. Vuthaluru, David H. French, Investigations into the air heater
ash deposit formation in large scale pulverised coal fired boiler, Fuel,
the spread of all-welded panels. Volume 140, 2015, Pages 27-33, ISSN 0016-2361,
https://doi.org/10.1016/j.fuel.2014.09.040.
[4] Robert Wejkowski, WacЕ‚aw Wojnar, Selective catalytic reduction in
CONCLUSION a rotary air heater (RAH-SCR), Energy, Volume 145, 2018, Pages 367-
Model of an air heater for research, of an air heater 373, ISSN 0360-5442, https://doi.org/10.1016/j.energy.2017.12.077.
channel made of corrugated sheets for research in a dusty [5] Sheng Shang, Xianting Li, Wei Chen, Baolong Wang, Wenxing Shi, A
flow have been developed. A research technique on a dusty total heat recovery system between the flue gas and oxidizing air of a
stream has been developed. gas-fired boiler using a non-contact total heat exchanger, Applied
Energy, Volume 207, 2017, Pages 613-623, ISSN 0306-2619,
https://doi.org/10.1016/j.apenergy.2017.05.169.
- 333 -
[6] Zixiang Li, Xinqi Qiao, Zhengqing Miao, Low load performance of
tangentially-fired boiler with annularly combined multiple airflows,
Energy, Volume 224, 2021, 120131.
https://doi.org/10.1016/j.energy.2021.120131
[7] G Sureandhar, G Srinivasan, P Muthukumar, S Senthilmurugan,
Performance analysis of arc rib fin embedded in a solar air heater,
Thermal Science and Engineering Progress, Volume 23, 2021, 100891,
ISSN 2451-9049, https://doi.org/10.1016/j.tsep.2021.100891
[8] Inderjeet Singh, Sachit Vardhan, Experimental investigation of an
evacuated tube collector solar air heater with helical inserts, Renewable
Energy, Volume 163, 2021, Pages 1963-1972, ISSN 0960-1481,
https://doi.org/10.1016/j.renene.2020.10.114
[9] Gagan Bansal, Chandra Kishore, R. Meby Selvaraj, Vijay Kumar
Dwivedi, Experimental determination of the effect of change in relative
roughness pitch on the thermo-hydraulic performance of air heater
working with solar energy, Materials Today: Proceedings, 2021, ISSN
2214-7853, https://doi.org/10.1016/j.matpr.2021.01.406.
[10] Refat Moshery, Tan Yong Chai, Kamaruzzaman Sopian, Ahmad
Fudholi, Ali H.A. Al-Waeli, Thermal performance of jet-impingement
solar air heater with transverse ribs absorber plate, Solar Energy,
Volume 214, 2021, Pages 355-366, ISSN 0038-092X,
https://doi.org/10.1016/j.solener.2020.11.059
[11] Agarwal, Shivansh Aggarwal, Performance evaluation of an arched
plate solar air heater with porous aluminum wire mesh cylindrical fins,
Energy Reports, Volume 6, Supplement 9, 2020, Pages 627-633, ISSN
2352-4847, https://doi.org/10.1016/j.egyr.2020.11.177
[12] Yogesh Agrawal, JL Bhagoria, Experimental investigation for pitch
and angle of arc effect of discrete artificial roughness on Nusselt
number and fluid flow characteristics of a solar air heater, Materials
Today: Proceedings, 2020, ISSN 2214-7853,
https://doi.org/10.1016/j.matpr.2020.09.249
[13] Naef A.A. Qasem, Mohamed N. Arnous, Syed M. Zubair, A
comprehensive thermal-hydraulic assessment of solar flat-plate air
heaters, Energy Conversion and Management, Volume 215, 2020,
112922, ISSN 0196-8904,
https://doi.org/10.1016/j.enconman.2020.112922.
[14] Anil Kumar, R.P. Saini, J.S. Saini, Experimental investigation on heat
transfer and fluid flow characteristics of air flow in a rectangular duct
with Multi v-shaped rib with gap roughness on the heated plate, Solar
Energy, Volume 86, Issue 6, 2012, Pages 1733-1749, ISSN 0038-
092X, https://doi.org/10.1016/j.solener.2012.03.014
[15] Lun Ma, Qingyan Fang, Chungen Yin, Huajian Wang, Cheng Zhang,
Gang Chen, A novel corner-fired boiler system of improved efficiency
and coal flexibility and reduced NOx emissions, Applied Energy,
Volume 238, 2019, Pages 453-465, ISSN 0306-2619,
https://doi.org/10.1016/j.apenergy.2019.01.084.
[16] Yacheng Liu, Weidong Fan, Yu Li, Numerical investigation of air-
staged combustion emphasizing char gasification and gas temperature
deviation in a large-scale, tangentially fired pulverized-coal boiler,
Applied Energy, Volume 177, 2016, Pages 323-334, ISSN 0306-2619,
https://doi.org/10.1016/j.apenergy.2016.05.135.
[17] Yukun Hu, Hailong Li, Jinyue Yan, Numerical investigation of heat
transfer characteristics in utility boilers of oxy-coal combustion,
Applied Energy, Volume 130, 2014, Pages 543-551, ISSN 0306-2619,
https://doi.org/10.1016/j.apenergy.2014.03.038.
[18] Suwarno Suwarno, Giri Nugroho, Andik Santoso, Witantyo, Failure
analysis of air preheater tubes in a circulating fluidized bed boiler,
Engineering Failure Analysis, Volume 124, 2021, 105380, ISSN 1350-
6307, https://doi.org/10.1016/j.engfailanal.2021.105380.
[19] Qiannan Zhang, Fengzhong Sun, Changxian Chen, Research on the
three-dimensional wall temperature distribution and low-temperature
corrosion of quad-sectional air preheater in larger power plant boilers,
International Journal of Heat and Mass Transfer, Volume 128, 2019,
Pages 739-747, ISSN 0017-9310,
https://doi.org/10.1016/j.ijheatmasstransfer.2018.09.006
[20] Subramaniyan, S. Venkatachalapathy, Heat transfer studies at different
speeds and loads of regenerative air preheater in thermal power plant,
Thermal Science and Engineering Progress, Volume 22, 2021, 100814,
ISSN 2451-9049, https://doi.org/10.1016/j.tsep.2020.100814
[21] Statistical yearbook of 2020. Statistical Agency of the Republic of
Kazakhstan. www.stat.gov.kz. [accessed 10.02.2021]
[22] Zhuravov A. A., PTVM-100 and PTVM-50 boilers reconstruction – the
real way to solve the problem of city heat power deficiency. Heating
supply news. 2000; 1, p.22-23.Power engineering 5, 1980, 1-30-02
- 334 -
Comparison of VARC DC Interrupting Module
Topologies for Enhancing Superconducting
VARC DCCB
Daeseong Jang and Hyosang Choi*
Chosun University, Korea
Email: [email protected]
Abstract— For the stable operation of high voltage direct allowing them to transmit current almost lossless under a
current(HVDC), research on circuit breakers with fast steady state in HVDC. However, if a fault occurs in the HVDC
interruption time is being done. Currently, hybrid DC circuit and a fault current exceeds the critical current of the
breakers(HDCCBs) are being researched, but they have slower superconducting element, the superconducting element can be
interruption times than semiconductor DC circuit breakers. quenched, causing a sharp rise in resistance and limiting the
Among the HDCCBs, we propose a superconducting vsc assisted fault current to rise. Through this, SFCL can reduce electrical
resonant current(VARC) DCCB combining a VARC DC energy burden as well as interruption time of the DC circuit
interrupting module with a superconducting fault current breaker. There are several types of SFCL, including resistive,
limiting(SFCL) module. The superconducting VARC DCCB
saturated iron core, and inductive [6]. Among them, the
can amplify resonant currents, causing the fault current to
reach an artificial zero point quickly through VARC DC
resistive type has a simple structure and is designed easily to
interrupting module. In addition, the superconducting VARC fit engineers’ needs. It also has so fast response time, that it
DCCB can reduce electrical energy burden by limiting fault can be quickly quenched to limit the fault current when a fault
current through the SFCL module when a fault occurs. In this occurs in the HVDC.
paper, the interrupting characteristics and electrical energy Interrupting technologies can be divided into resonant DC
burdens of superconducting VARC DCCB are compared circuit breaker (R-DCCB), solid-state DC circuit breaker (SS-
according to the VARC DC interrupting module’s topologies. As
DCCB), and hybrid DC circuit breaker (H-DCCB) [7]. R-
for the topologies, a full-bridge module in the VARC
DCCB is composed of a mechanical switch (MS), LC circuit,
interrupting module is classified into three circuit types: single,
series, and parallel. The operations of topologies are simulated and metal oxide varistors (MOV). It interrupts fault current
using PSCAD/EMTDC. Based on the simulation results, through a principle of LC series resonance. It has a simple
interruption times in the case of single and series are the same, structure and is economical, but it is not suitable for HVDC
interruption time in the case of parallel is about 2 ms slower than because of its slow interruption time and difficulty in selecting
others. Total electrical energy burdens are similar for single and the optimal capacity of LC elements. SS-DCCB is composed
series, and about 1.18 times greater for parallel than others. of a semiconductor switch such as an IGBT or SiC and a MOV.
It interrupts fault current through the turn-off operation of the
Keywords— Circuit breakers, fault current limiters, HVDC, semiconductor switch and has the fastest interruption time
PSCAD, superconducting devices among them. However, it is relatively inefficient because
resistance loss occurs in the ON state. H-DCCB is composed
I. INTRODUCTION of MS and semiconductor devices, which improve the
As the scale of renewable energy sources increases, high- disadvantages of other breakers. Recently, vsc assisted
voltage direct current (HVDC) transmission systems have resonant current (VARC) DCCB has been developed as one
been proposed as a suitable method for long-distance and of the H-DCCBs that combines an R-DCCB and full-bridge
large-capacity transmission [1], [2]. However, since there is (FB) module configured with IGBTs. It can amplify resonance
no zero point in the DC, it is more difficult to interrupt than through the FB module to bring a fault current to zero quickly
AC when a fault occurs. Therefore, research on DC protection [8], [9]. However, a lot of research is needed to enhance this
technology is essential for the reliable operation of HVDC. before it can be actively deployed in the real HVDC system.
Protection technology can be divided into current limiting In this paper, a study has been conducted to improve the
technology and interrupting technology. VARC DC interrupting module of superconducting VARC
A large inductor with a value of several hundred mH is DCCB. To this end, we modeled a superconducting VARC
commonly used in current limiting technologies of HVDC [3], DCCB that combines an SFCL module and a VARC DC
[4]. However, this can affect dynamic response and reduce the interrupting module using PSCAD/EMTDC. After that, three
stability of the system. In addition, it delays interruption time superconducting VARC DCCBs were constructed based on
by storing current energy. Superconducting fault current the topologies of the VARC DC interrupting module. The
limiter (SFCL) has emerged as one of the promising fault topologies are divided into single, series, and parallel,
current limiting technologies for HVDC [5]. Superconducting depending on the construction method of the FB module
elements have zero resistance in the superconducting state, within the VARC DC interrupting module. In addition, we
This work was supported by the Gwangju·Jeonnam local Energy Cluster Manpower modeled a fault simulation circuit diagram at 320 kV using
training of the Korea Institute of Energy Technology Evaluation and Planning(KETEP) PSCAD/EMTDC. The Interrupting characteristics and
grant funded by the Korea government Ministry of Knowledge Economy (No. electrical energy burdens were compared for topologies of
20214000000560) VARC DC interrupting modules.
This research was supported by Korea Electric Power Corporation [grant number:
R21XO01-32]
- 335 -
Fig. 2. Circuit diagram of the VARC DC interrupting module.
Fig. 1. Graph of quench resistance.
II. SFCL MODULE
The SFCL module is connected in series with the VARC
interrupting module. It is a resistive type that has a simple
structure and adjustable resistance. We modeled this in
PSCAD/EMTDC using (1) and (2). 𝑅𝑆𝐹𝐶𝐿 (t) is the resistance
of the SFCL, and 𝑅𝑚 is the maximum resistance of the SFCL
during the quench. 𝑅𝑚 is set to a value to limit the fault current
by more than 50 % [10]. 𝑇𝑆𝐶 is a time constant of the SFCL
module and means the time taken to reach 63.2 % of the 𝑅𝑚 Fig. 3. Circuit diagram of the VARC DC interrupting module
during the quench. The time constant varies depending on the topologies. (a) Single, (b) Series, (c) Parallel.
specifications of the superconducting elements but is typically
set to 1 ms or less [11]. 𝑡0 is the time that a fault occurs. Fig. 1 TABLE I. PARAMETERS OF VARC DC INTERRUPTING
MODULE TOPOLOGIES
shows the graph of a quench resistance curve of an SFCL
module modeled with PSCAD/EMTDC. In the test, the Topologies
maximum quench resistance and the time constant of the Parameters
Single Series Parallel
SFCL module were set to about 32 Ω and 0.411 ms. The
critical current of the SFCL module was set to about 2.0 kA MS delay [ms] 10 10 10
and if a fault current exceeds the critical current, the SFCL L [μH] 846.88 846.88 846.88
module is quenched to limit the rise of the fault current. C [μF] 30 30 30
𝑪𝑺 [mF] 10 10 10
𝑅𝑆𝐹𝐶𝐿 (𝑡) = 0 (𝑡0 > 𝑡) ()
Pre-charge source voltage [kV] 6 2*3 2*3
𝑡 Resonant frequency [Hz] 1,000 1003.02 999
𝑅𝑆𝐹𝐶𝐿 (𝑡) = 𝑅𝑚 √1 − exp (− ) (𝑡0 ≤ 𝑡) ()
𝑇𝑠𝑐 FB switching frequency [Hz] 1,000 1,000 1,000
MOV clamping voltage [kV] 480 480 480
III. VARC DC INTERRUPTING MODULE
VARC DCCB is an H-DCCB that combines an R-DCCB and 𝐶𝑠 in parallel. An IGBT switching frequency of the FB
with an FB module configured with IGBTs. It has no module is set to the same value as the resonant frequency of
resistance loss in On-State and amplifies resonance when a the LC elements. LC resonant frequency can be calculated by
fault occurs, bringing the fault current to zero quickly. In (4). The 𝐶𝑠 in the FB module is pre-charged by an external
addition, it is easy to select LC capacity because the amplitude power source. When MS opens, it discharges to match the
and frequency of the resonant current can be adjusted easily resonance of the LC elements, amplifying the resonant current.
by the FB module. The 𝑖𝑀𝑆 will be zero when 𝑖𝑟𝑒𝑠 reaches the value of fault
Fig. 2 shows a circuit diagram of the VARC DC current. At this point, the MS completes the opening operation,
interrupting module using PSCAD/EMTDC. VARC DC and the FB module stops switching. After that, when the
interrupting module consists of an interrupting circuit, an overvoltage is applied to the MS, energy is dissipated by the
absorption circuit, and a resonant FB circuit in parallel. The MOV.
interrupting circuit consists of a Mayr Arc model and an MS
connected in series. The MS was set to about 10 ms delay 1 𝑑𝑔𝑚 1 𝑖𝑠2
= 𝜏 (𝑃 − 1) ()
considering the real operation of MS. The Mayr Arc model 𝑔𝑚 𝑑𝑡 𝑚 0 ∙𝑔𝑚
was applied to the arc generated when MS opened. The Mayr
1
Arc model is shown in (3) [12], [13]. In (3), 𝑔𝑚 is the 𝑓 = 2𝜋 ()
conductivity of the arc, τ𝑚 is the time constant of the arc, 𝑖𝑠 is √𝐿𝐶
the current across the circuit breaker, and 𝑃0 is power loss. A
resonant FB circuit consists of LC elements and an FB module Fig. 3 shows the circuit diagrams of the VARC DC
in series. The FB module consists of IGBTs in the FB structure interrupting module topologies using PSCAD/EMTDC. Here,
FB modules in each topology are expressed as a symbol
- 336 -
Fig. 4. Fault simulation circuit diagram.
representing an IGBT. The topologies are categorized as (a)
single, (b) series, and (c) parallel depending on how the FB
module is configured in the resonant FB circuit. Fig. 3(a)
shows a single FB module, Fig. 3(b) shows three FB modules (a)
in series, and Fig. 3(c) shows three FB modules in parallel.
Table I shows the parameters of the VARC DC interrupting
module for each topology. In the case of series and parallel,
three 2 kV external sources are installed in each FB module to
equalize the total charge of the pre-charged 𝐶𝑠 s for each
topology. The charge on a capacitor can be calculated by (5)
Q=C∗V ()
IV. SIMULATION
A. Fault Simulation Circuit Diagram Modeling
Fig. 4 shows a fault simulation circuit diagram using
PSCAD/EMTDC. The DC voltage of the circuit diagram is set
to 320 kV, which is the DC rating of the INELFE system [14].
The reactance of the circuit diagram was set to 2 mH, the (b)
normal current to 1 kA, and the fault current to 10 kA when
no DCCB is installed.
B. Fault Simulation
Table II shows a comparison table of fault simulation
results for each topology.
Fig. 5(a) shows the interrupting graph of a
superconducting VARC DCCB for a single FB module. A
fault occurred at 0.1 s. A fault current rise should have risen
from about 1 kA to about 10 kA, but the SFCL module
quenched and limited the fault current to about 5 kA. MS
opens at 0.11 s after about 10 ms delay. When MS opens, arc
was generated at the MS contact, and a resonant current was
conducted by the LC series resonance. At the same time, the
pre-charged 𝐶𝑠 in the FB module discharged at the resonant (c)
frequency, amplifying the resonant current. The MS current
reached zero point at about 0.11115 s after being reversed two Fig. 5. Graph of fault current during the interruption by topologies. (a)
Single, (b) Series, (c) Parallel.
times by the amplified resonant current. And then, an
overvoltage of about 592.44 kV was generated between the TABLE II. COMPARISON OF SIMULATION RESULTS
MS contacts and the MOV was actuated to complete the
interruption at about 0.11117 s. The time from fault to Topologies
interruption was about 11.17 ms.
Single Series Parallel
Fig. 5(b) shows the interrupting graph of a
superconducting VARC DCCB for series FB modules. A fault Current limit rate [%] 50 50 50
occurred at 0.1 s. A fault current rise should have risen from
about 1 kA to about 10 kA, but the SFCL module quenched Reversals 2 2 6
and limited the fault current to about 5 kA. MS opens at 0.11 s
Reach zero point [ms] 11.15 11.15 13.15
after about 10 ms delay. When MS opens, arc was generated
at the MS contact, and a resonant current was conducted by Transient voltage [kV] 592.44 588.1 586.86
the LC series resonance. At the same time, the pre-charged 𝐶𝑠
in the FB module discharged at the resonant frequency, Interruption time [ms] 11.17 11.17 13.17
amplifying the resonant current. The MS current reached zero
- 337 -
Fig. 6. Graph of 𝐶𝑠 voltages by topologies during the interruption.
Fig. 7. Total 𝐶𝑠 discharge currents by topologies during the interruption.
point at about 0.11115 s after being reversed two times by the
amplified resonant current. And then, an overvoltage of about TABLE III. ELECTRICAL ENERGY BURDENS OF EACH
ELEMENT OF THE SUPERCONDUCTING VARC DCCB
588.1 kV was generated between the MS contacts and the
MOV was actuated to complete the interruption at about Electrical Energy Topologies
0.11218 s. The time from fault to interruption was about Burden [MW·t] Single Series Parallel
11.17 ms.
Fig. 5(c) shows the interrupting graph of a SFCL module 17,673,154 17,673,511 20,872,401
superconducting VARC DCCB for parallel FB modules. A Interrupting circuit 857 872 12,883
fault occurred at 0.1 s. A fault current rise should have risen
from about 1 kA to about 10 kA, but the SFCL module Resonant FB circuit 60,010 43,155 141,149
quenched and limited the fault current to about 5 kA. MS Absorption circuit 7,578 7,325 46,059
opens at 0.11 s after about 10 ms delay. When MS opens, arc
was generated at the MS contact, and a resonant current was Total 17,741,600 17,724,864 21,072,493
conducted by the LC series resonance. At the same time, the
pre-charged 𝐶𝑠 in the FB module discharged at the resonant B. Electrical Energy Burden Comparison
frequency, amplifying the resonant current. The MS current Table III shows the electrical energy burdens of each
reached zero point at about 0.11315 s after being reversed six element of the superconducting VARC DCCBs from fault
times by the amplified resonant current. And then, an occurrence to interruption. Electrical energy burden (EEB)
overvoltage of about 586.86 kV was generated between the means the integral of power burdens from fault occurrence to
MS contacts, and the MOV was actuated to complete the interruption by (7). In (7), W is an electrical energy burden,
interruption at about 0.11317 s. The time from fault to 𝑡0 is a time of fault occurrence, T is an interruption time, and
interruption was about 13.17 ms. p(t) is a power burden.
V. COMPARISON ANALYSIS 𝑇
𝑊 = ∫𝑡 𝑝(𝑡) 𝑑𝑡 ()
A. Interrupting Characteristic Comparison 0
Fig. 6 shows a graph of 𝐶𝑠 voltages by topologies. Here,
In the single case, the EEB of SFCL module was about
for series and parallel, they mean the 𝐶𝑠 voltages of one FB
17,673,154 MW·t, the EEB of interrupting circuit was about
module. For single and series, the rate of voltage discharge
was the same during the interrupting process. On the other 857 MW · t, the EEB of resonant FB circuit was about
hand, in the case of parallel, the rate of voltage discharge 60,010 MW · t, the EEB of absorption circuit was about
during the interruption was smaller than in the case of single 7,578 MW·t, for a total EEB of 17,741,600 MW·t.
and series. In the series case, the EEB of SFCL module was about
Fig. 7 shows a graph of total 𝐶𝑠 discharge currents by 17,673,511 MW·t, the EEB of interrupting circuit was about
topologies. For single and series, the rate of increase in total 872 MW · t, the EEB of resonant FB circuit was about
discharge current was the same during the interruption. On 43,155 MW · t, the EEB of absorption circuit was about
the other hand, in the case of parallel, the rate of discharge 7,325 MW·t, for a total EEB of 17,724,864 MW·t.
current increase was smaller than in the single and series
In the parallel case, the EEB of SFCL module was about
cases during the interruption. It can be confirmed by the
20,872,401 MW·t, the EEB of interrupting circuit was about
relationship between the current and voltage across the
capacitor, which is shown in (6). As a result, the interruption 12,883 MW·t, the EEB of resonant FB circuit was about
time was the same for single and series, and about 2 ms 141,149 MW · t, the EEB of absorption circuit was about
slower for parallel than others. 46,059 MW·t, for a total EEB of 21,072,493 MW·t.
In all cases, the SFCL modules performed well, with the
𝑑𝑣(𝑡) electrical energy burdens of them being about 99 % of the
𝑖(𝑡) = 𝐶𝑠 ()
𝑑𝑡 total electrical energy burdens. The total EEB increased in the
order of series, single, and parallel. Single and series cases
- 338 -
had similar EEB relatively. On the other hand, in the parallel [4] M. Wang, W. Leterme, G. Chaffey, J. Beerten, and D. Van Hertem,
“Multi-vendor interoperability in HVDC grid protection: State-of-the-
case, the EEBs of SFCL module, interrupting circuit, art and challenges ahead,” IET Gener. Transmiss. Distrib., vol. 15, no.
resonant FB circuit, and absorption circuit were greater by 15, pp. 2153-2175, 2021.
about 1.18, 14.89, 2.35-3.27, and 6.2 times than in other cases. [5] B. Ghatol, K. D. Thakur, “Application of Superconducting Fault
The total EEB in the parallel case is about 1.18 times greater Current Limiter (SFCL) for HVDC system with Mechanical Circuit
than in others. Breaker,” 2018 3rd IEEE International Conference on Recent Trends
in Electronics, Information & Communication Technology (RTEICT),
2018.
VI. CONCLUSION
[6] D. M. Larruskain, I. Zamora, and O. Abarrategui, “Fault current
In this paper, we compared the interrupting characteristics limiters for VSC-HVDC systems,” in Proc, 10th IET Int. Conf. AC DC
and electrical energy burdens of topologies of VARC Power Transmiss., pp. 1-6, 2012.
interrupting modules. The purpose of this study is to enhance [7] Q. Huo, J. Xiong, N. Zhang, X. Guo, L. Wu, T. Wei, “Review of DC
the superconducting VARC DCCB by studying the design circuit breaker application,”, Electric Power Systems Research, vol.
209, 2022.
direction of the VARC DC interrupting module. To this end,
a fault simulation circuit diagram, R-SFCL module, and [8] S. Liu, M. Popov, S. S. Mirhosseini, S. Nee, T. Modeer, L. Angquist,
N. Belda, K. Koreman, and V. D. Meijen, “Modeling, experimental
VARC DC interrupting module were modeled using validation, and application of VARC HVDC circuit breakers,” IEEE
PSCAD/EMTDC. Then, we divided topologies according to Transactions on Power Delivery, vol. 35, no. 3, 2020.
how the FB module is configured and simulated each case. [9] S. S. Mirhosseini, S. Liu, J. C. Muro, Z. Liu, S. Jamali, and M. Popov,
“Modeling a voltage source converter assisted resonant current DC
As a result, the SFCL modules performed as well as breaker for real time studies,” International Journal of Electrical Power
intended by modeling in the fault simulation. The fault & Energy Systems 117: 105678, 2020.
currents in all topologies were limited to about 50 % by SFCL [10] X. Pei, A. C. Smith, M. Barnes, “Superconducting Fault Current
modules. In terms of EEB, the SFCL modules account for Limiters for HVDC Systems,” Energy Procedia, vol. 80, pp. 47-55,
about 99 % more of the total EEB, which reduce significantly 2015.
EEB of the VARC DC interrupting modules. It also limited [11] R. P. R. Siddavatam, S. Kar, and K. Rajashekra, “Resistive SFCL
the fault current, which helped to reduce interruption time. integrated ultrafast DC hybrid circuit breaker for subsea HVDC
transmission systems,” IEEE Trans. Ind. Appl., vol. 58, no. 5, pp. 5977-
Interruption times were about 11.17 ms for single and 5986, 2022.
serial, and 13.17 ms for parallel with about 2 ms slower than [12] U. Havedank, “Application of a new arc model for the evaluation of
others. It is related to the magnitude of pre-charged 𝐶𝑠 short-circuit breaking tests,” IEEE Trans. Power Del. vol. 8, no. 4, 99.
1921-1925, 1993.
voltages of the FB module and the rate of discharge current
[13] L. Yuan, L. Sun, and H. Wu, “Simulation of fault arc using
rising. In the single and series cases, comprehensive 𝐶𝑠 conventional arc models,” Energy Power Eng., vol. 05, pp. 833-837,
voltages acting on the circuit diagram are about 6 kV, but in 2013.
the parallel case, it acts about 2 kV. In other words, the parallel [14] P. L. Francos, S. S. Verdugo, H. F. Alvarez, S. Guyomarch, J. Loncle,
case has a slower interruption time than others because of a “INELFE – Europe’s first integrated onshore HVDC interconnection”,
lower rate of rise in discharge current over time. 2012 IEEE Power and Energy Society General Meeting, pp. 1-8, 2012.
The EEB was significantly higher for parallel than single
and series. The total EEB was about 1.18 times higher for
parallel than others, and the interruption time was also about
1.18 times slower, indicating that the interruption time and
EEB are proportional.
Therefore, when designing superconducting VARC
DCCB, it is necessary to study fully the magnitude of pre-
charged 𝐶𝑠 voltage, which can affect interrupting time and
EEB. In addition, results show that connecting the FB
modules in series rather than in parallel will facilitate
increasing the magnitude of the 𝐶𝑠 voltage, which will have a
positive effect on the interruption time and EEB. However, in
the case of parallel, a fault current is divided to each FB
module by the number of parallel connections, which may
have a positive effect on the durability of FB modules. Thus,
this also should be considered for the successful design of
superconducting VARC DCCB. Furthermore, the study of
verifying the performance of this DCCB in real HVDC grids
is needed.
REFERENCES
[1] C. Humpert, “Long distance transmission systems for the future
electricity supply – Analysis of possibilities and restrictions,” Energy,
vol. 48, no. 1, pp. 278-283, 2012.
[2] A. Stan, S. Costinas, and G. Ion, “Overview and Assessment of HVDC
Current Applications and Future Trends,” Energies, vol. 15, no. 3, 2022.
[3] K. Shen, S. Wang, D.Zhao, and G. Zhao, “Influence of two-stage
coupled arm inductors on modular multilevel converter,” IEEE Access,
vol.7, pp. 162488-162499, 2019.
- 339 -
Resonance Frequency Analysis
for LC Resonant DC Circuit Breaker Design
Jihye Kim and Hyosang Choi *
Chosun University, Korea
Email: [email protected]
Abstract— This paper deals with the study of how to design
an LC resonant DC circuit breaker in a DC system. This study
aims to reveal the relationship between the frequency of the DC
system and the resonance frequency of the LC resonant DC
breaker, and to present a method to design the LC resonant DC
breaker based on it. Through experiments, we analyze the
operating characteristics of LC resonant DC breaker in a DC
system and measure the current flowing through the LC
resonant module. After that, Fast Fourier Transform analysis is
used to determine the dominant frequencies in the frequency
domain. Investigate the relationship between the resonance
frequency of the LC resonant DC breaker and the frequency of
a DC system in the frequency domain. As a result, we observe Fig. 1. DC system fault simulation circuit diagram
that resonance and main frequencies are largely proportional,
but singularities may exist. In addition, it is confirmed that the
resonance frequency, arc energy, arc duration time, and
interruption time are generally proportional.
Keywords— Arc discharges, Circuit breakers, Fourier
transforms, Frequency-domain analysis, Power system protection
I. INTRODUCTION
With the increasing demand for DC-based power and the
expansion of distributed power sources, there is growing
interest in DC systems as a solution to overcome the
limitations of traditional distribution networks [1]. Currently,
most commercially deployed DC systems adopt a 2-terminal
point-to-point configuration. However, the 2-terminal DC
system has the disadvantage of interrupting power flow when
the fault current is interrupted by controlling the power
Fig. 2. Experimental graph of operation characteristics of LC resonant
conversion devices. As a result, MMC(Modular Multi-level
DC circuit breaker
Converter)-based HVDC systems, capable of controlling
power flow, have gained attention and are evolving into Multi- the performance of the LC resonant DCCB. If the main
terminal DC (MTDC) systems [2-4]. However, MTDC frequency and the resonance frequency do not match, the
systems experience significantly larger fault currents performance of the DCCB deteriorates, and in severe cases, it
compared to 2-terminal DC systems. The control strategies of may fail to interrupt the fault current. Therefore, when
power conversion devices serve as alternatives to DC circuit designing an LC resonant DCCB, all of these factors need to
breakers but cannot fundamentally address the challenge of be considered to set the resonance frequency appropriately.
interrupting such high fault currents. In this regard, we are However, currently, the resonance frequency of the LC
studying the development of an LC resonant DC circuit resonant module is arbitrarily set due to the inability to find a
breaker (DCCB) capable of reliably interrupting DC fault resonance point where the main frequency of the DC system
currents. The LC resonant DCCB utilizes the resonance of an and the resonance frequency of the LC resonant DCCB
LC resonant module to create an artificial current zero- coincide.
crossing point, thereby interrupting the fault current. Properly
The purpose of this study is to establish a method for
setting the resonance frequency of the LC resonant module is
designing the LC resonant module to match the main
crucial for determining the performance of the circuit breaker.
frequency of the constructed DC system, moving away from
When operating a grid, power quality issues must be the existing approach of arbitrarily setting the resonance
considered. This is because a high amount of power electronic frequency. It elucidates the relationship between the main
converters required for grid configuration and weak frequency of the DC system and the resonance frequency of
connection points of the grid can cause harmonic excitation, the LC resonant DCCB. Additionally, the current flowing
leading to disturbances in the grid [5]. These disturbances through the LC resonant module is subjected to FFT analysis
affect the system's main frequency and, furthermore, impact to identify the resonance points.
• This work was supported by the Gwangju·Jeonnam local
EnergyCluster Manpower training of the Korea Insitute of Energy
Technology Evaluation and Planning(KETEP) grant funded by the
Korea government Ministry of Knowledge Economy (No.
20214000000560)
- 340 -
Fig. 4. 𝑖𝐿𝐶 by resonant frequency analyzed by Fourier spectrum
TABLE I. MAIN FREQUENCY ACCORDING TO RESONANCE
FREQUENCY
Resonance Main Frequency
frequency Frequency Subtraction
[Hz] [Hz] [Hz]
Fig. 3. Fourier spectrum of normal current and 𝑖𝐿𝐶 by resonance
frequency 1000 2187.5 1187.5
II. FFT ANALYSIS OF LC RESONANT DC CIRCUIT 1200 2187.5 987.5
BREAKER 1400 1875.0 475.0
A. Mechanism of LC Resonant DC Circuit Breaker 1600 2187.5 587.5
Fig. 1. is an experimental circuit diagram simulating the
1800 2500.0 700.0
occurrence of a failure to measure the arc signal generated in
the LC resonant DCCB. It consists of a DC power supply, LC 2000 2500.0 500.0
resonant DC circuit breaker, and fault occurrence device. The
LC resonant DC circuit breaker is composed of the
Mechanical Circuit Breaker(MCB) module, LC resonant By appropriately designing the inductance and capacitance of
module, and Surge Protective Device(SPD) module connected the LC resonant module, matching the frequency component
in parallel.The LC resonant DC circuit breaker is composed of (𝑓𝑀𝑀𝐶 ) contained in 𝑖𝐿𝐶 , resonance occurs, and an artificial
the Mechanical Circuit Breaker(MCB) module, LC resonant current zero-crossing point is created in the fault current.
module, and Surge Protective Device(SPD) module connected However, in practical systems, the frequency components
in parallel. contained in 𝑖𝐿𝐶 are determined by factors such as the
The operational mechanism of the LC resonant DC circuit operating frequency range, circuit quality factor (Q), material
breaker can be explained in two states: normal state and fault properties of components, component size, and various other
state. In the normal state, the MCB module remains closed, factors. It is extremely challenging to calculate the main
allowing normal current flow through the MCB circuit. In the frequency of the practical system accurately. Therefore, the
fault state, the MCB module detects a fault signal and opens. capacitance is adjusted until an artificial current zero-crossing
An arc occurs between the open contacts of the MCB, and due point is generated in the fault current, as part of the design
to the characteristics of the arc, the fault current flows through process for the LC resonant DC circuit breaker.
the LC resonant module. The current flowing through the LC B. Experimental Design and Method
resonant module can be calculated using equation (1) [6].
To simulate a DC fault, a system was constructed as shown
𝑑𝑢𝑎𝑟𝑐 in Fig. 1. The rated voltage was about 500 V, and the rated
−(0.5𝐿)( ) 1
𝑖𝐿𝐶 = 𝑖𝑡𝑜𝑡𝑎𝑙 {1 + 𝑒 𝑑𝑖𝑀𝐶𝐵 𝑡 sin (√ 𝑡)} () current was about 50 A. To simulate a fault condition, a fault
𝐿𝐶 impedance of about 1 Ω was designed to allow a fault current
of about 500 A to flow. The resonant frequency of the LC
In the case of an Modular Multi-level Converter(MMC) based resonant DC circuit breaker was divided into intervals of 200
HVDC system, when a fault occurs, the power electronic Hz from about 1000 Hz to 2000 Hz.
converters of the MMC undergo the interruption process,
resulting in frequency components in the fault current [7]. The Fig. 2. is an experimental graph showing the operating
frequency generated by the MMC can be calculated using characteristics of the LC resonant DCCB when a failure is
equation (2) [8]. simulated. Fig. 2. (a) represents the point at which the MCB
begins to open. Fig. 2. (b) represents the point at which the
2
current reaches the zero-crossing point. Fig. 2. (c) represents
1 1 𝑅𝐷𝐶 +𝑅𝑓
𝑓𝑀𝑀𝐶 = √ −( ) () the point at which the interruption is completed. The (a) to (b)
2𝜋 𝐶𝐷𝐶 (𝐿𝐷𝐶 +2𝐿𝑠 ) 2(𝐿𝐷𝐶 +2𝐿𝑠 ) interval corresponds to the zero-crossing point reaching
interval and is influenced by the LC resonant module.
According to the definition of a circuit breaker, the (a) to (c)
interval represents the rated interruption time. If an artificial
- 341 -
frequency. However, the relationship with the main frequency
couldn’t be determined.
III. CONCLUSION
In this paper, the aim was to address the issue of arbitrarily
determining the resonance frequency when designing an LC
resonant DCCB. To achieve this, a simulated DC fault circuit
was constructed, and the fault current flowing through the LC
resonant module was measured. The fault current was
analyzed using FFT to identify the main frequency. The
resonance frequency of the LC resonant DCCB was found to
be generally proportional to the main frequency. However,
there was a peculiar anomaly at 1400 Hz. To investigate the
relationship between the anomaly and the frequency, an
analysis of the arc was conducted. The resonance frequency
Fig. 5. Interrupting time and arc analysis by resonance frequency showed a generally proportional relationship with arc energy,
arc duration time and interruption time. However, the
current zero-crossing point is generated during the (a) to (b) resonance frequency when the arc duration time was the
interval, and the current reaches 0 A during the (a) to (c) shortest was about 1800 Hz, and the resonance frequency
interval, the interruption is considered successful. Therefore, when the interruption time was the shortest was about 1200
the current data flowing through the LC resonant module of Hz. These peculiarities are presumed to be due to the irregular
the experiment in which the blocking was successful is characteristics of the arc. In addition, LC resonance modules
obtained. A total of 10 experiments were repeated for each with a wider range of resonance frequencies should be
experimental condition, of which 3 current data were selected designed to analyze more reliable tendencies.
and averaged.
In conclusion, this research emphasizes the importance of
C. Experiment and Discussion resonance frequency in DC systems and presents methods
An arc is not defined as a simple harmonic wave with a related to the design of LC resonant DC circuit breakers.
constant period and amplitude. Non-harmonic waves, such as Therefore, such research is highly valuable for enhancing the
arcs, which exhibit irregular characteristics, can be analyzed stability and power quality of DC systems, and it can be
using Fourier Transform to obtain their representation. Fourier applied to future DC system design and operation.
Transform expresses any irregular motion as a sum of simple
harmonic waves. FFT analysis was conducted to analyze the
main frequency of the DC system constructed in the REFERENCES
laboratory.
Fig. 3. represents the Fourier spectrum of the normal [1] Ricardo Granizo Arrabé; Carlos A Platero; Fernando Á lvarez Gómez;
current and 𝑖𝐿𝐶 for each resonance frequency. In order to Emilio Rebollo López, "New Differential Protection Method for
Multiterminal HVDC Cable Networks", Advances in Energy Research:
know the main frequency, it was compared with the main 2nd Edition, 1-27, 2021.
frequency of normal current. The main frequency of the DC [2] Boyu Qin; Wansong Liu; Ruowei Zhang; Jialing Liu; Hengyi Li,
system is characterized by a concentration of energy "Review on Short-circuit Current Analysis and Suppression
compared to the normal current. Fig. 4. shows 𝑖𝐿𝐶 for each Techniques for MMC-HVDC Transmission Systems", Applied
resonant frequency as one Fourier spectrum. The results of Fig. Sciences, Vol. 10, Issue 19, 2020.10
4. are summarized in Table 1. The frequency difference is [3] Mohan Muniappan, "A comprehensive review of DC fault protection
calculated by subtracting the resonance frequency from the methods in HVDC transmission systems", Protection and Control of
Modern Power Systems, Vol. 6, Issue 1, 2021.01
main frequency. The resonance frequency and the main
[4] Lei Zhang; Yuntao Zou; Jicheng Yu; Jiangchao Qin; Vijay Vittal;
frequency generally showed a proportional relationship. George G. Karady; Di Shi; Zhiwei Wang, "Modeling, control, and
However, they did not increase uniformly, and there were protection of modular multilevel converter-based multi-terminal
anomalies, such as at 1400 Hz. This can be attributed to the HVDC systems: A review", CSEE, Vol. 3, Issue 4, 340-352, 2017.12
main frequency being influenced by factors other than the [5] A. Neufeld, D. Heide and L. Hofmann, "Calculation of Series
resonance frequency. Resonance Frequencies in Electrical Grids Using Resonance Mode
Analysis," 2019 IEEE PES Asia-Pacific Power and Energy
Fig. 5. is a graph summarizing arc power, arc duration time Engineering Conference (APPEEC), Macao, China, 2019, pp. 1-4
and interruption time for each resonance frequency. To [6] A. Mokhberdoran, A. Carvalho, H. Leite and N. Silva, "A Review on
investigate the relationship between the main frequency and HVDC Circuit Breakers," IET Renewable Power Generation, Jul. 27,
𝑖𝐿𝐶 , arc power, arc duration time, and interruption time were 2017.
analyzed. The arc detection time corresponds to the period [7] P. Qiu, X. Xuan, Y. Lu, H. Shen, L. Qi and X. Wei, "Design and
Application of Short Circuit Test Scheme for Hybrid DC Circuit
when the arc current is influenced by the resonance frequency, Breakers in Zhoushan VSC-HVDC Transmission Project," High Volt.
as observed in Fig. 2. (a)-(b). The interruption time refers to Eng., vol. 47, no. 4, pp. 1428-1435, Apr. 2021.
the rated interruption time during the periods depicted in Fig. [8] H. Shen et al., "Modeling of High-frequency Electromagnetic
2. (a)-(c). Arc energy, arc duration time, and interruption time Oscillation for DC Fault in MMC-HVDC Systems," in CSEE Journal
exhibited a proportional relationship with the resonance of Power and Energy Systems, vol. 9, no. 3, pp. 1151-1160, May 2023
- 342 -
Failure Analysis of Arc and Leakage
Generation Signal in DC Line
Han SoYeon1 , Bang SuChul1, Jeong SungIn2 and Yoon YongHo2*
1
Dept. of Future Technology Convergence Engineering Graduation School of Gwangju Univ, Korea
2*
Dept. School of Electrical & Electronics Engineering, Gwangju Univ., Korea
Email: [email protected]
Abstract— Distributed energy technology that combines occurred on. Accordingly, the Hall Type CT board was
multiple renewable energy sources to generate electricity is drawing manufactured to detect arcs by line (channel) as shown in
attention. This will develop near the detention area and does not Figure 2(b) by reading which line (channel) generated arcs
require a long-distance transmission network. As a result, new through Hall Type CT. Due to the characteristics of each CT,
construction costs can be reduced, while stability needs to be Hall Type CT has a magnetic core, making it difficult to
secured as it is located in an acceptable neighborhood that can cause accurately read arc generation and arc extinction due to
many casualties in the event of an electrical accident. Therefore, in saturation, and Rogowski CT is more sensitive than Hall Type
this paper, we want to analyze the failure of the line for occurrence CT. [4,5]. Therefore, two CTs were used to analyze arc-
of arcs and leaks on DC lines and secure stability through a
generated faults, and the information analyzed was designed
monitoring system.
to be checked by the monitoring system.
Keywords—Renewable energy, distributed energy, Arc and
leakage current in DC line, DC line fault analysis, monitoring system
I. INTRODUCTION
Distributed energy technology that combines multiple
renewable energy sources to generate electricity is drawing
attention. This has the advantage of reducing construction and
operating costs during new construction because long-distance
transmission networks are not required because they develop
in acceptable neighborhoods. However, since it is located near
the camp, electrical accidents can cause many casualties and
property damage, such as fires and electric shocks. A zero
point is formed when an arc is generated in the DC line, and
since the magnitude of the current operates within a normal
range, it is difficult to cut it off with an overcurrent breaker [1]. Fig. 1. DC arc generator
In addition, DC distribution lines need to secure safety in DC
distribution technology due to the obvious shortcomings of
ground electrode illumination sites [2]. Therefore, in this paper,
the monitoring system was designed through line failure
analysis when arc and leakage current occur.
II. FAULT ANALYSIS OF ARC AND LEAKAGE CURRENTS
IN DC LINES
A. FAILURE ANALYSIS METHOD AND DESIGN IN
CASE OF ARC GENERATION IN DC LINE
When an arc is generated on a DC line, the generation of (a) Hall Type CT and Rogowski CT
the arc was simulated using an arc generator satisfying
UL1699B standard as shown in Figure 1 for analysis of the
line failure [3].
In addition, CT was used for fault analysis in case of arc
generation as shown in Fig. 2. A fault signal for arc generation
was detected using Hall Type CT and Rogowski CT as shown
in Figure 2(a). An arc generation start point and an arc
extinction point are read using a Rogowski CT. In addition,
(b) Hall Type CT and Rogowski CT
arcs were detected for each channel so that the distributed
energy using multiple DC lines could check which line the arc Fig. 2. CT Used to Detect Line Failure Signals When Arcing Occurs
- 343 -
B. FAILURE ANALYSIS METHOD AND DESIGN IN III. EXPERIMENTAL RESULTS
CASE OF LEAKAGE CURRENT IN DC LINE
A. Fault analysis in case of arc on DC line
A leakage current generator is used for line failure analysis
when a leakage current occurs in the DC line, and a CT for The arc generation fault signals measured through Hall
measuring leakage current is used as shown in Fig. 4 to Type CT and Logoski CT can be identified as shown in Figure
measure a failure signal when the leakage current occurs. CT n. A signal measured through Hall Type CT in C1, a signal
for measuring leakage current can be measured up to -10 mA measured through Logoski CT in C2, a signal amplified by
to +10 mA and is outputted by signals of -5 V to +5 V. In order Hall Type CT in C3, and a signal amplified by Logoski CT in
to apply the leakage current failure signal of –5V to +5V C4 can be identified. In C4, it was confirmed that the value
measured through CT to 0-3.7V which is an input range of accepted by the current sensor reacts sensitively when
MCU, the signal was converted using a level shift system. As compared to the Hall Type waveform when arc is generated
shown in Fig. 5, signals (-) 5V to (+) 5V measured through CT on DC lines because it is composed of a coreless core due to
and signals (0 to (+) 3.7V) level-converted for input into the its characteristics of Logoski CT. On the other hand, in C3, it
MCU can be checked, and how much leakage current has been is confirmed that the occurrence from the start point of an arc
generated in which line between (+) and (-) lines through the to the extinguishing point due to residual magnetic flux and
monitoring system. saturation phenomenon is unknown due to Hall type CT
characteristics. Therefore, in this paper, the presence or
absence of arc generation (arc start and end) was identified
through Locoski CT, and the presence or absence of arc
generation locations by line (channel) was read through Hall
Type CT.
Figure 7 shows a monitoring screen when an arc occurs on
a DC line. 7(a) is a screen when no arc occurs on the line, and
7(b) is a screen where signals measured through Logoski CT
are analyzed for failure when an arc occurs on the line. When
an arc occurs, a "warning" message is displayed to check
whether the arc occurs. In addition, in Figure 7(c), the signal
measured through Hall Type CT is a failure analysis screen,
and on which line (channel) the arc occurred can be confirmed.
(a) 0mA (b) 10mA
Fig. 3. Leakage current generator
Fig. 7. Fault signals measured by Hall Type CT and Rogowski CT
Fig. 4. Leakage current generator
Fig. 5.
(a) Nomal (b) Reading for arc (c) Arc-generated
generation track (channel)
reading
Fig. 6. Leakage current generator Fig. 8. Monitoring Arc Occurrence Fault Analysis
- 344 -
B. Fault analysis in case of leakage current in DC line
A leakage current signal value generated between (-)10
mA to (+)10 mA is shown in Fig. 8. In Fig. 8(a), the measured
signal can be confirmed through CT, and it can be confirmed
that it is measured from (-)5V to (+)5V signals. In Fig. 8(b),
the signal output between 0 and (+) 3.7 V can be confirmed
with the level-shifted signal, and the result of converting the
reference point from zero point (0) to (+) 1.9 V can be
confirmed.
Therefore, the monitoring system can be designed through (a) -10mA (b) Nomal (c) +10mA
the analysis of leakage current generation failure and
confirmed in Figure 9. Fig. 9(a) shows a monitoring screen Fig. 10. Monitoring Leakage Current Occurrence Fault Analysis
when leakage current of 10 mA is generated on the (-) line,
and information about signal values of (-) 10 mA to 0 mA
corresponding between Fig. 8(b) and 0 V to (+) 1.9 V is shown.
Fig.9(b) is a monitoring screen in the normal state of a line IV. CONCLUSION
where no leakage current is generated. Fig.9(c) shows a The technology of distributed energy generation, in which
monitoring screen when leakage current of 10 mA is generated multiple renewable energies are combined to generate
on a (+) line, and information about leakage current generation electricity through renewable energy generation, attracts
signal values of 0 to 10 mA corresponding between Fig.8(b) attention and requires securing safety. Accordingly, in this
and (+)1.9 V to (+)3.7 V is shown. paper, the monitoring system was designed and organized as
follows through analysis of track faults in case of arc and
leakage on DC lines.
Real-time measurement of fault signals through
the occurrence of arcing and leakage currents in
DC circuits for the diagnosis of line safety
Identification of occurrence of arc and leakage
current in DC line through monitoring system
Improved fault diagnosis technology for DC
power systems such as renewable and
distributed energy
ACKNOWLEDGMENT
(a) Fault signal measured by CT for leakage current detection This research was supported by Korea Electric Power
Corporation. (Grant number:R21XO01-37).
REFERENCES
[1] G. S. Seo, “Study on Series Arc Fault Detection for DC Microgrids,” Ph.
D. thesis Seoul National University, 2015.
[2] Y. H Yoon, “Maintenance of Smart Grid System Performance by
Monitoring the Insulation Status of On-DC Lines,” Trans. KIEE, vol.
70, no. 8, pp. 1232 ~ 1237, 2021.
[3] Y. H. Yoon, “Safety Analysis of Smart Grid Lines According to DC Arc
Generation,” Journal of Electrical Engineering & Technology, Journal of
Electrical Engineering & Technology, vol. 18, no. 1, pp. 697~703, 2023.
[4] M. Rezaee and H. Heydari, “Design modification of rogowski coil for
current measurement in low frequency,” Iranian Journal of Electrical &
Electronic Engineering, vol. 6, no. 4, pp. 232~237, 2010.
(b) Level-shifted signal [5] J. H. Jeong and Y. H. Yoon, “Arc Generation Detection Analysis of Smart
Grid System Using a New Arc Detection Sensor of Current Measurement,”
Fig. 9. Leakage current generating waveform Journal of Electrical Engineering & Technology, vol. 17, no. 5, pp.
3077~3084, 2022.
- 345 -
Detection of Leakage Current Occurrence
Location of LVDC Lines through Level-Shift
Signal Processing
Su-Chul Bang 1, So-Yeon Han 2, Sung-In Jeong3 and Yong-Ho Yoon *
1
Gwangju University, Korea, Republic of
Email: [email protected]
Abstract— Recently, there has been a growing interest in II. DETECTION OF PROPOSED LEAKAGE CURRENT
smart grids, leading to an increased adoption of renewable GENERATION LOCATIONS
energy sources such as solar and wind power, which utilize low-
voltage direct current (LVDC). Additionally, there is a rising A. Current sensor for DC leakage current measurement
commercialization of high-voltage direct current (HVDC) for In order to detect leakage current, this study utilized a hall
efficient long-distance power transmission. The advancement of
sensor in the form of a current sensor. The measurement
these direct current distribution systems has highlighted the
importance of preventing faults and accidents, as well as the
method involved the passage of the (+) line and the (-) line
need for effective fault diagnosis. through a magnetic core with excellent self-recovery
performance. As shown in Figure 1, to determine which line,
Currently, traditional current relays face challenges in either the (+) or (-) line, experienced the leakage, a voltage of
detecting single-line-to-ground faults in LVDC lines. Therefore, ( 12V) was applied to the In terminal connected to the hall
in this paper, we propose using level shifting to input the leakage sensor in the form of a current sensor. The neutral point of the
current values measured by leakage sensors into an MCU resistor was measured as the output signal, allowing the
(Microcontroller Unit). By monitoring these values, our aim is measurement of the voltage corresponding to the occurrence
to diagnose of leakage..
Keywords—Smart Grid Distributed Power, Simulator, Failure
Analysis, Arc, Leakage Current, Signal Conversion
I. INTRODUCTION
As interest in smart grids has recently increased, interest
in DC distribution systems based on efficient HVDCs
generated from renewable energy sources such as solar and
wind power, and long-distance transmission such as
continents and continents.
Unlike AC distribution, low-voltage DC distribution has a
characteristic that the instantaneous value of voltage is fixed.
Consequently, the distribution line is divided into positive and Fig. 1. Leakage current measurement unit structure
negative depending on the potential of the electrode.
Therefore, all equipment connected to a DC power B. Proposed Level Shift Conversion
distribution system must take into account polarity [1]. if a
We intend to measure the leakage current generation
ground fault occurs in low-voltage DC distribution with these
characteristics and IT method, which is a non-grounding described in Section 2.A through the proposed Level Shift
method, the insulation resistance between the power line and conversion and perform fault analysis using MCUs. Due to
the earth decreases when a one-wire ground fault occurs. If the characteristics of the MCU, there is an acceptable voltage
one of the DC distribution lines is connected to the ground due limit (0~3.7V), and it goes through two conversions to
to insulation degradation, it is highly likely that it will lead to convert the voltage value with this as a goal, which is the
a two-middle ground fault accident, which forms a closed proposed level shift conversion process. A minimum value
circuit. In this case, unlike alternating current distribution, DC corresponding to a (-) line of the measured signal is set at 0
distribution systems maintain a certain value, making it V or more and a maximum value corresponding to a (+) line
difficult to extinguish natural arcs due to arc generation, which is set at 3.7 V or less, so that input voltage values are
can lead to fire and cause large-scale property and casualties converted into (-)3 V-(+)3 V through voltage distribution law
[2-4]. in primary conversion.
Therefore, this paper seeks to secure reliability by A signal value measured through a current sensor is
analyzing the condition diagnosis and safety of LVDC lines in designed to be output as an input waveform signal in Fig. 2.
ground fault accidents using simulated test equipment using The current sensors used in this paper can measure current
level shift in low-voltage DC distribution.
- 346 -
Fig. 2. The process of Level Shift conversion
TABLE I. LEVEL SHIFT CONVERSION VALUE
Current(mA) Voltage(V) Current(mA) Voltage(V) Current(mA) Voltage(V)
(-) 10 (-) 5 (-) 10 (-) 3 (-) 10 (-) 0
0 0 0 0 0 (+) 1.85
(+) 10 (+) 5 (+) 10 (+) 3 (+) 10 (+) 3.7
(a) (b) (c)
values from (–) 10 mA to (+) 10 mA, which can be expressed B. Result of Level Shift conversion
as values from (–) 5 V to (+) 5 V, as shown in Table 1(a).
Table 1(b) shows the waveform obtained by voltage
distribution and primary conversion of the input waveform
and the conversion value. Comparing Table 1(a) and (b), it
can be confirmed that the range of (–)5V to (+)5V is changed
from (–)3V to (+)3V. Subsequently, Table 1(c) confirms that
the proposed Level Shift conversion converts the range from
(–)3V to (+)3V into a range from 0V to 3.7V.
III. EXPERIMENTAL RESULT
A. DC Leakage current Generated Waveform
A current of (-)10 mA to (+)10 mA is generated through a
DC leakage current generator. A leakage current generated Fig. 4. DC Leakage current Level Shift Waveform
through a DC leakage current generator is applied to a current
sensor described in 2.A. Figure 3 shows that leakage has
occurred in the (-) and (+) lines and is being measured.
Fig. 5. Leakage current fault analysis monitoring
Fig. 3. DC Leakage current Generated Waveform
- 347 -
Fig. 4 is a graph in which leakage current values of (-)10
mA to (+)10 mA are changed in units of 2 mA through level
shift, and a waveform in which the measured values of (-)5 V
to (+)5 V are converted into (-)3 V to (+)3 V and 0 to 3.7 V is
shown. Finally, the signal values of 0-3.7 V will be analyzed
through the MCU, and as shown in Fig. 5, the monitoring
value for fault analysis due to leakage current generation will
be shown.
IV. CONCLUSIONS
With the development of DC distribution technology, low-
voltage DC systems are also expanding. In this situation,
accidents and fires caused by leakage of DC components are
also occurring frequently. DC leakage current cannot be
blocked in advance with existing circuit breakers and short
circuit breakers.
Therefore, this paper proposed a method for measuring
leakage current and diagnosing faults through the proposed
Level Shift conversion. Therefore, by monitoring the fault
diagnosis data through the proposed Level Shift
transformation, it can be organized as follows.
● Leakage of the photovoltaic system is measured by a
leakage sensor, and the energy leakage value can be corrected
to determine which line of (+) or (-) lines connected to the
photovoltaic system caused the leak.
● Accordingly, the manager can check where leakage
occurs in a wide range of electrical facilities of the
photovoltaic power generation system, and can quickly
handle and take measures for failure of the facilities.
Acknowledgements This paper was conducted with the
support of the National Research Foundation of Korea, funded
by the Ministry of Science and ICT. (No. NRF-
2022R1F1A1063116).
REFERENCES
[1] Dong-eon. Kim, “Design of Fault Tolerant and Pluggable DC
Distribution System,” Thesis, Seoul National University, 2019
[2] J. Clerk Maxwell, A Treatise on Electricity and Magnetism, 3rd ed.,
vol. 2. Oxford: Clarendon, 1892, pp.68–73.
[3] Armijo, K. M., Johnson, J., Harrison, R. K., Thomas, K. E., Hibbs, M.,
and Fresquez, A., Quantifying Photovoltaic Fire Danger Reduction
with Arc-Fault Circuit Interrupters, Progress in Photovoltaics,
December 2014, Amsterdam, NED.
[4] Armijo, K. M., Johnson, J., Hibbs, M., and Fresquez, A.,
Characterizing Fire Danger from Low Power PV Arc-Faults, 40th
IEEE PVSC, June 2014, Denver, CO, USA.
- 348 -
Using IoT technologies to prevent fires
Akyltay Burgegulov1, Gulzat Ziyatbekova1,3*, Тalgat Mazakov1,3*, Aigerim Sametova1,2*,
Magzhan Aliaskar1*
1
Al-Farabi Kazakh National University, Almaty, Kazakhstan
2
Almaty University of Power Engineering and Telecommunications named after Gumarbek Daukeyev
3
Institute of Information and Computational Technologies CS MES RK, Almaty, Kazakhstan
Email: [email protected]
Abstract-- This article explores various aspects of using demonstrates how the technology can be adapted and
the Internet of Things (IoT) to enhance fire safety. The main optimized to solve specific tasks and problems related to fire
goal of the article is to provide an overview of modern IoT safety. Highlighting potential limitations and drawbacks of
technologies used for fire safety, as well as discuss challenges, existing IoT devices and systems allows to identify areas for
standards, and potential development of this field in the further refinement and development of the technology. This
future. The introduction of the article presents the definition can encourage researchers, developers, and manufacturers to
of IoT and its role in the modern world, the importance of fire continue seeking new, more effective, and safer solutions for
safety and related challenges, as well as an overview of the ensuring fire safety using IoT. Overall, the article is a
article's structure and the goals of discussing examples. valuable source of information for all interested parties,
Contemporary IoT smoke and heat sensors, their operating including fire safety professionals, engineers, architects,
principle, as well as examples of such sensors on the market, building and facility managers, as well as legislators and
including Nest Protect and Halo Smart Smoke Detector, are regulatory bodies. It underscores the importance of
examined. Intelligent alert systems and connections with integrating IoT technologies into fire safety systems and
emergency services, such as Birdi Smart Alarm and ADT demonstrates how these innovations can help reduce fire risks
Smart Home, are discussed. The working principle of such and save lives.
systems and examples of their usage are examined. The
integration of IoT technologies into fire extinguishing Keywords: Internet of Things, IoT, fire safety, smoke
systems is analyzed. Examples of IoT fire extinguishing sensors, heat sensors, smart home.
systems on the market, such as Plumis Automist Smartscan
and Viking QRFS, are provided, and the advantages of using I. INTRODUCTION
IoT in fire extinguishing are discussed. The work is also The Internet of Things (IoT) represents a concept in
devoted to the regulation and standardization of IoT which physical objects, such as sensors, appliances, and
technologies for fire safety. The importance of machines, collect, exchange, and analyze data via the
standardization and regulation, as well as the main standards internet. IoT technologies provide the ability to manage
and regulatory bodies in this area, are discussed. The article
and monitor devices remotely, facilitating daily life and
concludes and discusses potential directions for the
development of IoT technologies in fire safety in the future.
promoting process automation in various sectors such as
The authors highlight the potential of IoT technologies for industry, agriculture, healthcare, and transportation.
improving fire safety and offer ideas for further development The Internet of Things (IoT) has a revolutionary
of this sphere. Overall, this article provides valuable
impact on many aspects of our lives, including the field of
information for fire safety professionals, engineers,
developers of IoT devices, and all those interested in the fire safety. Fires can be destructive and dangerous to
development and application of IoT technologies in fire human life and property, so it's important to utilize the
safety. This article serves as a valuable resource for those who latest technologies to prevent the occurrence of fires and
want to familiarize themselves with contemporary approaches mitigate their impact. IoT technologies offer new
and solutions related to using IoT to enhance fire safety, as
opportunities for fire safety, such as smart smoke and heat
well as for those who aim to develop and implement new IoT
technologies in their projects. In addition, the article can be sensors that can warn of potential fires and timely alert to
useful for legislators and regulatory bodies looking to create a fire occurrence.
or update norms and standards regulating the application of Fire safety is a crucial aspect in ensuring the security
IoT technologies in fire safety. Analysis of existing standards of life and property. The primary challenges associated
and regulatory bodies in this field can serve as a starting point
with fire safety include ensuring timely fire detection,
for determining the requirements for future IoT devices and
fire safety systems. Ultimately, this article represents a kind of effective alerting and response to fire, as well as the
landmark study that can stimulate further developments and application of modern technologies to prevent fires and
innovations in the field of IoT technologies for fire safety. It mitigate their impact.
draws attention to a number of important issues and The structure of the article is designed to provide
challenges facing this industry, and offers solutions. It is
readers with information about modern IoT technologies
expected that IoT technologies will continue to play an
increasingly significant role in fire safety in the coming years, applied for fire safety. The article is divided into sections,
and this article will help navigate this rapidly changing and each discussing different aspects of IoT technologies for
developing field. The article also highlights potential fire safety, such as smoke and heat sensors, intelligent
advantages and opportunities offered by the application of alert systems, fire extinguishing systems, and the
IoT technologies for fire safety in various sectors, such as
application of IoT in various sectors. The purpose of the
industry, housing, and transportation. Discussion of specific
examples of IoT solutions in each of these sectors article is to give readers an understanding of the
- 349 -
possibilities of IoT technologies for enhancing fire safety sensors, providing the ability for remote monitoring and
and examples of successful application of these alerting, reflected in the Figure 3.
technologies in practice. • Remote Monitoring: The RemoteLync Monitor listens
Smoke and heat sensors using IoT are devices that for alarm signals from existing smoke and carbon
detect smoke and temperature changes, indicating a monoxide detectors in the home. In case a sensor is
potential fire. These sensors are connected to the internet triggered, the device instantly sends a notification to a
and transmit data about potential fire hazards to the smartphone or another mobile device.
owner's mobile devices or smart home platforms, as well • Compatibility: The device is compatible with most
as to other connected devices. This allows owners and smoke and carbon monoxide detectors available in the
security services to be promptly and efficiently alerted to market, allowing it to be used with already installed
a fire and to take steps to extinguish it. security systems.
• Mobile Application: The Kidde RemoteLync Monitor
APPLICATION AND ANALYSIS OF IOT TECHNOLOGIES FOR has its own mobile application, allowing users to
FIRE PREVENTION:
customize notifications, check device status, and monitor
Nest Protect is a smart smoke and carbon monoxide smoke and carbon monoxide detection.
sensor that uses IoT technologies for enhanced safety, • Expansion of Security System Capabilities: The ability
shown in Figure 1. to connect with other smart home devices, such as
• Detects fast-spreading fires, slowly smoldering fires, cameras, lighting, and automatic fire extinguishing
and carbon monoxide. systems, to create a more comprehensive and effective
• Voice alert about the type of danger and location of the safety system [5].
fire.
• The ability to receive notifications on the owner's
mobile devices and interact with other smart devices in
the home [1].
Fig 3. Kidde RemoteLync Monitor Device [6]
First Alert Onelink Safe & Sound is a multifunctional
smart sensor designed for home safety. It performs smoke
and carbon monoxide detection functions, and also
Fig 1. Internal structure of the Nest Protect sensor [2] provides a smart home voice assistant and a speaker for
entertainment and communication, illustrated in the
The Halo Smart Smoke Detector is a smart smoke sensor Figure 4.
that combines photoelectric, ionization, and heat sensors,
as shown in Figure 2. • Combined Sensor: Onelink Safe & Sound detects both
• Internet connection and receipt of fire notifications on smoke and carbon monoxide, warning of potential fires or
the owner's mobile devices. carbon monoxide leaks in the home.
• Integrated weather sensor and warning of approaching
severe weather conditions. • Smartphone Notifications: The device connects to the
• Compatibility with many popular smart home platforms Wi-Fi network and sends notifications to a smartphone or
such as SmartThings and Iris [3]. tablet when smoke or carbon monoxide is detected, even
if the user is away from home.
• Amazon Alexa Voice Assistant: Onelink Safe & Sound
has a built-in Alexa voice assistant, which allows you to
control other smart home devices, listen to music, get
news and information, ask questions, and much more.
Fig 2. System Architecture for Integrating the Halo Smart Smoke
Detector [4] • Built-in Speaker: The device is equipped with a high-
Kidde RemoteLync Monitor is an innovative device quality built-in speaker, which can be used to play music
designed to enhance fire safety in the home. It serves as or other audio files via Bluetooth or Wi-Fi.
an addition to existing smoke and carbon monoxide
- 350 -
• Ease of Installation and Setup: Onelink Safe & Sound II. RESULT AND DISCUSSION
is easily installed on the ceiling and connects to the Wi-Fi However, instead, the systems use their own wireless
network, making it quick and easy to set up the device and communication technology to synchronize with other
start using its functions. detectors in the home. In addition to IoT smoke and heat
• Compatibility with Other Smart Home Devices: Onelink sensors, which provide early detection of potential fires,
Safe & Sound can be integrated with other smart home there are other technologies aimed at improving fire
safety. One such direction is intelligent notification
devices to create a more extensive and effective safety and
management system[7]. systems and communication with emergency services.
These systems not only promptly warn users about the
occurrence of a fire but also automatically notify
emergency services about the incident, which can reduce
response time and save lives.
Principle of operation of intelligent notification systems
Intelligent notification systems use IoT technologies for
monitoring potential dangers, such as fires or carbon
monoxide leaks, and automatically communicating with
emergency services upon detection of such events. These
systems can also alert owners and other authorized users
about the occurrence of danger, giving them the
opportunity to take necessary measures to reduce risks [9-
10].
Examples of IoT notification systems on the market
Birdi Smart Alarm is an intelligent notification system
that combines the functions of a smoke detector, carbon
monoxide detector, and air quality sensor, demonstrated
Fig 4. First Alert Onelink Safe & Sound Smart Sensor [8] in the Figure 5. It uses IoT technologies to provide
communication with emergency services and alert owners
All of the systems listed are advanced smoke and carbon about the occurrence of danger. Birdi also provides air
monoxide detectors, but like any other products, they have quality monitoring, warning about the presence of
their drawbacks. Here are some of them: harmful substances and allergens [11-12].
• High cost: these products usually cost more than
traditional smoke and carbon monoxide detectors, which
can be a hurdle for some users, especially if multiple
devices are required for installation in the home.
• Wi-Fi dependence: They rely on internet connection to
transmit notifications to the user's smartphone and to
integrate with other smart home devices. If the Wi-Fi
network is unavailable or there is an internet connection
failure, the functionality of the devices can be limited.
• Compatibility with home automation systems: may not
support some models of smoke and carbon monoxide
detectors and integration with all home automation
systems.
• Battery replacement: battery-powered versions require
battery replacement approximately every two years. Figure 5. Intelligent notification system Birdi Smart Alarm [13]
Although this is not a significant drawback, it can be Despite many positive aspects, this device also has some
inconvenient for some users. drawbacks:
• Lack of interconnect function: they do not support the • High cost: Birdi Smart Alarm may be more expensive
traditional interconnect function, which allows the smoke than other similar solutions on the market. This can be a
and carbon monoxide detectors in the home to trigger barrier for some consumers looking for more budget-
simultaneously. friendly options.
- 351 -
• Internet connection dependence: The device requires a • Long-term contract: ADT often requires its customers to
constant internet connection via Wi-Fi for its operation. In sign long-term monitoring contracts, typically for 3-5
case of internet connection issues, the device's years. If the contract is terminated, users may face
functionality can be limited, which can lead to delays in penalties and additional fees.
receiving fire hazard notifications.
• Professional installation: ADT Smart Home requires
• Setup complexity: Some users may experience professional installation, which can increase the cost of
difficulties setting up and installing Birdi Smart Alarm, the system and cause inconvenience for users who prefer
especially if they have no experience with similar devices. self-installation.
• Lack of support for the standard interconnect function: • Limited integration with third-party devices: While
Birdi Smart Alarm does not support the standard ADT Smart Home can integrate with some third-party
interconnect function that allows all smoke and carbon devices, it is not as versatile as some other smart home
monoxide detectors in the home to trigger simultaneously. platforms and may be incompatible with some popular
This can reduce the effectiveness of fire hazard products on the market.
notifications.
• Dependence on internet connection: ADT Smart Home
• Need for battery replacement: Unlike some other requires a constant internet connection to work. In the
devices with hardwired batteries, Birdi Smart Alarm event of a lack of internet connection, the functionality of
requires battery replacement every few years, which can the system can be reduced, which in turn affects fire safety
become an additional costly and inconvenient factor for and monitoring. So, we have reviewed various IoT smoke
users. and heat sensors, as well as intelligent alert systems and
communication with emergency services.
ADT Smart Home is a comprehensive home security
system that includes smart smoke and carbon monoxide However, the possibilities of IoT technologies for
detectors, surveillance systems, and access control enhancing fire safety are not limited to these solutions.
systems, depicted in the Figure 6. Upon detection of a Now let's move on to discussing the benefits of using IoT
hazard, such as a fire, the system automatically connects in fire extinguishing systems, which can also significantly
with emergency services and notifies the owners through increase the effectiveness of fire fighting and prevent
a mobile app. ADT Smart Home also allows owners to tragic consequences. The use of IoT technologies in fire
control and manage various security devices using a extinguishing systems can bring a number of benefits,
single app [14]. such as increasing the effectiveness of fire extinguishing,
reducing the response time to fire events, reducing risks
to people and property, and managing fire systems [16-
17].
Examples of IoT fire extinguishing systems on the
market. Plumis Automist Smartscan is an innovative fire
extinguishing system that uses IoT technologies for
automatic fire detection and extinguishing activation,
visible in the Figure 7. The system is equipped with an
infrared scanner that identifies the source of the fire and
activates a network of fine sprayers to direct water mist at
Figure 6. Comprehensive security system ADT Smart Home [15]
However, like any other device, ADT Smart Home has its
drawbacks:
• High cost: ADT Smart Home may be an expensive
solution, especially for those looking for budget-friendly
options. In addition to the cost of the equipment, users are
also required to pay a monthly subscription for monitoring
services, which can quickly add up over time.
- 352 -
the fire. Automist Smartscan can be integrated with other While the use of IoT technologies in fire suppression
smart home devices and allows owners to control the provides a range of benefits, such as rapid fire detection,
system's operation via a mobile application [18]. notification, and localization, as well as integration with
other smart home devices, it's also important to consider
Figure 7. Diagram of the Automist Smartscan system [19]
aspects of regulation and standardization. These factors
1. Stop valve, approved by the Water Regulations play a key role in ensuring the safety, reliability, and
Approval Scheme (WRAS) compatibility of IoT devices for fire safety, which in turn
2. Label and cable tie facilitates the application of these technologies across
3. Single non-return valve various sectors and spheres of life.
4. Filter and bracket
The Importance of Standardization and Regulation.
5. Pressure reducing valve
Standardization and regulation of IoT technologies for
6. Stainless steel intake hose
fire safety play a crucial role in ensuring compatibility,
7. Pump unit
safety, and reliability of devices and systems. Thanks to
8. Controller
standardization, manufacturers can create products that
9. Quick-connect junction with a control point for
easily integrate with other devices and systems, ensuring
a pressure gauge
efficient and safe operation. Regulation, on the other
10. High pressure discharge hose
hand, helps to control product quality and ensure
11. Assembled head
compliance with safety requirements.
12. Wireless multisensory alarm
13. Specified wired detector Overview of Key Standards and Regulatory Bodies.
Viking QRFS (Quick Response Fire Suppression) is a Different countries have different standards and
modern fire suppression system that uses IoT regulatory bodies responsible for controlling and
technologies for quick response to a fire. The system certifying fire safety devices and systems. Some of them
includes smoke, heat, and humidity sensors, as well as include NFPA (National Fire Protection Association), UL
devices for monitoring the operation of fire suppression (Underwriters Laboratories), EN (European Norms), and
pumps and valves. Viking QRFS allows for timely fire IEC (International Electrotechnical Commission). They
detection and activation of appropriate fire suppression develop standards that set the safety, performance, and
devices, as well as remote monitoring and management of reliability requirements for IoT technologies for fire
the system via the internet [20]. safety [21].
Despite their advantages, these systems also have several CONCLUSION
drawbacks:
After considering the regulation and standardization of
• High cost: Installation can be expensive, IoT technologies for fire safety, as well as their
especially compared to traditional fire application in various sectors, it's time to draw
suppression systems. This can make it less conclusions and outline potential directions for the
accessible for some consumers. • Complexity of development of this field in the future. The concluding
installation and maintenance: Installation and part of the article presents a brief summary of the main
maintenance require a professional approach, points discussed earlier, and possible trends and prospects
which can be inconvenient for users who prefer for the development of IoT technologies for fire safety are
to self-install and maintain their fire safety discussed.
systems. • Need for component replacement:
In this article, we have looked at various aspects of the
Like other fire suppression systems, it requires
application of IoT technologies in the field of fire safety,
regular replacement of components such as
the importance and use of the Internet of Things (IoT) in
pumps, nozzles, and sensors, which can increase
fire safety. The discussion included various types of
long-term costs.
devices, such as smoke and heat sensors, intelligent
• Potential compatibility issues: In some cases, the
notification systems and communication with emergency
systems may be incompatible with already
services, as well as the integration of IoT technologies
installed fire safety systems or other smart home
into fire suppression systems. Discussing different aspects
devices. In such situations, integration and setup
and examples of IoT technologies helps better understand
may require additional efforts and costs.
their potential and opportunities for improving fire safety
• Limited effectiveness under certain conditions:
in today's world.
Although the systems are usually effective in
quickly detecting and extinguishing a fire, their First, we defined the basic concepts of IoT and its role in
effectiveness may decrease under conditions the modern world, as well as outlined the issues and
where the fire occurs in enclosed spaces or when challenges of fire safety. We then looked at examples of
the fire spreads too quickly. IoT smoke and heat sensors, such as Nest Protect and
Halo Smart Smoke Detector, which can detect smoke and
- 353 -
temperature changes for timely notification of possible compatibility between different devices and
ignition. systems.
• Education and information: Development of
Next, we studied smart notification systems and training and information programs for users and
communication with emergency services, such as Birdi professionals in the field of fire safety
Smart Alarm and ADT Smart Home. These systems contributes to a wider dissemination of
provide quick communication with emergency services knowledge about IoT technologies and their
and responsive response to fire situations, significantly benefits for ensuring safety.
reducing risks to life and property. • International cooperation: Enhancing
Integration of IoT technologies into fire suppression international cooperation in the field of IoT
systems, such as Plumis Automist Smartscan and Viking technologies for fire safety can help develop
QRFS, also has a number of advantages, including common standards and approaches, as well as
increasing efficiency, accuracy, and automation of the fire exchange experiences and best practices.
extinguishing process. • It is also worth paying attention to the
development of environmental technologies in
The importance of regulating and standardizing IoT fire extinguishing, such as the use of safer gases
technologies for fire safety cannot be underestimated, as and chemical agents for extinguishing fires. The
they ensure the safety and compatibility of devices and interaction of IoT devices with such
systems. The article discusses the main standards and technologies can provide more effective and
regulatory bodies influencing the development of IoT environmentally friendly fire extinguishing.
technologies in this sphere, we touched upon the issues of
regulation and standardization of IoT technologies for fire For the successful development and implementation of
safety and considered examples of their application in IoT technologies in fire safety, continued cooperation
various sectors. between scientific institutions, manufacturers, regulatory
bodies, and experts in the field of fire safety is required.
In the future, IoT technologies for fire safety can continue This will allow taking into account all aspects and needs
to develop and provide even more innovative solutions. of the industry, developing standards and protocols to
Possible directions of development include: ensure the reliability and safety of IoT devices and
systems.
• More accurate and reliable sensors: Improved
smoke and heat sensors can detect fires at earlier IoT technologies for fire safety play an important role in
stages, allowing timely response to fire hazards the modern world and will continue to evolve to provide
and reducing the risk of property damage and better protection from fires and save human lives. The
injuries. implementation of these technologies in various sectors
• Integration with other technologies: IoT devices and at different facilities will contribute to improving the
for fire safety can integrate with other systems, overall level of fire safety and reducing the risk of fires
such as smart lighting, access control, and video and their consequences.
surveillance, creating a unified environment of a
smart building or city. The work was carried out at the Research Institute of
• Use of artificial intelligence: AI algorithms can Mathematics and Mechanics of the Kazakh National
analyze data from IoT devices for fire safety and University named after al-Farabi.
predict potential fire hazards, allowing for REFERENCES
preventive measures and preventing the
[1] https://store.google.com/product/nest_protect_2nd_gen
occurrence of fires. [2] https://eandt.theiet.org/content/articles/2014/05/teardown-nest-
• Improved notification and communication protect/
systems: Developing more intuitive and [3] https://www.halosmartlabs.com/products/halo-smart-smoke-co-
effective notification systems can simplify the alarm
process of evacuating people in case of a fire and [4] https://products.security.gallagher.com/security/medias/HALO-
Smart-Sensor-Integration-
provide more prompt response from emergency Flyer?context=bWFzdGVyfGRvY3VtZW50c3wxMzAzOTc3fG
services. FwcGxpY2F0aW9uL3BkZnxoOGYvaGYzLzEwODk5NzA4OT
Y4OTkwL0hBTE8gU21hcnQgU2Vuc29yIEludGVncmF0aW9u
• Increasing accessibility: Reducing the cost and IEZseWVyfDk3NWU5NGVmZWMzMjg0OWExYTM0MDkx
simplifying the installation and maintenance of MmU4NjgwNzIzYjE4NzVhNDk4MjEyZmNjNWQ1ZTFkYW
IoT devices for fire safety can make them more Y5M2M5MWVhYjE
accessible for mass use in various sectors and at [5] https://www.kidde.com/home-safety/en/us/products/fire-
safety/remotelync-monitor
different facilities. [6] https://www.kidde.com/home-safety/en/us/products/home-
• Development of standards and regulation: automation/remotelync-monitor/
Continuous updating of standards and regulation [7] https://onelink.firstalert.com/safeandsound
in the field of IoT technologies for fire safety [8] https://www.firstalertstore.com/store/images/products/large_ima
ges/1039102-onelink-safe-and-sound-smoke-and-carbon-
will help ensure a high level of protection and monoxide-alarm-with-amazon-alexa-4.jpg
- 354 -
[9] Deb, S., & Chatterjee, A. (2020). IoT-based smart emergency
management system. International Journal of Engineering and
Advanced Technology (IJEAT), 9(3), 3589-3593.
[10] Mazakov T.Zh., Kisala P., Jomartova Sh. A., Ziyatbekova G. Z.,
Karymsakova N.T. (2020). Mathematical modeling forecasting of
consequences of damage breakthrough. News of the national
academy of sciences of the Republic of Kazakhstan. Series of
Geology and Technical Sciences, 5(443), 116-124.
[11] https://getbirdi.com/
[12] Mazakov T.Zh., Jomartova S., Mukhaev D., Mazakova A.,
Tolegen G. (2021). Intelligent System for Assessing the Socio-
economic Situation in the Region. Wojtkiewicz K., Treur J.,
Pimenidis E., Maleszka M. (eds) Advances in Computational
Collective Intelligence. ICCCI. Communications in Computer
and Information Science. Springer, Cham, 1463, 437-447.
doi.org/10.1007 /978-3-030-88113-9_35
[13] https://www.electronichouse.com/smart-home/birdi-builds-
better-smoke-alarm-home-automation/
[14] https://www.adt.com/smart-home
[15] https://www.adt.co.uk/themes/ee/site/default/asset/img/common/
smart-home-slider-new.jpg
[16] Stojmenovic, I., & Wen, S. (2014). The Fog Computing
Paradigm: Scenarios and Security Issues. Federated Conference
on Computer Science and Information Systems, 1-8.
[17] T.Zh. Mazakov, Sh.A. Jomartova, G.Z. Ziyatbekova, T.S.
Shormanov, B.S. Amirkhanov, P. Kisala. The image processing
algorithms for biometric identification by fingerprints // News of
the National Academy of Sciences of the Republic of Kazakhstan.
Series of Geology and Technical Sciences, 2020. – Vol. 1, - No
439. – P. 14-22. https://doi.org/10.32014/2020.2518-170X.2
[18] https://www.plumis.co.uk/automist-smartscan
[19] https://plumis.co.uk/sites/default/files/2023-
02/Smartscan%20Hydra%20Handbook%20v3.1.pdf
[20] https://www.vikinggroupinc.com/products/viking-qr
[21] https://www.nfpa.org/Codes-and-Standards
- 355 -
Implementation of Real-time Water Quality
Monitoring for Energy Consumption
Optimization Utilizing IoT Sensors in a Smart
Fish Farm
Hunseo Lee 1, Eunsu Jung 1, Juhyeon Youk 1, June-young Park 2, Hyun-il Kang 1 and Young Park 1 *
1
Hanbat National University, Republic of Korea
2
CNC Aid Co., Ltd., Republic of Korea
Email: [email protected]
Abstract—This research paper presents the implementation consumption, ensuring that energy is utilized efficiently and
of real-time water quality monitoring in a smart fish farm, only when necessary.
focusing on energy consumption optimization through the
utilization of IoT sensors. The study aims to enhance the
efficiency and sustainability of fish farming practices by
continuously monitoring water quality parameters and
integrating renewable energy sources. The IoT-based system
provides real-time data on key water quality parameters such as
temperature, dissolved oxygen, pH, and salinity, enabling
prompt detection of any deviations from optimal conditions. By
automating the operation of water wheels based on real-time
Fig. 1. floating solar panels and ground-mounted solar panels
data, the system optimizes energy consumption while installed in a smart fish farm.
maintaining optimal water quality. The findings demonstrate
the effectiveness of the implemented system in reducing energy II. EXPLEMENTAL METHODS
consumption and ensuring optimal water quality for sustainable
fish farming. The experiment was conducted in a saltwater
aquaculture facility designed for the cultivation of whiteleg
Keywords—component, formatting, style, styling, insert (key shrimp (Litopenaeus vannamei) in Gomso, Korea. The
words) facility consisted of four saltwater ponds, each with
dimensions of 19834.71m². This paper presents the
I. INTRODUCTION installation of solar panels, processed with anodic aluminum,
In recent years, the integration of innovative with a width, height, and depth of 2,163 mm, 1,030 mm, and
technologies has revolutionized various industries, 35 mm, respectively, in a fish farm for monitoring
including aquaculture. The need for sustainable practices in environmental factors in a smart fish farm. The installation
fish farming has led to the development of smart fish farms capacity is a total of 45 kW, and the float, structure, module,
that utilize IoT sensors and renewable energy sources. and inverter are installed in order. After installation, the
Efficient management of water quality is crucial for the energy consumption in the demonstration complex is
success of fish farming operations, as it directly impacts the configured using a EMS, and wind power and ground-based
health and productivity of aquatic organisms. Real-time solar panels are installed to monitor them, as shown in Fig.
monitoring of water quality parameters enables proactive 2.
management and timely response to potential issues, The environmental monitoring equipment was analyzed
contributing to the overall sustainability of the fish farm.
This research paper aims to address the challenges of energy to evaluate the water quality safety of the operation of float
consumption optimization and real-time water quality ing solar panels installed in the fish farm. A variety of fact
monitoring in a smart fish farm through the implementation ors, such as the material characteristics of solar panels, ext
of IoT sensors. The IoT-based system allows for continuous
monitoring of key water quality parameters, including ernal temperature, humidity, harmful environment, and cor
temperature, dissolved oxygen (DO), pH, and salinity. By rosion due to salt, need to be analyzed to detect harmful fa
collecting and analyzing real-time data, the system provides ctors for water quality, and it needs to be periodically moni
insights into the current water quality conditions, enabling
prompt detection of any deviations from optimal ranges. tored to create a suitable growth environment for white shri
Furthermore, the integration of renewable energy sources, mp. We plan to report the results of the installation of float
such as solar and wind power, enhances the sustainability of ing solar panels in the fish farm through the analysis of the
the fish farm by reducing reliance on traditional energy
IoT-based monitoring system in the future.
sources. The system incorporates water wheels that are
powered by renewable energy, providing a more
environmentally friendly solution for water circulation and
aeration. By automating the operation of water wheels based
on real-time water quality data, the system optimizes energy
- 356 -
Fig. 2. The configuration diagram of the EMS applied to the smart fish
farm.
III. RESULTS AND FUTURE WORKS
This paper presents the installation of 45 kW of onshore
solar and floating solar panels for the construction of a ren
ewable energy generation facility in a smart fish farm. The
environmental monitoring equipment was analyzed based
on the installation of floating solar panels. Based on this, th
e results of the installation of floating solar panels in the fis
h farm will be reported through IoT-based remote monitori
ng, by utilizing the renewable energy generation facility an
d fish farming facility.
ACKNOWLEDGMENT
This work was supported by the Korea Institute of
Energy Technology Evaluationand Planning(KETEP) and
the Ministryof Trade, Industry& Energy(MOTIE) of the
Republic of Korea(No. 20213030160320)
REFERENCES
[1] Ibtissame Ezzahoui, Rachida Ait Abdelouahid, Khaoula Taji, and
Abdelaziz Marzak, “Hydroponic and Aquaponic Farming:
Comparative Study Based on Internet of things IoT
technologies.,"Procedia Computer Science, vol. 191, pp. 499-504,
2021.
[2] J A. Reyes Yanes, P. Martinez, and R. Ahmad, "Towards automated
aquaponics: A review on monitoring, IoT, and smart systems,"
Journal of Cleaner Production, vol. 263,2020.
[3] Malka N. Halgamuge, Alexe Bojovschi, Peter M.J. Fisher, Tu C. Le,
Samuel Adeloju and Susan Murphy, "Internet of Things and
autonomous control for vertical cultivation walls towards smart food
growing: A review,"Urban Forestry & Urban Greening, vol. 61,
2021.
[4] Boby Siswanto, Yasi Dani, Doni Morika and Bubun Mardiyana, "A
simple dataset of water quality on aquaponic fish ponds based on an
internet of things measurement device," Data in Brief, vol. 48, 2023.
- 357 -
Implementation of IoT-based smart elevator
remote regeneration control monitoring system
platform
Hyeong Ju Kwon, Hag-wone Kim and Sang-min Lee*
Korea National University of Transportation, Republic of Korea
Email: [email protected]
Abstract— This paper is about a ultra-high-speed, ultra-low-
latency hyper-connected technology has recently developed, various
ICT convergence technologies that reduce congestion, maintain safe
user movement in elevators and maintain physical distance are being
applied in the digital transition leading to the super-gap between
objects and technologies. In particular, in terms of remote ICT and
maintenance management, various methods are being attempted to
minimize elevator emergency field response, which is closely
related to public safety, and to cope with realistic maintenance
situations. Based on NB-IoT, LoRa, SigFox, etc., which enable low-
power long-distance data communication, various sensor
information is collected, stored, and used. GIS and remote
management system are applied to enable remote response in Fig.1. Smart elevator remote regeneration control structure.
elevator maintenance and emergency situations. We propose a smart
elevator regeneration control monitoring and control system based
on NB-IoT that can monitor and determine the situation by checking In addition, through minimization of overhead through
the situation as a whole and checking various information in real wireless interface, handover limitation, implementation of
time. appropriate security level, etc., it has achieved lower power
consumption and lower cost compared to Cat-M technology
Keywords— IoT,,regeneration control,,smart elevator,wireless [4]. To implement an elevator status information real-time
communication,energy collection management platform that can collect data on errors
that occur during elevator operation or remote regeneration
I. INTRODUCTION
control, provide a history for the collected errors, take
Recently, high-speed and hyper-connected technologies measures for the errors that have occurred, and then input
are rapidly emerging as new keywords in digital actions. In order to do this, it is necessary to propose a method
transformation, and various ICT convergence platforms that for using Controller Area Network communication that is
reduce congestion by maintaining safe user movement and vulnerable to RS232/RS485 communication speed and signal
physical distance are being implemented. In particular, in delay and noise, so event processing is possible, and
terms of ICT smart remote control maintenance, various communication speed is lowered while maintaining the same
methods are being attempted for smart maintenance according responsiveness and is more robust against noise.
to digital conversion and minimization of elevator emergency
site response, which is closely related to public safety. Based
on NB-IoT, LoRa, SigFox, etc., which enable low-power
long-distance data communication, we are going to propose
an IoT-based smart elevator regeneration control monitoring
system platform implementation to collect, store and analyze
various sensor information.
Fig. 1.
II. SYSTEM DESIGN
IoT designed a new channel structure to overcome the
problems of the existing Cat-M, and used 200 kHz, which is
Fig.2. Smart Elevator software system Interface.
less than 1.4 MHz of the existing Cat-M. Since the bandwidth
of 200 kHz used by NB-IoT is determined considering the The data is collected on errors that occur during elevator
existing GSM bandwidth, NB-IoT technology can operate in operation or remote regeneration control, and for the collected
stand-alone mode in the GSM frequency band. In addition to errors, a system protocol that connects the drive motor system
this, NB-IoT can be operated in both in-band mode and guard and mechanical system, such as control of each elevator
band mode by being allocated a part of the frequency domain composed of history and PWM converter and DC link voltage
of the LTE band. control, is established. Analyze Afterwards, the gateway
collects event information from the elevator control panel and
- 358 -
converts it into CoAP, MQTT, and HTTP message formats, III. SUMMARY
and the server stores and processes the collected information In this study, NB-IoT-based smart elevator remote
for remote monitoring service. regeneration control monitoring that enables low-power long-
distance data communication and systematic analysis by
collecting various sensor information in order to cope with
elevator emergency field response and realistic maintenance
situations closely related to life safety A method for
implementing a control system platform was proposed, and it
is expected that active research will be conducted on the
implementation of predictive maintenance management in
conjunction with an artificial intelligence system when
configuring a network in the future.
REFERENCES
[1] Lee Sang-min, “ICT Convergence and Connected Smart Elevator
Trend Opening the Future,” Weekly Technology Trend, 2022.
[2] Kim, B., Song, K., & Kim, S., “A Review of Data Planning
Fig.3. Smart elevator remote regeneration control monitoring system Methodologies”, ACM Transactions on Database System, Vol. 2, no.
structure 3, March 17, 1995, pp. 23-40.
[3] Korea Broadcasting and Radio Agency, “Research on a super-aging
The data on information is collected at each selected period, society support plan using narrowband IoT technology
and the data collected in the current period is compared with Gu”, 2018.
the data collected in the previous period, and only data with [4] Park Ok-sun, Hwang Hyun-yong, Lee Chang-hee, Shin Jae-seung, “5G
changed values is identified. When a data request is made Massive loT Technology and Standardization Trend,” Korea
from the browser, the data is retrieved from the web server Electronics and Telecommunications Research Institute, 2016.
into the database, and then the data is passed to the browser. [5] Seo Seok, Shin Eun-jung, Jo Kwon-do, “Narrowband Internet of
The NB-IoT controller installed in the elevator continuously Things Technology Trend”, Korea Electronics and Telecommunicati
accumulates data into the database and updates it with real- ons Research Institute, 2016
time data. The web browser is also designed to send real-time
data to the screen by continuously importing data.
- 359 -
An ontological model for automating the
process of preparing electronic courses in the
field of information and communication
technologies.
N.ZH. SABITOVA1, B.Sh. RAZAKHOVA2, S.O. GNATYUK2.
1
Doctoral student Eurasian International University named after. L. N. Gumilev,
Kazakhstan, Astana.
2
t.s.d, associate pofessor L. N. Gumilyov Eurasian National University,
Kazakhstan, Astana.
3
Foreign consultant, t.s.d., professor National Aviational University,
Ukraina, Kiev.
E-почта: [email protected], [email protected], [email protected]
Annotation. Intelligent information technologies disciplines related to the field of information and
(IT) influence the processes in the economic, scientific, communication technologies (ICT). Today, visual-
technical, humanitarian, educational and other fields of figurative components of thinking occupy a
activity. Informatization or digitalization of the world society predominant place in the worldview of modern
is focused on building an IT -oriented society. Knowledge
engineering technologies are becoming significant.
schoolchildren who have many gadgets, starting with
Ontological engineering can also be considered as a special smartphones and ending with high-performance PCs or
case of knowledge engineering. laptops. Which, accordingly, encourages teachers to
Keywords: intelligent technologies, subject area, present educational material in a visual expressive
ontology. form, involving a diverse color palette, animation
elements, dynamic illustrations, so characteristic of
Modern IT uses Knowledge Management. We many modern EC and/or EC.
can say that success in this direction is determined by The study of the ICT course should introduce
the level of efficiency of computer systems [1-5]. As students to the methods of the most effective use of
shown in [3], scientific research is being intensified on informatization tools in everyday practice. The
the creation of information systems based on examples given in the process of teaching students
knowledge bases, the development of methods for the should clearly demonstrate that ICT tools are able to
ontological analysis of subject areas to extract aspects increase efficiency in the study of most disciplines that
of the application of ontologies from them. The latter is are taught within the framework of secondary schools
of interest, for example, in the context of the and further colleges or universities.
development of electronic training courses (EC) and Considering the above, we can state that the
electronic textbooks (EC). relevance of new research in this direction is justified
At the current stage of IT development, it has by the need to develop effective methods of
become possible to bring the existing theoretical architectural and structural organization and synthesis
foundations of artificial intelligence systems closer to of relevant content for EC and EC in the field of ICT,
real implementation, while taking into account the especially for the contingent of secondary school
latest achievements of general engineering practice and students. Such methods, in our opinion, enable all
subsequent promising trends in the development of interested persons, starting with teachers in schools and
computer technology in the preparation of relevant colleges and ending with commercial structures
content of electronic courses in the field of ICT. offering their EC and/or EC on the educational services
It should be noted that computerization of the market, not only to design relevant EC and/or EC with
educational process already at the level of secondary the processing of subject knowledge, but also to
school or even primary classes, according to many digitalize the process of their synthesis based on
authors [1-3], is a large-scale innovation in the field of processing colossal volumes of textual information of
education. It is impossible to imagine teaching in a subject disciplines inherent in the content of the ICT
modern school if a teacher does not use a variety of field.
computer technologies and the Internet in his As a result of the research, the following main
educational work. This is especially true for disciplines results were obtained:
where content is updated rapidly, in particular,
- 360 -
a methodology for designing electronic
courses (EC) and/or electronic textbooks (EC) in the
field of ICT based on the application of an ontological
model is proposed;
It is shown that the ontological model
proposed in the article is intended primarily for the
implementation of information technology for
automated processing of ontologies of ICT subject
areas, which will simplify procedures related to the
implementation of EC and/or EC, based on relevant
content in the field of ICS, corresponding to the current
state of this subject area with the possibility of updating
this content as necessary.
References.
1. How E-learning Works by Lee Ann
Obringer. – Available at
http://communication.howstuffworks.com/elearning.ht
m.
2. E-learning 2.0 by Stephen Downes, National
Research Council of Canada October 17, 2005. –
Available at
http://www.elearnmag.org/subpage.cfm?article=29-
1§ion=articles
3. Tikhonov, U., Lakhno, V., Skliarenko, E.,
Stepanenko, O., Dvirnyi, K. Development of on
Tological Approach in E-learning When Studying
Information Technologies (2016) Eastern-European
Journal of Enterprise Technologies, 5 (2), pp. 13-20.
4. Sampson, D. G., Lytras, M. D., Wagner, G.,
& Diaz, P. (2004). Ontologies and the Semantic Web
for E-learning. Educational Technology & Society, 7
(4), 26-28. – Available at
http://www.miltiadislytras.net/8.pdf.
5. Ontologies: Structure E-Learning from
Information-based to Knowledge-based. Conference
Information. – Available at
http://www.editlib.org/p/15070.
- 361 -
A Study of Deep Learning Model for Color
Measuring Process in Industrial Facilities
Kyung Suk Kim 1, Sanghun Lee 2, Ki-Sub Kim 3*, and Jeong Won Kang 1,4*
4
Department of SMART Railway System, Korea National University of Transportation, Republic of Korea
2
Central Research Institute, KCC, Republic of Korea
3
School of Chemical and Energy Engineering, Korea National University of Transportation, Republic of Korea
4
Department of Transportation System Engineering, Korea National University of Transportation, Republic of Korea
Email: [email protected] (KSK), [email protected] (JWK)
Abstract— The paint manufacturing takes a complex
chemical process to turn a colorful liquid material into a solid.
Paint is used for beautification, protection, sanitation, and
identification. There are four main components in paint: resin,
additives, solvent, and pigment. The resin is the binder that
holds pigments together. Additives are used to enhance the
properties of the substance. In this work, we present a deep
learning color recognition model for color measuring process in
paint industry. Currently, the industry mainly uses
spectrophotometers for color measuring method because of
their accuracy. This method is measured by injecting the sample
manually into the spectrophotometer. Our proposed method is
using webcam with a deep learning model in stand of the
spectrophotometer. Deep learning models can be widely used for
the image and color detection. And in this study, the “You Only
Look Once (YOLO)” algorithm was applied to this study for
real-time detection of color samples. After training various
sample images through this algorithm, the model can detect the
sample area in real time through a webcam (Logitech BRIO 4K).
An Open Source Computer Vision library (OpenCV) was used
for the color recognition model and the detected RGB color
value was converted to the value of the International
Commission on Illumination color space (CIELAB), which are
mainly used in color measuring process. However, the color
value through the camera was difficult to implement accurate
thoughts due to the mirror-like reflection of light from the
surface (specular reflection). To address this problem, we were
comparing several specular removal methods and proposed the
most suitable model for the color recognition model of color
samples. In addition, the accuracy of the proposed model was
verified by comparing the colors of the various samples. Our
approach can easily detect samples and color values, which can
contribute a lot to automatically calculating the exact amount of
coloring for the target color.
Keywords—Deep Learning, Object detection, Color recognition,
Open Source Computer vision, You Only Look Once algorithm
ACKNOWLEDGMENT
This work was also supported by the Korea Agency for
Infrastructure Technology Advancement (KAIA) grant
funded by the Ministry of Land, Infrastructure and Transport
(Grant No. 22QPWO-C158106-04).
- 362 -
Research and development of an automated installation based on modern
technology to determine the degree of ambient air pollution
Madina Yessimkhanovna Mansurova, Ph.D., Associate Professor, Head of the
Department of Artificial Intelligence and BD, Al-Farabi Kazakh National University,
Almaty, Kazakhstan, [email protected]
Samal Beksultanovna Abdreshova, PhD, Acting Associate Professor, Al-
Farabi Kazakh National University, Almaty, Kazakhstan, [email protected]
Sandugash Kudaibergenovna Orazalieva, PhD, Associate Professor, Head of
the Department of Electronics and Robotics, Almaty University of Energy and
Communications named after G. Daukeev, Almaty, Kazakhstan, orazalieva-
[email protected]
Gulmira Nurlankyzy Abdreshova, PhD student, Al-Farabi Kazakh National
University, Almaty, Kazakhstan, [email protected]
Keyword: Ozon, microcontrollers, temperature sensors, humidity sensors, gas
sensors.
In commercial and industrial spaces where people work, there are strict
requirements for air quality. It is important to control harmful substances released
during technological processes. Ozone poses a high danger to the human body-the
alarm clock helps to control the amount of gas. Specially designed O3 sensors reliably
protect against harmful evaporation.
For the purpose of monitoring air quality in various premises of commercial and
industrial use, as well as for signaling about the excess of gas content above the
established limit parameters, automated air pollution devices are being developed.
Unique physical properties make it possible to use ozone for disinfection of
premises, in gas-producing, chemical, food enterprises, utilities.
The ozone detector is a wide functional equipment. The device not only monitors
the amount of harmful impurities, but also analyzes their composition.
The modern market offers signalmen for detecting hazardous gases in two
variations-stationary and portable models. A stationary ozone sensor in the Air provides
constant monitoring of the content of harmful substances. In rooms where periodic
inspection is required, portable devices are often used.
For ease of use, modern sensors are equipped with digital displays, a light and
sound indicator system, a durable case with moisture protection and interfaces for
connecting to electronic control and control systems.
The device we offer offers a cost-effective air quality control solution.
Environmental organizations have identified several main pollutants: ozone, airborne
particles, carbon monoxide, sulfur dioxide, and nitrogen oxide. The device can detect
all pollutants except sulfur dioxide. In addition, the device has a household gas
- 363 -
determinant that allows you to warn about a gas leak or the presence of combustible
gases. Also included is the temperature and humidity sensor.
This device consists of the following automation devices: Arduino UNO series
microcontrollers, temperature and humidity sensors, as well as gas sensors of various
series.
The wiring diagram is a general scheme that shows the operation of the
automated installation of harmful gases. If necessary (for any reason), most of the
digital and analog ports to which the sensors are connected can be changed; to do this,
you need to make changes to the provided program code.
To collect data on the concentration of particles in the air, we used dust sensors.
Each sensor uses an infrared LED and a photodiode to measure the concentration
of particles scattered in the air. The internal circuit converts the output signal of the
photodiode into digital signals. Usually there is a +5V signal at the sensor output, and
the sensor sends a low-voltage pulse when detecting particles. The period of time during
which the signal is low at the output, or" low pulse occupation percentile " (the
percentage of time during which the voltage level is low at the photodiode output) is
proportional to the concentration of particles in the air.
There is a wiring diagram for gas sensors and a stove plate of a humidity sensor
with temperature.
We used Surface sensors that can detect ozone and nitrogen dioxide in the air.
Sensors use an internal resistor in the sensor element.. The resistor resistance
should be between 10-20 kΩ.
A temperature and humidity sensor should be used, as temperature and humidity
control plays an important role in determining the concentration of gases. High
humidity and temperature have a significant impact on the measurement accuracy.
Therefore, it is very important to monitor these changing parameters. Temperature and
humidity can be controlled with one sensor at a time. According to the figure above,
the left pin is connected to the power, the middle pin is connected to the signal output,
and the right pin is connected to the ground. The signal from this sensor goes to the
Arduino digital port. In our code, it is assumed that the temperature signal will go to
the digital port 2. If necessary, it can be converted to another digital port; depending on
the selected port, appropriate adjustments must be made to the program code.
A code for obtaining source data from the device is fixed. This code prints sensor
resistance values, operating percentage of low pulse signals, and temperature and
humidity readings through a serial monitor on the computer. Also, the source data can
be viewed on the LCD display.
For the code to work properly, you must first download the libraries for the LCD
and temperature and humidity sensors.
- 364 -
Data mining of leading indicators
Author one: Laura Alimbayeva1, Author two: Kuralay Belgibayeva2
1
Author One’s Affiliation: Al-Farabi Kazakh National University, Kazakhstan
2
Author Two’s Affiliation: Kazakh University of Technology and Business, Kazakhstan
Email: [email protected]
Abstract — Modern time is characterized by the The impulse response functions pointed to the positive
accumulation of a large amount of data. To extract the impact of business activity sub-indices. Decompositions of
knowledge from these data necessary for making managerial variances revealed the timing of future changes in the turning
decisions, automatic data analysis, called Data mining, will points of business cycle phases in the short run.
provide indispensable assistance. Research in the field of
application of Data Mining technologies is the most in demand, Keywords— Data mining, leading indicators, business activity,
has a huge potential in any field of activity and a high level of model, predictive property.
practical significance.
INTRODUCTION
Such valuable and “raw, hidden” data include leading
Leading indicators first appeared in the United States
indicators. They characterize the state and dynamics of business
activity in the country's economy, as well as in the context of
around the beginning of the 20th century due to more frequent
sectors and branches of the economy. These indicators serve as crises. Then there was a need to indicate, assess the state and
an important guide: 1) for central banks when making decisions dynamics of business activity at the macro level within the
in the course of conducting monetary policy; 2) for participants economic cycle.
in the global and local markets when receiving operational Leading indicators include business activity sub-indices
information about the state of the country's economy; 3) for (BAS) and a composite leading indicator (CLI). Their main
interested international and national investors.
property is predictive, early detection signals, ahead of
The value of leading indicators lies in their main property - expectations about future and prospective changes in the
predictive, ahead of expectations about future and prospective business cycle before these changes occur in reality.
changes in the business cycle before these changes occur in The differences between BAS and CLI are as follows:
reality. BAS is calculated for four sectors (manufacturing, services,
construction and mining industry) and a value of more than 50
The novelty of this study is both the application of Data
points indicates an increase in sector activity compared to the
Mining technologies to leading indicators of the business cycle,
previous month, and vice versa, a value of less than 50 points
and the study of leading indicators - sub-indices of business
activity and a composite leading indicator in Kazakhstan.
indicates a slowdown in the economic activity. Whereas CLI
estimates are made for the real sector of the economy on a
In order to extract from the official information a “hidden” quarterly basis and the degree of activation is judged by the
but objective causal relationship between variables, the authors value: more or less than 100 points, respectively.
substantiated and applied for the first time vector error The purpose of this study is to identify and evaluate the
correction models (Vector Error Correction Model, VECM) to
relationship between leading indicators, to provide a forecast
the leading indicators of Kazakhstan. Sequential data checks for
of trends in the short term.
seasonality, outliers, stationarity, and cointegration were
applied. The scientific hypotheses of the present study are as
follows:
The purpose of this study is to identify and quantify the
hidden (previously not studied) relationships between the time 1. Hypothesis one: there is a cointegration relationship
series of business activity sub-indices of the four sectors and the between leading indicators.
composite leading indicator of the real sector of the economy, to
provide a forecast of their trends in the short term.
2. Hypothesis two: the predictive, prognostic property of
leading indicators of the business cycle is manifested in the
The objects of study are business activity sub-indices of four Kazakh economy.
sectors and a composite leading indicator of the real sector of The objects of study are business activity sub-indices
the economy. Study period: September 2016 to April 2023. (BAS) of four sectors of the economy and a composite leading
Number of observations is 80.
indicator (CLI) of the real sector of the economy. Study period:
Amazing results have been obtained. The developed
September 2016 to April 2023. The number of observations is
empirical VECM models, the calculated functions of impulse 80.
responses and decomposition of variances confirmed the Methods and materials: The following methods of
predictive property of business activity sub-indices for the cognition were used in the study: systematization and
Kazakhstan economy. In particular, the manufacturing generalization, dynamics series indicators, error correction
business activity sub-index predicts a change in the construction vector model (VECM), impulse responses, dispersion
business activity sub-index, and the construction business
decomposition.
activity sub-index predicts a change in the mining business
activity sub-index. The quality of the VECM models satisfies The information base was the official statistics of the
statistical tests, which characterizes the accuracy and reliability National Bank of Kazakhstan. The information is processed
of the results. and presented in the statistical package EViews.
- 365 -
For comparability of BAS and CLI indicators over time analysis of the data showed, in our study, not all time series
intervals, we recalculated data on CLI from quarterly to turned out to be stationary and, moreover, they do not move
monthly in EViews. too far from each other in the processes of growth and decline
in economic activity. To avoid false regression, we tried to
1. Literature review apply empirical VECM models for leading indicators, based
Leading indicators first appeared in the United States on the recommendations and experience of other scientists.
around the 1920s and 1930s. In connection with the increasing 2. Analysis of leading indicators
crises, it became necessary to indicate business activity at the
macro level in order to show, assess the state and dynamics Data visualization clearly demonstrates cyclical
within the economic cycle. The role, meaning, economic fluctuations for the period under review (Fig. 1).
interpretation of leading indicators are described in the works Fig. 1. Dynamics of BAS in the sectors of manufacturing, services,
of scientists: Carol M.Kopp. (2022), Nathan Reiff. (2023) construction and mining industry, as well as CLI
[1,2].
60 102
Many macroeconomic indicators are characterized by 55
non-stationarity, in which the values (numbers) differ greatly 101
50
from the average level and move away from it for an unknown 45 100
time. To stop non-stationary series, the procedure of taking the 40
difference as the difference of each subsequent and each 35
99
previous series members is applied. A linear combination of
30 98
several non-stationary integrated series is called series
January
January
January
January
January
January
January
September
May
September
May
September
May
September
May
September
May
September
May
September
cointegration.
Cointegration and error correction models were first
proposed by scientists Engle, R. F., Granger, C. W. J. (1987) 2016 2017 2018 2019 2020 2021 2022 2023
based on an empirical study of the non-stationarity property of
several economic time series, combined together and BAS in production BAS in services
BAS in construction BAS in mining industry
repeating movements one after another [3]. According to their CLI (right scale)
beliefs, the essence of the study lies in the fact that the
parameters of the cointegration vector are first estimated, Source: authors' calculations
which are then used in the VECM error correction model. This Note: CLI data for April 2023 is based on business expectations
method establishes a causal relationship between the studied
variables.
Depending on the amplitude of fluctuations, there is a
Following the ideas of Engle, R. F., Granger, C. W. J., change in phase shifts of economic cycles, successively
other scientists used the VECM model for cointegrated time alternating one after another with a new repetition. Economic
series: Hizir Sofyan, Eva Maulia, and Miftahuddin (2017, cycles of recovery (values above 50 or 100) are replaced over
2018) [4,5]; Bruce E. Hansen (2018) [6]; D. John Cochrane, time by recessive ones, which after a certain period of time are
John Cochrane (2006) [7,8]; Nosko V.P. (2021) [9]; Rudra P. replaced by depression ones (values less than 50 or 100).
Pradhan, Girijasankar Mallik, Tapan P. Bagchi (2018) [10];
Rudra P. Pradhana, Neville R. Normanb, Yuosre Badirc, Bele Since the beginning of the pandemic period in April 2020,
Samadhana (2013) [11]; Belinga Thierrya, Zhou Junb, Kazakhstan has experienced negative deteriorations in the
Doumbe Doumbe Ericc, Gahe Zimy Samuel Yannickd, Koffi BAS in four sectors of the economy. Thus, in the services and
Yao Stéphane Landry (2016) [12]; Nermin Yasar (2021) [13]; construction sectors, a decline was noted due to the
Chao Wang, Yul-Seong Kim, Chi Yeol Kim (2021) [14]; deterioration of all components (delivery times, employment
Farhad Taghizadeh-Hesary, Naoyuki Yoshino, Lisa Fukuda, levels, new orders and business activity). In manufacturing
Ehsan Rasoulinezhad (2021) [15]; Chaido Dritsaki (2013) [16] and mining industry, the deterioration was due to lower
and others. The followers applied the VECM model to other employment, new products and new orders.
objects of study, to other variables. It should be noted that only the manufacturing sector is
Thus, Hizir Sofyan, Eva Maulia, and Miftahuddin used dominated by the number of positive variables with BAS
VECM to study the relationship between tax revenues and values over 50, which determine the intensification of
inflation in the large administrative center of Indonesia - business activity: 75.0% (60 out of 80).
Banda Aceh for January 2013 to December 2015 [4]. In the service sector the trends are approximately the same
They explain the choice of this model of econometric as in the manufacturing sector, with some shifts in the months,
analysis by the fact that the stochastic trend in time series replacing the upward and downward economic phases. The
indicates the presence of long-term and short-term trends. The day of fall is evidenced by the minimum value of the BAS
VECM model is applicable to cointegrated rather than (34.9 in April 2020) in the service sector due to the
stationary data and is able to determine the short-term deterioration of all its components: delivery times,
behavior of a variable in the long run “due to a permanent employment levels, new orders. Having dropped to a
shock”. And further they state: “Cointegration is a long-term minimum, the BAS turns up, predicting some recovery in
relationship between variables that, although individually not business activity.
stationary, but a linear combination between these variables In the construction sector, the peak of growth characterizes
can be stationary” [3; 4, p.2]. the maximum value of the BAS (57.9) in September 2019,
In foreign and domestic literature, VECM models for associated with an increase in new orders and an increase in
leading indicators have not been previously used. As the employment. Having reached a maximum of growth, the BAS
- 366 -
turns down, heralding a reversal of the business cycle to the (Kurtosis) for all these series are not equal to 3. For the CLI
downside with some fluctuations in individual months. series, the kurtosis coefficient of the flat-topped distribution is
less than 3. For all BAS series, the kurtosis coefficients for the
Moderate exposure to cyclical changes is set for the sectors: peaked distribution exceed 3. The Jarque-Bera test confirms
construction and mining industry. The prevailing number of that all the considered series do not have a normal distribution,
BAS with low, negative values of business activity, less than because p-values do not exceed a significance level of 0.05.
50, affected: in mining industry 66.3% (53 out of 80), in These facts indicate that all series contain outliers (Table 6).
construction 57.5% (46 out of 80). At the same time, in mining
industry, the last economic phase of the decline began earlier For the analyzed period there are four outliers in each
than in other sectors: from June 2019 to April 2023. The main series (Table 7).
reasons for the decline are a decrease in the production of
crude oil, coal, natural gas, two lockdowns in March and May Stationarity check. When analyzing time series, it is very
2020, repair work at the Kashagan field, depreciation of the important to determine whether they are stationary. Since the
national currency against the US dollar and the Russian ruble, stationarity of a series can greatly affect its behavior and
price volatility in world commodity markets, aggravation of properties.
the geopolitical tension. Thus, the series of CLI and BAS in the services are non-
stationary, and BAS in manufacturing, construction and
In construction a decline in business activity was observed
from March 2020 to January 2023, caused by the onset of the mining industry are stationary (Table 8).
pandemic period, the termination of state programs for Checking for cointegration. Cointegration occurs when
preferential mortgage lending, and an increase in the threshold time series can be related by some long-term relationship,
for the adequacy of using part of pension savings to purchase from which these time series can deviate in the short term, but
housing at “Otbasy Bank” JSC. to which they must return in the long term.
3. Empirical models The concept of cointegration is applicable only to series
Before starting to build empirical models, it is important with a unit root. The idea of cointegration is that two or more
to determine the structure of the time series data: are there variables can change synchronously so that some linear
outliers, trends, seasonal components, etc. combination of them is a stationary process [17].
Seasonality check. Seasonality is a characteristic of a time In the study of time series, it was found that they are
series in which data undergo regular and predictable changes cointegrated. Therefore, further analysis is based on the vector
that repeat every calendar year. Any predictable fluctuations error correction model (Vector Error Correction Model,
or patterns that repeat or recur within the same year are called VECM).
seasonal. When building a VECM model, it is necessary that all
Seasonal effects differ from cyclical effects because variables be stationary at the level of first differences. Taking
seasonal cycles occur within one calendar year, while cyclical into account that the series of CLI and BAS in services are
effects can cover periods shorter than or longer than one non-stationary, we check them for stationarity at the level of
calendar year. first differences. The results of the Augmented Dickey-Fuller
test showed that the series of CLI and BAS in services are
Many data are affected by the time of year, and seasonal stationary at the level of first differences (Table 9), so VECM
adjustment means that more accurate relative comparisons can models can be built.
be made between different time periods. Seasonally adjusted
data smooths out periodic fluctuations in statistics or The choice of VECM model parameters was based on the
movements in supply and demand associated with the change best options for statistical tests, as well as the most optimal
of seasons. options for lags. At the same time, the initial data were cleared
of seasonality and outliers. The result is the following
All the series we studied have seasonality, which is proved dependency models:
by significant p-values at a significance level of 0.05 (Tables
1, 2, 3, 4 and 5). 1) BAS in manufacturing from BAS in services;
Outlier check. An outlier is a value or certain data that is 2) BAS in construction from BAS in production;
very different from the main dataset. 3) BAS in mining industry from BAS in construction.
In the case of outliers, a twofold choice arises: to keep or 3.1 Model of dependence of BAS in manufacturing on
exclude them from the entire data set. To do this, it is BAS in services
necessary to analyze the data with and without outliers, and as
a result, compare the results obtained. Let's use the Granger test to test the causal relationship
between the considered series. First you need to choose the
On the one hand, for a data set with outliers, the use of optimal lag.
classical methods of statistical analysis leads to incorrect
results. On the other hand, if the data set is not sufficient, then The optimal amount of lag is determined using the
eliminating outliers can have a significant impact on the result. smallest value of the criterion: the smaller the criterion, the
fewer errors and the higher the quality. Therefore, we choose
Therefore, it is important for us to evaluate and provide lag 2 (Table 10).
justification for keeping or removing outliers from the dataset.
The result of the Granger test showed that, at a significance
All the considered series do not have a normal distribution. level of 0.05, a series of BAS in services is the cause of a series
Skewness coefficients are negative, hence the distributions of of BAS in production according to Granger or in other words:
these series have long left tails. The kurtosis coefficients
- 367 -
BAS in services predicts a change in BAS in production There is no heteroscedasticity of residuals in the VECM
(Table 11). model (Table 19), since the probability value according to F-
statistics and Chi-Square-statistics is above the significance
Test statistics (Trace and Max-eigenvalue) indicate 2 level of 0.05 (Table 21).
cointegrating equations at a significance level of 0.05. Thus,
the null hypothesis of the absence of cointegration is rejected Figure 6 shows the function of impulse responses of BAS
in favor of the alternative one. Therefore, the series of BAS in in construction to changes in BAS in production. This feature
manufacturing and BAS in services are cointegrated (Table has a positive effect. If the change in BAS in production were
12). unexpected by 1%, then BAS in construction would change as
follows on average, all other things being equal: from 1 to 2
In the VECM model (Table 13), the residuals are normally months - an increase of 0.89%, from 2 to 3 months - a
distributed because the probability value (0.17) is above the slowdown from 0.89% to 0.65%, from 3 to 4 months - increase
0.05 significance level (Fig. 2). from 0.65% to 0.77%, from 4 to 5 months - slowdown from
The probability values for F-statistics and Chi-Square- 0.77% to 0.71%, from 5 to 6 months - an increase from 0.71%
statistics are above the significance level of 0.05 (Table 14), to 0.74%, from 6 to 7 months - a slowdown from 0.74% to
so there is no serial correlation of residuals in the VECM 0.73% and from 8 months - level stabilization (Fig.6).
model (Table 13).
Figure 7 shows the decomposition of the variance, which
There is no heteroscedasticity of residuals in the VECM shows how much the series of SDA in construction is
model (Table 13), since the probability value according to F- explained by its previous values and the values of the
statistics and Chi-Square-statistics is above the significance influence of SDA in production, respectively. If from the 2nd
level of 0.05 (Table 15). month the share of the influence of SDA in production will be
16.50%, by the 8th month it will increase to 49.09%, then by
In the interests of the economic substantiation of the the 10th month it will increase to 54.90%(Fig.7).
parameters of the VECM models, we use the impulse response
function and the decomposition of the variance. 3.3 Model of dependence of BAS in mining industry on
BAS in construction
Figure 3 shows the impulse response function of BAS in
manufacturing to changes in BAS in the service sector. If the The optimal lag is a lag equal to 1, therefore, we will
change in BAS in services were unexpected by 1%, then BAS choose this lag for our model (Table 22).
in manufacturing would change as follows on average, all The result of the Granger test showed that at a significance
other things being equal: from 1 to 2 months - an increase of level of 0.05, the BAS series in construction is the cause of the
0.49%, from 2 to 3 months - a slowdown from 0.49% to 0.44%, Granger BAS series in the mining industry or in another way:
from 3 to 8 months - an increase from 0.44% to 0.57%, and BAS in construction predicts the change in BAS in the mining
from 8 months - level stabilization (fig.3). industry (Table 23).
Figure 4 shows a decomposition of the variance that shows Test statistics (Trace and Max-eigenvalue) indicate 2
how much the BAS in manufacturing is explained by its cointegrating equations at a significance level of 0.05. In this
previous values and the values of the impact of BAS in connection, the null hypothesis about the absence of
services, respectively. If from the 2nd month the share of the cointegration is rejected in favor of the alternative one.
influence of the BAS in services will be 8.43%, then by the Therefore, the series of BAS in the mining industry and BAS
10th month it will increase to 45.21% (fig.4). in construction are cointegrated (Table 24).
3.2 Model of dependence of BAS in construction on BAS In the VECM model (Table 25), the residuals are normally
in production distributed because the probability value (0.59) is above the
The lag equal to 2 is optimal for our model (Table 16). 0.05 significance level (Fig. 8).
The Granger test showed that, at a significance level of The probability values for F-statistics and Chi-Square-
0.05, the BAS in manufacturing is the cause of the BAS in statistics are above the significance level of 0.05 (Table 26),
construction according to Granger, or in another way: BAS in so there is no serial correlation of residuals in the VECM
manufacturing predicts the change in BAS in construction model (Table 25).
(Table 17). There is no heteroscedasticity of residuals in the VECM
The Trace test statistic indicates 2 cointegrating equations model (Table 25), since the probability value according to F-
at a significance level of 0.05. Thus, the null hypothesis of the statistics and Chi-Square-statistics is above the significance
absence of cointegration is rejected in favor of the alternative level of 0.05 (Table 27).
one. Therefore, the series of BAS in manufacturing and BAS Figure 9 shows the function of impulse responses of the
in services are cointegrated (Table 18). BAS in the mining industry to the change in the BAS in
In the VECM model (Table 19) the residuals are normally construction. This function has a positive effect. If a 1%
distributed because the probability value (0.44) is above the change in the BAS in construction were unexpected by 1%,
0.05 significance level (Fig. 5). then the BAS in the mining industry would change as follows
on average, all other things being equal: months 1 to 2, an
The probability values for F-statistics and Chi-Square- increase of 0.72%; months 2 to 3, an increase from 0.72% to
statistics are above the significance level of 0.05 (Table 20), 0.92%, from 3 to 4 months - an increase from 0.92% to 0.98%,
so there is no serial correlation of residuals in the VECM from 4 to 5 months - an increase from 0.98% to 1% and from
model (Table 19). 5 months - level stabilization (Fig.9).
- 368 -
Figure 10 shows the decomposition of the variance, which variances revealed the timing of future changes in the turning
shows how much the series of BAS in the mining industry is points of business cycle phases in the short run.
explained by its previous values and the values of the
influence of BAS in construction, respectively. If from the 2nd The practical significance of the study lies in the
month the share of the influence of the BAS in construction possibility of the use by the regulator and management
will be 10.54%, by the 5th month it will increase to 28.77%, structures of sub-indices of business activity in the services,
then by the 10th month it will increase to 36.66% (Fig.10). manufacturing, construction, and mining sectors as the best
variables for predicting economic activity and restructuring
Results the real sector of the economy of Kazakhstan.
After analyzing the obtained impulse response functions, Our results suggest that sound economic regulation is
the following should be noted. about supporting services, manufacturing, construction, and
mining industry in Kazakhstan.
First, the BAS in services predicts changes in the BAS in
manufacturing and construction, and its influence has a REFERENCES
positive effect. This means that with an increase in BAS in Carol M.Kopp. ISM Manufacturing Index: Definition and How It's
services, both BAS in manufacturing and BAS in construction Calculated - Updated December 18, 2022.
will increase. The economy in the areas of manufacturing and (https://www.investopedia.com/terms/i/ism-mfg.asp)
construction will grow with the increase in services. Nathan Reiff. Leading Indicators: Definition and How They’re Used by
Investors. By February 15, 2023.
Secondly, BAS in manufacturing predicts a change in (https://www.investopedia.com/terms/l/leadingindicator.asp)
BAS in construction, and its influence has a positive impact. Engle, R. F., Granger, C. W. J. (1987). Co-Integration and Error Correction:
Therefore, an increase in BAS in manufacturing will lead to Representation, Estimation, and Testing. Econometrica, Vol. 55, No. 2
an increase in BAS in construction. The construction economy (Mar., 1987), pp. 251–276. © Econometric Society.
will grow with the increase in production. Hizir Sofyan, Eva Maulia, and Miftahuddin (2017). Structure analysis of tax
revenue and inflation rate in Banda Aceh using vector error correction
Thirdly, BAS in construction predicts a change in BAS in model with multiple alpha//Proceedings of the 13th IMT-GT
the mining industry, while its influence has a positive effect. International Conference on Mathematics, Statistics and their
Therefore, if BAS increases in construction, then BAS in the Applications (ICMSA2017)//AIP Conference Proceedings 1905,
050029 (2017). (https://doi.org/10.1063/1.5012248)
mining industry will increase. The mining economy will grow
Eva Maulia, Miftahuddin and Hizir Sofyan (2018). Tax revenue and inflation
with the increase in construction. This may be due to the fact rate predictions in Banda Aceh using Vector Error Correction Model
that the construction sector is a supplier of both building (VECM)// IOP Conf. Series: Materials Science and Engineering 352
materials and services required by the mining industry. (2018) 012056 doi:10.1088/1757-899X/352/1/012056.
Bruce E. Hansen (2018). Johansen’s Reduced Rank Estimator Is GMM//
Conclusion Econometrics 2018, 6, 26; doi:10.3390/econometrics6020026.
Based on the results of the study, the following regularities John Cochrane (2006). Prediction and impulse responses in linear
were revealed. systems//Quantile, 2006, issue 1, 21-26.
John Cochrane (2006). Prediction and impulse responses in linear systems
The composite leading indicator and business activity sub- (in Russian), Quantile, (1), 21-26 [Джон Кокрейн (2006).
indices of the four sectors in Kazakhstan interact in the same Прогнозирование и импульсные отклики в линейных системах (на
direction: higher values of one indicator correspond to higher русском языке), Квантиль, (1), 21-26].
values of other indicators, and low values of one indicator Nosko V.P. (2011) Econometrics. Book 2. Part 3.4: textbook / V.P. Nosko. -
M .: Publishing house “Delo” RANEPA, 2011. - 576 p. ISBN 978-5-
correspond to lower values of others. 7749-0655-3 [Носко В.П. (2011) Эконометрика.Кн.2. Ч.3,4:
The scientific significance of the study comes down to the учебник/В.П.Носко. – М.: Издательский дом “Дело” РАНХиГС,
2011. – 576с. ISBN 978-5-7749-0655-3].
fact that two scientific hypotheses have been proven.
Rudra P. Pradhan, Girijasankar Mallik, Tapan P. Bagchi (2018). Information
The first hypothesis that there is a cointegration communication technology (ICT) infrastructure and economic growth:
relationship between leading indicators has been confirmed. A causality evinced by cross-country panel data//IIMB Management
Review (2018) 30, 91–103. (www.elsevier.com/locate/iimb)
A unidirectional and positive causal relationship has been
Rudra P. Pradhana, Neville R. Normanb, Yuosre Badirc, Bele Samadhana
established in the business activity of Kazakhstan, going from (2013). Transport Infrastructure, Foreign Direct Investment and
services to production, from production to construction, from Economic Growth Interactions in India: The ARDL Bounds Testing
construction to the mining industry for the period from Approach// 2nd Conference of Transportation Research Group of India
September 2016 to April 2023. The reasons for the (2nd CTRG. - Procedia - Social and Behavioral Sciences 104 (2013)
unidirectional positive relationship are explained by the role 914 – 921. (www.sciencedirect.com)
and importance of the services in stimulating production, from Belinga Thierrya, Zhou Junb, Doumbe Doumbe Ericc, Gahe Zimy Samuel
Yannickd, Koffi Yao Stéphane Landry (2016). Causality Relationship
production to the supply of necessary building materials to the between Bank Credit and Economic Growth: Evidence from a Time
growth in the activity of construction. In turn, the growth in Series Analysis on a Vector Error Correction. Model in
activity in construction increases the activity of the mining Cameroon//12th International Strategic Management Conference,
industry in extracting and processing raw materials for ISMC 2016, 28-30 October 2016, Antalya, Turkey - Procedia - Social
construction in the long term. and Behavioral Sciences 235 ( 2016 ) 664 – 671.
Nermin Yasar (2021). The Causal Relationship Between Foreign Debt and
The second hypothesis that the predictive, prognostic Economic Growth: Evidence from Commonwealth Independent
property of leading indicators of the business cycle is States//Foreign Trade Review 1–15, 2021.
manifested in the Kazakh economy has been confirmed. Chao Wang, Yul-Seong Kim, Chi Yeol Kim (2021). Causality between
logistics infrastructure and economic development in China //
The impulse response functions pointed to the positive Transport Policy 100 (2021) 49–58.
impact of business activity sub-indices. Decompositions of Farhad Taghizadeh-Hesary, Naoyuki Yoshino, Lisa Fukuda, Ehsan
Rasoulinezhad (2021). A model for calculating optimal credit
- 369 -
guarantee fee for small and medium-sized enterprises// Economic
Modelling Volume 95, February 2021, pp. 361-373.
Chaido Dritsaki (2013). Causal Nexus Between Economic Growth, Exports
and Government Debt: The case of Greece //Procedia Economics and
Finance 5 (2013) 251 – 259. International Conference On Applied
Economics (ICOAE) 2013.
N. V. Artamonov, E. A. Ivin, A. N. Kurbatsky, D. Fantazzini. Introduction
to time series analysis: textbook for universities. Lomonosov Moscow
State University, Moscow School of Economics, Department of
Econometrics and Mathematical Methods of Economics. - Vologda:
VolNTs RAS, 2021. - 134 pp. ISBN 978-5-93299-496-2 [Н. В.
Артамонов, Е. А. Ивин, А. Н. Курбацкий, Д. Фантаццини.
Введение в анализ временных рядов : учебное пособие для вузов.
Московский государственный университет имени М. В.
Ломоносова, Московская школа экономики, Кафедра
эконометрики и математических методов экономики. – Вологда:
ВолНЦ РАН, 2021. – 134 с.ISBN 978-5-93299-496-2].
- 370 -
Sorry, my tables and graphs are not included in this
template, I attached a regular file in Word
- 371 -
Data-Driven Enhanced Travel Planning by
Utilizing Real-Time User Data
Sooah Park1 and Juhee Choi1*
1
Sangmyung Univ, Republic of Korea
Email: [email protected]
Abstract— We investigate a method to extract essential Our service aims to provide suggestions for tourist
information for the formulation of travel plans and creation of attractions and accommodation options tailored to the user's
itineraries among a variety of real-time user data. Firstly, we inputted address or specific themes, enabling domestic
collect various data types such as card issuer details, travelers to combine travel and work seamlessly. Such
telecommunication service usage records, popular tourist personalized services can play a crucial role for travelers,
destinations, accommodation options, and job opportunities. especially in the current situation where travel restrictions are
Next, the data is analyzed by Elastic Search and visualized gradually being eased post-pandemic. To support travelers in
through Kibana. Elastic Search enables swift storage and access their decision-making process, we conducted data analysis
to large-scale data, thereby facilitating a service that suggests
and will provide the obtained results.
tourist attractions and lodging options tailored to specific
themes. It also provides information on short-term workation II. MOVITATION
job opportunities, assisting in curating meaningful travel
itineraries. The findings from the data analysis are made Firstly, we tried to determine the average ratings of
available on a website, providing users with real-time access to specific locations based on addresses and identify their
the information they desire. This analytical approach and the strengths. It also included discovering the relationship
accompanying system leverage frequently updated data to gain between latitude, longitude, and ratings. To aim it, we
insights into user preferences and behavioral patterns. Such collected data related to accommodations, tourist attractions,
personalized services could be of significant importance for restaurants, and reviews from public sources.
travelers, particularly in the present context where travel
restrictions due to the COVID-19 pandemic are being eased. Fig. 1. shows visualizing each data point on a map based on
The primary focus is on utilizing data analysis to support addresses. The data analysis revealed that tourist attractions
travelers in their decision-making process. were distributed evenly throughout Jeju Island contrary to our
expectations. Therefore, we decided to adjust the scope of
Keywords—Tourism, Big data, ElasticSearch, Kibana, visualization more precisely and included latitude and
WebSite, AWS, Visualization longitude as numerical data. However, we still couldn't obtain
meaningful results.
I. INTRODUCTION (HEADING 1)
Therefore, a powerful tool is required to examine our data
“ Elasticsearch” is a distributed search and analytics more closely such as Elastic Search and Kibana
engine supporting real-time, fast, and scalable search
capabilities. It is suitable for processing large volumes of data
and provides advanced search functionalities such as full-text
search, filtering, and aggregation.“Workation” is a novel
workstyle that has gained popularity in recent years. The term
is a combination of "work" and "vacation," and it refers to a
practice where individuals engage in work activities while
enjoying a vacation-like setting.
During our data analysis research, we had the opportunity to
formulate hypotheses and conduct validation using data
related to Jeju Island. However, we encountered challenges in Fig. 1. Visualizing each data point on a map based on addresses
deriving meaningful results. In order to address these
difficulties and enhance the accuracy of our analysis, In this III. IMPLEMENTATION
regard, we found Elastic Search and Kibana[1] as suitable We converted the data stored in MySQL into CSV files
tools for our research. Additionally, during this exploration, and inputted it into Elastic Search. the errors of the timestamp
we discovered a novel work style known as "Workation." values were fixed. For Kibana, we adjusted the latitude and
The COVID-19 pandemic has brought about changes in longitude data to fit the 'Position' format for using maps. In
conventional work practices[2], leading to a growing addition, we learned that Kibana allows us to utilize 'text'
preference for various work styles such as co-working[3]. address data[6].
Unlike other countries, we found that there is a lack of a Therefore, we have implemented a feature for the desired
customized "Workation" destination recommendation website address and check the information about nearby
for Korea.(As of May 28, 2023, it has been confirmed that the accommodations and restaurants. Through the map, users can
Korea Tourism Organization is currently providing a similar provide the distribution of accommodations in the specific
service.)[4] As a result, we decided to create a data platform area and access details such as the average rating, average
about workation that allows users to access real-time price, and a list of accommodations. Plus, they explore the top
information based on their desired criteria.[5] 10 affordable accommodations, top 10 highly-rated
- 372 -
accommodations, top 10 expensive accommodations, and prices, accommodation listings by price range,
nearby highly-rated tourist attractions. accommodation listings by rating, and nearby recommended
Furthermore, the platform gives comprehensive information tourist attractions. It also gives a recommended tourist
on various types of accommodations, including hotels and attraction called 'Gamue-mul' near Namwon-eup.
traditional Korean houses (hanok), as well as restaurants and
tourist attractions. For job opportunities, the average hourly ACKNOWLEDGMENT
wages and working hours categorized by industry and location This research was supported by program for post-
is needed. Regarding transportation data, due to limited PRIME(Program for Industrial needs-Matched Education)
available data, we have utilized the 'Kakao Map API' to was conducted by the National Research Foundation of
provide features such as real-time route finding and distance Korea(NRF) funded by the Ministry of Education.
views. To make all these Kibana dashboards accessible to
everyone, we created a website and connected it to our REFERENCES
Kibana account. Subsequently, we distributed the website
using AWS servers. [1] A. Robles-Gómez, S. Ros, A. Martínez-Gámez, R. Hernández, L.
Tobarra, R. Pastor, J. Cano, “Using Kibana and ElasticSearch for the
Recommendation of Job Offers to Students.”2019, In LASI-SPAIN (pp.
93-99).
[2] R. Florida, A. Rodríguez-Pose & M. Storper, “Cities in a post-COVID
world.”. 2021, Urban Studies.
[3] M. Orel, “Coworking environments and digital nomadism: balancing
work and leisure whilst on the move.”2019, World Leis J 61 (3): 215–
227.
[4] Korea Tourism Organization , “Korea Gusuk-Gusuk”, Website URL :
https://korean.visitkorea.or.kr/main/main.do
[5] D. Shrestha, T. Wenan, N. Rajkarnikar, D. Shrestha & S. R. Jeong,
“Study and evaluation of tourism websites based on user perspective.” ,
2021, Journal of Internet Computing and Services, 22(4), 65-82.
[6] C. Conner, J. Samuel, A. Kretinin, Y. Samuel & L. Nadeau, “A picture
for the words! textual visualization in big data analytics.”, 2020, arXiv
preprint arXiv:2005.07849.
Fig. 2. Tourist attractions
Fig. 2 illustrates a dashboard for tourist attractions. It
provides insights into the number of visits and ratings for
attractions by region, as well as the distribution of popular
regions and attractions. Plus, they can identify the most
visited attractions and those with high search volumes. In the
dashboard, it is revealed that the most visited tourist attraction
in Jeju Island is 'Saryeoni Forest Road'.
Fig. 3. Accommodations
A dashboard for accommodations by region is shown in Fig.
3. Upon selecting a specific region, the screen provides the
relevant accommodation information including the
distribution of accommodations, average ratings, average
- 373 -
A Study on Reuse Policy for Worn-out Cells in
Non-Volatile Memories
Juhee Choi1 and Seonuck Paek1*
1
Sangmyung University, Cheonan, Korea
Email: [email protected]
Abstract— One of the main drawbacks of non-volatile A cell of MRAM is called as the magnetic tunnel junction
memories (NVMs) is the limited lifetime. After a certain number (MTJ), as shown in Fig. 1 [6]. Each MTJ consists of two
of write operations, an NVM cell is worn out, so that it cannot ferromagnetic layers such as 'pinned' layer and ‘free’ layer as
contain the information. To overcome this problem, we its name stands. The magnetization direction of the pinned
investigate the reuse possibility for the pre-defined pattern of layer is fixed, while the direction of the free layer can be
the data. In some fields, frequently used data patterns are rotated. A tunnel barrier is located between these two layers
consistent, so it is possible to store data matched to existing to separate them. The bit information in an MTJ cell is
worn-out cells. Our survey suggests the guide for maximizing
determined by whether the direction of the free layer is parallel
the potential.
or not to the direction of the pinned layer.
Keywords— non-volatile memories, limited lifetime, reuse II. REUSE OF WORN-OUT CELLS
policy, worn-out cells
Among various candidates of NVMs, PCM has emerged
I. INTRODUCTION as an appealing substitute for DRAM due to its distinct
Non-volatile memories (NVMs) have gained significant advantages and potential opportunities. Nevertheless, its
attention from researchers as an alternative of conventional limited write endurance presents a significant obstacle to its
memories such as static random access memory (SRAM) and practical implementation [7]. This problem signifies that some
dynamic random access memory (DRAM) [1-4]. The term memory cells may permanently become stuck at either '0' or
'non-volatile' means that each cell of NVM stores a bit '1' after a certain number of writes. To address this issue,
information instead of electric charges. This approach allows recovery techniques are required with minimal storage
NVMs to preserve information with minimal or no electrical overhead. Our study introduces a reuse mechanism to save the
power, resulting in a significant reduction of leakage energy data in the worn-out cells. If some cells are stuck at ‘0’s, when
consumption. the data consists of continuous ‘0’s, they are stored in the
failed cells. To aim it, we investigate the algorithm to find the
Plus, NVMs provides smaller unit memory cell size minimum shifting or flipping schemes which are have been
compared to SRAM. The last-level cache (LLC) or lower level studied [8-9].
caches, which typically occupy a large portion of the total chip
area, are generally made up of SRAM. Thus, the higher As the next step, we will study the optimal implementation
density of NVMs is an attractive prospect for chip policy for each workload. Since the data patterns of the
manufacturers. NVMs come in various forms depending on computation-centric workloads and communication-centric
the type of memory cell. Among them, phase change memory workloads differ, it is needed to detect the type of the
(PCM) and spin-torque transfer magnetic RAM (STT-MRAM) workload during program execution. Then, we will also devise
are the predominant areas of research. the mechanism to be applied for several memory subsystems
which usually have a significant influence on the performance.
The base of a cell in PCM [5] is usually made up of alloy
material like Ge2Sb2Te5. Each logic has two distinct states - ACKNOWLEDGMENT
the amorphous phase and the crystalline phase. These states This research was supported by KOICA(Korea
vary physically in their electrical resistance and optical International Cooperation Agency) through "Capacity
reflectivity. In logical terms, the amorphous phase represents Building Project for School of Information and
the RESET or "0" state, while the crystalline phase denotes Communication Technology at Mongolian University of
the SET or "1" state. The transition between these phases Science and Technology in Mongolia" (Contract No. P2019-
within a PCM cell is accomplished by the heat produced from 00124).
the electric current.
Fig. 1. Spin-Torque Transfer Magnetic RAM.
- 374 -
REFERENCES [5] Wong, H.-S.P., Raoux, S., Kim, S., Liang, J., Reifenberg, J.P.,
Rajendran, B., Asheghi, M., Goodson, K.E.: Phase change memory.
[1] Fantini, P.: Phase change memory applications: the history, the present Proceedings of the IEEE 98(12), 2201–2227 (2010)
and the future. Journal of Physics D: Applied Physics 53(28), 283002
(2020) [6] Zhu, J.-G.J., Park, C.: Magnetic tunnel junctions. Materials today 9(11),
36–45 (2006)
[2] Wang, K., Alzate, J., Amiri, P.K.: Low-power non-volatile spintronic
memory: Stt-ram and beyond. Journal of Physics D: Applied Physics [7] Ferreira, A.P., Zhou, M., Bock, S., Childers, B., Melhem, R., Moss ́e,
46(7), 074003 (2013) D.: Increasing pcm main memory lifetime. 2010 Design, Automation
& Test in Europe Conference & Exhibition, pp. 914–919 (2010)
[3] Pan, X., Ma, T.: Retention mechanism study of the ferroelectric field
effect transistor. Applied Physics Letters 99(1), 013505 (2011) [8] Bittman, Daniel, et al. "Designing data structures to minimize bit flips
on NVM." 2018 IEEE 7th Non-Volatile Memory Systems and
[4] Akinaga, H., Shima, H.: Resistive random access memory (reram) Applications Symposium (NVMSA). IEEE, pp. 85-90 (2018).
based on metal oxides. Proceedings of the IEEE 98(12), 2237–2251
[9] Luo, Xianlu, et al. "Enhancing lifetime of NVM-based main memory
(2010)
with bit shifting and flipping." 2014 IEEE 20th International
Conference on Embedded and Real-Time Computing Systems and
Applications. IEEE, pp. 1-7 (2014).
- 375 -
Development of an analytical computation
system and its application for automation of
equations derivation in robotics systems
Magzhan Aliaskar1, Sholpan Jomartova1*, Aigerim Mazakova1*, Talgat Mazakov1*,
Gulzat Ziyatbekova1,2*
1
Al-Farabi Kazakh National University, Almaty, Kazakhstan
2
Institute of Information and Computational Technologies CS MES RK, Almaty, Kazakhstan
Email: [email protected]
Abstract-- The advancement of automation techniques in systems emerged in the late 1960s, implementing various
the field of mathematical modeling has greatly facilitated the approaches to create universal computing systems.
process of constructing mathematical models. Symbolic
One approach involved using existing procedural
computations and computer algebra systems have played a
pivotal role in this regard. Mathematical modeling involves languages (such as C, Pascal, Fortran, Algol) or functional
the development of equations that describe the dynamics of languages (such as Lisp) as a base language for creating a
various systems, including robots and robotic systems. new language specifically designed for manipulating
By leveraging the capabilities of computer algebra algebraic objects. Another approach involved extending
systems, researchers can efficiently perform complex
the syntax and capabilities of an existing programming
computations involving matrices, series, simplification, and
the calculation of algebraic expressions. This automation language through symbolic manipulation modules. The
significantly reduces the time and effort required for deriving first approach provided a new language independent of
equations and enhances the reliability of the mathematical specific language translators but required efficient
model. implementation, learning, and support for the new
The dynamics of robots and robotic systems are of
language. Additionally, it suffered from slower execution
particular interest in the realm of mathematical modeling.
Analyzing the behavior and movement of robots involves speed compared to compiled packages and created
capturing their motion patterns, mechanical properties, and complexities when modifying syntax (requiring rewriting
interaction with the environment. Mathematical models all packages). The second approach was advantageous
enable the prediction and simulation of robot dynamics, due to high execution speed of packages, utilization of all
allowing researchers to optimize robot designs, control
programming language features, and package
strategies, and evaluate their performance in different
scenarios. independence from specific language translators.
The integration of automation techniques, symbolic Commercial packages often used the first approach to
computations, and computer algebra in mathematical provide a closed working environment without access to
modeling has revolutionized the field of robotics. It has led to source code due to the absence of language compiler
advancements in the development of efficient algorithms,
delivery. Experimental and university systems fully
control strategies, and system optimization. The utilization of
mathematical models has facilitated the design and embraced the second approach, providing openness and
implementation of robotic systems that exhibit enhanced implementation efficiency.
performance, precision, and adaptability. Currently, there are numerous analytical computation
This article explores the application of automation in systems, each with its own features and limitations. They
mathematical modeling, with a focus on the dynamics of
offer a wide range of functions and tools for working with
robots and robotic systems. It highlights the significance of
computer algebra systems in simplifying complex algebraic objects. However, certain limitations may arise
computations and discusses the implications of mathematical when using these systems, such as lack of access to source
modeling in the field of robotics. code, computation complexity restrictions, difficulties in
adaptation for learning, and the use of outdated concepts
Keywords: automation of mathematical model in implementation.
construction, symbolic computations, computer algebra,
mathematical model, dynamics of robots, robotic systems. Experimental symbolic computation systems have
also been created, utilizing non-traditional internal
I. INTRODUCTION expression representations (e.g., vector representation),
Analytical (symbolic) computation systems play a new processors (e.g., matrix processor), or writing styles
significant role in automating algebraic transformations, (e.g., functional style - Parsifal). The problem of
ranging from expression simplification to differentiation analytical computations has proven to be extensive, and
and integration. These operations are an integral part of each system only implements a few aspects of it. It should
the work of researchers in practically any mathematical be noted that recently, new promising directions have
model. As a result, the first analytical computation emerged. The first is the use of symbolic computations in
proving mathematical theorems [1-5], which require
- 376 -
extensive computations, leading to the proof of properties reuse, enable the efficient creation of various programs
of stereographic projection and lemmas leading to the focused on ease of interaction and quick results. For this
proof of properties of nowhere continuous Weierstrass purpose, it is necessary to have an accessible and user-
function. The second direction is the application of friendly interface between modules handling expressions
automated symbolic computations in questions related to and the programming language. Therefore, the system
programming the kinematics of mechanical robots [6], should be designed considering this aspect of usage,
where real-time solution of differential and other treating its core as a programming tool that includes a
equations is required during their training. For research library of symbolic algebraic transformation procedures
purposes, the application of various tools is necessary - available to developers.
from well-established systems to experimental and open Thus, the state of existing analytical computation
implementations. systems does not meet the requirements and the new role
Given the aforementioned limitations, there is a need they should play in the education system and problem-
for a new analytical computation system that overcomes solving. Therefore, there is a need for the development of
these problems and provides additional advantages. a new system that satisfies these requirements.
Further research will be devoted to the development of
such a system, as well as its application in the educational
process and for solving research tasks. II. MAIN RESULTS
The obtained results are a continuation of the authors'
PROBLEM STATEMENT: previous research published in [12-13].
Among the operations on expressions included in the The SAV package consists of a set of modules, each
set of core functions of an advanced analytical containing a set of data types, variables, and functions.
computation system [7-10], the following are desirable: The modules are divided based on their interdependence
and logical relationship to operations performed on the
User-controlled simplification.
expression tree. The package consists of 9 modules in
Substitution of various types. total: S_Init, S_Misc, S_Scaner, S_Tree, S_Sort, Intrface,
Ordering (reducing to a canonical form). S_Matrix, SAV.
Numerical computations (preferably without integer The developed core implements the main
range restrictions, with rational arithmetic). functionalities required to perform practical operations on
Set of operations on polynomials (GCD, expressions of a general form. An expression is described
factorization, decomposition into different bases). by a regular context-free grammar for algebraic
Differentiation. expressions.
The main (basic) operations include:
Integration.
Reading any text describing an algebraic expression
Matrix operations.
and detecting syntax errors.
It is desirable for the developed analytical
Constructing the expression tree from its textual
computation system to be modifiable and open. Often, a representation and detecting semantic errors.
large part of the functionality of such systems proves to Evaluating a numerical expression and detecting
be excessive for solving a specific problem, such as errors.
obtaining a complex differential equation in symbolic Modifying an expression, such as expanding
form. In such cases, only basic operations are required, parentheses or ordering.
such as simplification and differentiation without Simplifying any expression.
restrictions on expression complexity. Therefore, the ideal Substituting one expression into another expression.
set of capabilities of an analytical computation system Differentiating an expression with respect to a
should correspond to the level of problems being solved specified variable.
and the requirements of the educational system. Ease of Integrating certain specific cases of expressions.
learning the language and its functional sufficiency for Procedures have been developed to parse the input text
obtaining results from non-programmers are important string describing an algebraic expression and detect
criteria. syntax errors.
Furthermore, the widespread use of analytical A procedure has been developed to uniquely construct
computation systems is determined not only by their the expression tree based on the input string and detect
semantic errors.
functionality but also by the availability of
A function has been implemented to evaluate a
documentation, support, and ease of use during the initial
numerical expression, as well as the ability to substitute
and subsequent interaction with the system. Currently, variable values, perform partial evaluations, and detect
most existing systems do not meet these requirements, errors such as overflow, function argument error, and
indicating the need for the development of their division by zero, among others.
environment [11]. A procedure has been developed for expanding
Implemented and debugged code for handling parentheses and modifying the expression tree
algebraic objects is usually embedded in the considering its semantics.
corresponding system and is only used through it.
However, modern programming paradigms, such as code
- 377 -
Several types of substitutions of one expression into numerical expression, modify the expression, simplify the
another have been implemented, as well as checking for expression, substitute one expression into another,
cyclic dependencies between variables. differentiate the expression, and integrate specific cases.
A procedure has been developed to differentiate an The SAV system implements algorithms for syntactic
expression of arbitrary degree with respect to specified parsing of the input string, construction of the expression
variables. All functions are implemented as recursive tree, evaluation of the expression, differentiation of the
procedures working with the expression tree. expression, tree transformation, parentheses expansion,
The system of analytical computations implements the and simplification of the expression.
following algorithms: Additionally, a user-friendly interface has been
Syntactic parsing of the input string based on the created for working with expressions, allowing the input
regular representation of the set of algebraic expressions. of operator sequences in the program's special language,
Bottom-up construction of the expression tree reading data from the terminal or a text file, and
using recursive calls to procedures for reading displaying the results on the screen or saving them to a
non-terminal symbols of the grammar describing file.
algebraic expressions. Overall, the developed SAV package provides a
Evaluation of the expression by traversing the tree sufficient set of functionalities for performing operations
from left to right. on expressions and meets the requirements expected from
Differentiation of the expression by applying analytical computation systems.
differentiation rules and subsequent Here are some examples of the interpreter's work:
simplification. 1. Evaluating a complex expression:
Transformation of the tree into another f:=20^20*Sqrt(2*Pi*20)*Exp(-20+(1-1/30/20)/12/20);
semantically isomorphic tree. print f @
Expansion of parentheses in the expression. Output:
Recursive simplification of the expression by f= 2.43288E18
grouping and sorting of terms and factors, as well 2. Evaluating an expression with different variable
as evaluating subexpressions. values:
Recursive substitution with specifying the set of f:= Ln(x+Sqrt(x^2+1));
substituted variables and the set of target variables. x:= 2; Subst f to f2;
The next step in processing the expression is the x:= 3; Subst f to f3;
syntactic analysis of the token string, which Print f2 f3 @
simultaneously builds the expression tree in memory, Output:
hence called BuildTree. The procedure assumes the f2= 1.316
construction of expressions generated by the following f3= 1.818
context-free grammar, which is standard for ordinary 3. Computing derivatives of different orders:
algebraic expressions: y:= Tg(x);
E=E+T | E-T | T y1:= Diff(x)(y);
T=T*P | T-P | P y2:= Diff(x 2)(y);
P=P^L Print y1, y2 @
L = number | var | (E) | Func(E) Output:
y1= 1/(Cos x)^2
However, this grammar still contains direct left y2= 2*Sin x/(Cos x)^3
recursion and is not suitable for immediate 4. Computing partial derivatives of a multivariable
implementation of syntactic parsing. Therefore, by function:
factorizing it, a form is obtained that dictates the idea of a f:= x^4 + y^4 - 4*x^2*y^2;
top-down syntactic analysis algorithm: fx:= Diff(x)(f);
E = T E' | T fy:= Diff(y)(f);
E' = + T | - T | + T E' | - T E' Print fx, fy @
T = P T' | P Output:
T' = * P | / P | * P T' | / P T' fx= 4*x^3 - 8*x^y^2
P= L^ P|L fy= 4*y^3 - 8*x^2*y
L = number | var | (E) | Func(E) 5. Evaluating a function with different
substitutions:
The main modules of the SAV package are S_Init, f:= Sin(3*x/t);
S_Misc, S_Scaner, S_Tree, S_Sort, Intrface, S_Matrix, x:= t; Subst f to ft;
and SAV. These modules contain the necessary data x:=2*t; Subst f to f2t;
types, variables, and functions to perform operations on Print ft, f2t @
expressions. Output:
The developed SAV core implements the main ft= 0.0523
functionalities required to perform operations on f2t= 0.1045
expressions of a general form. Procedures have been 6. Computing the limit of a function using
developed to parse the input string, detect syntax and L'Hôpital's rule:
semantic errors, construct the expression tree, evaluate the f:= Sin(x); g:= x;
FG:= f/g; x:= 0;
- 378 -
Subst FG to Limit1; Print Limit1; Another example of the ease of use of library functions
df:= Diff(x)(f); dg:= Diff(x)(g); is the implementation of a function to construct a
dFG:= df/dg; Maclaurin series for an arbitrary differentiable function.
Subst dFG to Limit2; Print Limit2; The main part is given below:
Output: R:= SetNum(0);
!! Subst failed: Dividing by ZERO ff:= SetNum(1);
Limit2= 1 df:= CopyTree(f);
SetTo(x, 0);
SAV can be used to obtain a mathematical model of fact:=1;
robotic systems using Lagrange's equations of the second
kind. The Lagrange's equation of the second kind for FOR i:=0 to n DO BEGIN
describing the dynamics of an n-link manipulator is IF i<>0 THEN fact:= fact*i;
written in matrix form with the application of generalized ff:= SetNum(fact);
forces. R:= opetrees(R, '+',
Here is a fragment of a corresponding program text: opetrees(Substitute(df,[xc],[xc],[]), '*',
FOR j:=1 to n DO BEGIN opetrees(SetExpr(x+'^'+i2str(i), '/', ff)));
QP[j]:= SetNum(0); Diff(xc, df);
FOR i:=j to n DO Reduce(df,Ok);
FOR k:=1 to i DO BEGIN END;
QP[j]:=
OpeTrees(QP[j], The result is returned in the pointer variable R. The
'+', variable f contains the original function, x is the variable
OpeTrees(Track(MultMx(U[i,j], MultMx(H[i], name for which the expansion is performed, and n is the
TranspMx(U[i,k])))),'*', length of the series. The auxiliary variable df contains the
d2q[k])); current derivative of the original function.
END; Based on the core of the system and the interface
procedures between the core and the programming user, a
FOR i:=j to n DO knowledge testing program has been developed for
FOR k:=1 to i DO specific areas of algebra. The working scheme is as
FOR l:=1 to i DO BEGIN follows: the program and the test subject solve an
QP[j]:= algebraic problem, and then the program reads the result
opetrees(QP[j], provided by the individual and compares it with its own
'+', solution. The range of algebraic problems depends on the
opetrees(Track(MultMx(U[i,j], MultMx(H[i], functionalities implemented in the core system and
TranspMx(V[i,k,l])))), includes the following topics:
'*', Evaluation of expressions and function values
opetrees(dq[l], '*', dq[k]) )); Rules for simplifying algebraic expressions
END; Rules for differentiation with respect to one or
multiple variables
FOR i:=j to n DO BEGIN Rules for matrix operations: matrix arithmetic,
QP[j]:= opetrees(QP[j], '-', determinant computation, inverse matrix, minors,
opetrees(m[i], trace
'*', Function series expansions and operations with
Track(MultMx(G, MultMx (U[i,j],R[i])) ))); series
END; Computation of limits
END;
Solution of systems of linear algebraic equations
Integration of certain functions
It should be noted that the accumulation of summation
It is evident that additional problem classes based on
occurs in a variable of type pointer to the tree QP[j], and
the listed topics can be easily and quickly included.
the arrays dq, d2q, and m contain the first and second
derivatives with respect to time of the generalized CONCLUSION
coordinate functions and the masses of the links,
The program developed by the authors automates the
respectively. The advantage of obtaining all results in
process of deriving equations for the dynamics of robotic
symbolic form is indisputable since the obtained result
can be used for various substitutions without computing it systems. The implemented interface allows for interactive
communication between the program and the researcher,
from scratch each time.
who specifies the number of links and the characteristics
The efficiency of the developed system is
of each link. The user can provide some inputs
demonstrated by an example of a three-link robot
quantitatively and others symbolically.
manipulator, where the program Sav.exe is called. The
The program generates the Lagrangian function in an
input data for the output mode of the robot manipulator
automated mode and then derives the equations that
equations are placed in the file Robot3.Dat. The results of
describe the dynamics of a multi-link robot based on the
the program execution are outputted to the file
Lagrangian equation of the second kind.
Robot3.Out.
- 379 -
The research methods include an interpreter that [15] Shakin, V.N., Semenova, T.N., Semenova, T.I. Analytical
Computations and Examples of Electrical Engineering Problem
enables fast computation by directly executing operations Solving Using Symbolic Mathematics in Matlab. Moscow:
after reading the commands. The interpreter allows users Goryachaya Liniya - Telecom, 2023. 238 p.
to enter commands sequentially from the screen or from a [16] Dyakonov, V.P. Encyclopedia of Computer Algebra. Moscow:
text file. The computation results are displayed on the DMK Press, 2010. 1264 p.
screen or saved in a user-specified file. [17] Revinskaya, O.G. Symbolic Computations in MatLab. Moscow:
The program has been continuously improved over a Lan', 2020. 528 p.
considerable period of more than 20 years, confirming the [18] Mazakov, T., Wójcik, W., Jomartova, Sh., Karymsakova, N.,
Ziyatbekova, G., Tursynbai, A. The Stability Interval of the Set of
validity of the chosen implementation strategy as library Linear System. INTL Journal of Electronics and
functions. Telecommunications, 2021, Vol. 67, No. 2, pp. 155-161. DOI:
Users are provided with the following capabilities: 1) 10.24425/ijet.2021.135958
control over the sequence of computations, 2) [19] Mazakova, A., Jomartova, S., Mazakov, T., Shormanov, T.,
development of custom procedures and functions, 3) Amirkhanov, B. Controllability of an unmanned aerial vehicle.
IEEE 7th International Energy Conference (ENERGYCON),
design of custom interfaces, and 4) the ability to apply 2022, pp. 1-5. doi: 10.1109/ENERGYCON53164.2022.9830244.
library functions in their own research projects as external
modules.
The library functions of the proposed analytical
computation system can be effectively applied to
investigate various electro-technical systems, such as
studying the controllability or stability of a trajectory,
finding optimal controls, etc.
A promising direction for future work is the
development of the Symbolic Analytical Computation
System (SAC) for interval mathematics.
The work was carried out at the Research Institute of
Mathematics and Mechanics of the Kazakh National
University named after al-Farabi.
REFERENCES
[1] Kurbatova, N.V. Analytical Computations in Maple. Rostov-on-
Don: Southern Federal University, 2008. 146 p.
[2] Gubina, T.N., Andropova, E.V. Solving Differential Equations in
the Computer Mathematics System Maxima. Yelets: Elets State
University, 2009. 99 p.
[3] Sedov, E.S. Basics of Working with the Computer Algebra
System Mathematica. Moscow: National Open University
"INTUIT", 2016. 401 p.
[4] Osipov, N.N. Symbolic Computations in Mathematical Proofs
(Computer Assisted Proofs). Mathematical Education, 2020,
Issue 2, 42-47.
[5] Klepikov, P.N., et al. Symbolic Computations in Geometry and
Analysis Problems. Barnaul: Altai State University Publishing,
2015. 156 p.
[6] Grosheva, M.V., Efimov, G.B., Samsonov, V.A. History of Using
Analytical Computations in Mechanics Problems. Moscow:
Moscow State University, 2005. 87 p.
[7] Shilin, I.A. Computer Algebra in Problems. Moscow: Moscow
State Pedagogical University, 2018. 57 p.
[8] Goryushkin, A.P. Abstract and Computer Algebra. Moscow:
Yurait, 2021. 691 p.
[9] Voyevoda, A.A., Bobobekov, K.M. Symbolic Computations.
Collection of Scientific Papers of NSTU, 2017, No. 1 (87), pp.
23-37.
[10] Banshchikov, A.V., Burlakova, L.A., Irtegov, V.D., Titoreko,
T.N. Symbolic Computations in Modeling and Qualitative
Analysis of Dynamic Systems. Computational Technologies,
Volume 19, No. 6, 2014, pp. 3-18.
[11] Mukha, V.S. Computational Methods and Computer Algebra.
Minsk: BSUIR, 2006. 127 p.
[12] Nurmagambetov, A.A., Dzhomartova, Sh.A., Mazakov, T.Zh.
Development of an Analytical Computation System.
Computational and Information Technologies in Science,
Engineering, and Education. Ust-Kamenogorsk, 2013, pp. 143-
147.
[13] Dzhomartova, Sh.A., Karimsakova, N.T., Tursynbai, A.T.
Bulletin of KazATK, No. 2 (113), 2020, pp. 207-213.
[14] Yurevich, E.I. Fundamentals of Robotics. St. Petersburg: BHV-
Petersburg, 2010. 368 p.
- 380 -
MATHEMATICAL MODELING AND
DEVELOPMENT OF AUTOMATED FIRE
MONITORING SYSTEM
A.A. Sametova1,2* , T.Zh. Mazakov 1,3 , G.Z. Ziyatbekova 1,3 , A.D. Burgegulov1
1
Al-Farabi Kazakh National University, Almaty, Kazakhstan
2
Almaty University of Power Engineering and Telecommunications named after Gumarbek Daukeyev
3
Institute of Information and Computational Technologies CS MES RK, Almaty, Kazakhstan
Email: [email protected]
Abstract. Mathematical modeling and the development of The article discusses the problems of monitoring and
automated fire monitoring systems involve the use of mathematical forecasting forest and steppe fires, determining the
equations and computer algorithms to analyze data and predict the mathematical model used to prevent forest and steppe fires,
occurrence of fires. In this process, we create a mathematical model and developing an automated system. We all know about the
to simulate the behavior of fire and use it to develop an automated harm and consequences of a silent enemy. For this reason,
fire monitoring system. The first step in the process is to develop a forest fires are currently recognized as one of the most
mathematical model that describes the behavior of fire. The model important factors in changing landscapes. Each resulting fire
should take into account various factors such as fuel type,
is thrown into the atmosphere about a million combustion
temperature, humidity, wind speed, and direction. We can use
products, which are released many toxic substances and
different modeling techniques, such as computational fluid
dynamics (CFD) or finite element analysis (FEA), to develop a
chemical elements. Depending on the spread of fire and its
detailed and accurate model of fire behavior. Once we have dynamics, the description of mathematical models varies.
developed a mathematical model, we can use it to develop an Among the mathematical models of Fire Prevention, the most
automated fire monitoring system. The system should be able to important and complex are those that characterize the
detect and analyze fire behavior in real-time and provide early dynamics of forest and natural fires. Therefore, since fire
warning to authorities and individuals in the affected area. The prevention is a huge problem all over the world, we examined
system should also be able to predict the behavior of the fire and mathematical models used for the prevention and control of
provide recommendations for firefighting operations. emergency situations and analyzed fire control, forecasting
There are several types of sensors that can be used in an systems, their features in the article.
automated fire monitoring system, such as thermal imaging
cameras, smoke detectors, and infrared cameras. These sensors can The purpose of the work is to develop technology and
detect changes in temperature, smoke density, and radiation levels, software for automated control and forecasting of forest and
which can indicate the presence of a fire. The data collected by the steppe fires using mathematical modeling.
sensors is then processed by a computer system, which uses the
mathematical model to analyze the behavior of the fire. The system II. RESEARCH MATERIAL AND METHODS. DESCRIPTION OF
can also use machine learning algorithms to learn from past fire THE ALGORITHM AND MODULES
incidents and improve its predictions. In conclusion, mathematical Controlling and forecasting forest and steppe fires requires
modeling and the development of automated fire monitoring systems a multidisciplinary approach that combines various research
can help prevent and mitigate the damage caused by fires. By materials and methods. Here are some common research
developing accurate models of fire behavior and using advanced materials and methods used for the control and forecasting of
sensors and machine learning algorithms, we can provide early forest and steppe fires: Remote sensing involves the use of
warning and improve firefighting operations, ultimately saving lives
satellites, aircraft, and other remote sensing platforms to
and reducing property damage.
collect data on fires, weather conditions, and fuel moisture.
Keywords: fire, automated system, mathematical
modeling, fire monitoring systems.
This data is used to detect and monitor fires, as well as to
forecast their behavior. Geographic Information Systems
I. INTRODUCTION (GIS) - GIS is a computer-based tool that allows researchers
to analyze and visualize data related to fires, weather,
Fires can cause significant material damage and pose a topography, and vegetation [1]. This tool can be used to
serious threat to people's lives. As a result, it is very important identify high-risk areas and to develop fire management plans.
to have effective fire control measures in order to detect Fire behavior models are computer simulations that predict the
possible fires in a timely manner and react quickly. Fire behavior of fires based on environmental conditions, such as
control systems play an important role in fire prevention and fuel moisture, topography, and weather. These models can be
suppression. They must be carefully designed, installed and used to forecast the spread of fires and to develop effective
maintained to ensure their effectiveness in detecting and firefighting strategies. Meteorological data, such as
responding to fires. The development of a fire management temperature, humidity, wind speed, and direction, can be used
system can provide a number of advantages, such as to forecast fire behavior and to develop fire management
increasing safety, reducing damage and loss, and increasing plans. This data can be collected from weather stations,
property value. A well-thought-out fire management system satellites, and other remote sensing platforms. Fuel moisture
helps reduce damage, protect human lives, and comply with sampling involves collecting samples of vegetation to
safety regulations. measure their moisture content. This data can be used to
- 381 -
predict the potential for fires and to develop fire management plan should include procedures for alerting emergency
plans. Fire detection systems, such as cameras, smoke and services, evacuating people and animals, and
heat detectors, and automated weather stations, can be used to providing support to firefighters [3-4].
detect fires early and to provide accurate information to
emergency services. Fire management plans are developed Developing an algorithm for monitoring and predicting
based on research data and involve the implementation of forest and steppe fires requires a comprehensive approach that
measures to prevent and control fires. These plans may involves data collection, fire detection, fire behavior
include fuel reduction, prescribed burns, and the development prediction, fire danger rating, fire management, continuous
of fire breaks. monitoring, and response planning. This algorithm can help to
prevent fires and minimize their impact on human lives,
In conclusion, the control and forecasting of forest and property, and the environment. At the same time, the analysis
steppe fires require the integration of various research of the currently operating systems was carried out in the
materials and methods. These tools can help to detect and article. Although each system has its own advantages and
monitor fires, forecast their behavior, and develop effective disadvantages, each system has its own users. The analysis of
firefighting strategies. The use of these tools is critical in some automated systems used in the control and forecasting
protecting human lives, property, and the environment from of forest and steppe fires is shown in the table 1.
the devastating effects of fires [2].
TABLE I. SOME AUTOMATED SYSTEMS USED IN THE CONTROL
Using the materials and methods necessary to develop the AND FORECASTING OF FOREST AND WILDFIRES
above monitoring system, it is possible to effectively develop
Systems Description
a system for monitoring and forecasting forest and wildfires
based on the following algorithm. Developing an algorithm FireWatch FireWatch is a satellite-based fire detection system
for monitoring and predicting forest and steppe fires requires that uses a combination of infrared and visible light
sensors to detect fires in real-time. The system can
a comprehensive approach that takes into account multiple detect fires as small as 30 meters in diameter and
factors that can influence fire behavior. Here is a general can provide accurate location data to emergency
algorithm for monitoring and predicting forest and steppe services within minutes.
fires: Farsite Farsite is a fire behavior and growth model that
uses terrain, weather, and fuel data to simulate the
• Data Collection: Collect data on weather conditions, spread of fires. The model can be used to forecast
fuel moisture, vegetation, topography, and other the potential size and direction of fires, allowing
relevant factors. This data can be collected through emergency services to plan their response.
remote sensing, meteorological stations, fuel moisture MODIS (Moderate MODIS is a NASA satellite-based system that can
sampling, and other means. Resolution detect fires in near real-time. The system can
Imaging detect fires as small as 1000 square meters and can
• Fire Detection: Detect fires through various fire Spectroradiometer) provide data on fire location, size, and
detection systems such as cameras, smoke and heat temperature.
detectors, and automated weather stations. These Fire Danger Rating Fire Danger Rating Systems (FDRS) are used to
systems can detect fires in real-time and provide Systems assess the potential for fires based on weather
accurate information to emergency services. conditions, fuel moisture, and other factors. These
systems can help to predict the likelihood of fires
• Fire Behavior Prediction: Use fire behavior models to occurring and provide warnings to people living in
predict the spread of fires based on environmental high-risk areas.
conditions such as fuel moisture, topography, and Alertwildfire Alertwildfire is a camera-based system that
weather. These models can be used to forecast fire provides real-time video of wildfires to emergency
services. The system uses cameras mounted on
behavior and to develop effective firefighting towers or other high points to provide a panoramic
strategies. view of the area. This can help emergency services
to assess the situation and make decisions about
• Fire Danger Rating: Use fire danger rating systems to how to respond.
assess the potential for fires based on weather
conditions, fuel moisture, and other factors. These
systems can help to predict the likelihood of fires These are just a few examples of the automated systems
occurring and provide warnings to people living in for monitoring and forecasting forest and steppe fires that
high-risk areas. exist. Each system has its own strengths and weaknesses, and
• Fire Management: Develop fire management plans some may be better suited to certain types of fires or
based on the data collected in steps 1-4. These plans environments than others. It is important to choose the right
may include fuel reduction, prescribed burns, and the system for the specific needs of the area being monitored [5].
development of fire breaks. The plans should also One example of a mathematical model that can be applied
include a strategy for responding to fires when they to an automated system for monitoring and predicting forest
occur. and wildfires is the Rothermel model, which is a widely used
• Continuous Monitoring: Continuously monitor fire spread model. The Rothermel model is based on the
weather conditions, fuel moisture, and other factors physical processes involved in the spread of surface fires, and
that can influence fire behavior. This monitoring it takes into account factors such as fuel characteristics,
should be done in real-time to allow for immediate topography, and weather conditions.
action if conditions change. The Rothermel model uses a set of equations to estimate
• Response Planning: Develop a response plan that the rate of fire spread, fireline intensity, and other fire behavior
outlines the steps to be taken in the event of a fire. This
- 382 -
parameters. These equations take into account the following III. RESULT AND DISCUSSION
factors: Linking the Rothermel model to an automated forest and
Fuel Characteristics - The Rothermel model accounts for wildfire monitoring and forecasting system involves
the fuel characteristics of the vegetation being burned, integrating the Rothermel model into the overall system
including fuel load, fuel moisture content, and fuel particle architecture. Here are the steps involved in linking the
size. Rothermel model to an automated forest and wildfire
monitoring and forecasting system.
Topography - The Rothermel model accounts for the effect
of topography on fire behavior, including slope, aspect, and System design: The overall system architecture should be
the presence of ridges and valleys. designed to accommodate the Rothermel model. This may
involve selecting appropriate hardware and software
Weather Conditions - The Rothermel model accounts for platforms and designing the user interface.
weather conditions, including wind speed and direction,
temperature, humidity, and atmospheric stability. Data acquisition: The system should be able to acquire
data on fuel characteristics, topography, and weather
The Rothermel model is typically implemented in a conditions. This can be done through various means, such as
computer program that allows fire managers to input data on remote sensing, ground-based sensors, and weather stations.
fuel characteristics, topography, and weather conditions, and
then generates predictions of fire behavior, including fire Data processing: The data acquired by the system needs to
spread rate, direction, and intensity [6-8]. The model can be be preprocessed and organized for use in the Rothermel
used to develop fire management plans, to identify high-risk model. This may involve converting data to appropriate units,
areas, and to support decision-making during firefighting filtering out erroneous data, and calculating additional
operations. The following is the general formula for the parameters.
Rothermel model: Rothermel model implementation: The Rothermel model
needs to be integrated into the overall system architecture.
ERR= ρ*H**w**S**(1+β) () This may involve developing custom software or integrating
existing software tools into the system.
where:
Calibration and validation: The Rothermel model needs to
ERR is the energy release rate (kJ/m^2/s); be calibrated and validated for the specific area being
ρ is the density of the fuel (kg/m^3); monitored. This involves adjusting the model parameters to
match local fuel types and moisture conditions and comparing
H is the fuel bed depth (m); the predicted fire behavior to actual fire behavior observed in
w is the fuel moisture content (fraction); the field.
S is the wind speed (m/s); Visualization and reporting: The results of the Rothermel
model should be presented in a way that is easy for fire
β is a correction factor that accounts for the slope of the managers to interpret. This can include maps, graphs, and
terrain. other visualizations. The system should also provide reports
on fire behavior predictions, risk assessments, and other
The Rothermel model also includes equations to estimate
relevant information.
the rate of spread (ROS) and fireline intensity (FLI), which are
given by: Integration with other systems: The automated system
should be integrated with other fire management systems such
ROS=k*(ERR/ρ)n , () as fire detection and suppression systems. This will allow for
a comprehensive approach to fire management.
FLI=(ρ*H**w**c**(T-273))1.5*(ROS/60) () Overall, linking the Rothermel model to an automated
forest and wildfire monitoring and forecasting system requires
where: careful planning and design to ensure that the system is
k and n are empirical constants that depend on the fuel type reliable, accurate, and easy to use. The system should also be
and moisture content; regularly updated and maintained to ensure that it remains
effective over time.
c is the specific heat of the fuel (kJ/kg/°C);
IV. CONCLUSION
T is the ambient temperature (K).
We all know the harm and consequences of a mute enemy.
The Rothermel model is typically implemented in a One of the most pressing and difficult issues in the world is
computer program that allows fire managers to input data on the Prevention of fires in forest areas, forest countries, its
fuel characteristics, topography, and weather conditions, and prevention, protection from deforestation and deforestation.
then generates predictions of fire behavior, including fire Forest fires damage the economy of the area where the fire
spread rate, direction, and intensity. The model can be used to occurred, affected. Fire extinguishing costs can be a heavy
develop fire management plans, to identify high-risk areas, burden for locally allocated funds. To eliminate fires,
and to support decision-making during firefighting operations expensive equipment such as aircraft, helicopters, fire trucks
[9-10]. and tankers carrying cy is required. At the same time, it is clear
that biological restoration of the habitat will not be an easy
task either. To prevent the same damage, a system is being
developed that monitors, predicts forest and wildfires in real
- 383 -
time. The system provides information about the time of
occurrence of a fire, the level, the area of possible fire
coverage using space data. With the help of new software for
monitoring forests based on space data, it is possible to
significantly reduce the time of fire response and, as a result,
avoid large losses due to forest fires.
The work was carried out at the Research Institute of
Mathematics and Mechanics of the Kazakh National
University named after al-Farabi.
REFERENCES
[1] T.Zh. Mazakov., A.A. Sametova, A.S. Salimkhanova Monitoring the
development of forest and steppe fires // Bulletin of KazNITU, 2020.
№3. - P.175-179.
[2] Mazakov T.Zh., Kisala P., Jomartova Sh. A., Ziyatbekova G. Z.,
Karymsakova N.T. (2020). Mathematical modeling forecasting of
consequences of damage breakthrough. News of the national academy
of sciences of the Republic of Kazakhstan. Series of Geology and
Technical Sciences, 5(443), 116-124.
[3] T.Zh. Mazakov, A.A. Sametova. System for monitoring the
development of forest and steppe fires. VI - International Scientific and
Practical Conference "Computer Science and Applied Mathematics".
Institute of Information and Computing Technologies. Almaty, 2021.-
p.60-64
[4] Talgat Mazakov, Waldemar Wójcik, Sholpan Jomartova, Nurgul
Karymsakova, Gulzat Ziyatbekova, and Aisulu Tursynbai. The
stability interval of the set of linear system // Journal of electronics and
telecommunications, 2021. – Vol. 67. – No. 2. – Pp. 155-161. DOI:
10.24425/ijet.2021.135958
[5] G.A. Dorrer.Dynamics of forest fires. Novosibirsk: SO RAN, 2008. -
404 p.
[6] A. M. Grishin, Mathematical modeling of forest fires and new ways to
deal with them. / A. M. Grishin. - Novosibirsk: Nauka, 1992. - 408 p.
[7] G.M. Byram. Forest fire behavior. /G.M. Byram // Forest Fire: Control
and Use. - New York. -1959. - P. 113-126.
[8] I. P. Kurbatsky. Modern theory of the spread of forest ground fires. /
I. P. Kurbatsky, G. P. Telitsyn // Modern studies of forest typology and
pyrology. - Arkhangelsk. - 1976. - S. 90-96.
[9] Z.M. Mitchell. Rule of thumb determining rate of spread / Z.M.
Mitchell // Fire Control Notes. - 1937. - Vol. 6.-P.395-396.
[10] N.N. Krasovsky. Game problems about the meeting of movements /
N.N. Krasovsky. - M.: Nauka, 1970. - 420 p.
[11] G. A. Dorrer. Dynamics of forest fires. / G. A. Dorrer. -Novosibirsk:
Publishing house of SO RAN, 2008. - 404 p.
- 384 -
Phased Array Ultrasonic Inspection Model for
High-speed Train Wheels
Cha Jae Hwan, Kim Cheol Su*, Kang Jeong Hyung* and Kim Kyung Suk*
High-speed Vehicle Division at Korea Railroad Corporation, Republic of Korea
Safety Headquarters at Korea Railroad Corporation, Republic of Korea
Reliability Management Department at Korea Railroad Corporation, Republic of Korea
Email: [email protected], [email protected], [email protected], [email protected]
Abstract—Wheels is subjected to cyclical compression forces The purpose of this research is to study a model of
followed by slackening. This results in strains that can degrade phased array ultrasonic testing capable of detecting cracks
internal defects into fatigue cracks. This paper describes ultrasonic and defects that may occur both internally and externally in
model for Ultrasonic Testing-Phased Array to be demanded the the wheels of KTX trains running on high-speed tracks. The
wheel which is complicated in shape. scope of the investigation for the phased array ultrasonic
testing model for Monobloc Wheels of high-speed trains is
Keywords— ultrasonic inspection, phased array, Wheel Rim, as follows:
Wheel plate, array transducer, array probe, Wheel defect, Wheel
Scanning, focusing method, phased array ultrasonic testing First, Limited investigation focusing on the Monobloc
Wheel for KTX(drawing numbers TRP06572) and the series
INTRODUCTION of KTX-Sancheon(domestic developed high-speed trains).
Ultrasonic inspection is a non-destructive testing method Second, Selection of appropriate array probes for phased
designed to check the internal soundness of materials and array ultrasonic testing based on ultrasonic beam simulation,
detect defects on parts. The operations for wheels are extraction of phased array ultrasonic testing (UT-PA)
basically aimed at detecting internal defects that result in configuration parameters, and configuration of the scanning
fatigue cracks in the mass of the rim in service. A wheel rim method.
is basically inspected from the wheel tread.
Third, Standardization through the preparation of
However, finding defects locating on junction area, calibration test specimens and reference test specimens for
between rim and plate, it could incur wheel breakage with the equipment used in UT-PA.
catastrophic accident, needs simulation model for Ultrasonic
Testing-Phased Array (UT-PA). Through the research on the By introducing UT-PA technology, which is an advanced
ultrasonic detection model for Korea Train Express (KTX) non-destructive testing technique, for KTX and KTX-
wheels, we aimed to enhance the prevention of high-speed Sancheon high-speed train wheels, we have achieved the
train accidents in service and establish more objective and following benefits in terms of accident prevention and
reliable maintenance methods and procedures. operational safety:
Based on the technical documents of French National First, Establishment of a foundation for recording and
Railways (SNCF) using traditional perpendicular sounding managing ultrasonic detection results permanently using
methods, we established a non-destructive testing model digital image-based UT-PA methods.
using ultrasonic probe (or transducer) and phased array Second, The results of image-based UT-PA can be stored
ultrasonic testing advancements in software technology. For as digital files, enabling remote analysis and multi-party
complex-shaped Monobloc wheels, application of Phased analysis, enhancing objectivity and reliability by sharing the
Array Ultrasonic Testing allows more precise inspection. We interpretation results among ultrasonic testing operators and
introduce the results of this ultrasonic testing model. experts.
The inspection of ultrasonic detection presented in ISO Third, Ensuring reliability in the maintenance of high-
5848 focuses mainly on evaluation of rim area and in EN speed trains, thereby guaranteeing the safety of experts,
13632, the rim inspection is similar to ISO 5848. However, operators, and government officials involved.
when inspecting plate(or web) area inspections, considering
the curved shape, specific test specimens(or reference Fourth, Improving customer confidence in the safe
blocks) with artificial defects to detect flaws and procedures operation of high-speed railways, leading to enhanced
needed to be developed. Therefore, the design and customer satisfaction with railway vehicles.
procedures of the ultrasonic detection model are required.
We anticipate that the non-destructive testing technology
To inspect complex shapes accurately, such as curved of phased array ultrasonic testing for integral wheels of high-
plate areas of wheels, it is necessary to apply phased array speed trains will be expanded in the future to include axle
ultrasonic testing. Collaboration among operating companies, shafts, power transmission devices, axle bearings, and other
universities and specialized institutes is needed for the components of general railway vehicles and high-speed
development of professional models, including detection railway vehicles.
methods and specimen development.
- 385 -
THE REREARCH OBJECTIVES EN 13262 generally determines the internal soundness of
For this study, the investigation is limited to the phased- the wheel through ultrasonic inspection. The acceptance
array ultrasonic inspection model of the KTX wheel and the level of ultrasonic inspection for the rim area should not
integrated wheel of the KTX-Sancheon high-speed railway exceed the discontinuity indication representing internal
vehicle. The research objectives were set as follows: flaws, and it specifies the use of flat-bottomed holes with
diameters of 1 mm, 2 mm, or 3 mm (for wheels with
1. Reviewing domestic and international standard procedures operating speeds below 120 km/h). The plate area considers
related to ultrasonic inspection of high-speed vehicle the curvature when setting the sensitivity in the inspection,
wheels. and artificial dis-continuities (flat-bottomed holes) are
located on the surface for inspection. In this standard, the
2. Simulating ultrasonic beams using the CIVA ultrasonic
flat-bottomed holes in the plate area are machined
inspection program.
perpendicular to the plate surface and positioned at a depth
3. Selecting appropriate probes for phased-array ultrasonic of 5 mm from the surface or in the middle of the plate
inspection through simulation of ultrasonic beams. thickness, depending on the thickness of the plate. The basic
aspects of rim area ultrasonic inspection, such as contrast test
4. Confirming the insensitivity zone of the inspection blocks and sensitivity settings, are described in accordance
through simulation of ultrasonic beams. with ISO 5948.
5. Establishing a method for interpreting phased-array KS D ISO 5948 is a Korean standard that is a direct
ultrasonic inspection results. translation of ISO 5948:1994, which was published as the
6. Providing guidelines for the development of phased-array second edition. Although ISO 5948 was revised in 2018, the
ultrasonic inspection procedures. Korean standard does not reflect the revision.
KS R 9221:2021 includes "Ultrasonic Inspection" in
section 6.2.3. The content specifies that the inspection should
DOMESTIC AND INTERNATIONAL STANDARDS FOR be performed using ultrasonic waves with frequencies of 2
ULTRASONIC INSPECTION OF RAILWAY VEHICLE WHEELS MHz to 5 MHz ac-cording to KS B 0817. The timing and
The domestic and international standard procedures inspection area for ultrasonic inspection are limited to the
related to ultrasonic inspection of high-speed vehicle wheels rim area after heat treatment, and it is described simply as
are presented(see Table Ⅰ). These standards are intended for "there should be no harmful echo due to discontinuities."
Here, KS B 0817 is a standard for the general principles of
ensuring the quality assurance of newly manufactured
ultrasonic pulse re-flection and does not provide criteria or
wheels for high-speed vehicle wheels and not for
quantitative assessment standards for ultrasonic inspection.
guaranteeing the quality of wheels in use during operation.
However, since there are no specific standards for inspecting
wheels in use, the standard procedures for inspecting newly
manufactured wheels are being applied. SIMULATION OF ULTRASONIC BEAM FOR PLATE AREA
INSPECTION
TABLE I. RELATE INTERNATIONAL AND DOMESTIC STANDARDS Ultrasonic inspection of the plate area of high-speed
Standard
Revision vehicle integral wheels is only recorded in EN 13262, and
Category Standard Name Year(Estab this ultrasonic inspection should be performed by introducing
Number
lishment) ultrasonic waves perpendicular to the plate surface. However,
Railway rolling stock
ISO 5948 material - Ultrasonic
2018 such inspections are not possible when the wheels are
(1981) installed on the vehicle during operation. Therefore, to
acceptance testing
International inspect the wheels while they are mounted on the vehicle
Railway applications -
Standards
EN 13262
Wheelsets and bogies - 2020 during operation, simulations were conducted considering
Wheels - Product (2004) the restriction that ultrasonic waves should be introduced
requirements
from the back surface of the wheels.
Rolling stock materials
KS D ISO for railway vehicles - No revision 1. Input information for simulation (Probe: 5L32-A11)
Domestic 5948 Ultrasonic inspection (2002)
Standards test methods - Frequency: 5 MHz
Railway vehicle 2021 - Total array elements: 32 elements
KS R 9221
wheels (2000)
- Array pitch: 0.6 mm
- Active aperture array elements: 32 elements
International standard ISO 5948 provides detailed
information on ultrasonic inspection of the rim area of - Ultrasonic velocity: Longitudinal wave - 5,920 m/s
railway vehicle wheels and the axle (hollow and solid) of 2. Simulation of plate area inspection
railway vehicles. This standard defines the size and position
of the reference reflector to be included in the calibration test a) Perform S-scan simulation for the range of incident
block (contrast test block) for phased-array ultrasonic angles from 0° to -20° by contacting the array probe on
inspection of wheels. It also specifies the diameter of the flat- the back surface of the wheels.
bottomed hole used for setting sensitivity and establishing b) This is done to send ultrasonic waves to the apex of the
acceptance levels for flaw detection as 3 mm (for high-speed convex part of the plate area.(Position the array probe at
wheels, 2 mm is used, subject to inquiry or order). The the right side of the back surface 60 mm and perform 0°
specific size to be used within this range is determined by the to -20° S-scan)
ultrasonic inspection technique specification.
- 386 -
e) Simulations The initially received signals in the
ultrasound signal simulation for the notches depicted in
Figure 2 are the signals that are reflected and mode-
converted from the curved shape of the plate.
f) It is necessary to appropriately adjust the data signal
window to avoid collecting such mode-converted
signals.
g) Regardless of the depth of the notch, the ultrasound
signal with an angle of 14 degrees has the highest
amplitude, and the signal magnitude increases with an
increase in the notch depth.
TABLE II. ULTRASOUND SINALS FOR NOTCH DEPTH
Notch Depth(mm) 2 4 6
Increase in Signal
Reference Signal +12.2 +16.5
Amplitude (dB)
(a) Scan plan (b) Simulation results of ultrasound beam profiles
ULTRASOUND BEAM SIMULATION RESULTS
Fig. 1. Inspection of the wheel plate area
Ultrasound beam simulation results Ultrasound
inspection of the wheel rim area of high-speed vehicles is
c) The vertices of the convex area of the wheel plate are the performed by conducting vertical scanning in the axial and
points where the highest stress occurs during operation. radial directions, as shown in Figure 3(a). Based on this, a
simulation was conducted to determine the accessible area
d) Simulations were performed on ultrasound signals for assuming the inspection can be performed when the vehicle
artificial notches located at the vertices of the convex has completed its operation and the wheel assembly is
area in the middle of the plate. The notches have a installed on the axle, as shown in Figure 3(b).
length of 3 mm, width of 1 mm, and depths of 2 mm, 4
Furthermore, ultrasound should be incident in the vertical
mm, and 6 mm. (see Table Ⅱ).
direction on the surface of the plate area, as shown in Figure
3(c). However, such inspections are not possible while the
vehicle is in operation and the wheel is installed.
To inspect the wheel of a vehicle in operation, ultrasound
needs to be incident on the wheel surface. Therefore,
simulations were conducted on phased-array ultrasound
inspections with phased array probes contacting the wheel
surface, assuming various incident angles ranging from 0° to
-20°, as shown in Figure (d).
(a) Scan plan
(a) Scan plan (b) Simulation results of
ultrasound beam profiles
(b) Simulation results of ultrasound beam profiles
(c) Contact of phased-array (d) Simulation results of
probes on the wheel surface ultrasound signals with a notch
(0° to -20° longitudinal waves) depth of 2 mm and 4 mm
Fig. 3. Ultrasound Beam Simulation results - Rim and Plate area
(c) Simulation
Fig. 2. Inspection of the wheel plate area - Simulation results of
ultrasound signals
- 387 -
CONCLUSION
In this study, ultrasound simulations were performed REFERENCES
using the CIVA software, and based on the simulation results,
the feasibility of phased-array ultrasound inspection for
wheels was evaluated, and the inspection settings were [1] ISO 5948 “Railway rolling stock material – Ultrasonic acceptance
testing”, 2018
determined. Additionally, the calibration of equipment and
[2] EN 13262. “Railway applications - Wheelsets and bogies - Wheels -
setting of inspection sensitivity were performed by reviewing Product requirements”, 2020
and designing calibration test pieces and contrast test
[3] KS D ISO 5948 “Rolled materials for rolling stock - Ultrasonic
piece(reference block) before conducting the inspection. inspection test method”, 2017
These are summarized as follows:. [4] KS R 9221 “Wheels for railway vehicles”, 2021
1. For the inspection of the wheel rim area, it is [5] Railroad Equipment Construction Specification KRCS B 283 “High-
recommended to use a phased-array probe with a speed vehicle integrated wheel”
pitch of 1 mm, 32 array elements, and a frequency of [6] Korea Railroad Corporation, Phased Array Ultrasonic Inspection
(UT-PA) Procedure for Wheels of High Speed Vehicles (KTX, KTX-
5 MHz (such as the 5L32-A32 probe). Sancheon)
2. For the inspection of the wheel plate area, it is
preferable to use a phased-array probe with a pitch of
0.6 mm, 32 array elements, and a frequency of 5 MHz
(such as the 5L32-A11 probe).
3. The inspection of the wheel rim area is per-formed
with eight active elements, and the scanning is carried
out with 0° E-scan and 10° E-scan according to the
contact area.
In the future, it is necessary to verify the phase-array
ultrasound inspection sensitivity for the wheel plate region
using echoes generated by notches on the test specimen and
experimentally validate the detectable notch size. When
performing the inspection with a phased array ultrasonic
probe at a fixed probe position, it is observed that there are
regions of inspection insensibility due to the inability of
phased array ultrasonic inspection or the exclusion of signal
collection caused by mode conversion.
- 388 -
ICT Convergence Roof Installation Solar
Location Tracking Solar Power Generation
System
DeokGun Woo 1* , HyoJai Lee 2, ChoonWon Seo 3
1
Department of Intelligent Machinery, Korea-polytechnics, Rep. of Korea
2
Department of AI Automation, Korea-polytechnics, Rep. of Korea
3
Department of Development & Production Team, Neo Solartec Inc., Rep. of Korea
Email: [email protected]
Abstract—Starting with the Renewable Energy Mandatory II. ROOF-TYPE SOLAR TRACKING POWER GENERATION
Allocation System (RPS) in 2012, the spread of renewable energy SYSTEM STRUCTURE
has expanded, and the use rate of renewable energy is improving
worldwide through the RE100 (Renewable Electricity 100) A solar tracking system was applied to increase the power
campaign. Among them, solar power generation, which is less generation efficiency of solar panels. Solar tracking affects
affected by environmental destruction, is in the spotlight, and in power generation efficiency. The power generation efficiency
particular, a solar power generation system installed on the roof of the short-term solar tracking system and the fixed solar
of a building is in the spotlight without the use of a large space. power system to be applied in this study was referred to, and
However, the power generation efficiency is lower than that of it was confirmed that the power generation efficiency of the
other solar power generation facilities, and in residential areas, short-term solar tracking system increased compared to the
secondary damage may occur due to damage to solar panels. In fixed solar power system[4]. for accurate solar tracking, in
order to solve this problem, we propose an ICT convergence type addition to adding a separate sensor for solar tracking, a
short-term solar power generation system that has a simple tracking method using data from the Korea Astronomy and
structure and can increase solar power generation efficiency. Space Science Institute was compared and analyzed for solar
tracking [5].
Keywords— Solar energy generation, Solar tracking frame
structure, Rooftop, Solar tracking system, Structure. In case of the method of changing the angle of the solar
panel, the two-axis solar tracking method has a higher
I. INTRODUCTION efficiency than the one-axis method. However, the two-axis
The International Energy Agency (IEA) has predicted that solar tracking method is structurally complex because it
solar power will lead the market as the cheapest renewable requires two drive shafts, and it is burdensome to the weight
energy in its World Energy Outlook 2020 report. In line with when installed on the roof. Therefore, we selected the one-axis
this prediction, a wave of green energy is blowing, and tracking method, which is structurally simple and easy to
measures are being taken to achieve carbon neutrality around install and maintain on the roof, and safe in terms of weight,
the world, and solar power is the focus of attention[1]. The although it has a lower efficiency than the two-axis tracking
Korean government also joined the global carbon neutrality method. For the field test of the one-axis tracking method, we
movement and made it mandatory for public buildings to designed a 3kW solar power system with fixed and one-axis
obtain zero-energy building certification under the Green solar tracking methods and installed the power system on the
Building Support Act. From 2025, private buildings with a roof of a real building. Regardless of the angle of the building
floor area of more than 1,000 square meters, all public roof, the frame structure was designed as shown in Figure 2
buildings with a floor area of more than 500 square meters, and the driving unit adopted a worm gear structure in
and apartment buildings with more than 30 households will be consideration of the load and maintenance of the building roof
built as zero-energy buildings. From 2030, all buildings with while maintaining an angle for optimal power generation
a floor area of more than 500 square meters will be zero- efficiency. The worm gear structure has the advantage of
energy buildings[2]. Solar power is infinite, anyone can use being able to adjust the angle of the solar panel with relatively
it, and it has a high utilization rate because it can be installed little force due to its high deceleration ratio, and also to
close to life, as it has less noise than other renewable energy. prevent reversal, so it can maintain the angle even if there is a
In particular, power generation through roof-top solar panel load on the solar panel due to wind.
installation in residential areas is possible to utilize idle space
without allocating separate land, and it is free from regulations
on distance between facilities.[3]. This study conducted
research on a solar tracking power generation system that can
be installed on the roof of a residential area without allocating
a separate land area. ICT technology was also applied to
minimize the damage of solar panels from natural disasters in
residential areas.
Fig. 1 One-axis solar tracking frame structure and driving unit
- 389 -
III. SAFETY MEASURES FOR ROOF-TYPE SOLAR TRACKING IV. CONCLUSION
SYSTEM
Roof-type solar tracking power generation systems can be In the context of the global carbon reduction issue and the
vulnerable to natural disasters because they are additionally research on various renewable energy, we have conducted a
installed on the top roof of the building, and secondary study to improve the efficiency and safety management of
damage can occur if solar panels are damaged due to such roof-mounted solar power systems, which are the most
natural disasters. Therefore, in this study, ICT technology is popular solar power systems that are closely related to life.
applied to the roof-type solar tracking power generation
system to prevent damage to the solar power generation Roof-mounted solar power is a field that is expected to
system in advance. As shown in Figure 3, in order to prevent continue to grow due to factors such as the fact that it can be
damage to solar panels from natural disasters caused by strong installed without additional land and the phased mandatory
winds and snow, information on strong winds and heavy snow implementation of zero-energy buildings. In particular, in
was collected through weather information and an avoidance Korea, where there are many mountains and it is difficult to
driving method was implemented accordingly. In addition, it secure large pieces of land, the roofs of residential areas and
was intended to maximize the efficiency of the solar power industrial complexes have high potential for development.
generation system through regional and seasonal information However, the existing fixed method had a problem that the
such as sunset and sunrise by the Korea Meteorological efficiency was not high compared to the installation cost. In
Administration. this study, we solved this problem by using a short-axis solar
tracking system that is advantageous for application on the
roof. In addition, we proposed an ICT-integrated solar power
system that can prevent damage to solar panels and secondary
damage caused by it by using information from the Korea
Meteorological Administration, such as seasonal sunrise and
sunset information.
REFERENCES
Fig. 2 Avoidance drive for solar panel damage and power generation
efficiency increase [1] “World Energy Outlook,” International Energy Agency, 2020.
[2] J. W. Kim and S. K. Park, “A study on the establishment of statistics
for the package of building energy efficiency improvement policy,”
Korea Energy Economics Institute, June 2021.
Based on the Korea Meteorological Administration's [3] W. Moon, “Trends for separation distance regulations of photovoltaic
regional forecast, a scenario was created for heavy snow and facilities,” The Korean Institute of Electrical Engineers, pp. 23-27, June
strong winds that are expected to be damaged by weather. 2022.
When a snow forecast occurs, the solar panel's angle was [4] H. K. Song, K. S. Lee, and Y. S. Choi, “Comparison of generation
adjusted to the wind's angle to reduce damage. amount and operating time for fixed-concentrated type and single axis
trace type of photovoltaic”, Journal of the Korean Institute of Electrical
The strong wind avoidance mode to anticipate damage to and Electronic Material Engineers, vol. 28, no. 11, pp. 743-747,
the roof-type solar power generation system enables November 2015.
avoidance maneuvers when the wind speed over 25m/s, and [5] J. H. Hong and E. S. Kim, “Tracking method of inclination-dependent
the snow cover avoidance mode of the photovoltaic panels 2-axis solar tracker,” Journal of the Institute of Electronics Engineers
of Korea, vol. 50, no. 11., pp. 180-187, November 2013.
enables avoidance maneuvers when the snow cover over 3cm
above the solar panels. The system was developed, and Figure
3 is the avoidance maneuver algorithm and test by strong wind
and snow cover.
Fig. 3 Avoidance drive algorithm and field test
- 390 -
Study on the collapsing of artificial retaining
wall facility using acoustic emission
Kwang Bok Kim 1, Bong Ki Kim 1, Woon Kyung Kim 1, Tae Hun Kim 1,
Sung Hun Park 1, Saurabh Kumar 1, Jun Gil Kang 1 and Kyung-Wan Koo 2*
1
Integrity Diagnostic Korea Co. Ltd., Republic of KOREA
2
HoSeo University, Republic of KOREA
Email: [email protected]
Abstract- This paper is a study of the process of retaining wall preamplifier and amplified with a gain value of 40dB. The
collapse using the AET technique. Pile type AE Sensor specially AET system, equipped with two PCIe slot-mounted DAQ
designed to collect AE data during the collapse of the retaining wall boards (8-channel type), measured and automatically saved
was manufactured. The artificial reinforced retaining wall was made the signals flowing from the nine AE sensors to a PC. The
to be 27 square meter and height 3 meter. It consists of reinforced
soil and concrete blocks that are most commonly used for reinforced
data were collected at a maximum speed of 10Msps for each
soil retaining wall. The pile type sensor was installed vertically in sensor. The collected AE data were analyzed using the AE
retaining wall at a depth of 1 m and connected to the AET system Studio software developed by IDK, after removing
using a cable. AE data was collected and AE parameters were environmental noise. The timing control function values
converted by collapsing into forklifts. The signal strength in AE during data collection were set according to Table 1.
parameters showed the best AE characteristics during the collapse
process. After applying the 2D CNN model of artificial intelligence,
85% of the five-step collapsing process could be confirmed.
Keywords—Acoustic emission testing, retaining wall, data
collection, prediction
I. INTRODUCTION
The application of inclinometers, subsurface Fig. 1. Retaining wall testbed
inclinometers, displacement gauges, drones, and imaging
techniques in the prediction of wall and slope collapse has Table 1: AE Data collection parameters
been increasing. Real-time digital data conversion is
Parameter Value Parameter Value
necessary to enable the transition to early warning systems by
detecting small ground movements. To develop a prediction P/A gain (dB) 40 HDT (us) 400
technique for wall collapse using acoustic emission testing sampling rate
(AET), a pile-type AE sensor was developed [1]. This sensor 10 PDT (us) 200
(MHz)
is a real-time system that can detect ground movements by
drilling and inserting it deep into the ground and transmitting Pre-trigger 400 MHD (us) 10,000
micro-elastic waves to provide warnings of the collapse point. Threshold (dB) 30 HLT (us) 4,000
In this study, a reinforced soil wall was constructed in an
outdoor test bed as a preliminary step of empirical research, HPF (kHz) 50 Noise AVG TC (us) 100,000
and AE data related to the collapse process were collected to
LPF (kHz) 300 Signal AVG TC (us) 100,000
establish a collapse prediction based on AE feature values.
II. MAIN BODY B. characteristics during wall collapse
The AET system was installed in the outdoor test bed and
A. Reinforced retaining wall
observed using a real-time web monitoring system. CCTV
Reinforced soil walls were constructed in a designated cameras were installed for safety reasons, considering
area, with two sets of 2m x 2m and one set of 3m x 3m as potential hazards, and environmental data were collected
shown in Fig. 1. The exterior was made of concrete blocks, during the summer season, including typhoons, rainfall, wind,
and three types of reinforcing soil were compacted inside. etc. Subsequently, during the winter season and thawing
Pressure gauges were installed to measure the overall pressure period, the wall was artificially collapsed, and AE data were
of the soil. After soil compaction, nine 1m-long pile-type collected and analyzed in March 2023.
sensors were installed at 1m distance from each other. These
pile-type sensors were designed to detect micro-displacement- a) Environmental data and AE
induced elastic waves by attaching piezoelectric ceramics to The relationship between AE signal strength values and
the sensor line, which resonates at a maximum resonance at precipitation, wind and temperature were investigated. The
60kHz. AE signal strength values increased with precipitation, as the
a) AE Data collection top part of the pile-type sensors did not submerge entirely into
the reinforced soil. However, the AE values showed little
The acoustic emission testing collection device (IDK-
variation with wind and temperature.
PU16) developed by IDK was waterproofed and placed in a
waterproof box for real-time data collection. The data
collected from the file-type sensors were amplified through a
- 391 -
Fig. 2. AE data variation according to precipitation, wind and Fig. 4. Angle sensor and AE data relation
temperature
Figure 5 shows the trend of hit count intervals for
b) Wall collapse process generating the H index every 5 seconds during the collapse
process. The collected hit rates for each of the ten intervals
Figure 3 illustrates the images of the wall collapse process.
were used to calculate the HS index using Equation (1).
The wall was artificially collapsed using a forklift crane with
an 800kg load. AE data were collected throughout this
process.
Fig. 5. HS index for retaining wall collapse
Fig. 3. Retaining wall collapse procedure
As shown on the right side of Figure 5, the HS index
During the wall collapse, the signal strength data from provides a comprehensive view of the wall collapse process,
AE showed significant increases each time the forklift crane allowing prediction of the collapse process into six stages,
made contact with the wall and exerted force during the five labeled A to F.
stages of collapse, corresponding to the movement of the
reinforced soil. Figure 4 depicts the relationship between
slope displacement and AE data by installing an inclinometer III. CONCLUSION
sensor. The lower part of Figure 4 shows that when stress is
The entire process of wall collapse was performed in an
applied to the wall due to forklift crane contact, the initiation
outdoor test bed using acoustic emission technology. Despite
of horizontal loading, and the completion of horizontal
the influence of precipitation, wind, and temperature,
loading, the AE data increase vertically. This becomes
environmental factors were easily eliminated during AET
essential data for developing integrity indicators for wall
data collection. Clean AE data collected during the collapse
collapse.
process were secured by utilizing the AET system,
c) HS Index commercially developed by IDK, from collection to signal
The zonal intensity method evaluates the structural processing. The collapse algorithm was completed using AE
severity of defects by determining the intersection of two data from different collapse stages. Additionally, an integrity
independent parameters: the history index (HI) and severity index was proposed and developed, which is expected to
index (SI) [2]. contribute significantly to the future development of slope
collapse prediction technologies.
𝑁 ∑𝑁 𝑆𝑜ⅈ 1 ACKNOWLEDGEMENT
𝐻(𝑡) = 𝑁−𝐾𝑡=𝑘+1
∑𝑁 𝑆𝑜ⅈ
𝑆𝑎𝑣 = 10 ∑ⅈ=10
ⅈ=1 𝑆𝑜𝑖 (1)
𝑖=1 This work was supported by Korea agency for Infrastructure
Technology Advancement (KAIA) grant funded by the
Where N is the hit count up to time t, Soi is the signal strength Ministry of Land, Infrastructure, and Transport (MOLIT)
of the i-th hit, and K is defined as follows: (1615011892), Proof Test of real-time prediction technology
If N > 1000, K = N - 200 for retaining wall facility collapse using smart AET
If J ≤ N ≤ 1000, K = 0.8N technique).
If N < J, historical index not applied References:
J is a specific event count, which may vary depending on the [1] ASNT, NDT Handbook Vol 6, 315p
test material. [2] ASNT, NDT Handbook Vol 6, 121p
- 392 -
Technique for the use of matching circuits in
the transmit-receive path of the uav radio control
channel in conditions of icing of the antenna
device
Nurbolat Ondyrbayev1*, Nursultan Nurzhauov2, Anna Sutiko3, Pavel Boikachev4,
, Katipa Chezhimbayeva5, Asel Yerzhan6
1, 2, 5, 6
Almaty University of Power Engineering and Telecommunications named after Gumarbek Daukeev, Kazakhstan
3, 4
Military Academy of Belarus, Belarus
Email: [email protected]
The results of experimental studies of the antenna impedance technical characteristics [1]. Regardless of the methods used,
of the control channel, an unmanned aerial vehicle in icing the first stage of the synthesis of matching circuits is to study
conditions is presented. A mathematical model of the antenna of and set the initial properties of the matched load. As a rule, the
the control channel of an unmanned aerial vehicle in icing input characteristics of the matched load are presented in the
conditions and without them is presented. One of the approaches form of discrete points of the real and imaginary components
to the representation of the characteristics of the RTAs that are of the function describing the considered parameter of the
part of the RTAs (amplifiers, frequency filters, AD, etc.) is complex resistance – [2]. Such a representation is not always
analytical mathematical modeling based on the approximation of
informative enough and is not applicable for use in many
the results of experimental data. Such a mathematical model is
modern methods of synthesis of broadband matching and
based on a comprehensive analysis of the behavior of the system
and extensive use of the results of previously conducted statistical
frequency-selective devices that require equivalent circuits or
studies. The method of synthesis of broadband matching devices analytical representations of matched loads [3].
adapting radio engineering systems to varying load impedance in As a rule, the impedance characteristics of the matched
a wide range of natural conditions of their operation is shown. A loads are represented by numerical series of values of two
broadband matching circuit has been synthesized for the antenna quantities: independent (yi) and dependent (xi), each of which,
device of the control channel of an unmanned aerial vehicle in in general, in addition to a certain regular (deterministic)
icing conditions, which provides a gain in the potentially
component, may contain random components of a very
achievable radio line range of up to 25% relative to the antenna
different nature. This is due to both the statistical nature of the
device without a matching circuit.
processes under study and a variety of external factors of
Keywords – unmanned aerial vehicle, icing, broadband measurement and data transformation processes (noise,
matching, invariant, sensitivity, impedance, mathematical interference, destabilizing factors and measurement errors). In
modeling, synthesis technique. our case, these are temperature effects and icing. At the same
time, according to the central limit theorem, it is assumed that
I. INTRODUCTION the values of the random component of the dependent variable
Modern unmanned aerial vehicles (UAVs) are a complex (both own and "total") are distributed according to the normal
system that, depending on the tasks being solved, includes probability law [6].
control subsystems, navigation, reconnaissance, etc. These Based on this, [7] presents a methodology for determining
subsystems, being consumers or sources of information, have the required sample size N (the number of full-scale, semi-
antenna devices (AU) in their composition, providing control natural or model experiments conducted), in which the
of UAVs, exchanging video data with the operator, etc. The mathematical expectation of a particular characteristic is
impedance (input complex resistance) of many AU strongly determined with the specified values of reliability indicators
depends on disturbing influences [1]. The presence of various (characterized by the value of confidence probability) and
kinds of disturbing influences is especially important for accuracy (characterized by the value of relative error).
mobile systems. A striking example of such systems are
UAVs. A change in the impedance leads to a loss of power of Table I shows the required sample size for estimating the
the transmitted or received signal, which is the reason for the mathematical expectation of a normally distributed
lack of communication between the UAV and the control characteristic with a coefficient of variation equal to one.
point. Given the operating conditions of the UAV, one of the
most likely external influences are strong temperature TABLE I. THE REQUIRED SAMPLE SIZE TO ESTIMATE THE
MATHEMATICAL EXPECTATION WITH A COEFFICIENT OF VARIATION EQUAL
fluctuations and icing. TO ONE.
II. COORDINATION OF THE AD OF THE UAV The value Relative error value, %
of the
CONTROL CHANNEL AND THE RECEIVING AND confidence 30 25 20 15 10 5 1
TRANSMITTING MODULE probability
Modern methods of broadband matching are characterized 0,5 5 7 11 20 46 182 4556
0,6 8 12 18 31 71 282 7056
by significant opportunities for the development of radio
0,7 12 17 27 48 108 433 10820
engineering devices (RTU) for various purposes with high
- 393 -
0,8 18 26 41 73 164 657 16440 m1 = a0 + a2 s 2 + ... + a2 k s 2 k - the even part of the numerator
0,9 30 43 68 120 272 1080 26990
0,95 43 62 96 171 384 1537 38420 of the function f (s) ;
0,99 73 106 166 295 664 2654 66360
m2 = b0 + b2 s 2 + ... + b2 q s 2 q
- the even part of the
denominator of the function f (s ) ;
The totality of the data characterizing the RTU is
processed, i.e., the data obtained are presented in a definite n1 = a1s + ... + a(2 k −1) s (2 k −1)
and balanced system, which becomes the basis for further - the odd part of the numerator
meaningful analysis, interpretation and scientific conclusions, of the function f (s ) ;
and practical recommendations. As a result, we obtain data
from experimental studies of the RTU characteristic under n2 = b1s + ... + b(2 q −1) s (2 q −1)
- the odd part of the denominator
consideration (in the form of a table or graph), in which each of the function f (s ) .
value of the input factor (Xm) corresponds to the value of the
output factor (Ym), i.e. a pair of numbers (хmi, ymi) All coefficients for s must be real and non-negative, and
corresponding to a point estimate of the mathematical the higher powers of the polynomial of the numerator (k) and
expectation of the estimated characteristic is known – a denominator (q), as well as their lower powers, cannot differ
regular component of experimental data. As an example, by more than 1; zeros and poles of this function must alternate
Figure 1 shows the processed results of measurements of the on the imaginary axis of the plane s [11].
impedance of the pin antenna of the UAV control channel
For 2N-poles, the function f(s) must have the properties of
without icing and as a result of icing.
the input resistance Z(s) of these two-poles. The input
In practice, to solve engineering problems, the function functions of such four-poles belong to the class of so-called
values are also needed at other points other than those positive real functions (PRF), which satisfy the following
specified in the table. In this case, it is necessary to resort to additional condition: Re[f(s)] 0 or Re[Y(s)] 0 by s 0.
interpolation (or interpolation) – approximate finding of
unknown values of functions by its known values at specified Also, it is necessary to consider that positive real functions
points. always represent the ratio of two Hurwitz polynomials, i.e.,
the zeros and poles of such functions are located in the left
half-plane. In addition, if the PRF has poles or zeros on the
imaginary axis (including s = 0 and s = ∞), then these poles
and zeros are real and positive.
Based on the limitations presented above, to solve the
problem of approximation of the impedance characteristics of
complex loads, it is proposed to use a fractional rational
function of the form (1) as an approximating function [4].
Using a mathematical transformation, we will represent
expression (1) in the form even Ev f (s) and odd Od f (s) parts
from f ( s) :
Fig. 1. Graphical representation of the processed results of measuring the mm −nn
impedance of the pin antenna of the UAV control channel. 1, 3 – the real Ev f ( s) = 1 22 12 2 ;
part of the AD impedance without icing (under icing conditions); 2, 4 - the m2 − n2 (2)
imaginary part of the AD impedance without icing (under icing conditions).
n1m2 − m1n2
Od f ( s ) = .
Thus, the result of processing experimental data is a m22 − n22 (3)
reference mathematical model of the object of study,
presented in the form of an interpolation (with multiple By s = iω we have:
experiments) or smoothing (with a single experiment) cubic
Ev f ( s ) s=iω = Re f (iω)
spline, which most accurately describes the results obtained ;
during experimental studies and satisfies the requirements and Od f ( s ) s=iω = i Im f (iω)
limitations imposed on curves and surfaces describing .
functional dependence experimental studies. However, it is
Expressions (2) and (3) are proposed to be used as
necessary to pay special attention to the fact that such a
approximating functions and components of the input and
reference mathematical model cannot be used to solve various
transfer characteristics of the object of study. To find the most
engineering problems, because it does not correspond to the
optimal coefficients of approximating functions, it is
physical nature of the modeled process, which has the
necessary to use methods of nonlinear optimized procedures
following limitations [8].
using computers (for example, the Levenberg-Marquardt
Based on the conditions of physical feasibility [9], the optimization method [11], [12]). It allows you to form an
impedance characteristics of RTU are fractional rational analytical mathematical model (AMM) of the input and
functions of the form: transfer characteristics of the RTU with a high degree of
adequacy (with a confidence probability of at least 0.9 and a
a0 + a1s + a2 (s ) 2 + ... + ak (s ) k
f (s ) = , relative error of no more than 10%) [7].
b0 + b1s + b2 (s ) 2 + ... + bq (s ) q
(1) Based on the above, the technique for approximating the
Where impedance characteristics of radio engineering devices has the
following form (Figure 2).
s = iω - complex variable;
- 394 -
The essence of the technique is as follows: a3 52.215 b1 1.382 b5 1.187
• The requirements and limitations for the implemented
mathematical model are set. According to the
requirements, the required number of measurements TABLE III. COEFFICIENTS OF THE FRACTIONAL-RATIONAL FUNCTION
APPROXIMATING THE IMPEDANCE CHARACTERISTICS OF THE AD IN ICING
of the characteristics of the mouth is determined CONDITIONS
according to Table I;
Ratio Meaning Ratio Meaning Ratio Meaning
• After carrying out the required number of
a0 93.299 a5 0.407 b3 0.932
measurements, experimental data are processed and a
reference model of the object of study is formed a1 11.938 a6 45.561 b4 6.612
(transition from stochastic data to deterministic data), a2 267.507 b0 1.707 b5 0.461
which will allow to assess the quality of
a3 15.725 b1 0.455 b6 2.398
approximation;
a4 222.84 b2 5.907
• After that, considering the restrictions imposed on the
function describing the impedance characteristics of
the RTU, an approximating function is formed, the The frequency characteristics of the approximating
adequacy of which is estimated by calculating the functions of the input resistance of the AD are shown in Figure
absolute error relative to the formed standard. It 3.
should be noted that it is possible to find the error by
various methods presented in [10].
As an example, we investigate the possibility of
approximating the frequency characteristics of the AD using
the algorithm shown in Figure 2. The obtained AMM,
describing the real and imaginary components of the input
resistance of the AD without icing and under icing conditions,
have the form (1), the coefficients of which are presented in
Table II and III. a)
b)
Fig. 3. Functions of the input resistance of the AD without icing (a) and
under icing conditions (b) (points) and its calculated AMM (solid lines)
in the operating frequency band.
Figure 3 clearly shows that the calculated AMM of the
input resistance of the AD provides the required absolute
approximation error to the characteristics of the input
resistance and is less than 10%.
This approach allows not only to analyze any changes
in the load impedance depending on the operating conditions,
but also to apply various methods and methods of broadband
matching, which allow to ensure the maximum level of power
transmission when the operating conditions of the matched
device change.
In order to ensure the maximum level of power
Fig. 2. Method of approximation of impedance characteristics of radio transmission between the AD and the receiving-transmitting
engineering devices. module in icing conditions and without it, according to the
method of synthesis of MC [13], the block diagram of which
TABLE II. COEFFICIENTS OF A FRACTIONAL-RATIONAL FUNCTION is shown in Figure 2 and the method shown in Figure 4, a
APPROXIMATING THE IMPEDANCE CHARACTERISTICS OF AN AU WITHOUT
ICING ladder structure MC was developed (Figure 5), operating in
the range from 415 to 449 MHz with elements of a number of
Ratio Meaning Ratio Meaning Ratio Meaning nominal values E24. The expansion of the frequency range is
a0 94.617 a4 219.477 b2 0.279 due to the displacement of the resonance of the AD during
icing.
a1 37.184 a5 0.916 b3 3.282
a2 302.596 b0 0.359 b4 0.168
- 395 -
Fig. 6. The results of the experiment: a) in conditions of icing; b) without
icing; c) in the form of dependencies of CaVM on frequency.
REFERENCES
[1] Filippovich, G.A. Broadband matching of resistances // G.A.
Filippovich: monogr. / G. A. Filippovich. - Minsk: MA RB, 2004. -p.
175.
[2] Kaganov, V.I. Automatic control systems in radio transmitters / V.I.
Kaganov- M.: Svyaz, 1969. - 256.
[3] Sviridenko, A.A. The technique of broadband matching of loads with
concentrated parameters based on a generalized scattering matrix /
A.A. Sviridenko, M.A. Yantsevich // Information technologies in
education, science and production: VI International Scientific and
Technical Internet Conference, November 17-18, 2018[Electronic
resource]
Fig. 4. The method of synthesis of MC for broadband RED s with varying [4] Isaev V. O., P. V. Boikachev – a Way of finding an appropriate
load impedance. mathematical models of the wireless devices with the unstable
impedance // "Metrology and instrumentation" №1(92) 2021. – pp. 9–
16.
[5] Isaev V. O., P. V. Boikachev Analytical mathematical modeling of the
input characteristics of radio systems // "problems of physics,
mathematics and technology" №3(48) 2021. – pp. 88–95.
[6] Wentzel, E. S. Probability theory: ed. 10 / E. S. Wentzel. - M.:
Publishing Center "Academy", 2005. -p. 576.
[7] Kosachev, I. M. Methods of calculating the reliability and accuracy of
Fig. 5. The stair structure consisting of five elements. а – schematic the estimated tactical and technical characteristics of weapons, military
diagram; b – the appearance of the MC. and special equipment / I. M. Kosachev, D. S. Nefedov // Vetsnik of
the Military Academy of the Republic of Belarus. – Minsk, 2015. –
№1(46). – p. 107-135.
It should be noted that the implementation of the
[8] Isaev, V. I., Dubovik, I. A., Boikachev, P. V., Sutiko, A. A.
schematic diagram (Figure 5 a) in the protective housing Mathematical model of radio devices / V. A. Isaev, I. A. Dubovik, P.
(Figure 5 b) has insignificant dimensions and weighs 8 grams, V. Boikachev, A. A. Sutiko // II international Scientific and Practical
which is acceptable for use in the UAV control channel path. Conference "Endless Light in Science", the city of Nur-Sultan, 6-8 may
2020, Kazakhstan, 2020. – p. 27-33.
The use of SC in the receiving and transmitting path of [9] Carney, S., circuit Theory. Analysis and synthesis. - M. "Svyaz", 1973.
the UAV control radio channel provides the power -p. 269.
transmission level shown in Figure 6, where the solid line [10] Lanne, A.A. Optimal synthesis of linear electrical circuits / A. A.
shows the dependencies with the synthesized MC, and the Lanne. - M.: Svyaz, 1969. -p. 294.
dashed one without it. [11] Levenberg K. A method for the solution of certain problems in least
squares // Quart. Appl. Math., 1944. № 2. pp. 164-168.
After analyzing the dependencies presented in Figure 6
[12] Marquardt D. An algorithm for least-squares estimation of nonlinear
(c), it can be concluded that the use of MC provides a high parameters // SIAM J. Appl. Math., 1963. No. 11. pp. 431-441.
level of power transmission from the signal source to the load [13] Boikachev, P.V., Method of synthesis of broadband devices with
(reducing losses from 38% to 13%), both in normal conditions optimal characteristics of the power conversion coefficient matching
and in conditions of icing of the AD, which makes it possible the load impedance varying in time / I. A. Dubovik, P. V. Boikachev,
to provide a UAV control channel with stable radio V. O. Isaev // [Electronic resource] Journal of Radioelectronics. - 2021.
communication. - No. 5. - Access mode: https://doi.org/10.30898/1684-1719.2021.5.1
- 396 -
Designing a special mobility (hot air balloon)
communication interface
Seung-Hwan Byun1*, Kyung-Wan Koo2, Mun-Sung Kim3, and Jong-Gyu Kang4
1
Daejeon University, Korea
2
Hoseo University, Korea
3
Daejeon University, Korea
4
MetaCom Ltd., Korea
Email: [email protected]
Abstract—This paper designs and implements a to expand from $1815.5 million in 2019 to $220.8 million in
communication protocol that acquires real-time sensor data and 2026, with an average annual growth rate of 4.94 percent
provides it to ground control points in order to establish a safe (CAGR) from 2020 to 2026[1].
hot air balloon system. Real-time sensor data includes
temperature, wind direction, wind speed, altitude, speed, air B. the present situation
pressure, and fuel volume inside the hot air balloon, and a However, the current hot air balloon system is still being
database of data on the flight schedule and acquisition during adjusted in an analog way. Hot air balloons in the sky are
flight to establish a safe hot air balloon system where all flown only with the ability of pilots, and the support team on
equipment is sensed. The hot air balloon system, which is a the ground wants the balloon to land safely after about an hour
ground control, was constructed by limiting the maximum of flight while talking on the radio. The ground support team
transmission distance between the hot air balloon and the cannot directly check the condition or environment of hot air
mobile control station to within 6km.
balloons in the sky and cannot conduct integrated control.
Keywords—special mobility, hot air balloon, LoRa Gateway, Since hot air balloons can be used in various ways for events,
wireless, sensing tourism products, or environmental monitoring, there is no
system to communicate and monitor internal and external
I. INTRODUCTION situations of hot air balloons in flight in real time with the
ground, so it is necessary to develop a communication system
A hot air balloon is a device that freely flies propane liquid with these functions.
gas through the air heated by the flame of the burner to obtain
buoyancy through the rise and fall of the temperature inside
the Guppy. Hot air balloons are very simple in concept and
handling, consisting of a guppy and fuel system (frames and
burners to support the load) and finally a basket to carry
passengers and pilots. Goofy is manufactured using Nylon66
with high durability and strength, and the road tape made of
super-strong PE is designed to withstand the load or maintain
the pressure of the expanded hot air balloon. The burner has a
structure that increases efficiency by expanding the pressure
of the stored fuel (liquid propane). Currently, hot air balloons
are operated by skilled coordination of pilots, and only simple
information is provided by phone through radio with ground
control stations[1]. Fig. 1. Guppy Internal Structure - To be measured and Ready to Take Off
– To Determine the Situation
This paper designs and implements a communication
protocol that acquires real-time sensor data and provides it to This paper designed a system to cope with the problems of
ground control points to build a safe hot air balloon system in hot air balloons listed below by combining sashing technology
a hot air balloon system that has been in operation since 1783. and communication technology to cope with problems that
Real-time sensor data includes temperature, wind may occur when operating hot air balloons.
direction, wind speed, altitude, speed, air pressure, and fuel 1) weather limit
volume inside the hot air balloon. In order to establish a safe
hot air balloon system where all equipment is sensed, the - The ground wind speed at takeoff shall not exceed 15kt
control station has established a database on the flight history (27km/hor 7.5 m/s).
(passenger information, flight path, flight time, video - It is possible to fly only in very weak thermal conditions.
information, etc.).
- Flights should be prohibited in situations where active
gusts or gusts are expected.
II. NEED TO BUILD A SYSTEM 2) the condition of a hot air balloon
A. marketability - The hot air balloon shall stop taking off without delay if
According to QY Research's "Global Hot Air Balloon any abnormality is found in the pre-inspection.
Ride Market Size (Status and Forecast 2020-2026)" in - If severe damage to the Guppy due to obstacles or
September 2020, the global hot air ride market size is expected problems such as cable wires, road tapes, cylinders, and fuel
- 397 -
hoses of the burner occur before takeoff, take-off shall be If the maximum rising speed is from 3 m/s (600ft/min) to
stopped. 5 m/s (1000ft/min), it shall not exceed the limit temperature of
the Guppy internal thermometer. The maximum descent speed
- It is prohibited to fly without the maintenance of a shall also not exceed 5 m/s (1000ft/min). Within the allowable
designated technician. range of the internal thermometer, the Race-1(50, 60, 70)
series has a maximum rise and fall speed of 7 m/s
(1,400ft/min).
7) Internal temperature
During normal use, the maximum temperature inside the
guppy shall not exceed 120 ℃ (250 ℉ ). The Guppy
temperature close to the burner shall not exceed 130°C (266
℉).
Fig. 2. Hot air balloon bottom structure
3) Safety equipment (minimum equipment)
All of the following equipment must be in normal Fig. 3. Preparation for rise – Internal temperature and situation are not
condition and must be carried in a basket. numerically known
1- Safety certified 2kg (4.4lb) fire extinguisher or 2kg
In order to ensure the safety of the hot air balloon in flight,
(2.2lb) fire extinguisher (Use of fire extinguishers suitable for
the various hot air balloon flight conditions mentioned above
fire with valid hot air balloon required by the National Fire
should be checked in real time on the ground. As a general
Agency)
communication method, communication methods such as
2- Sufficient ignition aids (match, lighter, other igniter, WiFi, Bluetooth, and RF are far away, making it difficult to
etc.) communicate with hot air balloons floating in the air. LTE can
be used, but communication costs are high.
3- Leather Gloves for Pilot Protection
In this paper, a communication technology capable of
4- A (digital or cloth) Guppy thermometer indicating the supporting communication from hundreds of meters to
temperature inside the Guppy kilometers was designed and implemented a hot air balloon
5- Fuel residual gauge complete fuel tank monitoring system that supports real-time communication
between hot air balloons and mobile ground control stations
6- Means of measuring fuel consumption (clock or using LoRa (Long Range) communication.
stopwatch)
7- altimeter
III. DESIGN AND IMPLEMENT SENSING AND COMMUNICATION
8- Bariometer (elevation meter) showing rise and fall
PROTOCOLS
9- A thermometer to measure ambient temperature for In this paper, communication between hot air balloons and
flights over 300 meters and a loading chart provided by the ground control stations was designed as a LoRa
manufacturer communication system using a LoRa network with a public
4) Allowed weight (loading) frequency band of 900 MHz, not a commercial network.
The maximum take-off weight shall never exceed the limit - The data for real-time major from hot air balloon to
values given in the loading chart provided in Manual 5.2 ground are primarily information such as temperature, wind
provided by the manufacturer. There should be sufficient direction, wind speed, altitude, speed, air pressure, and fuel
space for the pilot to operate fuel and various systems to safely volume inside the hot air balloon. Thereafter, coordinate
take off and land. information, image information, route information, and the
like may be added.
5) a flight attendant
- A node board was designed to acquire sensor data
Minimum: 1 Pilot mounted on a hot air balloon and transmit data to a mobile
6) vertical velocity station on the ground located long distances.
- 398 -
- A monitoring system was designed to receive data D. Smartphone interworking through IoT G/W
transmitted from a hot air balloon and display it on a server or If you use the IoT board (Wemos D1R1) without using the
mobile terminal using a LoRa communication method. Unoboard on the Master node, you can transmit data to your
- Through the development of a prototype of the system as smartphone in real time with MQTT in a WiFi Internet
described above, it was mounted on a hot air balloon and environment. MQTT is a TCP-based prototype for the Internet
tested. of Things (IoT). It is designed to operate well even in low-
power and low-performance devices. Publisher/Broker
A. Hot air balloon radio monitoring system configuration /Subscriber is in the main configuration area[6][7][8].
diagram
The radio communication band of communication
between hot air balloons and ground sand petitions used the
900 MHz band as the common frequency band used by LoRa.
The terminal mounted on the hot air balloon was defined as a
LoRa node. The information obtained from the hot air balloon
is transmitted to the LoRa IoT gateway installed at the mobile
control station of the award through LoRa communication,
where data is transmitted to the IoT server or smartphone
terminal[2][3].
Fig. 7. Smartphone interworking through IoT G/W
When a publisher with sensor data transmits a value for a
particular Topic, the Subscriber who subscribes to the topic
through Broker receives the value and communicates. A
broker is a kind of server. It serves as a bridge that transfers
data between Publisher and Subscriber.
IV. CONCLUSION
Fig. 4. Hot air balloon radio monitoring system configuration diagram
In order to establish a safe hot air balloon system, this
B. Communication Module Design paper designed and implemented a communication protocol
that acquires real-time sensor data and provides it to ground
The development environment for implementing control points.
communication between LoRa nodes is shown “Fig. 5”.
Real-time sensor data includes temperature, wind
direction, wind speed, altitude, speed, air pressure, and fuel
volume inside the hot air balloon, and a database of data on
the flight schedule and acquisition during flight to establish a
safe hot air balloon system where all equipment is sensed.
In this paper, communication between hot air balloons and
ground control stations was designed as a LoRa
communication system using a LoRa network with a public
frequency band of 900 MHz, not a commercial network.
Fig. 5. Communication Module Design
REFERENCES
In the development environment, the H/W board [1] QYResearch, “Global Hot Air Balloon Ride Market Size, Status and
constituting each node used the Arduino board and the LoRa Forecast 2020-2026”, September 2020.
communication module (SX1276). A sensor was linked to the [2] Lee, Sung-Keun, “Design and Application of LoRa-based Network
LoRa Node on the Slave side to obtain data in real time. The Protocol in IoT Networks, ” The Journal of the Korea Institute of
software development environment applied the Aduido IDE. Electronic Communication Sciences, v.14 no.6, 2019, pp.1089–1096.
[3] M. Bor, J. Edward, and U. Roedig, "LoRa for the Internet of Things,”
C. Communication prototype In Proc. International Conference on Embedded Wireless Systems and
Networks, Graz, Austria, Feb. 2016. pp.361-366.
The communication prototype for testing LoRa
[4] Semtech Corporation, "LoRa Modulation Basics,” AN1200.22, May
communication is as follows[3][4][5]. 2015.
[5] N. Sornin, M. Luis, T. Eirich, and O. Hersent, LoRaWAN Specification,
Fremont, USA, LoRa Alliance, July 2016.
[6] ETRI, "LPWA IoT Network Technology Trends”", 2017.
[7] LoRa Alliance, "A technical overview of LoRa and LoRaWANTM”,
2015.
[8] SK telecom, "Working-level conference of IoT-LPWA”, 2016.
[9] Nable Communications,Inc. "Low Power Wide Area Technology for
IoT”, 2016.
[10] https://spri.kr/
Fig. 6. Communication prototype
[11] https://www.semtech.com/
- 399 -
VLAN Network Architecture for Trains to
Trains Communication
Youngjung Kim1 Marina Asankulova2 and Kyuhyoung Choi 1*
1
Seoul National University of Science and Technology, Korea
2
LLC Tendik Audit, Kyrgyzstan
Email: 1 [email protected], 2 [email protected] 1* [email protected]
Abstract— Urban railways are used as an integral means of to find the train ahead. This method causes network
transportation as cities develop. Train control systems have communication delays as the connection between the train and
been developed to operate urban railways safely and efficiently. the AP turns into a runner. This paper proposes a Virtual Local
With the development of science and technology, train control Area Network (VLAN) network structure suitable for T2T.
systems have gradually evolved from track circuit-based train
control methods to wireless communication-based
Communication Based Train Control (CBTC). The current
CBTC is a centralized train control system. Recently, research II. BACKGROUND
on train control using train-to-train communication has been A. Conventional centralized CBTC system
advanced. The following problems occur in the train-to-train
communication method in the existing network structure. A Urban railways have been used as an indispensable means
train wants to communicate with the preceding train. Trains of transportation as cities develop. Train control systems have
broadcast Address Resolution Protocol (ARP) to the network to been used in the past to utilize track circuit-based fixed blocks.
find the train ahead. This method causes network The fixed blockage system caused congestion problems with
communication delays as the connection between the train and saturated cities, so wireless communication technology-based
the AP turns into a runner. This paper proposes a Virtual Local CBTC was adopted to increase line capacity and reduce
Area Network (VLAN) network structure suitable for T2T. running time. The CBTC system is a mobile blockage system
that does not rely on existing track circuits.[2]
Keywords— CBTC, VLAN, ARP
The existing centralized Wireless Local Area Network
I. INTRODUCTION (WLAN)-based CBTC has a center receiving information
Urban railways are used as an integral means of such as train speed, direction and position, as shown in Figure
transportation as cities develop. Train control systems have 1. In addition, the center calculates the operating status in real
been developed to operate urban railways safely and time and transmits operating information such as train speed
efficiently. With the development of science and technology, and traveling direction to the following train. A zone
train control systems have gradually evolved from track controller (ZC) receives and processes information such as the
circuit-based train control methods to wireless speed and direction of travel of the preceding train. The ZC
communication-based Communication Based Train Control then transmits Automatic Train Protect (ATP) and Automatic
(CBTC). So far, CBTC is a centralized communication system Train Operation (ATO) data (train position, safe route, train
with train - track transformer - center - track transformer - train tracking, train operation restrictions, failure reports) to the
communication. Such a centralized train control system must train behind in cooperation with the interlocking device.
stop all trains for safety in the event of a failure of the central Automatic Train Supervision (ATS) serves to monitor and
center. To ameliorate such inefficiencies, train-track record ZC and interlocking devices.[2]
displacement-train communication requires a virtual train-to-
train virtual communication scheme. A recent study [1] also
advanced Train to Train (T2T) research using LTE PC5 Link.
The following problems occur in the train-to-train
communication method in the existing network structure. A
train wants to communicate with the preceding train. Trains
broadcast Address Resolution Protocol (ARP) to the network
Figure 2. LAN with a L2 Switch
Figure 1. CBTC Network Structure
- 400 -
Figure 5. Inter-VLAN Using Router
C. VLAN
For VLAN communication, it is necessary to physically
configure multiple LANs with multiple switches as shown in
Figure 3. Physical LAN Division and Logical LAN Division
Figure 3(a), and logically with one switch as shown in Figure
3(b). This method configures multiple VLANs. And VLANs
B. LAN can be divided by declaring VLANs on each port. In the
A LAN is a local area network, a network within the range declared VLAN10, only AP1, AP2, and AP3 can
of an ARP Request Message frame, as shown in Figure 2. The communicate with each other, in VLAN20, only AP4, AP5,
communication process of LAN communication is performed and AP6 can communicate with each other, and in VLAN30,
under the assumption that PC0 knows PC2's Internet Protocol only AP7, AP8, and AP9 can communicate with each other.
(IP). First, PC0 transfers data to PC2. PC0 looks up the route Therefore, a VLAN cannot send an ARP Request Message to
for PC2's IP address in its Routing Table and finds out what another VLAN.[3]
the Next Hop IP is. And PC0 does not have the Media Access 802.1Q is one standard for various VLAN schemes. This
Control (MAC) address of the Next Hop IP. Therefore, PC0 is a standard for decrypting and communicating VLAN IDs
sends an ARP Request Message to L2 SW to find out PC2's over trunk ports between switches. The header-added frame is
MAC address. The Layer 2 Switch (L2 SW) that received the a method of adding an 802.1Q header to the Ethernet frame
ARP Request Message looks at the source MAC address, header as shown in Figure 4, and data must be sent through
learns the MAC Table entry of PC0, and floods the ARP the port set as a trunk. The TFID in Figure 4 is information
Request Message to all devices connected to itself. PC2, that indicates the presence or absence of a tag, and has a value
which received the ARP Request Message, records the IP and of 0x8100 for tagged frames. TCI can choose VLAN priority
MAC address of PC0 in its own ARP Table, and transmits an and choose VLAN name from 0-4095.[4]
ARP Reply Message to the L2 SW. The L2 SW that received
the ARP Reply Message learns the MAC Table entry of PC2. D. Inter-VLAN
The L2 SW then forwards the ARP Reply Message to the Different VLANs cannot communicate with each other. At
Source MAC Table report PC0. PC0 that receives the ARP this time, in order to communicate between VLANs, a router
Reply Message registers the IP and MAC address of PC1 in that acts as a router, which is the upper layer, is required as
the ARP table. Then, the MAC address of PC2 is generated as shown in the Figure 5. Let's take a look at the process by which
the destination MAC of the frame to be transmitted to PC2, AP6, which belongs to different VLANs, communicates with
and the data is transmitted. L2 SW checks the frame header AP3. First, proceed assuming that AP6 knows AP3's IP. AP6,
sent by PC0, sees if it is in the MAC table, and forwards the after checking its own routing table, sees that AP3 does not
data to PC2. The PC2 that received the frame receives the data belong to its own LAN. AP6 tries to send data to AP3 through
because the destination MAC of the frame is the same as its Router. So AP6 sends an ARP Request Message to L2 SW
own MAC address. looking for Router's MAC Address. The L2 SW that received
the ARP Request Message learns the information of AP6 and
floods the port and Trunk port where VLAN20 is set. Router
sends ARP Reply Message to AP6. AP6, which received the
ARP Reply Message, learned the Router's MAC Address.
AP6 sets Destination Address to Router, Destination MAC
Address and IP to AP3, and Source MAC Address and IP to
AP6. AP6 then sends the frame to the Router. After receiving
the frame, the router checks the IP header and confirms that
the destination is AP3. At this time, it will send an ARP
Request Message to find AP3's MAC address. After receiving
the ARP Request Message, AP3 transmits an ARP Reply
Message to the Router. Router learns AP3's MAC Address
Figure 4. Ethernet Frame Header and 802.1Q Frame Header and information that it belongs to VLAN10 through ARP
Reply Message. Router deletes the existing frame header, sets
VLAN tag to VLAN10, Source IP to AP6, Destination IP to
AP3, Source MAC to Router, Destination MAC to AP3, and
then sends the frame. The L2 SW that has received the
- 401 -
modified frame correctly transmits the data to AP3 belonging receive the data because the Destination MAC and Destination
to VLAN10.[5] IP are the same as its own address.
IV. ACTUAL IMPLEMENTATION METHOD
III. PROPOSED VIRTUAL TRAIN-TO-TRAIN NETWORK VIA
VLAN Shinbundang Line is currently using a WLAN-based
centralized CBTC system as shown in the figure7. When inter-
A. Proposed network architecture train communication is applied to the existing system, the
A moving train transmits an ARP Request Message in a following problems occur. ARP request messages occur
broadcast format when the train and AP perform handover. frequently during train and AP handovers. Then, all L2 SWs
This method causes communication delays when there are are connected and send ARP Request Messages to all APs. In
many trains, such as rush hour. To solve this, group multiple this case, communication delays occur. To solve this problem,
APs and configure VLANs as shown in Figure 6. The router the proposed network structure uses VLANs as shown in
connects to the L2 SW as a trunk port for communication of Figure 8, and uses routers for communication between one
different VLANs. This is a way to reduce the data in the ARP VLAN and another VLAN. And Figure 8 shows that the rear
Request Message. train transmits a dada frame to the front train.
V. CONCLUSION
B. Novel communication scheme for virtual train-to-train Because trains move, handovers between trains and APs
Examine the communication process of Train1 (T1) and occur frequently. At this time, the ARP Request Message
Train2 (T2). Assume that T1 knows T2's IP. T1 sends data to generates a lot of data on the network and causes
T2. The message is that T1 sent an ARP request to AP6 communication delays. To solve this, this paper proposed a
because it does not know T2's MAC address. After checking network structure for train-to-train communication. This
its own MAC table, AP6 does not have T2's MAC address, so network structure used VLANs and Routers. VLANs reduce
it sends an ARP request message to L2 SW. After learning the the amount of data broadcast on the network. Therefore, this
information of T1, the L2 SW confirms the Tag of VLAN20 configuration reduces the probability of occurrence of
and Floods to the VLAN20 port. After receiving the ARP communication delays. Future work will proceed with
Request Message, AP4 checks its own MAC Table and sends simulations to examine the performance of the proposed
it to T2. T2 sends an ARP Reply Message to T1. T1 can network structure.
communicate by learning the MAC address of T2.
Examine the communication process between T2 and
Train3 (T3). Assume T2 knows T3's IP. T2 sends data to T3.
After checking its routing table, T2 sees that T3 does not
belong to his LAN. T2 would like to send data to T3 via
Router. T2 sends an ARP Request Message to AP4 to find out
the MAC address of the router. AP4 receives the ARP Request
Message, learns the information of T2, and transmits the
message to L2 SW. The L2 SW that received the ARP Request
Message learns the information of AP4 and floods the port and
Trunk port where VLAN20 is set. A router that receives an
ARP Request Message sends an ARP Reply Message to the
L2 SW. Upon receiving the ARP Reply Message, the L2 SW
learns Router information and transmits the ARP Reply
Message to AP4. AP4 learns the MAC address of L2 SW and
sends an ARP Reply Message to T2. T2 sets Destination
Figure 6. Proposed Network Architecture
Address to Router, Destination IP to T3, Source MAC
Address and Source IP to AP4, and then transmits the data
frame. AP4 then transmits the data frame to L2 SW. The L2
SW that has received the data frame transmits the data frame
to the Router. After receiving the frame, the router checks the
IP header and then confirms that the destination is T3. At this
point, send an ARP Request Message to find out the T3's
MAC address. After receiving the ARP Request Message, T3
sends an ARP Reply Message to the Router. The Router learns
the MAC Address of T3 and the information that it belongs to
VLAN10 through the ARP Reply Message. Router deletes the
existing frame header, sets VLAN tag to VLAN10, Source IP
to T2, Destination IP to T3, Source MAC to Router,
Figure 7. Sinbundang Line Network
Destination MAC to T3, and then transmits the data frame.
The L2 SW that has received the modified frame transmits the
data to AP3 belonging to VLAN10. AP3 checks its MAC table
and sends the frame to T3. T3 which received the frame will
- 402 -
[1] X. Wang, L. Liu, T. Tang and L. Zhu, "Next generation train-centric
communication-based train control system with train-to-train (T2T)
communications," 2018 International Conference on Intelligent Rail
Transportation (ICIRT), Singapore, 2018, pp. 1-5
[2] J. Farooq and J. Soler, "Radio communication for communications-
based train control (CBTC): A tutorial and survey," in IEEE
Communications Surveys & Tutorials, vol. 19, no. 3, pp. 1377-1402,
thirdquarter 2017
[3] Mathew, A. and Prabhu, S.R.Boselin, “A Study on Virtual Local Area
Network (VLAN) and Inter-VLAN Routing” , International Journal of
Current Engineering and Scientific Research (IJCESR), Volume 4,
Issue 10, 2017
[4] Li Xinzhan and Xu GenChao, "Research on double vlan scheme in
ethernet network," 2009 2nd IEEE International Conference on
Computer Science and Information Technology, Beijing, China, 2009,
pp. 127-129
Figure 8. Sinbundang Line Network with VLAN Applied
[5] Agwu Chukwuemeka Odi, Nweso Emmanuel Nwogbaga, and Ojiugwo
Chukwuka N., “The Proposed roles of vlan and inter-vlan routing in
effective distribution of network services in Ebonyi State
REFERENCES University,” International Journal of Science and Research (IJSR), vol.
4, no. 7, pp. 2608–2615, Jul. 2015.
- 403 -
August 22(TUE) - 25(FRI), 2023
Poster Session 2
14:30~15:10 Thu, August 24, 2023
Room : Building A 4th Floor
Chair: Prof. Jeamoon Kim
(Korea National University of Transportation, Korea)
Prof. Kyungwan Koo (Hoseo University, Korea)
Prof. Assel Mukasheva (AUPET, Kazakhsatn)
Prof. Juhee Choi (Sangmyung University, Korea)
- 404 -
A rectangular gate structure with attachable
sanitary lights and human monitoring camera
Woo-Young Kim1, Hyung-Kon Lee1, Young-Min Kim2, Seung-Youn Yang3, Jin-Tae Kim3, Kyung-Wan
Koo4, Min-Soo Kim5, Tae-Ho Kim6, Seong-Kweon Kim6, In-Choel Jeong7, Chang-Jun Ahn8,
Jae-Sang Cha8*
Boseung Electric Co., Ltd.1, Jooyoung F&G2, Fivetek. Co.ltd.3, Dongseoul Univ4, Hoseo Univ.5,
SNUST6, SungKongHoe Univ7, Chiba Univ. Republic of Korea & Japan8
*Corresponding author : [email protected]
Abstract— In this paper, we designed and implemented a the screen displayed by the camera in real time to confirm
rectangular type of gate structure with attachable sanitary that the human detection function is working properly.
lights and a human monitoring camera. Newly designed
structure in this paper is a specific type of entry/exit applications II. MAIN FEATURE OF PROPOSED STRUCTURE
for our experiments that incorporates sanitation and human
tracking & surveillance camera. Main features of proposed
structure in this paper could be summarized as : Rectangular A. Rectangular Gate Structure and Sanitary Lights
Gate Structure: The designed gate structure in this paper is The rectangular gate structure is designed with the
rectangular shape and serves as the physical barrier through assumption that it is a door for people to enter and exit. The
which individuals must pass. We implemented Gage frame with rectangular gate structure is combined with a sanitary light,
simple metal materials for our experiments. Sanitary Lights: and the sanitary light is made of a removable structure.[1]
The gate structure is equipped with attachable sanitary lights. The sanitary light has the function of a germicidal light that
These lights emit sanitary visible light radiation, which has outputs sanitary visible light that can sterilize the surface of
germicidal properties. When individuals pass through the gate,
an object, and the function of a general light that can be
the sanitary visible lights help disinfect their exposed surfaces.
Human Monitoring Camera: The gate structure features a
utilized in a home or factory. Fig. 1. shows the undetected
camera system for monitoring and tracking about human state of the sanitary light, and Fig. 2. shows the detected state
activity. The camera is typically positioned to capture the area of the sanitary light.
where individuals pass through the gate. The combination of a
rectangular gate structure, attachable sanitary lights, and a
human monitoring/ tracking camera offers several advantages:
Enhanced Sanitation: The attachable sanitary lights emit
sanitary visible light radiation that helps disinfect surfaces. As
individuals pass through the gate, their exposed body parts can
be subjected to the germicidal effects of the sanitary visible light,
contributing to improved sanitation. Human monitoring and
tracking: The human monitoring camera including computer
vision & IoT functions allows for real-time observation of
individuals passing through the gate. This enables Human
tracking and sensing and enforce hygiene compliance measures.
By using the proposed structure in this paper, disinfection is
performed using visible light harmless to the human body when
a person enters the room through a specific gate, and additional Fig. 1. The undetected state of the sanitary light
indoor hygiene management and security monitoring functions
are possible through continuous detection of the human body.
Keywords— Gate, Visible, Sanitary light, Camera
I. INTRODUCTION
In this paper, we present the design and implementation
of a rectangular frame structure that combines a camera and
an attachable sanitary light to secure a sanitary space in public
places. The attachable function of the sanitary light is
necessary for easy installation in public places, and the reason
for using visible light in the sanitary light is also considered
for safety issues that do not cause much harm to the human
body. Therefore, if the camera does not detect a human, the
sanitizing light will activate, emitting a sanitizing visible light
that disinfects the surface, and if the camera detects a human, Fig. 2. Shows the detected state of the sanitary light
the sanitizing light will stop, the sanitizing light stops
working and converts its function to a normal light. When a
person passes through the frame structure, the user can see
- 405 -
B. Human Monitoring Camera
The sensor capabilities of the camera were utilized to
detect humans. The cameras are positioned to view the area
through the rectangular gate structure and have the ability to
monitor human movement.[2] The images captured by the
camera utilize computer vision technology to display the
results of human monitoring on a screen, which can be
viewed in real time by the user on a monitor.[3] Fig. 3. shows
the camera module that can detect the human body, and Fig.
4. shows the appearance of the human body displayed on the
human body monitoring screen.
Fig. 5. An overall view of the system
III. CONCLUSION
Through this research, it is expected that this structure,
which utilizes computer vision technology using a camera
and light conversion technology according to the presence or
absence of a human body, will secure hygienic spaces in
public places.
ACKNOWLEDGMENT
This research was partially supported by project (#S2866225)
of Korea Technology & Information Promotion Agency for
SMEs, TIPA, Republic of Korea. and partially funded by a
Fig. 3. The camera module that can detect the human body
Grant of Scientific Research No. 22K04085 from the Japan
Society for the Promotion of Science (JSPS), the
Telecommunications Advancement Foundation (TAF), and
Suzuki Foundation.
REFERENCES
[1] Woo-Young Kim, Hyung-Kon Lee, Kyung-Wan Koo, Seong-Kwon
Kim, “Development of Vietnam certified connection-type LED and IT
local-type visible light sanitation equipment application products”
2023 The Korean Institute of Electrical Engineers
[2] D Arisandi, MElveny, R Rahayu, “Human Detection and Identification
for Home Monitoring System”, 5th International Conference on
Computing and Applied Informatics(ICCAI 2020)
[3] Woo-Young Kim, Hyung-Kon Lee, Min-soo Kim, Seung-Youn Yang,
Joo-Pil Jo, Jae-Sang Cha, “ Visible Light Sanitary Lighting
Fig. 4. The appearance of the human body displayed on the human body
Environment Sensing Type IoT Camera Image Processing Application
monitoring screen
Technology development”, 2023 The Korean Institute of Electrical
Engineers
C. Rectangular gate structure with attachable [4] Sung-Hoon Yoon, Kil-Soo Lee, Jae-Sang Cha, Vinayagam Mariappan,
Min-Woo Lee, Deok-Gun Woo, Jeong-Uk Kim, “Indoor Surveillance
sanitary lights and human monitoring camera Camera based Human Centric Lighting Control for Smart Building
Lighting Management”, Intenational Journal of Advanced Culture
Fig. 5. shows an overall view of the system. By utilizing Technology(IJACT 2020), Vol 8., No 1, 207-212.
a camera that can detect the human body, we designed a
structure that can select the function of the sanitary light
according to the presence or absence of the human body. The
system is implemented so that users can monitor the images
taken by the camera in real time. In this paper, we presented
the design and implementation of a rectangular frame
structure that combines a camera and a attachable sanitary
light to secure a sanitary space in public places.[4]
- 406 -
A development of IoT switching and control
box application used in various power source
choice and sanitary light control functions
Woo-Young Kim1, Hyung-Kon Lee1, Gyewon Choi2, Minsoo Kim3, Tae-Ho Kim4, Seong-Kweon Kim4,
Kyung-Wan Koo5, In-Cheol Jeong6, Chang-Jun Ahn7, Jae-sang Cha7*
Boseung Electric Co., Ltd.1, SKKU2, Dongseoul Univ.3, SNUST4, Hoseo Univ.5, SungKongHoe Univ6,
Chiba Univ.7
* Corresponding author : [email protected]
Abstract— In this paper, we designed and implemented an power block box is designed to be utilized for sanitary
IoT switching and control box application can be a versatile lighting. Fig. 1. shows a block diagram of the configured
solution for power source selection and sanitary light control system. [1]
functions. i.e., Implemented application’s structure in this paper
is an IoT switching and control box applications for our
experiments that incorporates various power choice and
sanitary light control. Main features of implemented application
in this paper could be summarized as : Power Source Choice:
The IoT switching and control box can be designed to connect
to different various power sources, such as the main electrical
grid, solar panels, or power bank batteries. And it can
automatically switch to the most cost-effective or sustainable
power source at any given situation. Sanitary Light Control: In
terms of sanitary light control, the IoT application can provide
smart lighting management in different areas or any space
where sanitation and cleanliness are crucial. And the application
can integrate with motion sensors, or other environmental
sensors. Additionally, the IoT application can offer features like
remote control function. Users can control the Sanitary lighting
settings by using mobile app or web UI, allowing them to turn Fig. 1. Block diagram of the configured system
on or off lights or sanitary mode changing. By using IoT based
remote control functions presented in this paper, various power B. IoT switching and control box application used in
sources of sanitary lighting can be easily selected. Additionally,
Sanitary LED light’s Power on/off mode or sanitary level status various power source choice and sanitary light
also easily changed
control function
Keywords— Gate, Visible, Sanitary light, Camera The configured system consists of multiple batteries for
always-on power and adaptability to different environments,
I. INTRODUCTION
a solar panel to charge the batteries, a controller and sanitary
In this paper, we present the design of IoT switching and light that can be controlled remotely. The power switch
control box application that can control sanitary light with a includes power sources that can be utilized in different
variety of power choices. It is designed to control various environments. In the event of a full-time power outage, the
power sources and to control sanitary light from a distance. batteries were being charged by the solar panel and the power
The designed switching module is composed of circuits that connection could be switched. The system is configured to
can respond to AC/DC power environments. Based on the remotely control the sanitary light to enable sanitizing
IoT communication system, a system that automatically functions and light dimming. Fig. 2. shows the structure of
switches power was realized by combining functions that can the IoT switching and control box application that allows for
perform IoT control. power selection and control of the sanitary light. [2][3]
II. MAIN FEATURE OF PROPSED STRUCTURE
A. Block diagram of proposed structure
The AC/DC power switchable module has a complex
power structure and uses a commercial power source AC. In
addition, when there is no commercial power source, the
electrical energy charged to the battery through the solar
panel is designed to be used as a power source by utilizing an
inverter. In addition, the IoT function is automatically
switched, and the power source utilized for sanitary lighting
is designed to be automatically switched. The independent
- 407 -
and general visible light functions. In addition, the brightness
of general visible light is configured to be dimming
adjustable from 0 to 100%. Thus, we introduced efficient
system that can automatically switch the power situation
according to the given situation in various spaces. Fig. 5.
shows the operation of various functions through the remote
control of the sanitary light.
Fig. 2. the structure of the IoT switching and control box application that
allows for power selection and control of the sanitary light
C. IoT switching various power sourcce
The connected power source is an auto-switching Power
Switch with IoT capabilities. Therefore, when connected to
the power source, the A part of the Power Switch lights up to
power the sanitary light. When the power is disconnected, the
switch is immediately repositioned and the B part lights up,
connecting to the battery that is being charged using the
existing solar panel, so that the sanitary light can be operated Fig. 5. the operation of various functions through the remote control of the
continuously. Fig. 3. shows when the power is always on. Fig. sanitary light
4. shows when the power is turned off and connected to the
battery. III. CONCLUSION
In this paper, we introduced a study on IoT-based remote
control of sanitary light with power switching function that
can provide the most suitable power for the surrounding
environment. Various light environments were provided for
users to choose, and various solutions for power supply
methods were provided due to different power control
methods in indoor and outdoor environments. This research
is expected to evolve into an IoT switching and control box
application that integrates various sanitary light controls.
ACKNOWLEDGMENT
This research was partially supported by project (#S2866225)
Fig. 3. the power is always on. of Korea Technology & Information Promotion Agency for
SMEs, TIPA, Republic of Korea. and partially funded by a
Grant of Scientific Research No. 22K04085 from the Japan
Society for the Promotion of Science (JSPS), the
Telecommunications Advancement Foundation (TAF), and
Suzuki Foundation.
REFERENCES
[1] A. Karthi, G. Sabaresh, “Design of Internet Controlled Switch Box
Using IoT”, Journal of Advanced in Chemistry, Vol 12. No 18.
[2] Woo-Young Kim, Hyung-Kon Lee, Jae-Sang Cha, Seung-Youn Yang,
“Development of solar power package box application technology for
outdoor visible light sanitary lighting power supply”, The Korean
Institute of Electrical Engineers.
[3] Woo-Young Kim, Hyung-Kon Lee, Jae-Sang Cha, Min-soo Kim,
Fig. 4. the power is turned off and connected to the battery
Seung-Youn Yang, Joo-Pil Jo, “Study for outdoor visible light sanitary
lighting flicker control IoT smart plug switch application development”,
In this study, the sanitary light fixture utilizes IoT The Korean Institute of Electrical Enginners
communication to remotely select the sanitizing visible light
- 408 -
On the Electrical Safety Analysis of Multi-chute
Classification System for Joint Courier Logistics
Base Infrastructure based STPA
*Note: Sub-titles are not captured in Xplore and should not be used
Sunwoo Hwang 1, Joouk Kim 2 and Youngmin Kim 1*
1
Ajou University, Korea
2
Korea Railroad Research Institute, Korea
Email: [email protected]
Abstract—This paper presents an electrical safety analysis of the space and tracked parts, finished products and facilities in
a multi-chute classification system designed for joint courier the manufacturing site through the simulation in order to smart
logistics base infrastructure using the Systems-Theoretic SCM (Supply chain Management) solution by the distributed
Process Analysis (STPA) approach. The multi-chute operating system based on NoSQL stored as big data [3]. Yu
classification system plays a critical role in automating the et al. identified the trends in intelligent robot technology in the
sorting and classification of courier packages based on their logistics field and studied intelligent robot technology for
destinations. However, ensuring electrical safety within such a logistics automation, and the results of this can be the basis for
complex system is of utmost importance to prevent accidents securing the reliability of robot development in the logistics
and hazards. The findings of this study provide valuable insights
field [4]. Choi et al. conducted overall robot research such as
into the electrical safety considerations of multi-chute
classification systems in joint courier logistics base
logistics robot operation technology and delivery robot
infrastructure. The proposed control actions can serve as technology, and identified key trends in robot technology
guidelines for system designers, operators, and maintenance related to the 4th industrial revolution [5]. Park et al.
personnel to enhance the electrical safety performance of such conducted an automatic classification logistics system study
systems. By integrating STPA into the analysis process, a using image analysis. This study is a study using a model, but
comprehensive understanding of the system's electrical safety since control using a communication device was implemented,
vulnerabilities and effective risk control strategies can be the meaning as a prototype was derived [6].
achieved, ultimately contributing to a safer and more reliable
logistics operation. III. METHODOLOGIES
A. Munlti-choute classification system
Keywords—System-Theoretic Process Analysis, Electrical
Safety Analysis, Joint Courier Logistics Base Infrastructure, Fig. 1 shows a schematic conceptual diagram of the
Electrical Safety Analysis, Logistics Automation delivery invoice information detection system. Research to
implement an actual physical system will be conducted. A
I. INTRODUCTION multidimensional tunnel image recognition system will be
With the growth of the online market and the resulting implemented as a subsystem by utilizing video and image
growth of the logistics market, delivery volume is increasing, acquisition sensors such as cameras, and a bottom line scanner
and various studies are being conducted to efficiently handle will be implemented under the path through which the courier
the increasing delivery volume. Among them, this is the point cargo passes. As a result of this, delivery invoice information
of time presented in the concept of point courier logistics will be transmitted through the interface with the joint courier
system that can jointly process multiple courier cargoes. In logistics system, which will be operated separately. In this
order to implement an efficient, safety logistics system, basic study, a study was conducted on ways to STPA analysis as a
research through various approaches and technology way to increase electrical safety this system.
development in which various fields are fused is essential.
Accordingly, this we conducted electrical safety analysis of
multi-chute classification system for joint courier logistics
base infrastructure based STPA (System-Theoretic Process
Analysis).
II. RELATED WORKS
Chae et al. reviewed cases of public logistics-related
business models and policy directions, and determined that a
rational and systematic management and support system is
necessary to implement a successful community logistics Fig. 1. Schematic diagram of the multi-chute system
business model [1]. Shin conducted a study in order to
applying the technology originated from the 4th industrial B. STPA(Systems-Theoretic Process Analysis)
revolution to the field of logistics technology, such as IoT, AI STPA, which stands for Systems-Theoretic Process
and robot [2]. Kwak et al. conducted a study that calculated Analysis, is a safety analysis method used to identify and
- 409 -
mitigate hazards in complex systems. It is a structured and Study identified electrical hazards of the Multi-chute
systematic approach that helps engineers and safety Classification System. Potential electrical hazards within the
professionals understand and address potential risks system may include electric shock, fire, and overvoltage. The
associated with the design, operation, and maintenance of study evaluated the system's electrical protection capabilities.
complex systems. STPA incorporates the principles of The effectiveness of electrical protection functions such as
systems theory to provide a holistic view of a system and its insulation, grounding, and protection devices implemented in
interactions with the environment, operators, and other the system was evaluated, and through this, the sufficiency of
components. The key idea behind STPA is to identify the protection function or the parts requiring improvement
hazardous scenarios by considering not only the immediate were derived. Proposals for safety control measures have been
causes of accidents but also the underlying system-level established. Based on the findings of the study, safety control
factors that contribute to them. Rather than focusing solely on measures have been proposed to prevent or reduce electrical
failures of individual components, STPA examines the hazards. For example, improvements in circuit design,
system's structure, interactions, and control processes to addition or upgrade of protective devices, improvements in
uncover potential vulnerabilities. maintenance procedures, etc. may be suggested. The study
evaluated whether the Multi-chute Classification System
complied with relevant regulations and standards. An
assessment was made of the system's compliance with the
appropriate regulations and standards to meet electrical safety
requirements.
V. CONCLUSION
This study deals with the importance of electrical safety
analysis of Multi-chute Classification System based on Joint
Courier Logistics Base Infrastructure. This study aims to
utilize the Systems-Theoretic Process Analysis (STPA)
technique to identify potential electrical hazards in a system
and propose safety control measures. The findings are
expected to aid in the design and operation of systems to
reduce electrical hazards. In this study, the electrical safety
analysis of the Multi-chute Classification System was
performed and the electrical risk of the system was evaluated
Fig. 2. Concept of Control structure for STPA by applying STPA. According to the research results, potential
hazards such as electric shock, fire, and overvoltage that may
IV. RESULT OF RESEARCH occur in the system were identified, and safety control
Table 1 is the result of deriving the UCA(Unsafe Control measures were proposed to control them. In addition, the study
Actions) of STPA from the viewpoint of securing the emphasized the importance of effective implementation and
electrical safety of the multi-chute classification system. In operation of electrical protection functions, and proposed a
this paper, the control action is set to operation. For each UCA method to protect against electrical hazards by applying
type, it was derived as follows: Not providing causes hazard, appropriate insulation, grounding, and protective devices to
“ A protective device or circuit breaker in an electrical system the system. The results of the study can be used as reference
fails to operate”, Providing causes hazard, “Circuit breakers in materials to improve the electrical safety of the Multi-chute
electrical systems are so sensitive that they trip unnecessarily Classification System, and are expected to help improve
even under normal operating conditions”, Too soon, too late, safety at the stages of system design, operation and
out of order, “ The per supply runs before the generator”, maintenance.
Stopped too soon / Applied too long, “ When a temporary
current peak occurs in the system, the overcurrent protection REFERENCES
device incorrectly recognizes it and cuts”. [1] M. N. Chae and H. K. Gwon, “Business models and policy directions
of joint logistics projects funded by government,” Korea Society of
TABLE I. UNSAFE CONTROL ACTIONS Transportation, vol. 13, no. 3, pp. 28–35, 2016.
[2] H. J. Shin, “A study on trends in the use of logistics technology based
Not Too soon, Stopped on the 4th industrial revolution,” The e-Business Studies, vol. 21, no.
Providing
Control providing too late, too soon / 2, pp. 17–27, 2020.
causes
action causes out of Applied [3] K. J. Kwak, S. G. Hwang, D. J. Shin, K. W. Park, J. J. Kim et al., “Study
hazard
hazard order too long of logistics object tracking service for smart SCM,” Journal of the
Operation A Circuit The per When a Korean Institute of Industrial Engineers, vol. 46, no. 1, pp. 71–81, 2020.
protective breakers in supply temporary [4] W. P. Yu, Y. C. Lee and D. H. Kim “Technical trends of robot task
device or electrical runs before current intelligence in intelligent logistics/agriculture,” Electronics and
circuit systems are the peak Telecommunications Trend, vol. 36, no. 2, pp. 22–31, 2021.
breaker in so sensitive generator occurs, the [5] S. l. Choi, D. H. Kim, J. Y. Lee, S. H. Park, B. S. Seo et al., “Logistics
an that they trip overcurrent and delivery robots in the 4th industrial revolution,” Electronics and
electrical unnecessarily protection Telecommunications Trends, vol. 34, no. 4, pp. 98–107, 2019.
system fails even under device cuts
to operate normal off the [6] C. H. Park, S. D. Bae, S. G. Choi, S. H. Choi, J. W. Choi et al.,
operating circuit “Automatic picking/classification system using video analysis,” in
conditions abnormally Proc. of the Korean Society of Computer Information Conf., Jeju,
Korea, vol. 28, no. 2, pp. 661–662, 2020.
- 410 -
The Remote Driving Control System Based on
Telepresence for Mini-tram
Jung-Ju Kim 1*, Gi-Won Kim 1, Jun-Seok Lee 1, Jae-Hun Choi 1, Hyung-Woo Jin 1, Chan-Jong Lee 1
1
Dept. of Robotics, Hoseo University, Korea
Email: [email protected]
Abstract—In this paper, we propose a method for public A. Configuration of remote driving control system for mini-
mobility remote driving control by converging the remote tram based on telepresence
driving control system and telepresence technology, In order to
remotely drive an autonomous vehicle such as a mini-tram,
Fig. 1 is a block diagram of the remote driving control
various visual and auditory information are required. To this system for AMR. Information such as visual, auditory, and
end, an omnidirectional camera with a built-in microphone is physical values obtained through sensors in the AMR are
manufactured, and by using it, visual and auditory information transmitted to the telepresence room. The transmitted
of the surroundings is obtained from a remote place. In addition, information is displayed in the telepresence room so that the
telepresence technology is used to transmit audio-visual user can fell the realism. To decrease delay time and increase
information obtained from the field to the remote driver so that reliability, we use WLAN to transmit the user's control
the same effect as the field can be obtained. Finally, a realistic command data and the information of AMR.
AR room capable of displaying audio-visual information is built
to provide information for mini-tram remote driving control.
The telepresence-based remote driving control technology that
can increase the safety and efficiency of autonomous mini-tram.
Keywords—remote driving control, telepresence, mini-tram
I. INTRODUCTION
Autonomous driving research is being actively pursued
around the world, and there is a wide range of research in the
field of public mobility[1-3]. Mini-trams are small-scale
public mobility for driving short distances in cities, which
have the advantages of reducing traffic congestion in urban
areas and being environmentally friendly. However, the
current remote driving control technology of mini-trams has Fig. 1. The diagram of remote driving control system based on telepresence
various challenges such as reliability and safety of the
communication system and the need for human intervention B. The telepresence room for mini-tram
in certain situations[4-7]. To overcome these challenges in Fig. 2 is a three-dimensional rendering of the telepresence
remote control technology, various studies are being room developed for this study. The telepresence room was
conducted using telepresence[8-10]. designed to allow one person to control a remote mini-tram
realistically within it. The floor area of the telepresence room
Telepresence is a technology that enables voice and video is 2.87(m2) and the height is 2.12(m).
transmission and interaction through two-way remote
communication. In this study, we propose a system that can
improve work efficiency and safety in a remote driving control
environment while building an environment where remote
users can experience the same immersive experience as
operating a mini-tram in real time by using telepresence
technology in the remote driving control system of a mini-
tram.
II. BODY
Through this study, we would like to integrate
telepresence technology with remote driving control
technology for mini-tram. In this study, we developed a
telepresence room and an omnidirectional camera with
microphone to enable real-time interaction between remote
users and mini-trams. This created a sense of immersion as if
a remote driver was actually driving an autonomous mini-tram.
We also developed an Autonomous Mobile Robot(AMR) that
Fig. 2. The 3D rendering of telepresence room
can be remotely controlled so that humans in telepresence
room can control the mobile robot according to the
surrounding environment where the mobile robot is located. C. Development of omnidirectional camera
We developed a camera capable of omnidirectional
photography of more than 300 degrees. The developed
- 411 -
omnidirectional camera consisted of eight CMOS USB web of the remote driving control system for mini-tram based on
cameras and two USB hubs. The omnidirectional camera is teleoperation can be measured by comparing the time
connected to a computer. Therefore we can store videos and between stop signal occurred and the time when the brakes
transmit them to telepresence room using WLAN. Fig. 3 is a start to operate by the user's stop command. From the
3D rendering of a portable omnidirectional camera developed measured delay time, you can get free running time and free
using eight 4K UHD cameras. The diameter of the developed running distance.
omnidirectional camera is 30(cm), and the weight of it is 2.5
(kg) A. Experiment of emergency stops
The emergency stop experiment is to determine how long
it takes for AMR to execute a user's stop command after
receiving a stop signal from a stop signal generator when the
AMR is driving at a predetermined speed. The stop signal has
a visual signal and an auditory signal. The visual signal
means that red color sign is output from a stop signal
generator installed in the driving direction of the AMR. The
auditory signal means that beep sound is output from a stop
signal generator. Fig. 5 shows the structure of the emergency
stop experiment.
Fig. 3. The 3D rendering of omnidrectional camera
D. Developemnt of autonomous mobile robot
In this study, we developed an AMR that can be remotely
controlled. Fig. 4 is a system block diagram of an AMR. The
embedded computer stores the videos obtained by the
omnidirectional camera and transmits them to the telepresence
room using WLAN. In addition, it receives the signal of the
remote controller manipulated by the user of the telepresence
room and transmits it to the MCU of AMR. The MCU directly
controls the motor, and it transmits the values of encoder and
an accelerometer sensor to the embedded computer.
Fig. 5. The method of emergency stop experiments
B. Result of experiments
Fig. 6 illustrates how to find the delay time of the remote
driving control system for AMR based on telepresence. In Fig.
6, the tsg is the time the stop signal generated and the tso is the
time the stop signal is displayed in telepresence room, and the
tuc is the time the user in the telepresence room manipulated
the stop command device. Also, the tro is the time the AMR
actually executes the stop command. Finally, the Td is a delay
time in the remote driving control system for AMR based on
telepresence. Td is obtained thorough (1). In the fig. 6 the tf is
free running time of AMR, and the tr is reaction time of AMR.
Fig. 7 shows result of some experiments. (2) is the formula for
calculating the free running distance.
Fig. 4. The system block diagram of autonomous mobile robot Td = tf - tr =(tro - tsg )-(tuc - tso ) (1)
III. EXPERIMENT
For verify the reliability of the remote driving control tf = Current Velocity × (Td + tr ) (2)
system for mini-tram based on telepresence, emergency stop
experiments were conducted. In the emergency stop
experiments, a stop signal is generated at a random time when
the mobile robot is driving at a constant speed. The delay time
- 412 -
Table 1 is a table showing the free running distance, free
running time, and delay time obtained through the result of
auditory and visual signal experiments conducted at each
speed. In Table 1, the average of free running times is
497.5(ms), the average of delay times in the auditory stop
signal experiments is 24.5(ms), and the average of delay times
in the visual stop signal experiments is 38.7(ms).
TABLE I. EMERGENCY STOP EXPERIMENT RESULT OF REMOTE
DRIVING CONTROL SYSTEM BASED ON TELEPRESENCE
Driving
Stop Free Running Free Running Delay time
Velocity
signal Distance(m) Time(ms) (ms)
(m/s)
Fig. 6. The analysis of free running time and reaction time 1.596 532 48
1.437 479 33
Visual 1.545 515 44
signal
1.569 523 39
1.888 513 40
3
1.458 486 24
1.497 499 24
Auditory 1.485 495 24
signal
1.443 481 25
1.545 515 25
2.045 552 36
1.960 490 55
(a) Experiment 1 : driving speed 3(m/s)
Visual 2.036 509 30
signal
1.880 470 28
1.888 472 24
4
2.904 509 25
2.988 489 25
Auditory 2.916 494 25
signal
3.036 469 25
2.868 472 25
3.337 562 41
2.928 488 28
Visual 2.958 493 35
signal
(b) Experiment 2 : driving speed 4(m/s)
2.838 473 41
2.958 493 55
6
2.904 484 25
2.988 498 23
Auditory 2.916 486 24
signal
3.036 506 24
2.868 478 25
IV. CONCLUSIONS
In this study, the remote driving control system applied
telepresence technology is proposed. The stability of the
(c) Experiment 3 : driving speed 6(m/s)
system is verified by calculating delay time through the free
running time and reaction time measured in the experiment.
Fig. 7. Result graphs of emergency stop experiments Existing remote driving control systems are difficult to receive
various information on driving site. However, our proposed
- 413 -
system provided a lot of information obtained in the driving an autonomous tram vehicle," Journal of the Electrical Society,vol. 71,
site to remote user through omnidirectional camera with built no. 3, pp.525-533, 2022.
in microphones and various sensors. The remote driving [4] R. Sharma, K. Kumar, and S. Vig, "DTMF based remote control
system," IEEE International Conference on Industrial Technology pp.
control system based on telepresence is expected to be 2380-2383, 2006.
effective in solving problems quickly in certain situations the [5] S. Lu, M. Zhang, T. Ersal, and X.J Yang, "Workload management in
require human intervention for mini-tram. teleoperation of unmanned ground vehicles: Effects of a delay
compensation aid on human operators’ workload and teleoperation
performance," International Journal of Human–Computer Interaction
vol. 35, no .19, pp. 1820-1830, 2019.
This results was supported by ”Regional [6] L. Kang, W. Zhao, B. Qi, and S. Banerjee , "Augmenting self-driving
Innovation Strategy (RIS)” through the National with remote control: Challenges and directions," In Proceedings of the
Research Foundation of Korea(NRF) funded by the 19th international workshop on mobile computing systems &
Ministry of Education(MOE)(2021RIS-004) applications, pp. 19-24, February 2018.
[7] C. E. Lin, C. C. Li, A. S. Hou, and C. C. Wu, "A real-time remote
control architecture using mobile communication," IEEE Transactions
REFERENCES on Instrumentation and Measurement, vol. 52, no. 4, pp. 997-1003,
[1] Jae-Hee Kim, Dae-Hyun Kim, Moo-Ryong Choi, Hyuk-Jin Yoon, "Bus 2003.
autonomous driving technology trends," World of Electricity, vol. 68, [8] J. V. Draper, David. B. Kaber, and John. M. Usher, "Telepresence,"
no. 6, pp.16-22, 2019. Human Factors, vol. 40, no. 3, pp. 354–375, 1998.
[2] Sung-Won Park, Kyung-Joon Choi, Hyeon-Chyeol Hwang, and Jae-Ho [9] Jung-Wook Hwang, Tae-Hyung Park, "Force feedback system of
Kwak, "A dynamic velocity profile generation method for autonomous telepresence robot for remote operation," Journal of the Electrical
driving of a tram vehicle,"Journal of the Electrical Society, vol. 69, no. Society, vol. 64, no. 4, pp. 586-591, 2015.
10, pp.1569-1577, 2020. [10] Van Kha Ly Ha, Rifai Chai, and Hung T. Nguyen, "A telepresence
[3] Sung-Won Park, Kyung-Jun Choi, Hee-Gyun Jeon, Hyeon-Chyeol wheelchair with 360-degree vision using WebRTC," Applied Sciences
Hwang, and Jae-ho Kwak, "Design of a velocity tracking controller for vol. 10, no. 1, pp. 369, 2020.
- 414 -
On the Derivation of Demand Time for
Response to Variable Function Focus Camera
Situation
*Note: Sub-titles are not captured in Xplore and should not be used
Min Joong Kim 1, Myung Sung Kim 1 and Young Min Kim 1*
Department of Systems Engineering, Ajou University, Korea
1
Email: [email protected], [email protected], [email protected]
Abstract—Today, as the spread of vehicles with autonomous particular, camera sensors are widely used not only for
driving functions increases, sensors are essential to stably detect detecting moving objects but also for recognizing traffic lights
and effectively track moving objects around. Various sensors and road signs. However, traditional camera sensors have
such as cameras, radars, and lidars are used to detect moving limitations in proactive perception due to their fixed field of
objects. In particular, camera sensors are widely used not only view. A variable field of view camera has been proposed to
to recognize moving objects but also to recognize traffic lights adaptively adjust its field of view according to the surrounding
or traffic signs. However, since the existing camera sensor has a situations and driving environment [4]. Authors conducted an
fixed angle of view, there is a limit to active recognition in the
analysis of the RSS model and its characteristics. They applied
surrounding situation. Therefore, a variable focus function
variables obtained from previous studies to derive the safe
camera with a different angle of view depending on the
surrounding situation or driving environment has been
distance. Additionally, they performed a timeline analysis of
proposed. In this paper, the RSS model was used to derive the each function of the variable focal length camera with the
range of angle of view required for situations occurring application of the RSS model during the operation of Highway
according to the driving environment and the situation response Driving Assist (HDA) [5]. In this paper, the required field of
time of the variable focus function camera. view range and response time of the variable field of view
camera are derived using the RSS model to address situations
Keywords—component, formatting, style, styling, insert (key that arise in different driving environments.
words)
II. BACKGROUND
I. INTRODUCTION
Nowadays, cars have become a common means of
transportation and have deeply ingrained themselves in our
lives to the point where it would be inconvenient without them.
In order to implement autonomous driving capabilities,
various perception sensors such as cameras, radars, and
LiDARs are utilized to detect vehicles and obstacles in the
front [1]. To demonstrate the safety and reliability of
autonomous vehicles (AVs), extensive testing is required. In
order to prove that AVs are more reliable and safe than
humans in terms of fatalities and injuries, it is necessary to
accumulate hundreds of millions of miles of driving data [2].
There are two main methods for testing the safety and
reliability of vehicles: using actual physical vehicles and
employing sophisticated simulation software programs. Using
actual physical vehicles entails high costs and risks. Therefore, Figure 1 Variable Functional Focus Camera Concept
reliable data can be obtained and the high costs and risks
associated with physical vehicle usage can be avoided by
A. Variable Function Focus Camera
utilizing autonomous driving simulation software. By
constructing road segments within the simulation software or As the name suggests, a variable function focus camera is
importing real road segments and creating vehicles, traffic a single camera that can adjust its focus and field of view
signals, infrastructure, etc., simulations can be performed. (FOV). Figure 1 illustrates a schematic diagram of the concept
Additionally, using virtual environments enables the testing of of a variable function focus camera. A variable function focus
various functionalities and application aspects that may not be camera length can change its field of view depending on
feasible to test in real-world environments. This can provide factors such as driving speed and environment. The FOV of a
more comprehensive guidelines and aid in the development camera with variable focal length is 28° (telephoto), 50° (mid-
and implementation of autonomous driving technology. As range), and 150° (wide-angle). For example, when driving at
the number of vehicles equipped with autonomous driving high speeds on a highway, the camera can measure distant
features increases, ensuring safety and reliability becomes an objects using a telephoto angle. In contrast, when driving at
important task. Recently, Mobileye proposed a white-box lower speeds in urban areas, the camera can use a wide-angle
mathematical model called RSS (Responsibility Sensitive perspective to detect a broader area.
Safety) to secure the safety of autonomous vehicles and
clearly assign responsibility in the event of an accident [3]. In
- 415 -
B. Responsibility Sensitive Safety approximately 60m for a FOV of 150 degrees and about 7.5m
The Responsibility Sensitive Safety (RSS) model is a for a FOV of 50 degrees. Therefore, when driving at 60 km/h
safety standard that regulates the behavior of autonomous at an intersection, if the camera's FOV is set to 50 degrees, the
driving systems, emphasizing responsibility-centered safety. maximum detectable distance of 7.5m is smaller than the RSS
The RSS model was developed to ensure a fair distribution of safety distance of 16m, which can be risky. However, if the
responsibility for potential accidents during driving. This FOV is set to 150 degrees, the detection distance of
model evaluates factors such as distance from surrounding approximately 60m ensures an adequate safety margin. Thus,
vehicles, collision potential, and maintains safe driving by at an intersection, a FOV of 150 degrees and a safety
requiring autonomous vehicles to adhere to a set of rules, assurance time of 1 second are derived, and if all procedures,
including maintaining appropriate speed and distance. RSS including braking, can be completed within this time frame, it
focuses on minimizing risks arising from interactions between can be considered safe.
autonomous systems and humans, enabling all participants on IV. CONCLUSION
the road to coexist safely. Here, 𝑣𝑣𝑟𝑟 and 𝑣𝑣𝑓𝑓 are the longitudinal
speeds of the vehicle, and 𝜌𝜌 is defined as the reaction time, With the advancement of the automotive industry, the
and if deceleration is made to 𝑎𝑎𝑚𝑚𝑚𝑚𝑚𝑚,𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏 by 𝑐𝑐𝑓𝑓 , 𝑐𝑐𝑟𝑟 must proliferation of vehicles equipped with autonomous driving
capabilities has increased, bringing the issue of vehicle safety
accelerate to 𝑎𝑎𝑚𝑚𝑚𝑚𝑚𝑚,𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎 during the response time and then
to the forefront. As a result, various research studies have been
decelerate to at least 𝑎𝑎𝑚𝑚𝑚𝑚𝑚𝑚,𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏 to avoid a collision. At this conducted to determine the safe distance for collision
time, the safety distance is as shown in Equation (1). avoidance with the preceding vehicle during driving. Recently,
Mobileye has introduced the RSS (Responsibility-Sensitive
1
𝑣𝑣𝑟𝑟 𝜌𝜌 + 𝑎𝑎𝑚𝑚𝑚𝑚𝑚𝑚,𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎 𝜌𝜌2 Safety) model to ensure the safety of autonomous vehicles and
1
𝑑𝑑𝑚𝑚𝑚𝑚𝑚𝑚 = � �𝑣𝑣 +𝜌𝜌𝑎𝑎 2 � (1) assess liability. A variable function focus camera is being
� 𝑣𝑣𝑓𝑓2
+ 𝑟𝑟 𝑚𝑚𝑚𝑚𝑚𝑚,𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎 − developed, which can change the field of view from 28° to 150°
2𝑎𝑎𝑚𝑚𝑚𝑚𝑚𝑚,𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏 2𝑎𝑎𝑚𝑚𝑚𝑚𝑚𝑚,𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏
+ depending on the surrounding situation and driving
𝑙𝑙𝑙𝑙𝑙𝑙
environment. In this paper, we integrated the RSS safe
It is defined here as 𝑣𝑣1,𝜌𝜌 = 𝑣𝑣1 + 𝜌𝜌𝑎𝑎𝑚𝑚𝑚𝑚𝑚𝑚,𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎 , 𝑣𝑣2,𝜌𝜌 = 𝑣𝑣2 − distance into a camera with variable focal length to derive the
𝑙𝑙𝑙𝑙𝑙𝑙
𝜌𝜌𝑎𝑎𝑚𝑚𝑚𝑚𝑚𝑚,𝑎𝑎𝑎𝑎𝑎𝑎𝑒𝑒𝑒𝑒 , where the two vehicles accelerate laterally to required field of view for different driving environments. We
𝑙𝑙𝑙𝑙𝑙𝑙
each other with 𝑎𝑎𝑚𝑚𝑚𝑚𝑚𝑚,𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎 , then decelerate until the lateral calculated the safe distance for intersection scenarios and the
𝑙𝑙𝑙𝑙𝑙𝑙
speed reaches zero with 𝑎𝑎𝑚𝑚𝑚𝑚𝑚𝑚,𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏 , and finally the distance perception range for laterally moving vehicles. This allowed
between the two vehicles must be 𝜇𝜇. At this time, the lateral us to determine the necessary safe distance and camera field
safety distance is as shown in Equation (2). of view for ensuring safety while driving at intersections.
Additionally, we derived the required response time for
2 ensuring safety in various situation-specific demands.
𝑣𝑣1 +𝑣𝑣1,𝜌𝜌 𝑣𝑣1,𝜌𝜌
𝜌𝜌 + 𝑙𝑙𝑙𝑙𝑙𝑙
2 2𝑎𝑎𝑚𝑚𝑚𝑚𝑚𝑚,𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏
𝑑𝑑𝑚𝑚𝑚𝑚𝑚𝑚 = 𝜇𝜇 + � 2 � (2)
𝑣𝑣2 +𝑣𝑣2,𝜌𝜌 𝑣𝑣2,𝜌𝜌
−� 𝜌𝜌 − 𝑙𝑙𝑙𝑙𝑙𝑙 � REFERENCES
2 2𝑎𝑎𝑚𝑚𝑚𝑚𝑚𝑚,𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏
+ [1] Chavez-Garcia, R. O., Burlet, J., Vu, T. D., and Aycard, O. "Frontal
object perception using radar and mono-vision." 2012 IEEE Intelligent
Vehicles Symposium, pp. 159-164, 2012.
[2] Kalra, N., and Paddock, S. M. “How many miles of driving would it
III. METHODOLOGY take to demonstrate autonomous vehicle reliability.” RAND
The variable function focus camera has a field of view and Corporation, pp. 1129-1134, 2016.
maximum detection range of 150 degrees (50m), 50 degrees [3] Shalev-Shwartz, S., Shaked S., and Amnon S. "On a formal model of
safe and scalable self-driving cars." arXiv preprint arXiv:1708.06374,
(100m), and 28 degrees (250m). For example, assuming a 2017.
vehicle speed of 60 km/h and a reaction time of 0.5 seconds. [4] Kim, M. J., Yu, S. H., Kim, T. H., Kim, J. U., and Kim, Y. M. "On the
The maximum acceleration and deceleration of the vehicle are development of autonomous vehicle safety distance by an RSS model
set to the values proposed by ISO 22179 international standard based on a variable focus function camera." Sensors, 21(20), 6733,
for the FSRA (full speed range adaptive cruise control) system, 2021.
which are 𝑎𝑎𝑚𝑚𝑚𝑚𝑚𝑚,𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎 = 4𝑚𝑚/𝑠𝑠 and 𝑎𝑎𝑚𝑚𝑚𝑚𝑚𝑚,𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏 = [5] Kim, M., Kim, T., Yu, S., and Kim, Y., “Safety Verification of RSS
𝑎𝑎𝑚𝑚𝑚𝑚𝑚𝑚,𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏 = 4.9𝑚𝑚/𝑠𝑠 [6]. By substituting the variable values Model-Based Variable Focus Function Camera for Autonomous
Vehicle”, SAE Int. J. of CAV 5(3), 2022.
into equation (1), an RSS (recommended safe stopping)
[6] Ahn D. R., Shin S. G., Park K H., Choi I. S. and Lee H. K. “Functional
distance of approximately 16m can be derived. Dividing the safety concept design and verification for longitudinal driving
safety distance by the current vehicle speed gives a safety assistance system of an autonomous vehicle.” Transactions of KSAE,
assurance time of about 1 second. vol. 26, no. 2, pp.149-158, 2018.
At an intersection, the maximum distance at which a
horizontally approaching vehicle can be detected is
- 416 -
Machine-Learning based Condition Diagnosis
Model of Heating, Ventilation and Air
Conditioning System for Train Cars
Juhee Choi1, Kangwoo Nam2 and Chul-Su Kim2*
1
Department of Smart Information Communication Engineering, Sangmyung University, Cheonan, Korea
2
School of Railroad Engineering, Korea National University of Transportation, Uiwang 16106, Gyeonggi, Korea
Email: [email protected]
Abstract—The air conditioning unit in railway vehicles is a appropriate control system is needed. Especially, it is required
device that integrates heating, cooling, and ventilation systems to adjust temperature and humidity control through sensors
for passengers. It is usually called the Heating, Ventilation, and directly.
Air Conditioning (HVAC) system, as the name represents its
functions. To check the status of this device, the train control
monitoring system (TCMS) continuously monitors the
operation of the device. However, the current system shows
irregularities and codes indicating system failures without
predicting the type and the timing of malfunctions. This study
develops a condition diagnosis model through the analysis of
actual failures from the last 6 years and the corresponding
remedial actions. The model includes the failure mode, effects,
and criticality analysis method (FMECA) with risk priority
number analysis.
Keywords—Machine-Learning, Condition Diagnosis Model, Fig. 1. Heating, Ventilation, and Air Conditioning System
Heating, Ventilation and Air Conditioning, Railway Vehicles
II. METHODS
I. INTRODUCTION
A. Failure Mode, Effects, and Criticality Analysis (FMECA)
In urban railway systems, the Heating, Ventilation, and
For the Failure Mode, Effects, and Criticality Analysis
Air Conditioning (HVAC) can be considered an important
(FMECA) of the HVAC, several criteria are required in terms
device for convenience and safety during operation[1-3].
of the severity of failure, the frequency of failure, and the
Considering the characteristics of subways, when lots of
detectability of failure [7-9].
people gather and trains are continuously running, the amount
of oxygen goes down while the heat and pollutants in the air Table I shows that the detailed criteria for the severity of
go up. failure. Severity rankings in the table mean that each failure
mode is assigned based on the effects, the overall system
Thus, the mechanical ventilation systems have two main
performance, and the safety.
jobs: they send fresh air around, so the commuters stay
comfortable while they're traveling, and they are important Next, the occurrence frequency evaluation is obtained
when it comes to keeping everyone safe in case of fire. from the failure rate values of the failure modes. If λ is the
Besides, having good ventilation is also important for keeping failure rate of a component, the mode failure rate is λμ and the
both tunnels and stations clean and safe. In addition, the calculation formula is as follows in Equation (1)
HVAC is required to meet the customer’s convenience in
terms of warmth. 𝜆𝜇 = α ∙ 𝜆 (1)
To overcome these problems, the sensors are installed on where α indicates the weight of the mode compared to other
the roof of the train to detect temperature and air quality failure modes.
sensors int the train control monitoring system (TCMS)[4-6]. Since there is no existing research on the detectability of
Based on the data of the sensors, the system makes internal failures, we propose a detectability criteria scale ranging from
temperature stay and maintain the amount of oxygen by 1 to 3 as shown in Table II.
recirculating air.
TABLE I. SEVERITY EVALUATION CRITERIA[9]
The main function of the HVAC is to pull in warm air from
outside and the refrigerant absorbs the heat. Next, it sends the Severity Criteria Evaluation
gathered heat outside the train. As a result, it provides the fresh None No signal 1
air with the passengers. When the sensors estimate the
temperature and quality of the air in the train, they check the Very Minor Comfort hurt 2
air to be recirculated and provides the filtered air inside. Minor Possible failure of one part 3
Each carriage is installed in the HVAC which gives air Very Low
Partial failure of
4
conditioning functions such as cooling and ventilation as functionality
shown in Fig. 1. To control these devices correctly, an
- 417 -
Severity Criteria Evaluation ACKNOWLEDGMENT
Functionality failure of one
Low
component
5 This work was supported by the Korea Agency for
Infrastructure Technology Advancement (KAIA) by the
Medium One function failure 6
Ministry of Land, Infrastructure and Transport under Grant
High Two functions failure 7 RS-2019-KA152194..
Very High All functions failure 8 REFERENCES
Warning Fire possibility 9 [1] F. Yangyang, et al, “Utilizing commercial heating, ventilating, and air
conditioning systems to provide grid services: A review,” Applied
Alarm Immediate danger 10 Energy, 2022, vol. 307, pp. 118133.
[2] Y. Yanzhe, et al, “A review on available energy saving strategies for
TABLE II. DETECTABILITY OF FAILURES[9] heating, ventilation and air conditioning in underground metro stations,”
Renewable and Sustainable Energy Reviews, 2021, vol, 141, pp.
Criteria Evaluation 110788.
[3] L. Guannan, et al. “Review on fault detection and diagnosis feature
Fully detectable 1
engineering in building heating, ventilation, air conditioning and
Partial detectable 2 refrigeration systems,” IEEE Access, 2020, vol. 9, pp. 2153-2187.
[4] L. Changyuan, L. Xiaoming, Y, Panpan, “Train control management
Undetectable 3 system safety assessment,” International Conference on Electrical and
Information Technologies for Rail Transportation (EITRT2013)-
Volume II, 2014. pp. 583-591.
Finally, as we have discussed so far, we propose Risk [5] M. Rekik, C.Gransart, M. Berbineau, “Analysis of security threats and
Priority Number (RPN) based on three criteria referring to the vulnerabilities for train control and monitoring systems,” International
SAE-J1739 standard[10]. It was calculated by multiplying the Multi-Conference on Systems, Signals & Devices (SSD). IEEE, 2018.
pp. 693-698.
occurrence frequency, severity, and detectability[9].
[6] S. H. Han, S. G. Lee, W. K. Kim, “Development of onboard train
RPN = Occurrence (O) × Severity (S) × Detectability (D) (2) automatic control system for Korean standard EMU,” In: ISIE 2001.
2001 IEEE International Symposium on Industrial Electronics
Proceedings. IEEE, 2001. pp. 1257-1259.
[7] D. Yongliang, L. Qiming, L. Ying, “A research on subway physical
B. Machine Learning based Condition Diagnosis vulnerability based on network theory and FMECA,” Safety science,
To build a model for prediction of the condition diagnosis 2015, vol. 80, pp. 127-134.
using machine learning algorithms, we utilize the history of [8] IEC 60812 Technical Committee, “Analysis techniques for system
the temperature change of electric vehicles. Raw data from the reliability-procedure for failure mode and effects analysis (FMEA),”
TCMS is converted to time series data. From the operation IEC 60812, 2006.
data of the electric vehicle that has been. We use the bagging [9] K. Nam, C. Kim, “A Study on the Condition Diagnosis Model of
HVAC System for Urban Railway Vehicles,” Master's thesis, Korea
tree, which is one of the schemes of machine learning to National University of Transportation, 2023.
analyze the data. To aim it, a preprocessing process is required [10] C. S. Carlson, “Effective FMEAs: Achieving safe, reliable, and
to modify the temperature history by date to match the economical products and processes using failure mode and effects
operation characteristics of the electric vehicle, which analysis,” John Wiley & Sons, 2012.
operates from dawn until dawn the next day.
Fig. 2. Example of merging data for machine learning
III. FUTURE WORKS
Even though this study found that R² score reached the
highest(0.8999) in 6 min and followed by 5 min, it is still
needed to investigate the further study to find the impact of
the other parts of the vehicles.
- 418 -
A Study on Route Selection Methods of
Autonomous Vehicles for Public Mobility
Jung-Ju Kim 1*, Gi-Won Kim 1, Chan-Jong Lee 1, Jun-Seok Lee 1, Jae-Hun Choi 1, Hyung-Woo Jin 1
1
Dept. of Robotics, Hoseo University, Korea
Email: [email protected]
Abstract—In order to introduce autonomous car to public the vision sensor. For the autonomous mini-tram to make
mobility prior to the development of fully autonomous vehicle, decisions and drive using this algorithm, the color indicative
a new system is needed to ensure driving performance and of the route it is to drive needs to be pre-established.
stability while minimizing human intervention. In order to solve
this problem, we introduced an autonomous guidance lane A. Operation of multi-complex routes for public mobility
system that displays the driving route of autonomous vehicle by A complex route is a route that overlaps with two or more
painting on the surface of the road. Public mobility used in single routes. To operate a multi-complex routes, we use
urban areas is operated in multiple routes in most cases, and different colors for the guidance lanes of each single route and
some sections of the multiple routes are operated by multiple complex route. Fig. 1 is a multi-complex routes map for test
cars in common. In this case, the autonomous mini-tram must operation of autonomous mini-tram for public mobility in an
recognize and drive its own route. In this paper, we propose a
industrial complex located in Seokmun-myeon, Dangjin-si,
method for autonomous car for public mobility to recognize and
drive on their own route by using autonomous guidance lane
Chungcheongnam-do. A total of three routes were set up for
system. test operation of autonomous mini-tram. Colored guidance
lanes were installed on the road surface of each route.
Keywords—autonomous vehicles, route selection, public
mobility, guidance lane for autonomous vehicle, mini-tram
I. INTRODUCTION
Traffic congestion and environmental issues are
constantly increasing in modern cities. As an alternative,
autonomous driving technology is attracting attention, and
various studies are being conducted to apply autonomous
driving technology in public mobility[1-4]. Mini-trams are
small-scale trams that can be used as a next-generation
mobility service in cities, and various studies are underway to
commercialize them[5-6]. However, most mini-trams
currently use the method to bury magnet on road method to
build infrastructure, which incurs high costs. In addition, when
multiple routes of mini-trams overlap, difficulties arise in
route identification and operation management, so various
studies are being conducted for accurate route selection[7-10].
In this study, we propose an autonomous driving method
for mini-tram to be utilized as public mobility through
Fig. 1. The multi-complex routes map for autonomous mini-tram
guidance lanes painted on roads. In particular, we propose a
route selection algorithm for operating multi-complex routes
of mini-tram. To this end, we developed colored guidance
lanes for formation of multi-complex routes. We also B. Guidance lane for autonomous vehicle
developed an algorithm that can select a route using the color The guidance lane for autonomous vehicle is a lane
difference of each route and verified its performance through installed on the road to select the route of the autonomous
experiments. mini-tram for public mobility. The guidance lane for
II. BODY autonomous vehicle can be set in various colors and forms
such as Fig. 2. The autonomous mini-tram recognizes
The CGL(Colored guidance lane) is necessary for the guidance lanes using vision sensor. Fig.3 is the multi-complex
operation of the multi-complex routes for autonomous public route consisting of two single routes and one complex route.
mobility. In this study, we installed colored guidance lanes on In the case of autonomous mini-tram that operate Route A, it
the road surface for the autonomous mini-tram. We developed is set to follow green and red colored guidance lanes, so it
a route selection algorithm that allows the autonomous mini- turns left along the red colored guide lane after going straight
tram to recognize the colored guidance lanes, select its route, along the green guide lane.
and drive accordingly. The route selection algorithm utilizing
colored guidance lanes recognizes the shape and color of the
colored guidance lanes extracted from the images acquired via
- 419 -
• Route selection using histogram of hue
Fig. 5 (a) is an image of colored guidance lanes
obtained through the preprocessing and lane
extraction algorithm from an image acquired using a
vision sensor. Fig. 5 (b) is a histogram of the hue
values extracted from the image of colored guidance
lanes. The autonomous mini-tram selects the route by
comparing the hue histogram extracted from the read
image using a vision sensor with the hue value set to
drive in advance.
Fig. 2. Guidance lanes for autonomous vehicle
(a) Recognized CGLs (b) Histogram of CGLs
Fig. 5. Histogram of colored guidance lanes
Fig. 3. Multi-complex routes using colored guidance lanes
III. EXPERIMENT
C. Route Selection using colored guidance lanes We evaluated the performance of route selection algorithm
The image of the colored guidance lanes detected by the and colored guidance lanes using an electric car. Experiments
vision sensor includes many noises due to the road surface or were conducted on pre-set multi-complex routes. The electric
surrounding environment. For route selection of autonomous car drove on predetermined routes through pre-set colors.
mini-tram, noises are removed through the preprocessing to A. Route selection at double junction
increase the recognition rate of the colored guidance lane.
Fig. 6 shows the route selection result of the autonomous
Next, the conversion of color space is operated to HSV color
vehicle driving on the colored guidance lanes installed on the
space from RGB color space.
road surface. Fig. 7 is a graph of distance-histogram generated
• Color space conversion from a video obtained by an vision sensor attached to
autonomous vehicle while driving along a guidance lanes. The
To recognize the color of the colored guidance lanes autonomous vehicle used in the experiment was programmed
from the preprocessed image, the hue value in the to drive on a blue single lane and a green complex lane. The
HSV color space is utilized. (1) and (2) are used to autonomous vehicle accurately selected its own route using
convert the RGB color space to the HSV color space. route selection algorithm and colored guidance lanes.
Fig. 4 shows the hue values in the HSV color space
obtained using (1) and (2).
V = max(R, G, B) ()
60(G-B)
if V = R
V- min(R, G, B)
60(B-R)
H= 120+
V- min(R,G,B)
if V = G ()
60(R-G)
{ 240+ V- min(R,G,B) if V = B
(a) Original image (b) Route selection result
Fig. 6. Result of route selection at double junction
Fig. 4. Color map of hue in HSV color space
- 420 -
guidance lanes and route selection algorithm based on
computer vision. The experiments confirmed that the vision
sensor can select the route to drive based on the color
difference of the guidance lanes. Existing autonomous mini-
trams require construction such as reclamation of magnet and
need various expensive sensors. However, since autonomous
mini-tram that use route selection algorithm using colored
guidance lane choose the driving route with a single vision
sensor, reclamation of magnet and expensive sensors are not
required, so it can be introduced at a lower cost than
autonomous mini-tram.
Fig. 7. The distance-histogram graph of double junction experiment
This results was supported by ”Regional
Innovation Strategy (RIS)” through the National
B. Route selection at triple junction Research Foundation of Korea(NRF) funded by the
Ministry of Education(MOE)(2021RIS-004)
Fig. 8 shows the route selection result of triple junction
experiment. The electric car is set to drive the single route in
yellow and the multi-complex route in green. Fig. 9 is a graph REFERENCES
of distance-histogram generated from a video obtained by an
electric car while driving along a guidance lanes. Fig. 9 shows [1] M. Salazar, F.Rossi, M. Schiffer, C. H. Onder, and M.Pavone, "On the
that the correct route selected at the triple junction. interaction between autonomous mobility-on-demand and public
transportation systems," In 2018 21st international conference on
intelligent transportation systems(ITSC), pp. 2262-2269 ,IEEE,
November 2018.
[2] A.Bucchiarone, S.Battisti, A.Marconi, R.Maldacea, and D.C.Ponce,
"Autonomous shuttle-as-a-service (ASaaS): Challenges, opportunities,
and social implications," IEEE Transactions on Intelligent
Transportation Systems, vol. 22, no. 6, pp. 3790-3799, 2020.
[3] J.Grippenkoven, Z.Fassina, A.König, and A.Dreßler, "Perceived safety:
A necessary precondition for successful autonomous mobility
services," Proceedings of the human factors and ergonomics society
Europe, pp. 1-15, 2018.
[4] E.Kassens Noor, Z.Kotval Karamchandani, and M.Cai, "Willingness to
ride and perceptions of autonomous public transit," Transportation
Research Part A: Policy and Practice, vol. 138, pp. 92-104, 2020.
[5] Chan-Soo Kim, Kwang-Sup Lee, Rak-Gyo Jung, Seok-Won Kang,
"Development of a Korean minitram vehicle system," World of
Electricity, vol. 64, no. 9, pp. 14-20, 2015.
[6] Yoon-Seop Byun, Rak-Kyo Jung, Hee-Chang Moon, and Hoon-Je
Woo, "Development of a driving control system for minitrams,"
Electricity in the World, vol. 64, no. 9, pp. 21-28, 2015.
(a) Original image (b) Route selection result
[7] Dong-Hoon Lee, Dong-Sik Kim, Jong-Ho Lee, and Dong-Won Kim,
Fig. 8. Result of route selection at triple junction "Real-time Hybrid Path Planning Algorithm for Mobile Robot,"Journal
of the Electrical Society, vol. 63, no. 1, pp. 115-122, 2014.
[8] M.Liu, X.Deng, Z.Lei, C.Jiang, and C.Piao, "Autonomous lane keeping
system: Lane detection, tracking and control on embedded system,"
Journal of Electrical Engineering & Technology, vol. 16, pp. 569-578,
2021.
[9] C.Katrakazas, M.Quddus, W.H.Chen, and L.Deka, "Real-time motion
planning methods for autonomous on-road driving: State-of-the-art and
future research directions," Transportation Research Part C: Emerging
Technologies, vol. 60, pp. 416-442, 2015.
[10] D.H.Lee, and J.L.Liu, "End-to-end deep learning of lane detection and
path prediction for real-time autonomous driving," Signal, Image and
Video Processing, vol. 17, no. 1, pp. 199-205, 2023.
Fig. 9. The distance-histogram graph of triple junction experiment
IV. CONCLUSION
In this paper, we propose a route selection method for
autonomous mini-trams to be used in public mobility, using
- 421 -
A Model for Implementing Network-Connected
Systems to Train Control Systems
Gyuoyoung YEON 1 , Jaeho LEE2 and Jongwoo LEE 1*
1
Graduate School of Railway, Seoul Nation University of Science and Technology
2
Korea Railroad Research Institute, Korea
Email: [email protected], [email protected], [email protected]*
Abstract -Train control systems are composed of different following train, and system operation as shown Figure 1. Each
signaling subsystems which are connected each other through systems works with a kind of states information and yield
hard wired communication cables. In virtue of internet control information. The subsystems being part of a system
technology progressing, the internet communication systems are only output its state information such that track circuits. Some
gradually introducing to the train control systems. This paper subsystems receive the output of exterior systems and output
reviewed how internet communication does implement to the control signal such that interlocking systems. [1]
train control systems by considering system architecture
configuration, data structure, address allocation and
communication speed.
Keywords-train control system, IP system, system architecture,
data structure
I. INTRODUCTION
As a result of Internet systems having been rapidly
progressing, hyper connected society comes to true, which all
systems would be connected through internet systems, which
it is called IP based systems. In railway systems, many
domains become informatized such as information systems,
ticketing, scheduling, resource management and so on…
The IP based systems are applied to train control systems
which are safety critical systems. Railway systems are
composed of a lot of various systems, which coexist as past,
present and future systems. It means that it is very difficult to
get to computerize all systems at once. Most railway systems
are to be IP-based systems with incremental renovation.
Fig. 1. Train Control System Architecture
Train control systems consist side-by-side of mechanical,
electronic and computerized systems. Nonetheless many train The train control systems have to exchange the states and
control systems get to be informatized, but all systems aren’t control information through communication systems. In the
converted to IP based systems. past, the communication has achieved with analog system and
hard wired system. Nowadays, by virtue of computer-based
Train control systems are safety critical systems which communication and optic-fiber systems, train control systems
ensure train control safety with railway system states and and communication systems become intelligent electronic
control. The safety-critical states and control information flow devices. The intelligent train control systems get high
in closed communication networks to keep its security. But the throughput for data communication and have a various
electronic systems connect one to one for each other with hard functions keeping its fundamental functions. [2]
wired systems for data communication. The electronic
systems easily maintain its security because the Railway lines are divided into two regions, one includes
communication lines are reserved only for its communication. turning switches, another no switches. The interlocking
system controls the region having switches. The interlocking
If IP-based systems are implemented for train control system control the routes and train speed in the region by the
systems, the IP-based systems are required to get the same control of points and lineside signals. The region without
level or the above of the conventional system safety. switches are controlled by block systems which consist of
In this paper, we considered how to implement IP-based track circuit and speed control logic equipment. The block
systems to train control systems. system controls train speed.
Train control systems carry out the functions which are The interlocking system and the block system cooperates
train routes control, spacing between precedent train and to set routes and train speed control. They exchange states
- 422 -
information and control the equipment such that point communication system would be adopted ad hoc
machines and lineside signals to open routes and to control whenever the need arose. [3]
train speed.
All networked systems are rapidly converted to IP-based
The function blocks in Figure 1 can be represented in systems. In the world of networking, an IP address is a unique
systems, we can define the I/O between the systems in Table numerical label assigned to each device connected to a
1. The systems are represented in row and column to define computer network that uses the Internet Protocol for
input and output between the systems in Table 1. The communication. Dedicated IP addresses also allow for
shaded(upper) rectangle in the table interior is the output of customizable signal system settings. With a shared IP address,
systems in row and input for the systems in column. The the signal system settings are controlled by the signal engineer
normal(lower) rectangle is the output of the systems in column and cannot be changed. However, with a dedicated IP address,
and input for systems in row. the user has full control over the signal system settings and
can make changes as needed. This allows for greater
TABLE I. I/O DATA BETWEEN SYSTEMS flexibility and can be especially useful for businesses that need
to make frequent changes to their signal settings.
II. 3. NETWORK BASED TRAIN CONTROL SYSTEMS
A. 3.1 IP-based train control systems
IP-based signaling systems have to install NIC(Network
Interface Card) and communicate each other with internet
protocol to get control command and states data. Computer-
based train control systems can be converted to IP-based
systems by implementing NIC and SW for IP communication.
The converted systems have to connect network and to
allocate its own IP address to communicate the other IP-based
systems.
B. 3.2 Communication network architecture for
Simplified input/output data between the systems
interlocking systems
represents in the Table 1. To transmit the data between the
systems, serial communication such as RS 232, RS 422, Interlocking systems as a system managing train operation
RS482 and giga-Ethernet have been widely used in close are connected to most of signaling systems and play a role for
range. The lineside signals and its LEU (Lineside Electronic data collecting hub. The interlocking systems places in station
Equipment) are located in long range. The communication premises because the interlocking systems run manually or
between LEU and control is used modem systems rather than automatically.
the serial communication. In recent, many fiber-optic Interlocking systems are integral to the management of
communication introduces in train control system. The fiber- train operations. They are connected to a wide range of
optic system is resilient to heavy noise environment and high signaling systems within a railway network and serve as a
speed communication. Most of railway train control systems central hub for collecting data. The primary function of an
implement fiber-optic communication phase-in. interlocking system is to ensure the safe and efficient
movement of trains by coordinating signals, switches, and
track sections.
Interlocking systems can operate in either manual or
automatic modes, depending on the specific configuration and
requirements of the railway network. In manual mode,
operators stationed at the interlocking system manually
control the routing of trains, monitor track conditions, and
manage the overall train movement process. In automatic
mode, the interlocking system uses predefined rules and
algorithms to make decisions and control train movements.
Due to their critical role in managing train operations,
interlocking systems are typically located within station
premises. Placing them in station premises offers several
advantages. Firstly, it provides easier access for operators who
need to interact with the system. Secondly, it allows for better
coordination between the interlocking system and other
station-based operations, such as ticketing, passenger
information, and station management. Lastly, centralizing the
Fig. 2. An Example of Mixed Communication System interlocking system within the station premises enables
efficient communication and integration with other signaling
Figure 1 An Example of Mixed Communication System and control systems.
This type of communication system as shown Figure 2 By being positioned within the station premises,
have been phasing communication system in. This new interlocking systems can effectively interface with various
signaling systems, collect essential data related to train
- 423 -
movements, and facilitate the necessary control and • SCI-IO (General IO): Defines the interface for each
coordination to ensure safe and efficient train operations. lineside device.
The closed network is dedicated to connecting safety- • SCI-LS (Light Signal): Defines the interface for light
critical signaling devices. It ensures secure and reliable signals.
communication between these devices, which are crucial for
maintaining the safe operation of the railway system. This • SCITDS (Track Detection System): Defines the interface
closed network is designed to handle and transmit safety- for track detection systems.
critical information between the interconnected devices. • SCI-LX (Level Crossing System): Defines the interface
On the other hand, the open network is utilized for for level crossing systems.
transmitting information that is not directly related to safety. • SCI-P (Point): Defines the interface for points (switches)
This network facilitates the flow of non-critical information, in the railway system.
such as operational data, maintenance updates, or non-safety-
related communications, between various devices within the • SCI-CC (Command Control System): Defines the
interlocking system. interface with the controller.
By separating the communication network into these two • SCI-RBC (Radio Block): Defines the interface for the
subsystems, EULYX's architecture provides a structured and radio block and interface center.
secure approach to manage both safety-critical and non- In order to ensure the integrity of communication data
safety-related information within the interlocking systems. between the transmission layer and the application layer, the
SCI utilizes a railway safety communication application
protocol called RaSTA (Railway Safe Transport Application).
This protocol has been developed to comply with standards
such as EN50159 and IEC62280, which are relevant to
railway safety networks. EULYNX prescribes the use of the
RaSTA protocol in its standards and provides equipment
development guidelines accordingly.
C. 3.3 Data Frame
IP communication involves the transmission of data
frames, which are structured units of data used for
communication. The data frame consists of OSI layer 7 and is
illustrated in Figure 4. [5][6][7]
Fig. 3. EULYNX Interlocking Interfacing Architecture [4]
EULYNX, the standardization initiative for railway
signaling and control systems, defines the interface between
the electronic interlocking device subsystem and the device-
specific subsystem using a naming convention called CI-xx.
The Standard Communication Interface (SCI) within
EULYNX is a telegram-based interface that consists of two
layers: the transport layer and the application layer. The
transport layer handles the transmission of telegrams between
the electronic interlocking device and the device-specific
subsystem, ensuring reliable communication. The application Fig. 4. IP communication layer
layer defines the structure and content of the telegrams
exchanged between the two subsystems. Architecture of RaSTA is implemented between a typical
communication stack’s application layer and the transport
It's important to note that only telegram definitions layer, as pictured in Figure 5. There are only a few
specified within the EULYNX standard can be used in the SCI. requirements on the transport layer, making RaSTA suitable
This means that any telegrams or interfaces that are not for different scenarios: It must be possible to send messages
explicitly defined by EULYNX cannot be employed within to specific receivers. The network, including the
the SCI framework. The standardization ensures communication partners, must process the messages in a best-
interoperability and compatibility between different systems effort manner. It is unnecessary to enforce deadlines, but
that adhere to the EULYNX specification. safety reasons imply fast transmission and adequate quality.
It appears that EULYNX has defined several standard Commercial of the shelf transport services such as UDP or
communication interfaces within the Standard TCP are well suited.
Communication Interface (SCI) framework. These interfaces
include:
• SCI-ILS: Defines the interface with adjacent interlocking
devices.
- 424 -
Kbytes. 100M bps of switch and routers is sufficient for the
train control systems.
Fig. 5. Architecture of the RaSTA Protocol Stack
In Figure 6, SCI-xx.PDI message stack is represented.
Fig. 7. A Network Configuration for a IP based Interlocking System
The connection between interlocking and subsystems uses L3
switches for the closed network and MPLS(Multi Protocol
Label Switch) for the open network. The closed network shall
be constructed as shown in Figure 8.
Fig. 6. SCI-xx.PDI Message Stack
Fig. 8. Switches Allocation to Subsystems
III. AN EXAMPLE OF NETWORK-BASED TRAIN CONTROL
SYSTEM The connection between interlocking systems is illustrated
in Figure 9. All devices in stations are linked with L2 switches.
A. 4.1 Interlocking data processing speed It means that each interlocking system has its own LAN. The
The required speed can be adopted according to each linking between interlocking systems applies routers.[8][9]
signaling system. The processing speed of the interlocking
system as shown Table 1 is represented in Table 2.
TABLE II. REQUIRED COMMUNICATION TRANSMISSION TIME
B. 4.2 Interface device between interlocking and Fig. 9. Closed network for train control Systems
subsystems
The interlocking system and the subsystems can be C. 4.3 IP address
constructed as shown in Figure 6. The train control system can IP address is a kind of private address for the closed
be cut off from the outside by forming a LAN network using networks. The IP address allocation is to be followed to its
an L3 switch. own principle which the addresses allocate. [10] In train
control system, address number is relatively small and
The data amount of control and states information of the communication reaching area is small with compared to
train control systems isn’t big which is less than several
- 425 -
public internet system. We can easily make routing tables for
routers.
IV. CONCLUSION
We investigated and analyzed the relationship between the
network configuration and the railway signal system to realize
the IP-based interlocking device, and presented the IP-based
network.
Ethernet communication is required to connect signaling
equipment and network equipment. Although it is necessary
to consider the safety of the NIC used for the traffic control
purpose, the basic concept for IP-based system is to configure
a basic network by connecting NIC for train control and the
L2 switch.
Since the train control system is physically grouped, it is
necessary to build a LAN network to connect this group. A
physical configuration of a LAN network can be configured
using an L2 switch.
A network construction of a train control system needs to
be constructed using a router that can connect each LAN. A
train control network consisting of large stations requires
multiple routers, and networks in the same station is treated as
LAN and must use the same protocol.
The network of the train control system needs to construct
a hierarchical network. This hierarchical network becomes a
topological tree-like structure and is easy to control.
Since the address assignment of the train control network
is a closed network, a separate address assignment method can
be constructed. The network address for train control needs to
be constructed as the public network address system, which is
easy to group signaling systems. This address method can
easily construct a LAN using a router.
REFERENCE
[1] Gregor Theeg, Sergej Vlsenko, “Railway Signalling and Interlocking ”,
International Compendium, Eurail press
[2] Marie Le Bliguet and Andreas Andersen Kjaer, “Modelling
Interlocking Systems for Railway Stations”, Kongress Lyngby 2008
[3] Bruce Chang, “Railway Communication Systems”,
http://www.oringnet.com
[4] Frans Heijnen, “EULYNX The next generation signalling
strategy for Europe”
[5] CISCO, “Internetworking Technologies Handbook”
[6] Larry L. Peterson and Bruce S. Davie, “Computer
Networks- a system approach”, mkp.com
[7] Behrouz A. Forouzan, “Data Communications and
Networking”, McGraw Hill
[8] CISCO, “IP Addressing : IP4 Addressing Configuration
Guide”
[9] Kevin R. Fall and W. Richard Stevens, “TCP/IP Illustrated
Volume 1”, Addison-Wesley
[10] Richard T. B. Ma, “Forwarding and Routing-PPT”,
[11] https://electronicspost.com/forwarding-and-routing-in-
network-layer/
- 426 -
A Study on Transient Relief Methods through
Characteristic Analysis of Electric Railway
Vehicle Spark Gap
Jongrok Choi 1, Chinyoung Chang 1, Jaemoon Kim 1, Jaeyoung Kim 1, Hojun Kwon 1
(Use * to indicate the correspondence author)
1
Korea National University of Transportation Smart & Safe Management System for Advanced Rail Transport, Korea
Email: [email protected]
Abstract—In accordance with the demand for the relief of In addition, a spark gap is a device that induces a voltage
traffic congestion in large cities, high-speed trains, drop due to insulation breakdown between electrodes in the
rationalization of energy use, and eco-friendly transportation, air. It performs a similar function to an arrester, but unlike an
the electrification project began in 1968, starting with the arrester that has stable voltage suppression and current
central line and the Taebaek line, which are industrial lines. It limiting capabilities, the ambient conditions (temperature,
contributes a lot to mass transportation, logistics cost reduction, humidity) and pollution level Depending on the environment,
and management improvement. the magnitude of the breakdown voltage varies, so the
prediction range of the suppression voltage is widely
Recently, faults such as burnout of instrument
transformers and lightning arresters have repeatedly occurred
distributed.
in insulation sections and steep gradient sections of domestic
In Figure 1, when the surge current flows, the limit voltage
commercial and industrial lines, and abnormal voltages caused
by harmonic power such as power converters have become a of the arrester is generated, and the limit voltage and the
cause of failure of electric railway vehicles. In this study, we transient voltage generated by the ground potential rise are
would like to analyze the operating characteristics of the discharged in the spark gap to protect the main transformer.
lightning arrestor and spark gap of the high-speed rail vehicle
against the abnormal voltage.
Keywords— abnormal voltage, spark gap, Insulation
Cooperation
I. INTRODUCTION
Domestic disorders such as transformers and lightning
instructors are generated repeatedly in the intersection of
domestic electric railways and industrial lines and industrial
lines, and it is causing disorders of electric rail vehicles due to Fig. 1. Composition of KTX-Sancheon Lightning Arrester SG
abnormal voltage caused by harmonics in power conversion
devices such as power conversion devices there is. In this B. KTX-Sancheon Lightning Arrester and SG Abnormal
paper, we will analyze the operation characteristics of the Operation Occurrence Status
abnormal voltage of KTX-Sancheon's lightning group and After the introduction of KTX-Sancheon, an analysis of
spark gap. the abnormal operation of the spark gap up to now confirmed
that the spark gap of the power car on the opposite side of the
II. MAIN SUBJECT power supply car (pantograph rise) operated in about 86% (12
A. Arrester and Spark Gap out of 14 times) of the total. This seems to be due to mutual
influences such as the resonance of the roof line of the front
Lightning arresters are divided into gap type and gapless and rear power cars, the lightning arrester, and the SG when
type. Spark gap is the most basic of gap type and economically the MCB is input after the pantograph rises.
used as a device that can protect the system from overvoltage.
It has been used for a long time. Due to the improvement in C. Simulation
performance, the gapless lightning arrester is in charge of the
main protection function against abnormal voltage, and the Figure 2 is an AC electric railway power supply system of
spark gap mainly performs the back-up protection function. the AT power supply method. Since the % impedance changes
In the KTX-Sancheon high-speed vehicle, a spark gap (SG) according to the position of the electric vehicle, which is the
is installed in parallel with the lightning arrester in the circuit load in the AT power supply system, simulation was
configuration of the high-voltage bushing of the main performed to analyze the operation characteristics of the
transformer, so that the surge voltage flowing into the main arrester and SG according to the resonance position.
transformer is primarily limited by the arrester, and this
limiting voltage is It serves to protect the main transformer by
grounding the secondary surge voltage in the connected spark
gap.
- 427 -
(a) Lightning Arrester + No Spark Gap (b) Arrester + With Spark Gap
Fig. 5 Surge voltage simulation result (%Z=0.5).
Fig. 2. AT-powered alternating current electric railway Here, it can be seen that in the case of no insulation
coordination, the Vpeak value increases by 168% from
· Simulation of the presence or absence of insulation the reference value, whereas in the case of insulation
coordination coordination, it increases only by 124% from the
reference value.
In order to review the surge voltage generated when
the MCB of KTX-Sancheon is turned on, the change · A graph within a graph is an “inset”, not an “insert”.
in % impedance of the transformer, the use of lightning The word alternatively is preferred to the word
arresters, the voltage amplification phenomenon due to “alternately” (unless you really mean something that
resonance, and the application of RC circuits were alternates).
confirmed in case studies and vehicle load modeling.
· Do not use the word “essentially” to mean
Vpeak[kV] Vpeak[kV] Vpeak[%] Vpeak[%] “approximately” or “effectively”.
%Z
Case @PC1 @PC2 @PC1 @PC2
(pu)
w/o SA w/o SA w/o SA w/o SA
· In your paper title, if the words “that uses” can
Case 1 0.1 64.480 64.598 165.84 166.15
accurately replace the word “using”, capitalize the “u”;
Case 2 0.2 64.677 64.806 166.35 166.68
if not, keep using lower-cased.
Case 3 0.3 64.855 64.994 166.81 167.17
Case 4 0.4 65.019 65.166 167.23 167.61
· Be aware of the different meanings of the homophones
Case 5 0.5 65.168 65.324 167.61 168.01
“affect” and “effect”, “complement” and
“compliment”, “discreet” and “discrete”, “principal”
Fig. 3. Simulation result according to %Z change (without arrester) and “principle”.
%Z
Vpeak[kV] Vpeak[kV] Vpeak[%] Vpeak[%] · Do not confuse “imply” and “infer”.
Case @PC1 @PC2 @PC1 @PC2
(pu)
w/ SA w/ SA w/ SA w/ SA
· The prefix “non” is not a word; it should be joined to
Case 1 0.1 48.169 48.561 123.89 124.90
the word it modifies, usually without a hyphen.
Case 2 0.2 48.187 48.567 123.94 124.92
Case 3 0.3 48.204 48.572 123.98 124.93 · There is no period after the “et” in the Latin
Case 4 0.4 48.220 48.577 124.02 124.94
abbreviation “et al.”.
Case 5 0.5 48.234 48.581 124.06 124.95 · The abbreviation “i.e.” means “that is”, and the
Fig. 4. Simulation result according to %Z change (with arrester) abbreviation “e.g.” means “for example”.
Figures 3 and 4 are simulation results of surge voltage An excellent style manual for science writers is [7].
generation according to the increase in % impedance
III. CONCLUSION
of each transformer with and without lightning
arresters. As %Z increases, the peak voltage tends to In this study, the operating characteristics of the arrester
increase slightly. and spark gap applied to KTX-Sancheon were analyzed.
Therefore, in the KTX-Sancheon high-speed vehicle in which
· Surge voltage simulation result power cars are configured before and after and power is
Through Figure 3, Figure 4, Figure 5, and Figure 6, it supplied to the power cars at both ends through one collector,
can be seen that the surge voltage simulation result the lightning arrester performs the primary protection function
value changes depending on the presence or absence of against the transient voltage, but the lightning arrestor protects
the arrester and the spag gap when the main circuit the vehicle opposite to the collector power car. It can be seen
breaker is turned on. that the protection effect against fatigue can be enhanced by
using a spark gap for additional protection beyond the
operating capacity.
REFERENCES
[1] Kim Jae-moon, “Electric railway power supply system voltage drop
(a) Lightning Arrester + No Spark Gap (b) Arrester + With Spark Gap compensation using fuel cell system”, 2015
Fig.54 Surge voltage simulation result (%Z=0.1) [2] Korea Railroad Corporation, ‘High-speed Vehicle Maintenance
Manual’, 2013
[3] Choi Jong-rok, “A Study on the Insulation Cooperation of Lightning
Arrester and Spark Gap of High-Speed Rail Vehicle”, Master's Thesis
2016
- 428 -
Development of Trolley for Detection
Installation Errors of Catenary System
Chulmin PARK1*, Dongkue Kim1, Sangjin SON2
1
Korea Railroad Research Institute, Korea
2
Myeong-sung Rollingstock & Plant Co. Ltd, Korea
Email: [email protected]
Abstract— Electric railways use a catenary system to supply
electricity required for the operation of rolling stock. For the
safe operation of this catenary system, errors in each component
are measured and maintained from the construction stage. In
order to measure these construction errors and maintain
facilities, we are carrying out catenary system measurement
work that relies on manpower. As a basic research for
measuring catenary construction errors, we are building an
intelligent composite sensor of LiDAR, Stereo Camera, and
Encoder. In this paper, the development of a mobile trolley
equipped with an intelligent composite sensor system for (a) Human-centered measurement (b) Composite sensor
measuring catenary construction errors was described. The Measurement(plan)
Fig. 1. Development of construction error detection of catenary system
unmanned/automatic moving trolley under development is used
to measure catenary system, reducing measurement time by
manpower, suppressing safety accidents, solving manpower
shortages through the development of automatic measuring II. DEVELOPMENT OF MOBILE TROLLEY FOR INTELLGENT
devices, and improving train operation efficiency. The trolley COMPLEX MEASUREMENT SYSTEM
developed in this study is equipped with an intelligent composite
sensor-based instrument and uses a battery-powered motor to
move unmanned/automatically along the track and is used to A. Overview of composite measurement system
measure the construction error of the catenary. If a
sophisticated 3D catenary map is built using the developed
unmanned/automatic moving trolley and a geographic LiDAR and stereo cameras are required to accurately
information system (GIS) is built, it is expected to be greatly measure the construction error of catenary, and mobile
utilized for construction errors and maintenance. trolley is needed to measure the components (contact line,
messenger wire, droppers, etc.) of catenary system installed
Keywords—Trolley, Automatic measurement, Catenary long in the rail direction. In addition, an odometer such as an
system, Contrition error, Maintenance encoder is required to continuously measure the three-
dimensional shape in the driving direction of the trolley. To
measure the three-dimensional shape of the catenary system,
I. INTRODUCTION a 360-degree LiDAR was installed in the forward direction
of the bogie as shown in Figure 2. Using a stereo camera
installed at the midpoint of both rails of the trolley, images
Catenary system is installed in all sections as a key of the catenary components are acquired and a depth image
element of the electric railway, and train operation is delayed according to the movement of the mobile trolley is generated.
in the event of a breakdown. In addition, since it is installed Finally, the 3D shape of the catenary system is reconstructed
on all lines along with the rail, the failure frequency is high. using the 3D point cloud information obtained from LiDAR
This is why it is necessary to accurately evaluate construction and the depth image obtained from the stereo camera, and it
errors for catenary elements from the construction process. is used as basic shape data for evaluating construction errors.
Currently, construction evaluation is carried out through The developed composite sensor based catenary inspection
human-centered work at heights, so there is a possibility of system is expected to be used to improve the stability of the
human accidents, and human errors can be included in the railway by accurately evaluating construction errors and
evaluation and measurement process[1]. To solve this maintenance information of the catenary system.
problem, we are developing a composite sensor-based
measurement system that can minimize worker access to the
site and evaluate with high precision through catenary
measurement and automation of construction errors. This
paper describes the design/manufacture of a trolley that can
automatically measure the construction error of catenary
while moving unmanned/automatically with such a composite
sensor-based measurement system. It is expected that the
construction error of the catenary using the manufactured
trolley will be used for maintenance of the catenary by
utilizing the automatic measurement technology. Fig. 2. Image of LiDAR installed Trolley
- 429 -
B. Development of mobile trolley catenary system. In the future, we plan to develop an
algorithm that can reconstruct a precise 3D shape by reflecting
the curve and cant of the line using GPS, IMU, etc., and
In this paper, an unmanned/automatic mobile trolley measure the construction error of the catenary through this.
equipped with a complex sensor-based construction error
detection system was developed to precisely measure the
construction error of the catenary system. This trolley is
designed/manufactured considering the installation location
of each sensor. To measure the deviation of the contact line,
the center of the stereo camera is placed in the center of the
rail, and the LiDAR stands on a separate pole at the end of the
top plate to prevent interference with other sensors. On the
other hand, springs are installed on the wheels of the trolley so
that it can run without twisting on the curved roads of the rail,
and driving information is recorded using an encoder-
integrated motor that can be moved with a battery.
III. MOBILE TROLLEY DETAILED DESIGN AND PRODUCTION
A. Detailed design of mobile trolley
In order to precisely measure the construction error of the
catenary system, we developed a trolley that moves at
constant speed by a motor, not a trolley that moves by
manpower. Construction errors are precisely analyzed using
the SLAM(Simultaneous Localization And Mapping)
method that detects the exact position of the trolley by
utilizing GNSS(Global Navigation Satellite System) and
motor-integrated encoders that confirm the location of
catenary components. Since LiDAR has a difference in
accuracy depending on the distance to the measurement target, Fig. 4. Completed mobile trolley and 3D map creation result
the trolley was designed to be located at the highest possible
position and the position of the stereo camera to be located in
the center of the track. The battery to supply the power IV. CONCLUSION
necessary for the movement of the cart and the operation of
the instrument is heavy, so it is located at the bottom center In this paper, the development of a mobile trolley that can
of the trolley for safety. A detailed drawing of the trolley is be equipped with a composite sensor-based measurement
shown in Figure 3. system that can detect construction errors of catenary system
is described. The currently produced trolley is the first
prototype, and we plan to develop a lighter and more
convenient trolley through further testing and
supplementation. If this trolley is used for construction errors
and maintenance of catenary system, it will greatly contribute
to improving the safety of electric railways.
ACKNOWLEDGMENT
This research was supported by a grant from R&D
Program (A study on operation efficiency improvement of
Fig. 3. Detailed design of Trolley(side view and top view) railway electricity utilizing smart composite sensors based
geometric information system, PK2303E2) of the Korea
Railroad Research Institute.
B. Fabrication and testing of moblie trolley
The mobile trolley was successfully fabricated according
to the design using aluminum for field application. A REFERENCES
composite sensor was mounted on the trolley to verify the
performance of the trolley through test driving. In this study, [1] Dongkue Kim, Kiwon Lee, Chulmin PARK, Young Park “Basic
basic 3D shape information as shown in Fig 4 was obtained research on the measurement of construction errors of catenary systme
based composite sensors for electirc railways”, Korean Institute of
by designing and configuring hardware for measuring Electrical Engineers Spirng Conference 2023. p.182-185. 2023.
- 430 -
Research on the development and application of
lightweight catenary poles for emergency
recovery
Ho Ryung Cho1, Jong Nam Kim 1, and Hong Kuk You1, Young Jae Na1, and Jae Moon Kim2*
1
Department of SMART Railway System, Korea National University of Transportation, Republic of Korea
2
Department of Transportation System Engineering, Korea National University of Transportation, Republic of Korea
Email: [email protected](HRC), [email protected](JKK), [email protected](HKY), [email protected](YJN),
[email protected](JMK)
Abstract— It is necessary to install steel pipe catenary poles
for emergency recovery for the restoration of catenary lines in
the event of loss of roadbed on railroad track or damage to II. Main subject
catenary poles due to train derailment due to typhoon and
heavy rain damage caused by recent abnormal weather.
1. Review of application cases of catenary poles for
Existing emergency steel tube columns are made of steel and
are heavy and have a large length per unit, so they are emergency recovery at home and abroad
inconvenient to transport and install, so there is a risk of
occupational accidents due to musculoskeletal disorders. In 1) Review of cases of application of steel pipes for
addition, it was manufactured to be installed on a gravel emergency recovery in Korea
roadway, so there was a problem that it was impossible to
In the case of Korea, as shown in Figure 1, in order to
install on a concrete roadway. In this study, an emergency
recover from the failure caused by the damage to 5 catenary
recovery utility pole using lightweight materials was developed,
poles in the freight train derailment accident that occurred in
and a slim rail fastening device applicable to concrete
2006, as shown in Figure 3, after fixing the pipe structural
roadways was researched and developed, assembled and
beam to the lower part of the rail, steel emergency recovery
installed on a test bed, and then tested to confirm safety. It is
This is a case of assembling and installing a steel pipe column
expected that the product developed through this study will be
on the side part [2].
able to safely and quickly restore catenary lines, thereby
minimizing disruption to train operation and preventing
industrial accidents of related workers.
Keywords : catenary pole for emergency recovery, light weight,
glass fiber reinforced plastic (F.R.P), aluminum, rail fastening
device
I. Introduction
In the recent 6 years from 2016 to 2021, there were 19 train
derailment accidents on domestic railways, an average of 3.2
cases per year. situation [1]. The existing product used as an
emergency steel pipe column for accident recovery is made of Shintanjin ~ Maepo freight Photo of checking the
steel and weighs about 360 kg and has a large unit length. Since train derailment scene photo damage of the catenary pole
it must be constructed after construction, there is a high risk of
industrial accidents in the musculoskeletal system of workers, Figure 1. Gyeongbu Line (Shintanji-Maepo) Freight Train
and there is a disadvantage in that excessive recovery time is Derailment Accident (Train pole: 5 damaged) - 2006. 03. 11.
required.
Figures 2 below are about the installation method of steel
In this paper, after confirming the feasibility through the pipe columns for emergency recovery. First, after digging out
characteristics analysis and design review of the lightweight about 60cm of the part between the sleepers at the installation
material for the light weight of the emergency recovery location for the installation of emergency steel pipe columns,
catenary pole, and then manufacturing the prototype and and secondly, pipe structural beams for emergency steel pipe
structural test, the final aluminum material was selected and column installation. is fixed to the bottom of both rails, and
the lightweight prototype was manufactured. In addition, the then, thirdly, assemble the movable bracket ring and the upper
rail fastening device for assembling and installing lightweight feeder support insulator assembly bracket to the emergency
catenary poles was also developed for concrete roadbeds, and steel pipe column consisting of three units, and then assemble
the safety test through on-site installation test was conducted. the lower part of the emergency steel pipe column to the side of
the pipe structure beam to make the steel pipe The column is
- 431 -
assembled and installed at a right angle, and fourth, the material Steel Steel Al alloy
construction of the emergency steel pipe column is completed
by connecting wire ropes to the left and right at the upper two
places and firmly fixing it to the ground [3]. Apply gravel Gravel ballast
Gravel ballast
track track application
track application
After the installation of the emergency steel pipe column
is completed, assemble the movable bracket including the long It is a method of It is a method of
It is a method of
insulator for the installation of the catenary and clamshell, assembling and assembling and
assembling and
installing 3 units installing three
install the feeder, catenary and clamshell to connect the tension of an integrated
installing three
adjusting device, and then install the dropper and perform Charac- truss structures truss structures on
teristic (pipe) structure the lower side of a
adjustment work to the train is ready to run. on the lower side
in a pipe beam rail (1 piece). The
of a rail (1 piece),
structure at the movable bracket
and is
bottom of the for general use is
characterized by
rail (2 pieces), applied as it is.
applying a
and the general
simple-shaped
movable bracket
movable bracket.
is applied as it is
weight About 360Kg - About 130Kg
< Excavation work between <Pipe Structural Beam
the sleepers > Assembly >
Photo
< Assembly and installation < Installation of the movable 2. Design of lightweight catenary pole and rail
of emergency steel pipe bracket >
fastening device for emergency recovery
columns >
Figure 2. Installation of steel pipe columns for emergency 1) Design of lightweight catenary pole for emergency
recovery recovery
First, the strength condition (bending stiffness) of the
2) Review of cases of application of steel pipes for existing steel emergency recovery steel pipe column must be
emergency recovery overseas equal to or higher than that of the existing steel emergency
recovery steel pipe column. The design was reviewed in
In French SNCF, as shown in Table 1, after assembling consideration of the fact that the conditions for ease of
and installing an emergency catenary pole in the form of a operation should be satisfied.
truss (steel) structure on one side of the rail and installing a
branch line in the direction of the track, the movable bracket is (1) Examination of bending stiffness according to material
a straight line with a simple structure that can be used for weight reduction
temporarily. It is a structure in which insulation insulators are
attached to the lower part of a beam and catenary and jog Since the light weight emergency recovery catenary pole is
wires are assembled, and a method of connecting a diagonal an important element that must satisfy both bending rigidity
support wire to the upper part of an emergency truss catenary and light weight conditions, the materials were made of two
pole is applied [3]. materials, F.R.P – Fiberglass Reinforced Plastics and
aluminum (A6063-T6), which are composite materials. Based
In China, as shown in Table 1, a truss (aluminum alloy) on the same shape and size (external dimensions) as the steel
structure is installed on the lower side of the rail in the outer pipe column for emergency recovery of the first, a comparison
direction of the rail, and a branch wire is installed in the was made as shown in Table 2.
direction of the track and at a right angle to the rail to stably
install the emergency catenary pole, and then the movable In order to satisfy the bending stiffness of the existing steel
bracket is actually operated. It is a structure that applies the pipe column for emergency recovery, a change in thickness
general type used in lines. And the weight of the truss- dimension was required based on the bending strength of
structured emergency catenary pole is about 130 kg [3]. aluminum (A6063-T6) and FRP materials, so the dimensions
were adjusted. ), FRP was selected as ∅165 × 7.5T and
Table 1. Comparison of installation methods for catenary aluminum pole as ∅161 × 10.5T through calculation.
poles for emergency recovery at home and abroad
A. Calculation of the maximum bending stiffness (M :
index Korea France China moment) of the existing catenary pole
- 432 -
B. Maximum bending stiffness (moment) calculation of
FRP catenary pole: ∅165 × 7.5T
C. Maximum bending stiffness (moment) calculation of
FRP+AL catenary pole: ∅165 × 8T
D. Calculation of maximum bending stiffness (moment) of
aluminum catenary pole: ∅161 × 10.5T
Figure 3. Al Alloy Pole & FRP Pole design
2) Review of the design of lightweight catenary poles for
emergency recovery
Lightweight emergency recovery catenary poles were
Table 2. Check the bending stiffness (moment) and weight analyzed using the Solidworks stress analysis program based
comparison of lightweight materials (FRP, aluminum) on the design drawings of the existing emergency recovery
catenary poles compared to existing catenary poles catenary poles and lightweight catenary poles made of
aluminum and FRP materials to review the design safety. The
following figures 4 and 5 As shown in , it was confirmed that
Category Steel FRP FRP+AL AL
all three types ensure structural safety.
Dimension ∅139.8×4.5t ∅165×7.5t ∅165×8t ∅161×10.5t
bending 5,456,167
stiffness 5,705,885 5,382,452 6,141,003
[100%]
(kg.mm)
15.015kg/m 7.422kg/m 9.068kg/m 13.19kg/m
weight
[100%] [49.40%] [60.39%] [87.8%]
(2) Catenary pole structure and shape design for
lightweight emergency recovery
The design of the light pole for emergency recovery was
designed as shown in Figures 3 in consideration of the
improvements of the existing product based on the dimensions
of the material (aluminum and FRP) determined above.
Figure 4. Result of structural analysis of existing steel
emergency recovery pole
- 433 -
3. Prototype production and bending characteristics
verification test
After producing prototypes for each type of lightweight
emergency recovery pole, a test for the most important
characteristic, bending stiffness, was performed as shown in
Figure 9 for the prototype of Type 3 compared to the existing
product, and the results shown in Table 3 were obtained. For
the bending test, a bending load of 450 kg was applied to the
intermediate pipe of the prototype for each material, and the
amount of deformation was measured [5].
Figure 5. Result of structural analysis of aluminum
<Steel Pole> <Al Alloy Pole >
emergency recovery utility pole
Figure9. Bending property test of prototypes by Steel or Al
3) Rail fastening device design Table 3. Bending property test results for prototypes for
each material
The rail fastening device for the installation of catenary
poles for emergency recovery is a pipe beam-shaped rail
fastening device that is installed after excavating more than displacement remark
60cm of existing rail gravel, and is designed to be slim so that it test load check (mm) displacement
sample
can be applied to concrete roadways in the city center as shown (kgf) (mm) Length/weight
in Figures 6 and 8 [4]. Height Height (kg)
(free) (Loading)
FRP
Ⅰ(Filament 485.48kg 1890 1675 215 2650mm / 38
Winding)
FRP Ⅱ
485.48kg 1900 1690 210 2650mm / 45
(Prepreg)
A6063-T6 490.48kg 1875 1800 75 2650mm / 40
STEEL 491.48kg 1885 1820 65 2600mm / 65
Figure 6. Existing pipe beam type rail fastening device
4. Field installation test of lightweight catenary pole for
emergency recovery
In order to verify the assembly characteristics and safety
of the prototype for each material of the lightweight emergency
recovery catenary pole and the existing steel pipe pole for
emergency recovery in the actual field, after assembling and
installing it on the test bed as shown in Figure 10, the load
applied during actual wiring work (2 workers – 200kg) was
confirmed by applying tension weights, and there was no
Figure 7. New slim rail fastening device design abnormality in all three types as shown in Table 4 and Figure
10[5].
Table 4. Test results for safety verification during field
installation
concrete road surface
MMinimum installation height Height
NO. Type Load (Kgf) Result
⇒ 55 mm (mm)
FRPⅡ 200Kgf Loading
1 200 7900
Figure 8. Slim rail fastening device Dimension check (Prepreg) safety
- 434 -
200Kgf Loading
2 AL Alloy 200 7900
safety
200Kgf Loading
3 STEEL 200 8000
safety
<Slim rail fastening device>
Figure 12. Slim rail fastening device
III. Conclusion
In this paper, in order to improve the problem that the
existing steel pipe column for emergency recovery is made of
steel, it is heavy and the unit length is large, so it is
<Steel Pole> <Al Alloy Pole > <FRP Pole> inconvenient to transport and install, which poses a risk of
industrial accidents, and it is for gravel road installation, so it
Figure10. Safety confirmation test for 3 types Pole
is impossible to install on concrete roadways. We reviewed the
research and development of a 47% lightweight aluminum
5. Verification of improvement plan for slim rail fastening
emergency recovery utility pole and a slimmer rail fastening
device
device that can be safely installed even on concrete roadways in
the city.
Workability and safety by fabricating and assembling and
Since the prototype developed through this research and
installing a prototype as shown in Figure 11 regarding the
development was assembled and installed on a test bed, and
improvement plan of the slim rail fastening device so that it
even tested to verify safety, it is expected that it will be possible
can be installed on the concrete roadway in the downtown
to quickly and safely restore catenary tracks in the future to
section of the lightweight emergency recovery railroad pole
minimize disruption to train operation. .
and minimizes the excavation work even on the gravel roadway.
confirmed.
The developed product is difficult to assemble firmly [Reference literature]
because the existing pipe beam type rail fastening device
inserts insulation material under the rail and assembles bolts [1] Aviation ․ Railroad Investigation Committee, (2017),
as shown in Figure 12, and there is a problem that the bolts “Investigation report on freight train derailment between
loosen due to train vibration. Insulation material (engineering Shintanjin Station and Maepo Station on the Gyeongbu Line”.
plastic) was applied to the middle joint to ensure strong [2] CHINA Patent, CN 102642482 A - “Urgent repair
assembly, and no insulation material was applied to the rail construction method for anchor mast of catenary” 2012.08.22
assembly part to ensure fundamentally safe fastening [6]. [3] Office of International Cooperation, (2021. 04. 29.),
In addition, guide blocks are placed on both sides of the “Results of overseas railway case studies related to emergency
lightweight catenary pole assembly part of the rail fastening steel pipelines”, Korea Railroad Corporation.
device to prevent falling when assembling the lower electric [4] D2 Engineering, (2022. 03.), “Joint Research Performance
train part, so as to improve the safety and assembly Report (1st)”.
convenience of workers. [5] D2 Engineering, (2022. 08.), “Joint Research Performance
Report (2nd)”.
[6] Cho Ho-ryung, Kim In-cheol, So Kwan-young, Lee Byung-
gon, Nam Han-seok, “Study on construction methods
according to the development of lightweight catenary poles for
emergency recovery,” Proceedings of the 2022 Korean Institute
of Electrical Engineers Fall Conference, pp. 75-76, 02, 2023.
Figure 11. Photos of existing rail fasteners
- 435 -
A Study on the Characteristics of Power System
Protection System According to Regenerative
Braking of Urban Railway Vehicles
Jong Hak Park 1,2, Jeong Hyung Kang 3,4, Seung Ho Song 1,4,
Sung Cheol Han 1,2, and Jeong Won Kang 1,3*
1
Department of Transportation System Engineering, Korea National University of Transportation, Republic of Korea
2
Seoul METRO, Republic of Korea
3
Department of SMART Railway System, Korea National University of Transportation, Republic of Korea
4
KORAIL, Republic of Korea
Email: [email protected] (JWK)
Abstract— This study examines the influence of regenerative Additionally, the impact of the overhead line voltage on the
power generated through regenerative braking on the power catenary line and the power system was investigated. The study
system, specifically focusing on the transition from resistance also examined the protection system for the train and power
and chopper control electric vehicles used in urban railways to system, focusing on the inverter protection system and the
VVVF electric vehicles. Urban railways in our country are power system protection system. A comparison and analysis of
predominantly built underground, limiting the utilization of the suitability of the mutual protection system settings revealed
natural lighting to about 20% of the total system. Consequently, a discrepancy in protection coordination due to differing
electricity plays a vital role in powering trains, as well as various parameter values. Accident case analysis was employed to
passenger amenities (E/S, E/V), lighting, and auxiliary facilities, highlight failures that may occur due to disparities in protection
which collectively account for over 95% of the total energy coordination settings between power systems.
consumption. Train operation itself constitutes approximately
70% of the energy consumption. Moreover, with the addition of Within the inverter protection settings of VVVF electric
safety features such as screen doors and the expansion of vehicles, further assessment is required to determine the
customer convenience facilities, the energy demand continues to adequacy of the low voltage threshold for the overhead line
rise, while the energy-saving rate declines. Thus, it becomes (850±10%), as well as the increased capacity of surge arresters
imperative to enhance the system to significantly reduce and energy storage devices (ESS) to ensure stability in energy
electricity consumption, particularly in electric vehicles that efficiency and overhead line voltage within substations.
consume more than 70% of the total energy. Additionally, a thorough discussion is needed regarding the
reuse of system protection to address this issue. Specifically,
Since 2000, ○ ○ Corporation has been progressively electric vehicle manufacturers should focus on developing
replacing the initial resistance and chopper control trains with technical measures to reduce the error range, enabling the
VVVF trains, thereby achieving improved energy efficiency inverter-based decompression system to operate at an overhead
through regenerative braking, combined with the installation of line voltage of DC1,800 [V].
energy storage devices. However, this transition has resulted in
unexpected phenomena, including increased overhead wire
voltage during regenerative braking, abnormal power system
voltage, and an upsurge in vehicle failures. To address these Keywords—Power System, Protection System, Braking, Urban
issues, the VVVF control vehicles underwent modifications, Railway Vehicles
such as a 10% weight reduction compared to the initial VVVF
control vehicle, the adoption of LED lighting, improved ACKNOWLEDGMENT
regenerative braking cutoff speed (from approximately 12 km/h This work was also supported by the Korea Agency for
to 3 km/h), and the utilization of regenerated power from the Infrastructure Technology Advancement (KAIA) grant
motor car in the trailer car (M-T method), wherein 35% of the funded by the Ministry of Land, Infrastructure and Transport
electric power consumption of the electric car is repurposed as (Grant No. 22QPWO-C158106-04).
regenerative power.
After the transition period, which involved the mixed
operation of resistance control, chopper control, and VVVF
control vehicles, the current state of regenerative power and
energy storage devices was analyzed. A comprehensive
evaluation was conducted on the effect of the catenary and
power system by assessing the monthly maximum catenary
voltage in two main line substations and one vehicle depot
substation on the circular lines of ○○ Corporation Line 2.
- 436 -
A Study on Stabilization of Code Frequency
Control Between Wayside and Onboard Signal
System of AF Non-insulated Track Circuit
Dong Hoon Park 1, Ho Hyun Han 1, and Seon Kyo Kim 1, Jeong Won Kang 1,2*
1
Department of SMART Railway System, Korea National University of Transportation, Republic of Korea
2
Department of Transportation System Engineering, Korea National University of Transportation, Republic of Korea
Email: [email protected] (PDH), [email protected] (HHH), [email protected] (KSK), [email protected] (KJW)
Abstract— In urban metro railways, Audio Frequency (AF)
non-insulated track circuit that electrically separates rails to
determine the movement location of trains, adjust travel
intervals, and control speed method is adopted and used.
While the train is starting and running, the code Frequency
for controlling the operating, speed of the corresponding section
is transmitted, and after arriving at the station, the code
frequency for controlling the door opening in the direction of
getting on and off is changed and transmitted to the onboard
signal system on the wayside.
In the process of switching from the speed control frequency
to the access control frequency, passenger safety accidents often
occur due to opening of the door on the opposite side due to
malfunction.
In this paper, the causes of frequency instability control in
the process of switching door opening direction control
frequency from the operating speed frequency were analyzed
and an alternative was proposed.
First, the carrier frequency form of the rate code control
method is analyzed. Second, the principle of transmitting
frequency information from the wayside signal system to the
onboard signal system is confirmed, and the process of switching
from train operation speed control to door open control is
analyzed. Third, research on improving the reliability of the
speed code track circuit system by analyzing the possibility of
malfunction of train door direction control (resonant circuit
characteristic change, frequency strength, frequency noise,
appropriate processing time for onboard signal system
frequency) .
Keywords—AF track circuit, code frequency track, signal system,
train control system
- 437 -
A Study on the Integrated Operation of VHF
and Complex Communication Facilities
Seon Kyo Kim 1, Dong Hoon Park 1, and Ho Hyun Han 1, Jeong Won Kang 1,2*
1
Department of SMART Railway System, Korea National University of Transportation, Republic of Korea
2
Department of Transportation System Engineering, Korea National University of Transportation, Republic of Korea
Email: [email protected] (KSK), [email protected] (PDH), [email protected] (HHH), [email protected] (KJW)
Abstract— The purpose of this paper is to present a Keywords—VHF, LTE-R, leaky coaxial cable, complicated
communication facilities
method for integrated operation of the train wireless
communication system, into a unified operation system
which is currently used in combination newly extended
urban railway line,
The need of improve the unified wireless communication
command system has been raised to support cooperative
work for the disaster control towel in case of the national
event. In addition the VHF and TRS systems is no longer
used after 2029 due to changes of government policy. Thus,
the LTE-based LTE-R systems are being built in to replace
the VHF and TRS in railway sector.
The policy changes caused the mixture use of the VHF and
LTE-R which is inefficient for the operators in railway
communication system the tunnels and underground
sections. Especially the mixture use system also potential
threat to safety of train operation in disaster and emergency
rescue.
This study is about an alternative method for mixture use
of the VHF and LTE-R to unified railway communication
system using a leaky coaxial cable with a similar frequency
band and coupling device for complicated communication
facilities. Through the study, it might contribute to railway
operators to improve in effeciency and safety especially in
tunnels and underground sections using the VHF and LTE-
R together, without loss of communication among the OCC,
train crew, and maintenance worker.
- 438 -
A Study on CBM Application of Power Capacitor
for SIV in Railway Vehicles
Hyo-Seok Oh 1, Yong-Eun Choi 1, Byung-Min Moon 1, Seung-Pyo Jeon 1, Jae-Moon Kim 1*
1
Korea National University of Transportation, Republic of Korea
Email: [email protected]
Abstract—This paper proposes a Condition-Based
Maintenance (CBM) system for the SIV capacitor of railway
vehicles. In actual operation, a Data Sensing Module (DSM) is
installed on the vehicle and a capacitance estimation method
that considers the internal structure and characteristics is
proposed. This method is believed to be an efficient maintenance
approach by monitoring the condition of the capacitor.
Keywords— SIV, CBM, DSM, capacitance
I. INTRODUCTION
The power capacitors used in the power conversion
devices of railway vehicles experience a decrease in
capacitance due to aging caused by various environmental
factors and usage time, resulting in a deterioration in the
performance of the power conversion device. However, the Fig. 1. SIV circuit and DSM system
existing maintenance system for railway vehicles is based on
mileage and usage time, making it impossible to determine the Most capacitor status diagnosis research primarily focuses
condition of components during vehicle operation. To solve on electrolytic capacitors, using capacitance and equivalent
this problem, it is necessary to introduce a condition-based series resistance (ESR) estimation to determine their status.
maintenance system that can monitor and analyze the However, the DC-link capacitor used in the SIV utilizes a more
condition of components in real-time.[1][2] stable metalized polypropylene film (MPPF) capacitor instead.
In this paper, we have selected the SIV as the target among Estimating the status of MPPF capacitors poses challenges due
the power conversion devices of the vehicle and have installed to their extremely low ESR component. Therefore, we selected
a Data Sensing Module (DSM) on the vehicle during actual capacitance as the diagnostic factor for assessing the capacitor's
operation to sense the device's data. Additionally, we propose status [4]. Additionally, measuring the current flowing through
a CBM approach that can provide more accurate information the capacitor is not feasible due to the SIV's internal structure.
on the status of power capacitors by considering the SIV's The converter's output current consists of direct current and
internal structure and characteristics. ripple, with a significant portion of the ripple component
passing through the capacitor. By extracting the ripple
II. SIV ENVIRONMENT AND CBM METHODS component using frequency analysis methods like Fast Fourier
Fig. 1 presents a block diagram of a capacitor condition Transform (FFT), we propose an indirect method to estimate
diagnosis system implemented through an SIV circuit and the capacitor's capacitance, as shown in equations (1) and (2).
DSM. The SIV receives power from the pantograph and 𝐼̂𝐶 = 𝐼𝑐𝑜𝑛_𝐹𝐹𝑇 (1)
undergoes a step-down process through a transformer.
Subsequently, the power is converted into a stepped-down Here, 𝐼𝐶 represents the indirectly derived capacitor current,
direct current by the AC/DC converter, and a stable voltage is and 𝐼𝐶−𝐹𝐹𝑇 represents the current value in the frequency band
then supplied to the inverter through the DC-Link capacitor. of the converter's output ripple.
The inverter, in turn, provides power to AC loads such as air
𝐼̂𝑐
conditioning and heating devices in the passenger 𝐶̂ = 𝐼 (1)
compartment. Additionally, there is a circuit that converts 2 × 𝜋 × 𝑓 × 𝑉𝐶_𝐹𝐹𝑇 𝑐𝑜𝑛_𝐹𝐹𝑇
power to direct current, which is supplied to DC loads
requiring direct current for lighting and control devices within Here, C represents the estimated capacitance, f represents
the vehicle.[3] The SIV has the capability to store various data, the frequency range of the acquired voltage and current, and
including operation information and failure records, in 𝑉𝐶−𝐹𝐹𝑇 represents the voltage value in the frequency band of
conjunction with the Train Control & Monitoring System the converter's output ripple.
(TCMS) in the event of an electrical accident [3]. However,
when a failure occurs, only simple data and failure codes are It is recommended to replace MPPF capacitors when their
stored, which makes it difficult to determine the status of the capacitance has decreased by 2% compared to their rated
device using TCMS data alone. Therefore, in this paper, we capacitance. This is because once the capacitance has dropped
installed voltage and current sensors and a Data Sensing by 2%, the capacitance experiences a sharp decrease, greatly
Module (DSM) inside the SIV to acquire capacitor data. affecting the performance of the power conversion device.
Therefore, it is reasonable to monitor the timing of capacitor
- 439 -
replacement using the proposed capacitance estimation
technique for the implementation of CBM in the capacitor.
III. RAILWAY VEHICLE TESTING AND DATA ANALYSIS
Fig. 4. Estimated capacitance results after applying the filter
The table presents the error between the capacitance
estimated using the proposed technique and the capacitance
measured using the LCR meter. It indicates a maximum error
Fig. 2. SIV circuit and DSM system
rate of approximately 2.5%.
Fig. 2 is a photograph of the installed DSM and sensors
TABLE I. ESTIMATED RESULT OF CAPACITANCE OF SIV
within the SIV. The DSM was securely mounted using a
bracket to prevent any structural deformation. The current LCR meter Estimated value
sensor was installed along the rigid line of the AC/DC capacitance(uF) Capacitance[uF] Error[%]
converter's output stage, while the voltage sensor was bolted Data 1 5,066,17 0.23
together with the DSM voltage sensor, utilizing the existing Data 2 5,063.93 0.27
DC link voltage sensor terminal of the SIV for safety … … …
5,078
Data 162 4,951.55 2.49
assurance. To verify the proposed capacitance estimation … … …
technique, data was acquired while the railway vehicle was in Data 800 5,051.36 0.52
a stationary state with the power running. Analysis of the
capacitance estimation based on the power supplied to various IV. CONCLUSION
loads such as lighting, air conditioning, and heating devices is In this paper, we installed a DSM on a vehicle that is
crucial for the SIV. Therefore, the load of the SIV was currently in operation to acquire data and conducted a
intentionally varied during the process of collecting MPPF verification of the proposed capacitance estimation technique.
capacitor status monitoring data. The results showed a low error rate between the estimated
capacitance and the capacitance measured by the LCR meter,
with a maximum error of approximately 2.5%. However, it
should be noted that the performance of capacitors is
influenced by temperature and environmental factors, and for
reliable application of CBM, it is necessary to collect
temperature data of the capacitors and additional vehicle
information.
ACKNOWLEDGMENT
This research is supported by the Korea Agency for
Infrastructure Technology Advancement (KAIA) by the
Ministry of Land, Infrastructure and Transport under (Grant
RS-2019-KA152194)
Fig. 3. Estimated capacitance results graph (Load variation) REFERENCES
[1] Sresakoolchai, Jessada, and Sakdirat Kaewunruen. “Railway
Fig. 3 shows a graph of capacitance estimation data with Infrastructure Maintenance Efficiency Improvement Using Deep
varying loads, indicating changes in the ripple of the converter Reinforcement Learning Integrated with Digital Twin Based on Track
output current. This change is attributed to the decrease in the Geometry and Component Defects.” Scientific Reports 13, no. 1
(February 10, 2023): 2439.
ripple component as the load decreases. Consequently, when
[2] Sedghi, Mahdieh, Osmo Kauppila, Bjarne Bergquist, Erik Vanhatalo,
the ripple value is low, it can be observed that the calculated and Murat Kulahci. “A Taxonomy of Railway Track Maintenance
capacitance is also influenced by electrical noise inside the SIV. Planning and Scheduling: A Review and Research Trends.” Reliability
However, considering that a significant change in Engineering & System Safety 215 (November 2021): 107827.
capacitance during the normal operation of the SIV is unlikely [3] Korea Railroad Corporation, "Maintenance Manual for 128 Electric
Railway Vehicles", 2019
unless caused by mechanical factors, a filter and tolerance
[4] Buiatti, G M, J A Martín-Ramos, C H R García, A M R Amaral, and A
range for ripple current were implemented for the capacitor's J M Cardoso. “An Online and Noninvasive Technique for the
capacitance data. Furthermore, a moving average method was Condition Monitoring of Capacitors in Boost Converters.” IEEE
employed to obtain a stable capacitance value for accurate Transactions on Instrumentation and Measurement 59, no. 8 (August
capacitance estimation. 2010): 2134–43
- 440 -
A Study on SoH of Lithium Polymer Battery
Applied to Railway Vehicles
Hyo-Seok Oh 1,Yong-Eun Choi 1, Byung-Min Moon 1, Seung-Pyo Jeon 1, Chin-Young Chang 1,
Jae-Moon Kim 1*
1
Korea National University of Transportation, Republic of Korea
Email: [email protected]
Abstract— In this paper, a battery with the same specificatio of battery loads under normal operating conditions, based on a
ns as a railway vehicle was constructed, and the load characterist standard configuration of one train consisting of 10 cars in the
ics of the vehicle battery were analyzed and applied to the experi metropolitan area. The details are presented in Table II.
ment. Additionally, for the estimation of SoH, we performed tem
perature correction and trend analysis to examine the relationshi TABLE II. BATTERY LOAD TYPE AND CAPACITY OF RAILWAY VEHICLE
p between changes in battery capacity and internal resistance.
Power Battery
Quantity Operational
No Device Name Consumption Load
Keywords— railway vehicle, load characteristic, SoH (1 Car) per Car(W) rate
Power (W)
1 SIV Controller 3 900 0.70 630
I. INTRODUCTION Initial
2 236 6,372 0.05 319
The battery for railway vehicles is used as a power source cabin lights
Continuous
to maintain starting, control power, and emergency power for 3 80 2,160 0.95 2,052
cabin lights
devices. In the past, lead-acid batteries and nickel-cadmium … … … … … …
batteries were used in electric railway vehicles, but recently, 65 I/O module 10 1,000 0.70 700
66 Circuit breakers 2 156 1.00 156
lithium polymer batteries are being applied to increase safety Indicator Lamp
67 2 100 1.00 100
and efficiency. The Korean railway industry has been using
Battery load capacity 13,413
lithium batteries since 2017, but there is a lack of research on
battery capacity changes and performance degradation. In
The analyzed battery load power is 13,413W, supplied by
addition, most lithium battery research focuses on cells, and
three battery packs at 4,471W each. Each battery pack has a
research on battery packs used in railway vehicles is
current capacity output of approximately 50A, calculated
insufficient. In this paper, the battery charging method and
based on the rated voltage of the battery. In situations where
load type applied to railway vehicles were analyzed, and a
power reception through the pantograph is unavailable,
battery system with the same specifications was established.
railway vehicles rely on battery power for all internal DC
In addition, data such as voltage, current, and temperature
loads during startup. The auxiliary air compressor raises the
were collected through the BMS(Battery Management System)
pantograph using pressure from the auxiliary air duct to
to analyze the aging characteristics of the battery. Correlation
establish contact with the trolley wire. The battery is used to
with internal resistance was analyzed to estimate the
charge the auxiliary air compressor, which typically takes
SoH(State of Health) of the battery.
around 8 minutes and occurs once a day on average.[3] In
II. BATTERY SPECIFICATIONS AND LOAD CHARACTERISTICS addition, when an urban railway vehicle traverses an insulated
section where the catenary does not supply electric power
The battery used in the experiment had the same
during operation, the DC load is solely powered by the battery.
specifications as those currently used in actual vehicles, as
The insulated sections were analyzed based on the operational
shown in Table I. Furthermore, We analyzed load types and
conditions of urban railway vehicles in the metropolitan area.
operational rates for batteries in railway vehicles and
As a result, the average daily time spent passing through
established a standard for the battery discharge current.
insulated sections was approximately 3 minutes. Combining
TABLE I. RAILWAY VEHICLE BATTERY SPECIFICATIONS
this with the initial startup time of 8 minutes, resulted in an
analyzed battery usage time of approximately 11 minutes.
Classification Specifications
Lithium polymer battery
Therefore, in this paper, we conducted experiments with a
Battery type
Module 1 : 2P9S
discharge current of 50A for 12 minutes, applying a 10%
Configuration Module 2 : 2P9S margin to the 11-minute usage time.
Module 3 : 2P7S
Nominal voltage DC 91.25V (3.65V X 25Cells) III. BATTERY AGING EXPERIMENT
Maximum charging voltage DC 106.25V (4.25V X 25Cells)
Rated capacity 106Ah (56Ah*2P)
SoH is an indicator of the aging state of a battery and is used
to determine when the battery needs to be replaced. The SoH
The battery of the railway vehicle is primarily used to supply of a battery is typically calculated using equation (1), which
power for the initial startup and control devices, operating compares the initial capacity to the current capacity due to
under a DC load of DC 75V to 100V. The battery is charged aging. A battery is considered to have failed when its capacity
using the constant float charging method, and power is directly falls to 80% of its initial capacity.
supplied to the DC load through the charger during railway 𝐶𝑐𝑢𝑟𝑟𝑒𝑛𝑡
operations. In this paper, we analyze the types and capacities 𝑆𝑜𝐻𝑐𝑎𝑝 [%] = × 100 (1)
𝐶𝑖𝑛𝑖𝑡𝑖𝑎𝑙
- 441 -
Fig. 1 is an experimental configuration diagram. The
bidirectional power supply is controlled through a computer,
and battery voltage, current, and temperature data are
collected at 1-second intervals from the BMS.
Fig. 1. DIAGRAM OF BATTERY EXPERIMENT
Fig. 2 shows the waveform of the experimental results,
applying the analyzed load characteristics of the railway Fig. 3. AGING EXPERIMENTAL RESULTS
vehicle battery. Discharge was carried out in constant current
mode at DC 50A for 12 minutes, and charging was done in Fig. 3 presents the data and trends of battery capacity and
constant voltage mode at DC 100V at the end of discharging. internal resistance obtained from the battery experiments
The same conditions as the battery charging method of the conducted with the railway vehicle environment, incorporating
railway vehicle were applied, and a rest time for battery temperature compensation. After 300 cycles, the battery
stabilization was included. capacity decreased by approximately 1.7% compared to the
initial capacity, resulting in a calculated SoH of 98.3%.
Additionally, the internal resistance was estimated to have
increased by approximately 1.7% compared to its initial value.
IV. CONCLUSION
In this paper, the environment of railway vehicles was
applied, and partial discharge experiments were conducted
considering load characteristics. Additionally, a temperature
correction curve was derived and applied by conducting the
experiments at room temperature. The relationship between
internal resistance, an indicator of battery aging state, and SoH
was analyzed, confirming an increase in internal resistance as
Fig. 2. BATTERY EXPERIMENT WAVEFORM capacity decreases. It is expected that SoH prediction can be
achieved through analysis of various battery aging state
The battery experiment in the railway vehicle environment indicators and the use of machine learning techniques.
involved three stages: discharge, charging, and rest. During
discharge, there is an initial voltage drop due to internal series ACKNOWLEDGMENT
resistance(Rs) described by equation (2). The discharge This research is supported by the Korea Agency for
process exhibits nonlinearity due to polarization effects, Infrastructure Technology Advancement (KAIA) by the
gradually discharging based on the load level. The internal Ministry of Land, Infrastructure and Transport under (Grant
resistance of a battery increases with age, but its lifespan lacks RS-2019-KA152194)
a clear standard, unlike SoH.[4] In this paper, we conducted a
REFERENCES
partial discharge degradation experiment in a railway
environment for 300 cycles, measuring battery capacity every [1] Michael Brady, “Assesment of Battery Technology for Rail Propusion
Application”, U.S Department of Transportation FRA(Federal
100 cycles. We analyzed the relationship between capacity Railroad Administration), pp. 4-68, 2017
change and internal resistance based on the experimental data. [2] H.S. Oh, et al. "A Study on the SoC of Lithium Battery Applied to the
∆𝑉 Operating Environment of Urban Railway Vehicle." The Transactions
𝑅𝑠 = (2) of the Korean Institute of Electrical Engineers, vol. 72, no. 5, pp.
𝐼 662~668, 2023
[3] J. Y. Kim, "A Study on SoH Estimation of Lithium Battery
In the experiments, the battery was kept at room Considering Load Characteristics of Electric Railway Vehicles," Ph.D.
temperature, similar to its installation in the vehicle. However, dissertation, Korea National University of Transportation, 2023.
battery performance varies with temperature, so temperature [4] A. Nuhic, T. Terzimehic, T. Soczka-Guth, M. Buchholz, and K.
compensation is necessary for accurate estimation. A temper- Dietmayer, “Health diagnosis and remaining useful life prognostics
ature correction curve was calculated using manufacturer- of lithium-ion batteries using data-driven methods,” Journal of Power
Sources, vol. 239, pp. 680 ~ 688. October 2013.
provided characteristics and measured values and applied to
battery capacity and internal resistance.
- 442 -
Pantograph Charging System for Electric Bus
Seungkwon Shin1*, Kiwon Lee1, Hyungchul KIM1, Rakgyo Jeong1, Sangryul Choi2
1
Korea Railroad Research Institute, Korea
2
Induck University, Korea
Email: [email protected]
Abstract—Electric buses are widely used as a II. MULTI-CHARGING PANTOGRAPH SYSTEM
countermeasure against environmental pollution. It is
necessary to solve the problem of inefficiency in manual A. Domestic Electric Bus Charging System
charging of existing electric buses and to develop to localize In a domestic electric bus charging market, a
and standardize pantograph automatic transfer charging
pantograph type charging system is in a starting stage, a
technology applied with pantograph-type electric bus
charging technology recently introduced in North America CCS’s (combined charging system) fast charging system is
and Europe. Therefore, we are planning to form a consortium established, and plug type charging methods for high
and carry out demonstration projects with large bus carriers current charging are occupied for the majority.
to successfully develop and respond to standardization of Plug-in charging is currently widely used in electric
electric bus transfer type automatic charging system with roof vehicles, and the charging speed is slower than other
connection. This is expected to contribute to improving the methods and requires a dedicated manpower for charging.
spatial/temporal efficiency of bus depot for electric bus Wireless charging method is a technology that embeds
charging. This paper aims to deal with the technology a power rail in the lower part of a bus route and charges
development trend of the pantograph type electric bus
electricity using a magnetic field in a collector window
charging system.
under the bus vehicle. Electric buses can be charged while
on the go, but the initial charging infrastructure cost is the
Keywords—Electric bus, Charging method, PLUG-IN,
highest. Non-contact charging also reduces charging speed
Wireless Charging, Pantograph down, Pantograph-Up and power efficiency.
TABLE I. ELECTRIC BUS CHARGING STATUS IN KOREA
I. INTRODUCTION
As the 'Paris Agreement' was adopted in 2015 as a Type PLUG-IN Wireless Charging
measure against the acceleration of global warming due to
the increase in greenhouse gases caused by the use of fossil
fuels by humans, 195 countries must implement measures
to reduce greenhouse gases from 2020. Recently, in order to Representative
Picture
reduce greenhouse gases, each local government has
introduced and operated electric buses in accordance with
the government-led policy to promote the introduction of
eco-friendly and carbon-free transportation. The electric bus - Builds about 5-15%
- Various charging electricity supply
charging technology can be divided into a plug-in charging methods make it infrastructure for all bus
method, a battery exchange method, a pantograph charging inconvenient to use routes
method, and a wireless charging method depending on the (CHADEMO, - Only OLEV (Online Electric
GB/COMBO I, II, etc.) Vehicle) vehicles are
power supply method. However, the problem of the plug-in Key Features - It is a manual charging recognized, Reducing standby
type charging system applied to the existing electric bus is method, and operating power loss by applying power
the charging connector part and appearance are frequently costs are borne due to supply segmentation
damaged, the charging time is long and it takes 70-80 damage to charging guns - When charging in a non-
and cables and the need for connected wireless charging
minutes to fully charge. In addition, dedicated personnel are
dedicated personnel method, the speed and power
required for charging, and allowances are incurred as a efficiency are lowered.
result, which is a burden on the management of the electric
bus. Therefore, it is necessary to secure the pantograph-type - It is the most charging
charging technology currently being introduced in Europe - Sejong City, Gumi City
method in Korea and is
and North America, and to develop the pantograph-type - Pilot projects and
Operation Status applied to 1:1, 1:N public
demonstration projects are in
large-capacity automatic charging technology suitable for and private bus garages.
progress
the domestic garage environment.. - buses, cars, trucks, etc
- Development of a fast - Charging efficiency
charging method of 300- maximization technology
400 KWh or more - Development of Failure
Development
- The PLUG-IN charging Response Technology for
Direction
method is being improved Buried Chargers
through an automatic - Wired and wireless
socket method or a robot. integration technology
- 443 -
B. Overseas Electric Bus Charging System
Overseas rooftop charging systems are being actively
developed by ABB and Siemens for efficient charging
infrastructure by using the pantograph down method (Pole-
mounted) and the pantograph up method (Roof-mounted).
The pantograph down method, in which the charging
infrastructure contacts the roof of the vehicle to charge, and
the pantograph up method, which connects the pantograph
installed on the roof of the electric bus. The pantograph
charging method can be charged at both the garage and the
bus stop, but the initial charging infrastructure construction
cost is high. Fig. 1 is a prototype of a top-down pantograph
method, and Fig. 2 is a conceptual diagram of charging four Fig 1. Prototype of top-down pantograph method
vehicles with one pantograph.
TABLE II. OVERSEAS ELECTRIC BUS CHARGING STATUS
Pantograph-Down Pantograph-Up
Type
(Pole-mounted) (Roof-mounted)
Representative
Picture
Fig 2. Schematic diagram of automatic positioning multi-charging
pantograph system
- Adverse to vehicle III. CONCLUSION
- Advantage of vehicle
layout/weight (pantograph In order to prevent global warming and spread low-
layout/weight (pantograph
located on vehicle roof) carbon transportation, it is necessary to increase the
located on pole)
- Advantageous for
- Advantages to vehicle
infrastructure construction operating efficiency of electric buses. As a result of the large
cost and supply capacity of electric bus charging infrastructure, it is required
Key Features and supply
- Use of wireless (Wifi)
and wired (PLC) between
- Use of wired (PLC) to develop a charging method for safety and automation of
between vehicles and charger and charging interface. In addition, it is required to
vehicles and infrastructure
infrastructure introduce a new fast charging method to effectively respond
- Mainly applied to station
- Applied to station / garage
charging
charging
to the growth of high-capacity battery-powered vehicles.
And, to reduce electric bus costs and ease of maintenance,
the development of a garage-type intensive automatic
charging system to charge a number of vehicles in the
-Hamburg(Siemens Corp.)
- Schiphol Airport, garage of bus companies should be promoted by applying a
Eindhoven, Norway, pantograph down method. In addition, more research and
Operation Status Luxembourg(ABB Corp.,
Sweden, Singapore, etc development is needed to advance overseas.
Heliox Corp.)
(Heliox Corp.)
- Recently, it has been REFERENCES
actively carried out by a [1] Aoife Foley, Beatrice M.Smyth, Tmmislav Pukšec, Natasa
- Recently, it has been
number of charger Markovska, and Neven Duić, “A review of developments in
actively carried out in some
companies, mainly in technologies and research that have had a direct measurable impact
charger companies, mainly
Europe on sustainability considering the Paris agreement on climate change”,
in Europe
Development - Suitable in terms of Renewable and Sustainable Energy Reviews, Vol.68, No.2, pp.835-
- Suitable for ultra-large
Direction pantograph cost and failure 839, 2017.
vehicles such as trams,
when a large number of
BRTs in terms of [2] Seungkwon Shin, Kiwon Lee, Chulmin PARK, Jongyoung Park,
electric buses operate on
pantograph cost and fault Juuk Kim, Hyungchul KIM, “Pantograph Chargeing System for
urban routes, with a focus
response Electric Bus”, The 52th Korean Institute of Electrical Engineers
on various transportation
Sunner Conference 2021, No.7, p.1896, 2021.
companies
[3] Koon-Hwan Oh, Ki Hyeon Kim, Sang-Taek Lee, “Introduction of
the trend of charging types for electric bus and power management
needs of charging system”. Korean Institute of Electrical Engineers
Information and Control Conference 2021, pp.252-253, 2021.
[4] Press release Deutsche Post DHL, December 2014
- 444 -
Preparation Analysis of the Reduction Effect the
Increase Harmonic Resonance in the AC Electric
Railway Feeding System
Guk Hyun Son 1, Tae Hoon Kim 1, and Jae Moon Kim 1,2*
1
Department of SMART Railway System, Korea National University of Transportation, Republic of Korea
2
Department of Transportation System Engineering, Korea National University of Transportation, Republic of Korea
Email: [email protected] (JMK), [email protected] (GHS), [email protected](THK)
Abstract—The harmonic generation characteristics harmonic measurement location in the power supply
according to the running of the railway vehicle were confirmed substation.
by measurement at the sectioning post, which is the end of the
power feeding system. In particular, the phenomenon of
enlarged resonance of harmonics was confirmed through
measurement. As a solution, the harmonic resonance
suppression effect according to the separation or connection of TABLE I. IMPEDANCE AND FACILITY CAPACITY OF THE MEASURING LINE
the R-C bank was compared and presented. In this paper, it was
possible to quantitatively show the effect of reducing harmonic A downstream
Section An upward track
track
resonance through actual measurement.
Distance(㎞) 30.23 ㎞ 30.23 ㎞
Keywords— Source of harmonics, Harmonic resonance, Z 23.23 24.02
Harmonics reduction, R-C Bank
θ 70.73 70.73
Impedance
R 7.67 7.93
(TF-R)
X 21.93 22.67
INTRODUCTION
Z/㎞ 0.768 0.794
M.TR 45/60[MVA]
This study measured the number of vehicles put into the Facility
relevant power supply section on the actual train operation AT 7500[KVA]
capacity
track of the AT power supply system and the degree of RC-Bank R:550[Ω], C: 0.21[μF]
harmonic generation due to train operation. As a result, the
amount of harmonics generated was quantitatively confirmed. Theoretically, when the impedance of the sectioning post at
In addition, the power supply circuit displayed by the
the location of the electric train is , it is as follows.
distribution constant circuit is amplified by parallel resonance
at capacitance and higher order frequencies.[1] Accordingly, (1)
the reduction effect could be confirmed quantitatively by
comparing measurements for each condition using RC-BANK, It can be expressed like from this
the countermeasure facility. This study measured the AT
transformer supply section where various vehicles, including equation (2)
KTX, are actually operating, and confirmed the effect of
That is, if impedance is equal to characteristic
reducing resonance occurrence through R-C BANK
impedance using R-C BANK, it becomes and
connection or separation.
the characteristic impedance does not cause resonance
HARMONIC MEASUREMENTS IN THE FEEDING SECTION regardless of the location of the electric railway vehicle. If the
harmonic resonance is short-circuited at the supply end with a
resistance of the same value as the line's characteristic
Table 1 shows the system impedance and facility capacity impedance, a suppression effect will appear.[2]
of the relevant power supply section, and Figure 1 shows the
- 445 -
Accordingly, harmonic analysis equipment was installed amplification compared to cases where RC-Bank is not
on the control panel PTT of the power supply station to connected.
measure harmonic (THD) in order to analyze the amount of
TABLE III. RC-BANK ON
harmonic generated and resonance in the power supply section.
Section Minimum Maximum
Voltage THD[%] 1.17 9.32
Fig. 1. Harmonic (THD) measurement location in the sectioning post
Fig. 3 Feeding Feeding voltage and Harmonic in case of RC- Bank ON
C. Comparison of harmonic reduction effects
HARMONIC MEASUREMENTS BY CASE
Figure 4 shows a comparison of the measurement results for
A. RC-BANK not connected the harmonic size for each condition before and after RC-Bank
input. Before RC-Bank, the harmonic size was indicated as A,
It is a graph measuring the voltage THD according to the and the harmonic size was indicated as B and C (dotted line)
movement of the vehicle in the corresponding section while
the RC-Bank is not connected.
The measured harmonic magnitude is shown in Table 2, and
it may be seen that the harmonic generated from the electric
railway vehicle is amplified and very high.
TABLE II. RC-BANK OFF
Section Minimum Maximum
Voltage THD[%] 1.47 21.65
Fig. 4 Comparison of THD before and after RC-Bank
CONCLUSION
Recently, the reduction effect of harmonic generation and
resonance expansion due to the operation and stop of RC-
BANK on the AT power supply system operation line
operated by various vehicles was compared. Before the
connection of the RC-Bank, the voltage THD was up to
Fig. 2 Feeding voltage and Harmonic in case of RC- Bank OFF 21.65%, but a significant reduction effect was confirmed to be
up to 9.32% in the RC-Bank operation. In order to increase the
effect of reducing harmonic resonance, it is necessary to apply
customized R-C values based on the harmonic source and the
B. RC-BANK connected impedance of the system through field measurement.
Figure 3 shows the result of measurement in a normal state
by connecting the R-C Bank of the power supply substation.
Changes in harmonic voltage significantly reduce harmonic REFERENCES
- 446 -
[1] Kim Jeong-cheol, “ Analysis of power supply system and
understanding of Korea Electric Railway,” pp. 556, 2020. 8
[2] Railway Technical Research Institute, electrical system technical
course, pp. 12, 1997. 10.
- 447 -
A Study on the Train Position Detection for the
Integrity of Freight Train Based on ETCS Level
3 With GPS and 6-Axis Sensors
Jeon, Seong Yong1 Nurlan Berdigozhin2 and Choi, Kyu Hyoung1*
Seoul National University of Science and Technology, Korea
1
2
Hua Xia Company, Kazakhstan
Email: [email protected] [email protected] [email protected]
1
Abstract— Accurate detection of the precise position of status of train position detection both domestically and
regular and inspection trains plays a vital role in enhancing the internationally, and conduct tests by creating hardware and
safety, efficiency, and convenience of railway systems. In software for freight train position detection technology. We
particular, in ETCS Level 3, which supports automatic train design algorithms for software development and proceed with
operation, monitoring and detecting the position of trains software development. We develop hardware that can be
running without track circuits is a key technology. Currently, tested by applying the developed software. As it is challenging
with regular trains, it is possible to monitor the status of the to conduct tests directly in railway environments, we conduct
formation as power is supplied to all vehicles. However, for tests on the road using regular vehicles.
freight trains, power is only supplied to the locomotive, which
limits the ability to monitor the status of train formation. II. TRAIN POSITION DETECTION TECHNOLOGY AND
Although it is possible to verify the status through brake pipe NAVIGATION SYSTEMS
and air pressure, there are limitations in ensuring safe and
efficient operation. Shift2Rail Safe Train Positioning (TD2.4) aims to develop
safety devices and a multi-sensor train positioning system. In
In this study, we investigated the research status of domestic particular, efforts are being made to replace various sensors
and international train position detection for accurate detection with a trackside train monitoring system in order to improve
of freight train integrity based on ETCS Level 3. We proposed the quality of train position and integrity information while
an algorithm to determine the position information calculated
reducing costs.
through real-time information from GPS and 6-axis sensors.
Furthermore, we conducted field tests using a prototype Figure 1 represents the basic structure proposed by Shift2Rail.
incorporating the proposed algorithm and analyzed the test It utilizes GPS information, IMU (Inertial Measurement Unit)
results, providing insights and discussions based on the collected data, and track information to detect train position.
data.
Keywords—ETCS Level3, 6-Axis Sensors, GPS, Freight
Train, Train Position Detection
I. INTRODUCTION
Trains that transport people or goods and travel along a
fixed rail typically consist of multiple passenger or freight cars
connected together. Detecting the position of these trains
allows for controlling the spacing between trains and prevents
accidents between preceding and following trains. Therefore,
train position detection is crucial for controlling the train
spacing between stations and ensuring proper train routing
within the station.
In the case of regular trains, if a coupling between cars
becomes disconnected due to an accident during operation, the
train separation signal line is interrupted, allowing for the
detection of the train separation situation. On the other hand,
electric or diesel locomotives used for freight operations are
operated in non-fixed formations, and power is supplied only A. Tachometer
to the leading vehicle, with brake pipes connected between In railway vehicles, the speed is calculated by the pulse
adjacent cars that do not require power connection. count of the tachometer installed on the wheels and the
calculation based on the wheel diameter. The calculated speed
However, in some cases of freight trains, when the
and the distance traveled are used to determine the position of
pressure in the brake pipe decreases due to car separation, the
the train. However, due to the nature of trains running on steel
filling speed of air in the brake pipe to replenish the pressure
tracks, there are factors that can introduce errors, such as
may occur rapidly, resulting in the scale change on the brake
slippage and wheel slip on the track, as well as errors caused
pipe gauge not being perceived. As a result, trains have been
by wheel wear. Therefore, it is necessary to develop methods
known to continue operating for a significant amount of time
to compensate for these errors and improve the accuracy of
in a separated state. In this study, we investigate the research
train position detection.
- 448 -
B. Satellite navigation system celestial navigation is employed to determine an accurate
The Global Navigation Satellite System (GNSS) is a position, from which dead reckoning is then resumed.
satellite-based system that provides three-dimensional To determine the position using dead reckoning, devices
position and precise time information by utilizing the such as a plotter (a protractor attached to a straight edge) or
positions of multiple satellites in Earth's orbit and distance calculation tables are used. Nowadays, smaller aircraft or
measurements using radio waves. GNSS is a type of radio vessels often utilize these devices, while larger aircraft or
navigation system that is synchronized with atomic clocks, naval vessels have been mostly replaced by one or more
and it determines the receiver's position using the time computers that receive inputs of direction and speed (wind
difference of arrival (TDOA) technique, which measures the speed can be manually entered).[15]
difference in arrival time of signals transmitted from
navigation satellites.
In the field of railway transportation in Korea, GNSS III. TRAIN LOCATION DETECTION METHOD USING 6-AXIS
systems are used to provide operational information to train SENSOR AND GPS
operators, such as wireless VHF, ASTRO, and TETRA ETCS L1/L2 Electronic Interlocking System (EIS) uses
communication frequency change notifications, as well as track circuits to verify the position of trains. However, in the
section and train station arrival notifications. They are also case of ETCS L3-based Electronic Interlocking System, train
employed in maintenance and inspection vehicles for position information is obtained wirelessly from the Radio
detecting anomalies in equipment and track infrastructure and Block Center (RBC), and a system configuration based on
for verifying the location of detected anomalies. Virtual Block is implemented within station areas. This
requires the development of train position detection
technologies in locations where track circuits are not available,
C. Inertial navigation system as well as train route control technologies based on the
The Inertial Navigation System (INS) is a navigation concept of moving block.
system that utilizes inertial sensors, such as gyroscopes and
A. Configure train location detection
accelerometers, to calculate the orientation, velocity, and
position of a moving object by integrating the outputs of Regarding the train position detection device, an algorithm
angular velocity and linear acceleration. Unlike GNSS, INS is developed and tested to track the position of the train using
operates independently of external infrastructure and can the signals from the 3-axis accelerometer and 3-axis
provide navigation information without relying on GPS gyroscope sensors mounted internally when GPS signals are
signals. not received in tunnel areas or GPS shadow zones. The
algorithm utilizes the acceleration vector signals from the
With advancements in Micro Electro Mechanical Systems accelerometer, as well as the angular velocity signals from the
(MEMS) technology, MEMS-based inertial sensors have been gyroscope, to calculate the tilt correction values of the train's
developed, offering improved performance at a lower cost. plane and gravity plane. These corrected values are then used
INS is not only used in aerospace and military applications but to track the train's position.
also integrated with GPS to enhance the position accuracy of
vehicle navigation systems. To implement train position detection, the GPS
information and data obtained from the 6-axis sensor are
In the railway industry, INS can be employed to improve utilized to generate the final position information using a
the accuracy of vehicle navigation systems, providing precise Linear Kalman Filter. In open areas or areas with GPS
positioning and velocity information without the need for coverage, the position is determined primarily based on GPS
external infrastructure support.[1] information. However, in tunnel sections where GPS signals
may be unavailable or unreliable, the sensor data from the 6-
axis sensor is used to estimate and correct the position using
D. Dead reckoning system the Kalman Filter. This corrected information is then used as
Dead reckoning or Dead reckoning navigation is a method the basis for determining the train's position in tunnel sections.
of navigation where one estimates their position and navigates
based on calculated course and speed from a known starting
point (datum). The estimated position (EP) is determined by
considering various factors that may affect the navigation,
such as the characteristics of the moving vehicle and
environmental conditions (e.g., wind pressure
differentials).[15]
The navigation route, speed, estimated distance based on
these factors, known starting point, and recorded or estimated
deviation are used to determine the estimated position. In
practice, almost all navigation devices utilize dead reckoning
to some extent. Some navigators distinguish between the dead
reckoning position (DRP), which is the position calculated B. Algorithm for Freight Train Position Detection Using 6-
based on the course and estimated speed through the water, Axis Sensor and GPS
and the estimated position, which is the dead reckoning Train position detection is divided into four stages:
position adjusted for other factors such as tidal currents and stationary stage, movement stage, gradient movement stage,
wind. As the uncertainty increases with time and distance, and position estimation stage. In the MCU, during the
- 449 -
stationary stage of the train, the device plane is established, e) The GPS coordinates are represented as a 2-dimensional
and the inclination between the gravity plane and the train's value (x1, y1).
plane is calculated. In the movement stage, the correction
acceleration for the gravity axis is calculated. In the gradient f) The process proceeds to the stationary stage.
movement stage, when the train is moving on a gradient, the
inclination for latitude and the slope angle for altitude are
computed. In situations where the train cannot receive GPS 2) Train Progress Stage
location information, the 6-axis sensor is used to perform the a) It is executed when the State Flag is set to Forward.
coordinate estimation stage. b) When the GPS Receiver receives GPS location
information and confirms the train's movement, the
following process is triggered.
c) If the train's speed is greater than 5 m/s, it is
determined that the train is in motion.
d) If the train is moving, the Z-axis corresponding to the
first angular velocity (h1) of the first acceleration
signal (r1, q1, h1)ω received from the 6-axis sensor
is selected to obtain the rotation angle (R) of the gyro
sensor.
e) Applying the correction value of the tilt
compensation for the device plane (PlaneD) to the Z-
axis, which is the device's gravity axis, the
compensated acceleration value is calculated.
f) Obtaining the GPS coordinates 2 (x2, y2)GPS from
the GPS Receiver, and the second acceleration vector
signal (x2, y2, z2)A and the second angular velocity
signal (r2, q2, h2)ω from the 6-axis sensor. The
compensated plane of gravity is calculated by
reflecting the correction value of the tilt
compensation to the second acceleration plane vector
signal (x2, y2) of the second acceleration vector
signal A.
g) Based on the compensated plane of gravity, the
second X-axis acceleration vector (x2) and the
second Y-axis acceleration vector (y2) of the second
acceleration plane vector signal (x2, y2, z2)A are
vector summed to create the forward direction axis.
Fig. 3. Program Flow
h) During the train's progress, the State Flag is changed
to Incline, and the Incline process is carried out.
1) System Initialization Stage
a) When the program starts, the State Flag is changed to
3) Gradient Progression Stage
"Init," and the sensor inputs are checked for normal
a) It is executed when the State Flag is set to "Incline."
operation.
b) Continuously check the movement of the train while
b) The 6-axis sensor consists of an accelerometer sensor
receiving GPS signals.
and a gyroscope sensor. The accelerometer sensor
outputs acceleration vector signals in 3 axes (X, Y, Z), c) Determine if the train's movement speed is greater than
and the gyroscope sensor outputs angular velocity 5 m/s and if the number of satellites exceeds 10 by using
signals in 3 axes (R, Q, H) every 100ms. The program GPS position information.
sequentially averages the 10 data signals to calculate the
average value every 1 second. d) Store the obtained gyro rotation angle (R) and the
compensated acceleration value for the gravitational axis
c) The first acceleration vector signal (x1, y1, z1)A and the (Compensated acceleration value) obtained in the train
first angular velocity signal (r1, q1, h1)ω are obtained movement stage.
from the 6-axis sensor. Here, the first acceleration vector
signal (x1, y1, z1)A represents the acceleration vector e) If the train's velocity is greater than 5m/s and the number
values, and the first angular velocity signal (r1, q1, h1)ω of satellites exceeds 10, the algorithm calculates the
represents the angular velocity vector values. rotation angle (D) using the inclination (ϴ) with respect
to latitude. The rotation angle (D) represents the
d) The GPS receiver checks if the position information is difference between the previous slope (ϴ0) and the
received correctly every 1 second. current slope (ϴ1) with respect to latitude. This
calculation can be illustrated using Figure 4, and the
corresponding formula is Equation (1).
- 450 -
𝑡𝑡𝑡𝑡𝑡𝑡𝑡𝑡 𝑛𝑛𝑛𝑛𝑛𝑛 = 𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶 𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠(∅1 ) +
𝑇𝑇ℎ𝑒𝑒 ℎ𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜 𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟 𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎 𝑜𝑜𝑜𝑜 𝑡𝑡ℎ𝑒𝑒 𝑔𝑔𝑔𝑔𝑔𝑔𝑔𝑔𝑔𝑔𝑔𝑔𝑔𝑔𝑔𝑔𝑔𝑔 𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠(𝑅𝑅)
i. (3)
e) The device's incline angle (Incline angle_new)
is calculated by adding the calculated incline
angle from the slope progression (α) and the
pitch rotation angle (q) from the gyroscope
sensor. The equation for this calculation is as
follows.
f) Rotation angle (D) for latitude.
𝐼𝐼𝐼𝐼𝐼𝐼𝐼𝐼𝐼𝐼𝐼𝐼𝐼𝐼𝐼𝐼𝐼𝐼𝐼𝐼𝐼𝐼𝐼𝐼𝐼𝐼𝐼𝐼𝐼𝐼 = 𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠 𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝(𝛼𝛼) +
𝑡𝑡ℎ𝑒𝑒 𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝ℎ 𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟 𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎(𝑞𝑞) (4)
𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅 𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎(𝐷𝐷) = 𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶 𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠(∅0 ) −
𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃 𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠(∅1 ) (1)
f) The speed of the train (v_new) is calculated
using the current speed based on GPS location
information (V_GPS) and the compensated
g) To determine if the difference between the rotation angle acceleration value stored during the slope
(D) for latitude inclination and the gyroscope sensor progression stage. The equation for this
rotation angle (R) is less than or equal to 5 degrees, and calculation is as given in Equation (5).
if it is, calculate the current inclination for latitude (ϴ) Additionally, by accumulating the acceleration
and the slope angle for altitude (α). If the difference is over 1 second, it becomes the speed.
greater than 5 degrees, the state flag remains unchanged,
and the slope progress process is repeated. Equation (2)
represents the calculation for the slope angle for altitude.
𝑣𝑣_𝑛𝑛𝑛𝑛𝑛𝑛 = 𝑣𝑣_𝐺𝐺𝐺𝐺𝐺𝐺 + 𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶 ∗
𝑡𝑡(= 1) (5)
the slope angle for altitude (𝛼𝛼)
= 𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐 𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠 𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎 𝑓𝑓𝑓𝑓𝑓𝑓 𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎(𝛼𝛼1 )
− 𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃 𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠 𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎 𝑓𝑓𝑓𝑓𝑓𝑓 𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎(𝛼𝛼0 ) g) Using the tilt_new of the device's horizontal
(2) plane and the v_new speed of the train, the
coordinates on the GPS are calculated.
h) If the difference between the rotation angle (D) for h) If location information is received through GPS
latitude inclination and the gyro sensor rotation angle (R) reception, the State Flag is set to Normal, and if
is less than or equal to 5 degrees, the State Flag is location information is not received, the State
changed to "Normal," and the Normal process is Flag is set to Tunnel to proceed with the process.
executed.
IV. TESTING AND EVALUATION
4) Position Estimation Stage A. Testing Methods
a) This process is executed when the State Flag is
set to Normal or Tunnel. In this study, the system under investigation involves
connecting a 6-axis sensor and GPS to a moving vehicle to
b) When GPS information is received, the State track its trajectory. In open areas, the vehicle's movement path
Flag is set to Normal, and the process continues. is determined through GPS information. When entering a
The position of the train is determined using the tunnel, the values from the 6-axis sensor are utilized in
inclination angle for latitude and the position conjunction with GPS to calculate and display the vehicle's
obtained from GPS. movement path using a cargo train position detection
algorithm.
c) If GPS information is not received, the
inclination angle for the device's horizontal
plane (tilt_new), the incline angle for the
device's vertical plane (Incline angle_new), and 1) Test Environment Configuration
the train's velocity (v_new) are calculated. The GPS antenna is externally exposed on the vehicle to
ensure good reception of GPS signals. The test environment is
d) The inclination angle for the device's horizontal configured with the vehicle traveling between the open area
plane is calculated using the following equation (where GPS signals are well received) and the tunnel (where
GPS signals are not received). Inside the tunnel, the latitude
and longitude coordinates calculated through the 6-axis sensor
- 451 -
are stored on a laptop. The stored values are then displayed on acceleration_X value increases, the speed also increases, and
a map to verify the results. as the acceleration_X decreases, the speed decreases as well.
In the underground track or tunnel sections, we can observe
a) Test location that the GPS velocity is displayed as 0, while the acceleration-
This is the road section for on-site testing. velocity values do not show a value of 0 and continue to
indicate speed.
Fig. 4. Test location
Fig. 5. Upstream acceleration sensor information
TABLE I. KEY TEST ENVIRONMENT INFORMATION
No Item Key Content
1381-30 Iui-dong,
1 Starting Point Yeongtong-gu, Suwon-si,
Gyeonggi-do
1382-3 Yeongtong-gu,
2 Destination Point
Suwon-si, Gyeonggi-do
3 Straight distance About 3.2Km
4 Lane distance About 3.7Km Fig. 6. Downstream acceleration sensor information
5 Open land About 2,985m
The X, Y, and Z values from the gyroscope sensor were
6 Tunnel section About 100m
used to generate graphs as shown in Figures 8 and 9. The
Underground lane gyroscope Z-axis determines the orientation of the device. It
7 About 615m
section was observed that during left turns in the upward direction, the
The tunnel section has a length of 100m. gyroscope Z-axis increased, while during right turns in the
The underground roadway section has a length of 615m. downward direction, the gyroscope Z-axis decreased. This
The underground roadway provides a favorable testing indicates that the gyroscope Z-axis exhibits normal changes
environment for measuring direction and speed in curved during left turns at intersections.
sections compared to straight sections.
B. Analysis of Test Results
After aligning the device plane and the vehicle plane using
acceleration_X and acceleration_Y with respect to
acceleration_Z as the center, the results obtained through the
algorithm proposed in this study can be summarized as
follows. The following data graph was derived using the X, Y,
Z values from the acceleration sensor, sensor velocity, and
GPS velocity.
The acceleration Z value represents gravity, while
acceleration_X indicates changes in speed. By observing the
graphs in Figures 6 and 7, it can be seen that as the
- 452 -
• A train position detection algorithm utilizing GPS and a 6-
axis sensor was proposed.
• A prototype was developed to apply the algorithm and
performance evaluation tests were conducted using a
regular vehicle. Data was analyzed and discussed.
• The algorithm and prototype effectively detected the
position of underground tracks and tunnels.
• Through this study, it was observed that the accelerometer
sensor values vary significantly due to road conditions
and vehicle vibrations, leading to speed errors. On the
other hand, the gyroscope sensor provided relatively
Fig. 7. Upstream gyro sensor information accurate directional information.
In ETCS Level 3, it is necessary for the leading and trailing
ends of freight trains to be within a certain distance to ensure
train integrity. While the leading train can receive electric
power and external signals such as tachometers to determine
its position, the trailing train, being non-powered, cannot be
equipped with such devices.
By applying the proposed algorithm to accurately
determine the position of the freight train's trailing end without
relying on external inputs, efficient integrity detection of
freight trains can be achieved.
Fig. 8. Downstream gyro sensor information
REFERENCE
The values from the accelerometer sensor were found to
be too sensitive, leading to errors in speed calculation. [1] Korea Railroad Research Institute, Kyungpook National University.
Vibrations from the vehicle, such as when crossing speed (2012). "Development Planning Research Report for Railway
bumps or encountering poor road conditions, caused Transportation Position Detection System." [CODIL] Korea Railroad
inaccuracies and introduced errors in the speed calculation. On Research Institute Academic Information, pp. 101-122.
the other hand, the gyroscope sensor provided relatively [2] Korea Rail Network Authority. (2017). "Track Circuit General
accurate values and did not pose significant issues in Information." Korea Rail Network Authority, July 2017.
determining the direction. [3] Korea Railroad Research Institute. (2019). "Development of
Tachometer for Automatic Train Control (ATC)." [NTIS] Korea
To address the identified issues, several alternative Railroad Research Institute, pp. 17-18.
solutions can be considered. One approach is to modify the [4] Lee, S., Yoon, J., & Byeon, W. (2021). "Trends in Satellite Navigation
algorithm by adjusting the application timing of the Kalman Systems and Technologies." Journal of Electronics and
filter, both in the prediction and update stages. Additionally, Telecommunications Trends, Vol. 36, No. 4, pp. 61-71.
the acquisition time for accelerometer sensor values can be [5] "GPS," [Online]. Available:
https://en.wikipedia.org/wiki/Global_Positioning_System.
modified. Currently, the data is sampled every 100ms and
[6] Kim, H. (2020). "A Study on Train Position Detection Using Wireless
averaged over 1 second. By changing this acquisition time, Communication." Master's Thesis, Seoul National University of
more accurate speed values can be obtained. Science and Technology.
Furthermore, by storing the GPS coordinates of the [7] Park, J., Hong, W., Jeon, B., & Railroad Signal Engineering. Same
Publisher, 2001.
railway track in tunnel and GPS signal weak areas within the
[8] Types of Track Circuits, Korea Rail Network Authority: Dec 2020.
device, it would be possible to calibrate the sensor data based
on those GPS coordinates. This would result in a more [9] Shin, K., & Lee, J. (2012). "Position Detection Technology for Railway
Transportation." Railway Journal, Vol. 15, No. 1, pp. 16-21.
accurate determination of the train's position. The sensor data
[10] Ahn, D., Lee, B., & Nam, H. (2012). "Development of KTX Track
acquired from the 6-axis sensor can be further refined using Defect Diagnosis Program using Track Circuit Modeling." Journal of
GPS reference values and Kalman filters to reduce error Lighting & Electrical Engineering, Vol. 26, No. 10, pp. 44-51.
frequency and enable effective train position detection. [11] Kim, S., Park, S., Lee, J., Kwon, M., & Kim, H. (2014). "Preliminary
Results of Train Position Detection System Using Position Detection
Overall, these modifications are expected to improve the Sensors." Proceedings of the Korean Society for Railway 2014
accuracy of the speed calculation and facilitate more effective Conference, pp. 861-865.
train position detection. [12] Lee, S., Lee, W., & Lee, B. (2013). "Study on Vehicle Control
Technology using Train Position Detection." Proceedings of the
Korean Institute of Electrical Engineers 2013 Fall Conference, pp. 281-
282.
V. CONCLUSION [13] Lee, J. (2016). "Study on Indoor Localization using Sensor Fusion
The obtained research results from this study are as based on Dead Reckoning." Master's Thesis, Soongsil University,
Seoul.
follows:
- 453 -
[14] ISMB. "GNSS Algorithms Design." Satellite-based Signaling and
Automation Systems on Railways along with Formal Method and
Moving Block Validation.
[15] "Dead Reckoning," [Online]. Available:
https://en.wikipedia.org/wiki/Dead_reckoning.
- 454 -
An Analysis of the Changes of Regenerative
Energy Utilization Facilities by the Tap
Switching of Urban Railway Transformers
Chi-Myeong Yun 1, and Hosung Jung 2*, Hanmin Lee 2, and Hyungchul Kim 2
1
University of Science and Technology, South Korea
2
Korea Railroad Research Institute, South Korea
Email: [email protected]
Abstract— As climate change is progressing rapidly,
numerous efforts are being made worldwide to reduce 𝑪𝑶𝟐
emissions. In the field of rail transport, numerous studies are
being carried out on the use of regenerative energy, based on the
properties of urban railways, which often involve acceleration
and braking. It offers many advantages as the power
consumption of urban railway substations can be effectively
reduced by using regenerative braking energy appropriately.
However, existing urban railway substations are not energy
efficient as they maintain a limited voltage range via diode
rectifiers with fixed tabs of transformers. If an active energy
efficiency improvement method using an On Load Tap Changer
(OLTC) or the like is applied, energy efficiency can be improved
by changing the input voltage. Therefore, this paper analyzes
the characteristics of the system change according to the tap Fig. 1. Typical DC traction power system of urban railway system
adjustment of the transformer, and further introduces a method
of adjusting the threshold voltage of the regenerative energy 3√2
utilization facilities. E = ×𝑉 ()
𝜋
Keywords—DC railway, Regenerative Energy, OLTC, E is a DC voltage, V is an AC voltage. The supply voltage
Energy Efficiency, Transformer of the system shall maintain a voltage in accordance with the
international standards of IEC 60850.
I. INTRODUCTION
The load of a railway train is characterized by a large
Due to rapid climate change, many efforts are being made
variation. That is, the voltage of the system varies greatly due
to reduce carbon worldwide. In the field of railway
to the change in load, and in severe cases, there is a possibility
transportation, many studies are being conducted using
that sufficient power may not be supplied to the train.
regenerative energy generated using the characteristics of
urban railway system that frequently accelerate and brake[1,2]. If high voltage is continuously supplied by adjusting the
However, existing urban railway systems are not energy- transformer tab in consideration of the overload situation, the
efficient because they maintain a limited range of voltage insulation of train-related parts is weakened. It is possible to
through diode rectifiers[3]. Therefore, in this paper, changes reduce the voltage fluctuation rate by reducing the
in power consumption and regenerative energy were transformer % impedance, but the size of the short-circuit
simulated and analyzed in the event of a change in the primary current will increase and the breaking capacity of the rectifier
voltage of the transformer through adjustment of the and DC circuit breaker will also increase.
transformer tab.
Therefore, if the voltage supplied from the substation
II. MODELING AND ADVANTAGES DESCRIPTION increases and decreases according to the situation rather than
supplying the voltage at a fixed value, the problems specified
A. Configuration of DC Electric Railway System above do not occur, and energy utilization can be increased.
Figure 1 shows a simplified diagram of the DC power The power consumption of the substation is calculated as
supply system of a general urban railway. DC power supply Equation (2) after representing the DC electric railway feeding
systems are usually converted through silicon rectifiers that system in which one train is operating between the two
receive AC from power suppliers and convert it into DC. Since substations as an equivalent circuit.
the diode of the rectifier is an uncontrolled element, there is a
2 𝑔2 +𝑉 𝑔 𝑉 𝑔 )
disadvantage in that it is impossible to adjust the catenary (𝑉𝑠𝑎 𝑎 𝑠𝑎 𝑎 𝑠𝑏 𝑏
𝑃𝑠𝑎 = 𝑉𝑠𝑎2 𝑔𝑎 − −
voltage. The equation of DC voltage and AC voltage using the 2(𝑔𝑎 +𝑔𝑏 )
diode rectifier is shown in (1). 𝑉𝑠𝑎 𝑔𝑎 √(𝑉𝑠𝑎 𝑔𝑎 +𝑉𝑠𝑏 𝑔𝑏 )2 −4(𝑔𝑎 +𝑔𝑏 )𝑃𝑣
()
2(𝑔𝑎 +𝑔𝑏 )
- 455 -
After the train power consumption and admittance are set
to a fixed value, the voltage can be variably changed when the
input voltage of both substations is changed to achieve higher
energy utilization.
B. Simulation results
In order to derive changes in power consumption and
regenerative energy according to changes in the tap voltage of
the transformer, a power analysis program PSCAD/EMTDC
was used to simulate the operating conditions of a substation
and a vehicle in the DC 1,500V DC electric railway system. Fig. 3. Power changes with 22.9 kV tab settings
To confirm changes in various voltage ranges, the
transformer tab showed a total of eight levels by changing the
primary voltage from 18.9 kV to 25.9 kV. Figure 2 shows the
result of the resulting power loss. It can be found that the loss
becomes relatively low if the receiving voltage is lowered by
changing the tab.
Fig. 4. Power changes with 23.9 kV tab settings
III. RESULTS
In this paper, the change of the gas efficiency of the urban
iron voltage according to adjust the change of the urban steel
Fig. 2. Simulation result of power loss calculation
according to the change of the change of the urban philosophy
tab adjustment.As a result, it was relatively low loss, the
relatively low loss occurred.When the current 229kV, it was
used to increase the current voltage of the vehicle resistance
If there is a resistance in the railway train, it may further in the vehicle braking time, it was used to increase the
increase the use of regenerative energy according to changes. efficiency of the vehicle braking time.
Figure 3 and 4 shows changes when it is set to 22.9kV and
23.9kV. In the case of the existing 22.9kV system, when REFERENCES
regenerative energy is generated, the catenary voltage rises
higher than the threshold voltage of the resistor in the train, [1] C.M. Yun, G.J. Cho, H.C. Kim, H.S. Jung, A Study on the Train Brake
and the resistor operates. On the other hand, as a result of Position-Based Control Method for Regenerative Inverters. Energies 2022, 15,
changing the tab to 23.9 kV, even if the threshold voltage of 6572
the resistor is the same as that of the existing system, the [2] H.S. Jung, An Optimal Charging and Discharging Scheduling Algorithm of
catenary voltage is lowered and does not operate, so all the Energy Storage System to Save Electricity Pricing Using Reinforcement
Learning in Urban Railway System
generated regenerative energy is used in the regenerative
[3] H.S. Kim, The study on Reduction of Demand Power in Urban Railway
energy utilization facility. using OLTC, The Journal of the Korea Institute of Electronic
Communication Sciences v.11 no.10 , 2016, pp.963 - 968
- 456 -
A Network Architecture for A Substation
Automation Based On IEC 61850
Kyungsoon Park 1 Bekzhan Rysmendeev2 and Jongwoo LEE 1*
1
Seoul National University of Science and Technology, Korea
2
National Company Kyrgiyz Temir Jolu, Kygyzstan
Email: [email protected], [email protected], 1*[email protected]
Abstraction-In virtue of Ethernet technology progress, equipment for active and reactive power control. The third is
SCADA systems for substations become more powerful for measurement functions such as current, voltage, active power
protection, control, monitoring, metering, testing, and so on. and reactive power measurement. The last on is analysis
The various SCADA types supplies in the world market. Each functions which evaluate load factor, load value (%), current
SCADA system consist of various sensors and controllers and value.
different communication protocol for the devices interfacing.
IEC 61850 standard was originally addressing applications and Power SCADA used in railways is also similar to SCADA
communications for the substations. IEC 61850 can provide systems in terms of functions and equipment used in power
comprehensive and accurate information models for various systems. The SCADA of the railway power system follows
components of distribution automation systems, while providing the power generation direction of the SCADA system of the
an efficient solution for this naturally multi-vendor power systems. Although there are unique installations and
environment. The IEC 61850 standard allows for methods of operation for rail power systems, they are largely
communication between devices within a substation where a similar to existing power systems.
peer-to-peer model for Generic Substation Events (GSE)
services is used for fast and reliable communication between The modern power SCADA system is based on IEC 61850
Intelligent Electronic Devices (IEDs). This poses new and uses standards such as IEC 61970/61968, IEC 60870-101,
requirements on products and systems to ensure timely 60870-104, and communication protocols such as Modbus
handling of real-time data with much higher bandwidth needs and DNP. Based on these standards, power systems are
for proper performance of the substation protection, developing their products and software.
automation and control functions. This paper shows how a given
architecture and devices consumes time to determine whether Railway power systems should be followed in the same
the consuming time meets the required time constraints. direction as the SCADA system based on IEC61850 for
general power system. The use of common components and
SW can lower the operating costs in railway power systems.
Keywords-Digital substation, IEC 61850, fault clearance time,
Therefore, it is necessary to build a SCADA system for the
system architecture railway power system based on the IEC61850 system.
I. INTRODUCTION II. 2. SUBSTATION FUNCTION
Electrical energy which is convenient to transmit has A. General substation functions
played an indispensable role in the development of modern Implicative power system operation is illustrated in Figure
industry. Electrical energy is transmitted to consumers 1.
through several stages. The transmission phase consists of
power generation, transmission and distribution phases. Each
stage has substations, which are a power transformation
facility.
In order to supply stable power, substations must be
maintained in good condition, and in the event of a failure or
accident, appropriate measures must be taken promptly to
recover from the failure and prevent the spread of the accident.
SCADA (Supervisory Control and Data Acquisition)
system has been introduced and operated to maintain the
soundness of substations and power facilities.
The main functions of the power SCADA system can be
divided into 4 functions. The first one is power equipment
monitoring which survey monitor circuit breaker, transformer, A substation is a part of power generation, transmission
etc. The second is controlling that manipulate such as circuit and distribution. Its main function is to step up and step down
breaker, transformer tap control, STACON and Shunt Reactor voltage, and performs various functions such as other
- 457 -
interruptions, voltage regulation. [1][2] In point of view, the
substation functions classify into three categories. The first is
a defensive function that monitors health and load. The second
includes control functions such as voltage regulation, reactive
power control, and switch operation. The third function is the
operation functions.
• The states of power systems are allocated as follows:
normal, alarm, emergency, extreme emergency, and
multiple folded states. [3]
• Normal state: all system variables are within the
configured range
• Alarm state: when the variable values deviate from the
set range due to disturbance. In this case, emergency
power generation or the like can be performed.
• Emergency state: As a serious disturbance, it is
necessary to isolate the failure area, stop power
generation, and reduce the load.
• Extreme-emergency state: A large-scale blackout may
occur in the form of sequential generator step-out. In this
case, load shedding and system separation lights are
performed.
Fig. 3. An Architecture of SCADA Systems
• Recovery State: Recovery state is under that all systems
are reconnected and system load is restored. IEC 61850 provides communication protocols for
intelligent electronic devices at electrical substations. The
abstract data models defined in IEC
61850 can be mapped to a number of protocols. Current
mappings in the
standard are to Manufacturing Message Specification
(MMS), GOOSE (Generic Object Oriented System Event)
and SV (Sampled Values) or SMV (Sampled Measure Values).
III. IEC 61850 BASED SUBSTATION AUTOMATION
A. Substation Operation
Substation operation generally includes state surveillance,
load surveillance, voltage control, reactive power control,
Fig. 2. State Transition Diagram switching, switch handling and analysis. The actions require
different time response. A SCADA system based on
B. SCADA systems IEC61850 is illustrated in Figure 4.
SCADA systems for Substation Automation System has a
lot of different architecture and consists of different type of IEC61850 defined time constrain in the Generic Object
sensors, but they work for the above mentioned functions. Oriented Substation Event(GOOSE) being designated in the
IEC61850 part 7-1 as shown in Table 1. The GOOSE
IEC 61850 is a standard for communication networks and transmission period is determined by minimum and maximum
systems used in the automation of power generation, retransmission period. [4]
transmission, and distribution systems. It was developed by
the International Electrotechnical Commission(IEC) to This message is classified within type1 performance that
provide a common framework for communication between means high-speed message to deliver status and event changes.
devices in the substation automation domain. The standard Among these classified messages are type 1-A which is
defines a set of protocols, data models, and services that mission critical that require less than four milliseconds ETE
enable interoperability between devices from different delay. The built-in reliability for GOOSE come from
vendors, making it easier to integrate and manage complex retransmission of message frame until end of its life according
power systems. to predefined field named Time Allowed To Live (TATL
property).
- 458 -
information transmission model, where a continuous flow of
data is maintained among LNs to increase communication
reliability. Each message in a GOOSE transmission sequence
has an attribute called Time allowed To Live (TTL) that
informs the receiver about the maximum time to wait for the
next transmission.
C. Time budget analysis
A basic scheme of the time budget available from a fault
inception in a power system until the fault is cleared
physically is given in [7].
Figure 6 shows the time budget, which is composed of
several artefacts. Main components are the fault recognition
by the protection equipment. If differential applications are
involved – which are not scope of this paper – then time for
transmitting and receiving information from a remote location
needs to be taken into account as well. From the instant a fault
is detected until the physical outputs of a protection relay
operate is called the relay decision time. In this time budget
we also considered the time for an auxiliary tripping relay,
which is typically used today and is located between the IED
and the trip coil of the circuit breaker. The final part is the
Fig. 4. An Example of SCADA System based on IEC61850 operating time of the circuit breaker until the arc is
extinguished.
TABLE I. TIME CONSTRAINTS FOR IEC 61850 PROCESS BUS
MESSAGES[5]
Fig. 6. Protection Communication System
We need to estimate time consumption in a proposed
SCADA system based on IEC61850 whether it satisfies or not.
IV. IEC61850 PERFORMANCE PROPERTIES
A. Protection Model
B. Communication Protocols
Protection system for substations comes up with many
The IEC standards group provided a standard (IEC 870-5- different models. An proposed protection architecture in
104) that outlines how to encapsulate the IEC 870-5-101 over substations shows in Figure 7. The protection devices which
Ethernet. Hence that, the IEC 61850 standard allowed using are IED, CIT and MU are connected with L2 switch.
of Ethernet technologies for GOOSE, SV and MMS
communications [6].
Fig. 5. Protocol suite and functional diagram
Fig. 7. An Example of Protection System
The SV protocol works on a periodic information
The protection scenario is measuring current, abnormal
transmission model, regularly sending messages at a fixed rate.
state detection and current cutoff using proposed devices. The
For protection purposes, the default rate is 4000 or 4800
information flow is shown in Figure 8.
messages per second for, respectively, 50 or 60 Hz power
systems. The GOOSE protocol operates in a sporadic
- 459 -
https://www.watelectrical.com/electrical-substation-components/
[2] AEMO,“Power System Requirement”, https://www.aemo.com.au
[3] Guilherme Moutinho Ribeiro, et al., “ Expert System Substation
Operation and Control ” ,
https://www.researchgate.net/publication/301627481
[4] Salman Mohagheghi, J. Stoupis, “Applications of IEC 61850 in
distribution automation”, DOI: 10.1109/PSCE.2011.5772491
[5] Substation Automation LAN-CISCO
[6] IEEE PSRC H6 Special Report, Application Consideration of IEC
61850/UCA 2 for Substation Ethernet Local Area Network
Communication for Protection and Control, IEEE WGH6, 5-05-2005.
[7] Stefan Meier, Thomas Werner, OP060 –
Performance considerations in digital substation
applications, library.e.abb.com
Fig. 8. Data flow for current cutoff
B. Protection System Time Consumption
A particular interest in this setup is a fault clearance time,
which is the time from analog data captured by sensors,
transferred to the protection IED where the analog quantities
are analyzed and produce a GOOSE trip information is sent to
the breaker IED, the circuit breaker opens and clear faults. [6]
The time consumption of the subsystem for the protection is
shown in Table 2.
TABLE II. DEVICES TIME CONSUMPTION
The time consumption can decrease through using rapid
response systems and by changing system architecture. To
meet a requirement for time requirement, it also needs to
consider data size and communication methods such as VLAN
implementation.
V. CONCLUSION
We reviewed IEC61850 standard to automatize
substations. IEC61850 provides a lot of object oriented classes
to implement in IED devices having logical devices and
logical nodes. The IED devices can work stand-alone if well-
organized logical devices and logical nodes would be
provided in the IED devices.
An architecture for fault clearance is required in order to
fulfill the time consumption, and more specifically if
performance should not be degraded where network delays are
irrelevant due to the fact that devices incorporate all
functionality, from data acquisition to issuing trip commands.
Techniques such as high-performance computing and
computational intelligence will enable these systems to
behave intelligently, handling more devices with lower
response time and with the ability of making failure
predictions in the system components.
REFERENCE
[1] Dave, “Electrical Substation Components & their Workings”,
- 460 -
A Study on the LCC Analysis for High-Speed
Train On-Board Signaling Systems
Jungjae LEE 1 , Meruyert Zhumakhanova2, and Jongwoo LEE 1*
1
Seoul National University of Science and Technology, Korea
2
JSC National Company Kazakhstan Temir, Kazakhstan
Email: [email protected], [email protected], 1*
[email protected]
Abstract - LCC analysis was used to evaluate the economic documentation and result review, life cycle cost analysis
aspects of the high-speed train system, but previous research secretion, and implementation and monitoring of life cycle
neglected the on-board signaling systems. This study fills that cost analysis.
research gap by conducting a detailed LCC analysis for on- The IEC 60300 standard outlines a structured approach to
board signaling systems and estimating a total LCC cost of
performing life cycle cost analysis. This method consists of
14,508,137,795 won. The findings highlight the importance of
minimizing operating losses to reduce overall life cycle costs and five main steps: setting the organizational context, planning
provide valuable insights for future research and evaluation of the analysis, defining the analytical approach, performing the
LCC during system introduction. analysis, and completing the analysis. The application of this
method can vary depending on the industry and project being
Keywords—LCC, High-Speed Train, On-Board Signaling, analyzed.
IEC 60300 The first step in the IEC 60300 LCC analysis method is to
establish the organizational context. This involves
I. INTRODUCTION
considering internal and external factors that impact the
The background of the study highlights the significance of organization, such as population growth, demographic
high-speed railways in South Korea, particularly the KTX information, levels of automation, and obsolescence.
system, which has experienced continuous growth and Alternative scenarios should be identified to assess their cost
expansion. The study emphasizes the importance of the train implications.
control system, including the signaling system, in ensuring the Planning the analysis involves defining the scope of the
safety of railway vehicles and passengers. Currently, there is
study, including the item under analysis, life cycle duration,
a high reliance on foreign technology for train control systems,
leading to high costs. To reduce this dependency and save operating environment, and maintenance considerations. The
costs, the Korea Railroad Corporation (KORAIL) is purpose of the analysis should be clearly defined, and the
developing the Korean Train Control System (KTCS). necessary resources, personnel, and methods should be
However, there has been limited research on the cost- determined. Conditions, assumptions, limitations, and
effectiveness of the signaling system compared to other constraints should also be identified and documented.
components of the high-speed railway system. Prior to performing the analysis, the rules and methodology
to be used must be defined. Factors to consider include
The scope of the study focuses on the ETCS L1/High- whether an absolute or comparative analysis will be
Speed ATC/ATS signaling device used in high-speed railways, conducted, the level of accuracy required, available data, and
and the analysis period is set for the 30-year service life of
available resources. A suitable LCC model should be selected
high-speed railway vehicles. The study aims to analyze the
cost factors of the train control system, establish a detailed or developed to achieve the analysis purpose.
cost structure for each life cycle phase, and develop an The analysis phase involves estimating the cost elements
analysis model based on the cost structure to conduct a life based on available data. Parametric equations or similar cost
cycle cost (LCC) analysis of the signaling device. estimation methods can be used when data is scarce. Cost
data should be classified and applied to the detailed cost
structure, allowing for the calculation of the total life cycle
II. LCC ANALYSIS PROCESS cost. Sensitivity analysis should also be conducted to assess
the robustness of the model and the impact of uncertainties.
LCC (Life Cycle Cost) is a method used to compare and The accuracy and integrity of the analysis results should be
evaluate costs over an extended period, encompassing reviewed and evaluated. This includes assessing whether the
acquisition costs, operation costs, maintenance costs, and purpose and scope of the analysis were properly addressed,
disposal costs. It provides objective data for assessing the adequacy of the model and data used, and whether the
investment responsibility, cost effectiveness, and return on results align with the original goals. If necessary,
investment. LCC analysis is commonly employed during the modifications or additional alternatives should be considered.
project preparation stage for cost evaluation and financial All analysis results should be thoroughly documented and
management. provided to stakeholders and decision-makers. The
The process of analyzing life cycle costs involves several documentation should include supporting evidence for the
steps. These steps, as per the life cycle cost guideline, include results, and any follow-up measures or recommendations
plan analysis, model selection and development, resulting from the analysis should be outlined.
- 461 -
When analyzing the life cycle cost of a train control system, ② Focus on the ETCS L1/High-Speed ATC/ATS train
it is crucial to recognize that the cost incurred in each life protection systems used in high-speed railways,
cycle stage may vary depending on the specific system. considering the system's service life and the lifespan of
Therefore, a detailed analysis of the cost generation specific high-speed railway vehicles.
to the system is essential to accurately assess the life cycle ③ Establish a detailed cost structure for the train protection
cost. system.
④ Use parametric estimation, analogous cost estimation,
III. TRAIN CONTROL SYSTEM LCC ANALYSIS and engineering cost estimation methods classified by
A. Signaling system LCC analysis cost factors.
Train control systems, including ATS, ATP, and ATC, are ⑤ Use estimated data for undisclosed or confidential
essential for safe train operations. These systems can vary in information, providing justification for the estimates.
configuration based on the railway system, but their main ⑥ Conduct LCC analysis for the additional introduction
purpose is to ensure train safety by detecting train positions, scenario of the train protection system.
transmitting track information, and controlling train speeds.
ATO and CBTC functionalities are often integrated into these 3) Definition of Analysis Methods
systems. Railway systems require continuous improvement, a) Establishment of Rules and Methodologies:
new construction, and sometimes replacement due to factors ① This study focuses solely on the life cycle costs of the
like rolling stock changes and technological advancements. ETCS L1/High-Speed ATC/ATS train protection system
Train control systems have limited application and involve a for high-speed railways and does not compare it with
small number of authorized operators and manufacturers. other train protection systems.
Changes in the railway system often require corresponding
modifications to the train control systems. Railway operators ② Both publicly available data and estimated data will be
utilized in this study.
need to acquire train control systems in four cases: new
implementation, additional implementation, replacement due ③ In case of overlapping cost factors, they will be
to the end of service life, and complete replacement due to distinguished and allocated according to the respective
changes in operational scenarios. cost structures.
The detailed cost structure for each scenario has been
b) Selection of LCC Model:
analyzed and the results are presented in TABLE 1.
LCC estimation methods can be classified into two
approaches: deterministic life cycle cost analysis and
B. On-board signaling device LCC analysis for high-speed probabilistic life cycle cost analysis. Deterministic analysis
rail relies on deterministic data obtained from statistical sources
1) Situation Setting or subjective opinions of experts. Probabilistic analysis, on the
The utilization of high-speed railway vehicles, such as other hand, converts uncertain data into probabilistic
KTX and KTX-Sancheon, is increasing in South Korea. The characteristics, using statistical data or expert opinions. Since
high-speed railway industry continues to grow with the the cost factors and data for the existing train protection
introduction of new high-speed railway vehicles like KTX- system in high-speed railways are relatively determinable, we
EMU260 and the development of KTX-EMU320, a will use the deterministic life cycle cost analysis approach.
distributed power high-speed railway vehicle. In this study, The LCC can be expressed as follows:
we aim to conduct a life cycle cost (LCC) analysis of the train LCC = Cost of Acquisition + Cost of Operation and
control system, specifically the automatic train protection Management + Cost of Maintenance + Cost of Disposal
system.
c) Definition of Cost Structure
2) Analysis Plan
If we establish the detailed cost structure of the train
a) Scope and Objectives : protection system, we can divide it into 65 cost elements.
LCC analysis has been widely applied in various areas of Following the approach applied in the train control system, we
the railway industry, but the analysis of life cycle costs can categorize these elements into four scenarios: new
specifically for the train protection system, especially the introduction, additional introduction, replacement based on
automatic train protection system, has not been adequately the lifespan, and full replacement. TABLE 2. presents the cost
discussed. In this study, we intend to establish a cost breakdown for the additional introduction scenario, which is
classification system for the acquisition, operation, the focus of this study based on the defined scope of analysis.
maintenance, and disposal costs of the train protection system
used in the existing KTX high-speed railway vehicles. We will TABLE I. ON-BOARD SIGNALING SYSTEM COST BREAKDOWN
focus on the cost analysis of the additional introduction STRUCTURE BY CASE
scenario among the four identified scenarios. New
Additiona
Replacem
l Replacem
Cost Factors Implemen ent of
b) Definition of Analysis Tasks tation
Implemen ent of Life
tation
Scenario
To analyze the life cycle costs of the train protection 1 Basic plan establishment cost O △ X △
system for high-speed railways, the following tasks will be 2 Consulting fees O △ X △
3 Feasibility study execution cost O △ X △
carried out: 4 Scenario design cost O X X △
① Analyze the total life cycle cost of the train protection 5 Hardware design cost O △ △ △
6 Software design cost O △ △ △
system for high-speed railways.
- 462 -
7 Mechanism design cost O △ △ △ 5
Sales loss cost O O O O
8
8 Prototype production cost O △ △ △
5
9 Product testing cost O △ △ O Train delay compensation O O O O
9
1 6
Environmental testing cost O △ △ O Disposal permit cost O O O O
0 0
1 6
Certification consulting fees O △ △ O Dismantling cost O O O O
1 1
1 6
SIL certification cost O △ △ O Waste disposal cost O O O O
2 2
1 6
Formal approval cost O △ △ O Environmental conservation cost O O O O
3 3
1 6
Major device production cost O O O O Restoration cost O O O O
4 4
1 6
Auxiliary device production cost O O O O Disposed asset loss cost O O O O
5 5
1 [ O : Essential / △ : By Case / X : Non Essential ]
Spare parts cost O O O O
6
1
Quality inspection cost O O △ O
7
1 IV. ANALYSIS IMPLEMENTATION
Inspection agency inspection cost O O △ O
8
1
Major device installation cost O O O O A. Establishing Methods for Cost Estimation
9
2
Auxiliary device installation cost O O O O In this study, cost elements are estimated using parametric
0
2 estimation, analogous cost estimation, and engineering cost
Transportation cost O O O O
1 estimation. Parametric equations are primarily used to
2
2
Administrative expenses O O △ O estimate labor-related costs among various cost elements.
2
Completion document production cost O O △ O They are applied when labor costs are included in other cost
3
2 elements such as production or installation costs, or when
Other miscellaneous expenses O O △ O
4 there is no clear criterion for estimation. Analogous cost
2
5
Education expenses O O △ O estimation is used when actual data that is not publicly
2
6
Training expenses O O △ O available or inaccessible, such as trade secrets, cannot be
2
Operations personnel wages O O O O
obtained, but cost elements such as labor costs or
7
2
improvement costs can be estimated based on standardized
8
Management personnel wages O O O O norms, related studies, or reports. Engineering cost estimation
2
9
Outsourced labor service fees O O O O is applied to cost elements for which actual data is available.
3
0
Building usage cost O O O O B. Data Collection for Cost Estimation
3
1
Facility usage cost O O O O
1) The collected data for each cost element is presented
3
2
Safety management cost O O O O in TABLE 2.
3
Environmental improvement cost O O O O
3
3
TABLE II. TOTAL COST OF EACH CATEGORY
Data management cost O O O O
4
3 Main Occurrence
Energy cost O O O O Subcategory Subdivision Detailed Item Cost Element Cost
5 Category Time
3 Development Certification Certification
Communication cost O O O O Design cost 27,453,476 acquisition
6 cost fees consulting fees
3 Major device
Water cost O O O O Major device
7 production 394,000,000 acquisition
production cost
3 cost
Maintenance personnel wages O O O O
8 Direct Auxiliary
3 Production device Auxiliary device
Management personnel wages O O O O 134,000,000 acquisition
9 cost production production cost
4 cost
Replacement work outsourcing cost O O O O
0 Acqusition Spare parts
Spare parts cost 73,707,000 acquisition
4 cost Production cost
Repair work outsourcing cost O O O O
1 cost Quality inspection
4 1,000,000,000 acquisition
Environmental improvement outsourcing cost O O O O Inspection cost
2
cost Inspection agency
4 Indirect 2,258,876 acquisition
Building usage cost O O O O inspection cost
3 Production
4 cost Major device
Facility usage cost O O O O acquisition
4 Installation installation cost
49,453,496
4 cost Auxiliary device
Safety management cost O O O O acquisition
5 installation cost
4 Operations
Environmental improvement cost O O O O Operations
6 personnel 3,839,705,100 annual
4 personnel wages
Data management cost O O O O wages
7 O&M cost Wages Wages
Management
4 Management
Building repair cost O O O O personnel 383,970,510 annual
8 personnel wages
wages
4 Maintenance
Facility repair cost O O O O Maintenance
9 personnel 19,359,110 annual
5 personnel wages
System improvement cost △ △ △ O wages
0 Management
5 Management
Energy cost O O O O personnel 1,935,911 annual
1 Wages Wages personnel wages
wages
5 Replacement work
Communication cost O O O O
2 Outsourced outsourcing cost 31,023,446 replace
5 Maintenance Labor serivece
Water cost O O O O cost fee Repair work
3 7,248,205 repair
outsourcing cost
5
Replacement part cost O O O O Replacement Replacement part
4 416,024,000 replace
5 part cost cost
Repair cost O O O O Part cost Part cost
5
Repair cost Repair cost 31,201,800 repair
5
Education expenses O O O O
6 Education Education Education Education
5 1,909,433 acquisition
Training expenses O O O O expenses expenses expenses expenses
7
- 463 -
Training
Training expenses 62,854,251 acquisition Delay cost Delay cost Sales loss cost Sales loss cost 2,754,087,477 delayed
expenses
Dismantling Dismantling
Delay cost Delay cost Sales loss cost Sales loss cost 1,311,102,096 delayed 38,705,649 disposal
cost cost
Dismantling Disposal Disposal Waste Waste
Dismantling cost 20,000,000 disposal Disposal cost disposal
cost cost cost disposal cost disposal cost
Waste Waste disposal EnvironmentalEnvironmental 1,935,282
Disposal Disposal disposal
Disposal cost disposal cost cost conservation conservation disposal
cost cost
Environmental 1,000,000 cost cost
Environmental
conservation disposal Total Cost 14,508,137,795
conservation cost
cost
[ CURRENCY : KRW ]
Total Cost 7,808,206,710
[ CURRENCY : KRW ]
V. ANALYSIS REVIEW
C. Present Value of LCC Analysis Upon reviewing the life cycle cost analysis of the ETCS
LCC analysis, due to its relatively long duration, cannot L1/High-Speed ATC/ATS train protection system for high-
simply be represented by the sum of monetary amounts. speed railways, the following points are observed:
Therefore, to evaluate the costs of alternative options, a
concept of time value needs to be applied to assess them in
terms of present value. In this study, the present value method ① Detailed cost structure for each life cycle phase of the
will be used among various economic theories, and the high-speed train protection system has been established.
procedure is as follows:
② Detailed cost structure for different implementation
PV (Present Value) = initial cost + recurring cost + cases of the high-speed train protection system has been
nonrecurring cost developed.
TABLE 3. presents the total costs derived from TABLE 2., ③ c. Determination of inclusion/exclusion of cost analysis
converted into present value. for each cost element during the simple replacement of
the train protection system based on its durability.
TABLE III. TOTAL COST OF EACH CATEGORY BY PRESENT VALUE
Main Occurrence ④ Collection and estimation of cost data for analysis
Subcategory Subdivision Detailed Item Cost Element Cost
Category Time purposes.
Certification
Development Certification
cost
Design cost
fees
consulting
fees 27,453,476
acquisition
⑤ Calculation of costs for each cost element and total costs.
Major device Major device
production production
394,000,000
acquisition ⑥ Calculation of the present value of total costs.
cost cost
Direct Auxiliary Auxiliary
Production device device
acquisition
cost production production 134,000,000
cost cost A. Evaluation of Achievement of Analysis Objectives
Spare parts Spare parts
cost cost 73,707,000
acquisition Through the life cycle cost analysis of the ETCS L1/High-
Acqusition Quality Speed ATC/ATS train protection system, the total costs per
cost inspection acquisition
Production
cost cost 1,000,000,000 formation due to the simple replacement of the train protection
Inspection
cost
Inspection system upon reaching its durability have been calculated and
agency
inspection 2,258,876
acquisition converted into present value. Thus, it can be evaluated that the
Indirect
Production
cost analysis objectives have been achieved.
Major device
cost
installation acquisition
cost However, there are limitations in evaluating cost elements
Installation
cost
Auxiliary
49,453,496 with duplications or those with difficult estimation, as well as
device
installation
acquisition not covering all aspects related to various implementation
cost cases, which is a point of regret.
Operations Operations
personnel personnel 8,159,678,951 annual
wages wages
O&M cost Wages Wages
Management Management
personnel
wages
personnel
wages
815,967,895 annual REFERENCES
Maintenance Maintenance [1] John Vail Farr, Isaac Faber. (2019). “Engineering Economics of Life
personnel personnel annual
wages wages 41,139,650 Cycle Cost Analysis”, CRC Press
Management Management [2] B.S. Dhillon. (2010). “Life Cycle Costing for Engineers”, CRC Press
personnel personnel annual
wages wages 4,113,965 [3] Ooi Chu Hui, Abdul Hakim Mohammed. (2015). “The Role of Cost
Wages Wages Replacement Breakdown Structure in Life Cycle Cost Model”, Jurnal Teknologi,
work May 2015
60,039,130 replace
Outsourced outsourcing
Labor serivece cost [4] Megan Davis, Regina Coony, Scott Gould, and Allan Daly. (2005).
Maintenance fee Repair work “Guidelines for Life Cycle Cost Analysis”, Stanford University Land
cost outsourcing 15,403,013 repair
cost and Buildings, October 2005[5] IEC. “IEC 60300-3-3”, IEC 2017
Replacement Replacement
805,123,943 replace [5] Korea Development Institute. “2021 Preliminary Feasibility Study
part cost part cost Report Detailed Guidelines for Conducting Preliminary Feasibility
Part cost Part cost
Repair cost Repair cost 66,306,308 repair Study Research on Roads and Railways”, KDI Public Investment
Management Center, 2021
Education Education
1,909,433 acquisition
Education Education expenses expenses [6] Korea Railroad Research Institute. “Type Approval, Producer
expenses expenses Training Training
62,854,251 acquisition
Approval, and Completion Inspection Fees for Railroad Cars”, Korea
expenses expenses Railroad Research Institute, 2016
- 464 -
[7] Jeong Chang-mo. (2020). “Analysis of railway electrification cost
using LCC technique”, unpublished master’s thesis, Seoul National
University of Science and Technology, Seoul, pp. 10 to 13
[8] Daeseong Kwon. (2020). “Study on Improvement of Urban Rail
Vehicle PCB Reliability Using FMECA”, unpublished master's thesis,
Seoul National University of Science and Technology, Seoul, pp. 51
[9] Ming Chuin Teng. (2013). “Life Cycle Cost for Modification Project”,
Master’s Thesis, University of Stavanger, Stavanger, Norway, pp. 4 to
5
[10] Seongsu Han. (2011). “A study on the proper manpower management
of urban railway drivers”, unpublished master’s thesis, Woosong
University, Daejeon, pp. 29-30
[11] Kim Seok-ho. (2016). “Reliability-based light rail vehicle LCC
analysis plan”, unpublished master’s thesis, Korea National University
of Transportation, Chungju, pp. 10 to 14
[12] Jeongjin Jang. (2019). “Estimation of Life Expectancy through
Reliability Analysis of Train Control PCB”, unpublished master’s
thesis, Seoul National University of Science and Technology, Seoul,
pp. 30
[13] Korea Railroad Corporation. “Rules for maintenance and repair of
rolling stock”, Korea Railroad Corporation, 2020
[14] Korea Railroad Vehicle Engineering. “Railroad Vehicle Complete
Vehicle Inspection Fee”, Korea Railroad Vehicle Engineering, 2016
[15] Seongsik Kim. (2021). “A Study on Risk Analysis of On-board
Signaling System (ATC) of Railroad Vehicles on Line 3”, unpublished
master’s thesis, Seoul National University of Science and Technology,
Seoul, pp. 51
[16] Korea Railroad Corporation. “Guidelines for appropriation of vehicle
replacement investment in railway business”, Korea Railroad
Corporation, 2014
[17] Korea Railroad Corporation. “Fixed Asset Accounting Guidelines
(Fixed Asset Classification Items and Useful Life)”, Korea Railroad
Corporation, 2012
[18] Korea Railroad Corporation. “2021 Korea Railroad Corporation
Financial Statement”, Korea Railroad Corporation, 2021, pp. 88
[19] Ministry of Strategy and Finance. “Public institution management
information disclosure system (www.alio.go.kr)”, Ministry of Strategy
and Finance, 2021
- 465 -
MCB Interrupting and Sequential Energizing of
EMU in DC and AC Transition Section
Dalseo PARK 1, Hyosang KIM1, and Jaeseok SHIM 1*
1
Seoul National University of Science and Technology, Korea
Email: [email protected], [email protected], [email protected]
Abstract-SeoulMetro and Korail commonly operate Line 1 and L4.
The lines have different traction power supply system. The power
supply systems of SeoulMetro’s line is DC 1500V, Korail AC
25,000. At the connection location, dead section was implemented
not to flow abnormal current to each different power system. The Fig. 2. SeoulMetro EMU
MTUs are required to switch DC to AC or vice versa. The drivers On the other hand, Korail EMU are configured differently,
or automatic devices convert MTU’s power system to power with 6 pantographs attached to 3 out of 10 cars.
system being supply. The drivers and the automatic sometimes
commit error for dead section passing caused from dead section
contamination, switch mal-manipulation, and harsh environment.
This paper identified the problems and proposed the solution.
Keywords—MTU, dead section, MCB, DC1500V, AC25,000V
Fig. 3. Figure 1 Korail EMU
A pair pantographs of two different type EMUs on a car
I. INTRODUCTION
are electrically connected, currents can flows.
Due to the expansion of metropolitan areas, the urban
railway system is frequently operated in conjunction with II. DEAD SECTION INSULATION
conventional railways, which two systems would be different The dead section insulator is composed of FRP, and
systems, which are different signaling systems or power Korail's standard is that the insulation resistance have to be
supply systems and so on. As the same case, Seoul Metro lines 2000 ㏁ or more per 30cm.[3]
are connected to Korail lines.
The contact strip of the pantograph is manufactured with
When a railway vehicle enters an AC section from a DC copper and carbon as lubricant. The copper and graphite dust
section, the vehicle must enter section 4 with the parking worn by the contact strip adheres and accumulates to the FRP
short-term cut off in the DC section, and turn on the parking catenary system such as contact wire, messenger wire, hangers,
short-term when entering the AC section. Conversely, when and insulators. Finally the FRP catenary insulation shall be
the vehicle enters the DC section from the AC section, the four
deteriorated. The insulation resistance in this contaminated
sections are operated in the same procedure. [1]
section was measured at 1M Ω or less. It means that the
contaminated dead section is no more a perfect insulator. [5]
Fig. 1. DC and AC Transition Section
The dead Section as shown Figure 1 exits in Line 1 and
Line 4 operating by two authorities, one is SeoulMetro and
another one Korail.. The driver manually switches the Fig. 4. An Example of Contaminated FRP
transition switch (ADS) in Line 1, and in Line 4, the switching
carries out automatically by beacon tag.
The MTUs of SeouMetro and Korail commonly pass in
the three transition sections.
MTUs operating in the AC insulation section of Lines 1
and 4 are composed of 10 MTUs as shown in Figure 2, and 10
pantographs are installed in 5 of the 10 MTUs of Seoul
Transportation Corporation, and the minimum distance
between MTUs is 6.5 m as shown in Figure 2. [2]
Fig. 5. Figure 2 An Example of Contaminated FRP Fixed By Metal Jig
- 466 -
III. COUNTERMEASURES AGAINST ENTERING POINT PASSING
AND MCB MALFUNCTION
The distance between pantographs of two consecutive
A. 1 States of FRP Catenary
motorized cars is 6.5m and the distance between pantographs
on the same car is 13.m as shown in Figure 4. The total We checked how much the insulation resistance provokes
distance of three consecutive pantographs becomes 20m. If the problems of entering point passing and MCB malfunction
the resistance of the 6.5m FRP at the beginning of FRP dead section with Physical security information
insulator is deteriorated under the predefined value, the two management(PSIM). [8]
consecutive pantographs become connected state and the car
already in dead location becomes flowing current. The point
which the car is under no current comes to 33.5m as shown The consequences result in as follows :
Figure 6. The MTUs shall get no current state for 2.01s which
is not sufficient time to cut off MCB. The MTUs shall get 1) DC arrester was destructed in the case that FRP
current before current cutoff. An inadequate power supplies to resistance is below 9,278 Ωby flowing AC current.
the MTUs. The MTUs comes to failure. [4][6]
2) MCB malfunction occur in the case that FRP
resistance is below 9,278 Ωby flowing AC current.
3) MTU main fuse burning takes place in the case that
FRP resistance is below 7,378 Ω by flowing DC
current.
4) MCB cutoff appears in the case that FRP resistance
is below 1,870 Ω by flowing AC current. The above
phenomena is summerized in Table
Fig. 6. EMU entering into Dead Section from Energized section
As the MTUs pass the dead section, the evoked problems
classify into two problems. One is the MTUs enters TABLE I. LIMIT VOLTAGE IN CONTAMINATEDD FRP TO WORK
inadequate power systems, another MCB cutoff problem. MCB
In case, the MTUs run the transition location, Proper
switching is not to occur problems.
A. entering into dead section
Either manual switching by the driver or automatic
switching by RFID Tag can be performed immediately before
passing the dead section.
When the MTUs enter into the AC or DC section under
the AC/DC switching fails, the case the vehicle passes from
the DC section to the AC section damages DC arrester, the B. 2. Countermeasure Against Entering Point Passing
another case destroys main fuse caused from overcurrent The first solution is that the MCB cutoff can automatically
flow(400~500A). [7] be realized by lengthening FRP insulator with ensuring
sufficient cutoff time more than 2.05s. In this case, The more
dead section lengthens, the more MTU speed increases to
avoid MTU stopping in the dead section.
B. Exiting from dead section The second one introduces air section in FRP insulator as
The MCB of the MTUs is open during the dead section. shown in Figure Double air insulation structure diagram of
The FRP insulation at the exit point of the dead section is FRP insulation section
deteriorated, current can flow the MTUs.
The above case repeats more than three times, then the
MCB of the motorized car becomes open. The MTUs lose a
part of power. all MCB of the MTU have to be reactivated to
recover normal power as illustrated in Figure 7.
Fig. 8. Air Section Implemented in Dead Section
Fig. 7. EMU exiting into Energized section from Dead Section
- 467 -
When the second pantograph of the first motorized car under harsh railway environment to meet railway
passes the air section, the two pantograph of the consecutive requirements.
car doesn’t flow current. The MCB automatically cuts off
and proctect DC arrester and main fuse melting from fault
current.
The third one implements beacon systems to inform the
MTUs of approaching the dead section. The MTUs cut off REFERENCE
MCB to remove the malfunction. [1] “Seoul Metro Line 1 VVVF Rotem ADV (2001)”, pp.3, pp.7
C. Countermeasure Against MCB malfunction [2] “Seoul Metro Driving Regulations (2014)”, pp.10
[3] ″Seoul Metro Electrical Facility Management Regulations (2009)″,
The air section installs at the point of Exiting from dead pp.18
section to avert the MCB malfunction. The MCB functioning [4] Baek Gwang-seon, Lee Jang-moo, Koo Se-wan. (1997) “Identification
carry out after 3s of the condition that adequate power supplies. of the cause of main circuit breaker (MCB) insufficiency in the
The 3s ensures to run 49.8m and then the normal power is Gwacheon line”, pp.15~16., pp.20~23.
furnished to the MTUs. CT has to be installed in main [5] Songbae Choi. (2016), ″Study according to the wear of each material
transformer to get the 3s to work time relay. of the wet plate of the urban railway MTU″, pp.11~14.
[6] Koo Hee-soo. (2015), ″Teaching transfer switch operation check
circuit″, Journal of the Korean Society of Railways, pp.35
[7] Jeong Jong-deok, Editing. (2008). “A Study on Improvement of Safety
IV. Ⅳ EXPERIMENT for Urban Railroad Vehicles Passing through Insulated Section”, Korea
Railroad Research Institute Urban Railroad Standardization Research
SeoulMetro adopted the beacon, CT, and time relay. The Center, pp.9~10.
example of the CT and time relay installation is illustrated in
[8] Baek Kwang-seon, Kim Myeong-ryong, Kim Jin-hwan (2008).
Figure 9 and 10. “Analysis of MCB Input Phenomenon in Four Sections of Urban Rail
Vehicle Teachers”, Journal of the Korea Electric Association, pp. A
Study on Allocation of Maintenance Cycle for Railway Signal
Facilities”, Journal of the Korean Railroad Association, pp.8~9
Fig. 9. Figure 3 A Diagram For 3s Delay At Exiting Point
Fig. 10. Figure 4 Beacons for MCB Interrupting in Dead Section
As the result, the problems of the Entering Point Passing
and MCB malfunction was effectively eliminated.
V. CONCLUSION
To remove the problems of the Entering Point Passing and
MCB malfunction, the beacon and time relay was installed.
The result is completely satisfied.
But the additional devices can arise another problems
caused from their failure. So it needs to select robust devices
- 468 -
Specification Analysis of Reducer-less Motor for
Railway Vehicle Door
Byung-Min Moon 1, Hyo-Seok Oh 1, Yong-Eun Choi 1, Seung-Pyo Jeon 1, In-Soo Chung 1,
Jae-Moon Kim 1*
1
Korea National University of Transportation, Republic of Korea
Email: [email protected]
Abstract— Railway vehicle doors are of utmost importance
for passenger safety and reliability due to their close proximity TABLE I. MAIN SPECIFICATIONS OF THE ELECTRIC DOOR
to passengers. In this paper, we propose the application of a
Classification Specification
reducer-less motor to enhance the reliability of the doors. We
validate the motor specifications through simulations. Door system Pocket sliding
Operating voltage 70 VDC ~ 125 VDC
Keywords— door, reducer-less, motor specifications 2 × 650mm = 1300mm
Opening width (stroke)
I. INTRODUCTION Opening time 2.5 ± 0.5 sec(2~3sec)
Urban railway vehicles typically have eight doors per car. Closing time 3.0 ± 0.5 sec(2.5~3.5sec)
The doors of railway vehicles are critical devices that ensure DCU + DC Motor
passenger safety during the opening and closing operations at
station stops. If any of the doors experience a malfunction Operating components Emergency release unit
while the vehicle is in operation, it is transported to the Interlock switch (Bypass switch)
maintenance depot for inspection and reopening tests before
resuming service. However, the use of diverse components The DC drive motor used in the system employs a DC
and their limited production quantities often lead to supply permanent magnet electric motor. It is installed on the base
issues, making it challenging to address failures promptly and plate and connected to the drive screw through a coupling and
effectively.[1] a reducer. The drive motor is equipped with an encoder that
In this paper, we propose a method to remove the reducer detects the motor speed, and it controls the motor speed by
of the motor to minimize the failure range of the door. To adjusting the pulse width modulation (PWM) waveform of
analyze the motor specification with the motor reducer the input voltage. [4]
removed, the operation sequence of the door was
implemented through simulation. Then, the motor data was
analyzed to consider the appropriateness of the motor
specification.
II. STRUCTURE AND OPERATIONAL CHARACTERISTICS OF
ELECTRIC DOOR
Most door systems used in urban railway vehicles are
pocket sliding doors. These doors communicate component
information to the driver through the Train Control &
Monitoring System (TCMS) and are controlled by the Door
Control Unit (DCU) based on commands received from the
driver. Furthermore, safety features such as obstacle
detection and emergency manual opening/closing devices are
implemented to ensure passenger safety. The opening and
closing speed can be adjusted by modifying the parameters of
the DCU [2].
Table I presents the key specifications of an electric door Fig. 1. DC motor status data [4]
system, consisting of the drive mechanism, known as the door
engine, and the door panel. The door engine enables the Figure 1 depicts the key status data (speed, current, distance)
movement of the door panels by receiving a DC voltage of of the DC motor during the closing process. The door operation
100[V] and opening/closing signals. The rotational motion of is determined by the limit switches (DLS, DCS) when the door
the motor is converted into linear motion through a screw and is fully closed. The motor speed is controlled through
nut mechanism, resulting in the lateral opening and closing of acceleration, constant speed, and deceleration phases. During
the doors. The door panel is composed of stainless-steel acceleration, the motor current increases, while it stabilizes
inner/outer panels, an aluminum honeycomb core, a stainless- during the constant speed phase. Analyzing the motor's driving
steel frame, and a liquid adhesive. Typically, each panel has current without the reducer allows us to assess the
a width of 650[mm]. appropriateness of the motor specifications.
- 469 -
III. REDUCER-LESS MOTOR SIMULATION
Table II shows the specifications of a conventional motor
and a motor with the reducer removed. The configuration of Door Door Door Door
a motor generally consists of a motor, reducer, coupler, and Opening Open Closing Close
spindle gear, and the rated power of the motor determines
whether the motor has a reducer or not. High-speed motors 650[mm]
that use a reducer are controlled by matching the
deceleration/gear ratio, so they are error-prone structures. In
this paper, a low-speed motor without the reducer was 0[mm] 0[mm]
implemented to reduce the scope of door malfunctions, and
simulations were conducted to verify the suitability of the Max : 1.2 [A]
motor specifications prior to implementation.
TABLE II. SPECIFICATIONS FOR EXISTING MOTOR AND REDUCER-
LESS MOTOR.
As-Is
Fig. 3. Door simulation waveform
Figure 3 shows the operational sequence of the door and
presents simulation waveforms for the motor without the
reducer. The door's travel distance is based on one door panel
(650mm) and calculated using the lead pitch of the screw.
The lead pitch represents the door's travel distance per one
Configuration DC Motor + Reducer + Coupler + Spindle Gear. revolution of the motor and is typically designed and applied
Rated Power 100[W], 100[V] within the range of 36-54[mm]. Therefore, reaching a travel
3,150[RPM] distance of 650[mm] indicates full opening, while 0[mm]
Rated Speed
signifies door closure. The motor current exhibits a peak
To-Be value during the initial acceleration phase of the door opening
and decreases during the deceleration phase. The maximum
current recorded is approximately 1.2[A], indicating that the
maximum power consumption of the motor is below 120[W].
Based on this analysis, it can be concluded that the
specifications of the motor are suitable for the application of
the door.
Configuration DC Motor + Coupler + Spindle Gear IV. CONCLUSION
Rated Power 150[W] , 100[V]
In this paper, the application of a motor with a reducer
Rated Speed 1,000[RPM] removed is proposed to reduce the door failure range. In
addition, by analyzing the structure and characteristics of the
Figure 2 shows the control system of the door, which is door, the door operation sequence was implemented through
designed to take all hazards into account for the safety of simulation. Through the door driving simulation, the motor
passengers and vehicles. The operation of the door is specifications were applied and the current data was analyzed
performed upon receiving the open/close signal from the to verify the adequacy of the motor specifications.
vehicle. The DCU receives the signal and, based on factors ACKNOWLEDGMENT
such as vehicle speed, door operability, door blocking status,
and emergency manual opening device status, verifies the This research is supported by the Korea Agency for
safety of the door operation before determining the opening Infrastructure Technology Advancement (KAIA) by the
or closing action. Additionally, the DCU controls the speed Ministry of Land, Infrastructure and Transport under (Grant
and torque of the DC motor through a PWM converter. RS-2022-KA165903)
REFERENCES
[1] B. M. Moon, Y. E. Choi, H. S. Oh, and J.M. Kim, "Specification
Analysis of Electric Door Systems for Urban Railway Vehicles," The
Korean Institute of Electrical Engineers Conference Proceedings,
pp. 242-243, 2022.
[2] S. H. Kim, "A Study on Failures Prediction of EMU Plug-in Type Door
using Moving Average Method," Ph.D. dissertation, Graduate School
of Transportation, Korea National University of Transportation, pp. 34-
53, 2020.
[3] Korea Railroad Corporation, "General Electric Multiple Unit for
Gwacheon-Ansan Line," 2020.
[4] I.K. Kim, " A Study on the Evaluation of Conditions Using Electric
Door Data for Electric Vehicles," Master's thesis, Graduate School of
Fig. 2. DOOR CONTROL SYSTEM Transportation, Korea National University of Transportation, 2020.
- 470 -
Analysis of Platform Screen Door Using
Reliability Block Diagram Analysis Tool
Jong Nam Kim 1, Tae Gil Ha 1, Guk-Hyun Son 1, and Jeong Won Kang 1,2*
1
Department of SMART Railway System, Korea National University of Transportation, Republic of Korea
2
Department of Transportation System Engineering, Korea National University of Transportation, Republic of Korea
Email: [email protected] (JWK)
Abstract— The purpose of railway construction in a railway
project is to provide safe and convenient railway service for
passengers. In this study, we predicted the RAM target value
using a software tool to operate a railway system without failure
and degradation after construction. RAM prediction verifies the
RAM target before equipment installation. Later, a railway
operator validates the RAM target during the operating stage as
a key performance indicator (KPI). A railway system in the
project stage which is centenary, transmittal communicator
system, and railway control system. This paper, a validated
railway platform door (PSD) only RAM prediction and to
achieve international criteria, which can call the automatic door
on a platform when a train departs or arrives. The process of
RDB modelling inputs the device quantity and failure value in a
software tool and makes the RDB with serial, parallel, and
multiple and used a graphic user interface. These outputs of
modeling can be used to judge whether a design achieves the
requirements and RAM target. Meanwhile, we change the link
between devices and add device redundancy if we cannot
achieve the RAM target. In addition, we have to exchange the
different device if we still cannot meet the requirement. The
RAM target value of PSD was 99.932%, which complies
international standards. This prediction of the RAM in the
design stage is to reduce the installation and to keep the project
period.
Keywords—Railway System, RAM Target, RBD(Reliability
Block Diagram), Reliability, Availability
ACKNOWLEDGMENT
This work was also supported by the Korea Agency for
Infrastructure Technology Advancement (KAIA) grant
funded by the Ministry of Land, Infrastructure and Transport
(Grant No. 22QPWO-C158106-04).
- 471 -
A Study on High-Power IGBT Gate Drivers for
Railway Vehicle
Seung-Pyo Jeon 1, Hyo-Seok Oh 1, Yong-Eun Choi 1, Byung-Min Moon 1, Jeong-Won Kang 1
Jae-Moon Kim 1*
1
Department of Transportation System Engineering, Korea National University of Transportation, Republic of Korea
Email: [email protected]
Abstract— Gate drivers in railway vehicles commonly rely Fig. 1 depicts the gate driver consisting of an isolated
on self-developed products or commercially available cores, power supply, desaturation detection circuit, protection circuit,
leading to limitations in circuit configuration and tuning, posing and gate resistor. The basic operating principle of the IGBT
challenges in supply and technical application. This paper gate drive circuit can be divided into two stages: providing a
proposes a high-power IGBT that addresses these limitations,
offering compatibility with existing railway vehicle power
voltage above the threshold voltage to the gate for turn-on and
converters and improved reliability. applying a voltage below the threshold voltage to the gate for
turn-off. Typically, a voltage of 15V or higher is applied
Keywords— Gate driver, Railway vehicle, High-power IGBT during turn-on, while a negative voltage is applied during
turn-off to ensure proper blocking operation.
I. INTRODUCTION
IGBT is an important device that has high-speed switching
capability of MOSFET, high current density of BJT, and small
on-state voltage drop, and plays a key role in configuring
various power electronic systems. The gate driver must be
designed to provide the correct output to ensure stable IGBT
operation. The use of a gate driver that produces incomplete
or abnormal output can potentially damage the IGBT and its
surrounding circuits. Gate drivers for railway vehicles must be
specifically designed to operate in a different environment
compared to ordinary products. Railway vehicles operate in
harsh conditions, including high voltage, various forms of Fig. 2. IGBT's switching waveform (inductive load)
electromagnetic interference, vibrations, and temperature
fluctuations. Therefore, having highly durable gate drivers A. Isolated power supply
capable of withstanding these conditions is crucial. However, Railway vehicles typically operate at voltages of 1500V or
when low-cost products are used, the compromise on higher, with an insulation voltage requirement of 6000V
durability and reliability can lead to problems such as product specified by UL. To meet this requirement, an isolated power
damage and frequent malfunctions requiring frequent supply (Isolated DC/DC converter) is used to provide power
replacements. In this paper, we propose high-power IGBTs to each channel of the gate drive circuit. Design considerations
with protective circuits that are compatible with the power include ensuring stable driving power by considering factors
converter of existing railway vehicles, aiming to minimize such as IGBT gate charge capacity, operating frequency, and
issues such as input voltage fluctuations, output load power requirements.
fluctuations, and overcurrents that may occur in the operating
environment of railway vehicles.[1]
II. GATE DRIVER
Fig. 3. Example of flyback power supply operation
Fig. 1. Gate Driver Diagram
- 472 -
B. Desaturation detection circuit REFERENCES
To prevent excessive voltage (VCE) in the IGBT caused [1] Korea National University of Transportation LINK 3.0, “Project Plan
by short-circuit or abnormal operation, a desaturation for the Development of Gate Driver of High-Power devices for
Railway Vehicles”, 2023.
detection circuit is necessary. This circuit detects the VCE
value based on the collector-emitter current (IC) and triggers
protective measures if it exceeds a certain threshold, thus
preventing component damage. Typically, a diode-based
circuit is used for direct VCE detection. The diodes are
carefully chosen in series to ensure their breakdown voltage
exceeds the operating voltage, preferably with fast-recovery
capabilities.
Fig. 4. Desaturation detection operation
C. Protection circuit
Fluctuations in input power are unavoidable in railway
vehicles, and there are instances where the gate driver operates
at less than 70% of its nominal voltage of 15V or lower. This
poses a risk of component and device damage due to heat and
malfunctioning. To prevent this, a protective circuit is used
below a certain voltage threshold. The circuit incorporates a
hysteresis loop to prevent chattering and set upper and lower
limits at the reference voltage.
D. Gate resistor
The gate resistor plays a crucial role in controlling the
turn-on/off characteristics, speed, and switching loss of the
IGBT. Proper selection of resistance is essential to prevent
surge voltage caused by inductance effects from the power
device stack configuration and gate driver connection method.
Research can help establish criteria for selecting the
appropriate resistance for devices with specifications of
3300V and 1500A.
III. CONCLUSION
In this paper, we propose a highly reliable high-power
IGBT that is compatible with the existing railway vehicle
power converter while minimizing issues such as voltage and
load fluctuations, and overcurrent. To achieve this, we have
designed an insulated power unit, a desaturation detection
circuit, a low voltage protection circuit, and a gate resistor.
This enhances the reliability and durability of gate drivers for
railway vehicles and reduces costs associated with product
damage and failure.
ACKNOWLEDGMENT
Following are results of a study on the "Leaders in
INdustry-university Cooperation 3.0" Project, supported by
the Ministry of Education and National Research Foundation
of Korea
- 473 -
Full-Stack Simulator for Non-Volatile
Memories-based Computing Elements for
Railway Safety
Zheng Wang1, Seong Tae Jhang2, and Juhee Choi3*
1
Linyi University, China
2
The University of Suwon, Republic of Korea
3
Sangmyung Univ, Republic of Korea
Email: [email protected]
Abstract—Specialized hardware for machine learning has SMAUG suggested the optimized the deep learning camera
been developed for both academic and industrial fields, usually pipeline for enhanced SoC.
referred to as a neural network accelerator. Although
researchers have proposed a significant number of ideas, one of
their main challenges is finding a cost-efficient method to prove
the efficiency of their schemes. Accurate estimation requires
Register Transfer Level (RTL) simulation with place and route
implementation, which demands considerable time and
resources. The introduction of a full-stack simulator aims to
solve this problem by abstracting the detailed hardware logics.
Among several simulation tools available for researchers,
SMAUG has emerged as a comprehensive full-stack simulator.
Additionally, further studies have been conducted to evaluate
the extended lifespan and power consumption of non-volatile
memory-based systems. In this research, we have improved
SMAUG-variant simulation tools specifically for the railway
domain, based on EN 50129 standards. Since train cars have
strict requirements defined by the railway system's standards, a
new method has been added to evaluate whether the proposed
scheme meets the tolerance requirements. Our experiments
show that the confidence level of this novel simulation tool is
high enough to be used with negligible additional time.
Keywords— Railway Safety, EN 50129, Full-Stack Simulator,
Non-Volatile Memories
I. INTRODUCTION
Simulators play an important role for researchers in the
field of computer systems to measure the performance of their
proposed idea. The most accurate measurement method is to
carry out a design at the RTL (Register Transfer Level)
directly, or HLS (High-Level Synthesis)[1,2].
However, this method requires significantly execution Fig. 1. Overview of SMAUG’s execution flow[8]
time, so a pre-RTL level simulator has been proposed instead.
Among several simulators, gem5 has been widely used by II. MOTIVATION
researchers as a cycle-accurate simulator[3]. Another reason
why gem5 has become the most popular simulator is its ability Even though the SMUAG has many advantages, it is not
to directly run deep learning frameworks like PyTorch[4] or suitable for non-volatile memory(NVM)-based IoT systems.
TensorFlow[5] as the entire system including the kernel. Therefore, we modified the SMUAG with infrastructure
considering the characteristics of NVMs[9]. However, it is not
Another kinds of simulators such as DNNBuilder[6] and directly applied to the safety-related electronic systems in the
MAGNet[7] allow for the execution of deep learning railway signaling field.
accelerators up to the RTL level. On the other hand, they
require an RTL or equivalent model for the entire system; Most error correcting codes (ECC) in today's computing
which implies that it is difficult to implement the proposed systems operate on the principle of single error correction
system to verification. double error detection (SECDED). This approach is cost-
effective for the past manufacturing environments. On the
Finally, an end-to-end simulator for deep learning called contrary, semiconductor companies tries to reduce a cell size
SMAUG (Simulating Machine Learning Applications Using to mitigate the pressure of the power consumption. It makes
gem5-Aladdin) was introduced as shown in Fig. 1[8]. Based the existing error detection methodologies inadequate for the
on gem5-Aladdin, SMAUG can execute full-stack workloads narrow channel cells[10]. In particular, this problem becomes
with capability of cycle-level accuracy. The authors of the
- 474 -
even more pronounced in the field of rail transport defined by
EN 50129 [11].
Therefore, we are investigating the source code and
analytic models for the railway requirements to compensate
the drawbacks of NVMs. Our current progress is inserting the
monitors for the expected remaining lifetime based on the
error patterns and failure histories. In the future, the SMAUG-
R(SMAUG with railway system) will be suggested to support
the designers of SoC makers for the electric systems in the
train vehicles.
ACKNOWLEDGMENT
This work was supported by the National Research
Foundation of Korea (NRF) grant funded by the Korea
government(NRF-2021R1G1A1004340).
REFERENCES
[1] L. Piccolboni, P. Mantovani, G. Di Guglielmo and L. P. Carloni,
"Broadening the exploration of the accelerator design space in
embedded scalable platforms," 2017 IEEE High Performance Extreme
Computing Conference (HPEC), pp. 1-7, 2017.
[2] T. Liang, L. Feng, S. Sinha and W. Zhang, "PAAS: A system level
simulator for heterogeneous computing architectures," 27th
International Conference on Field Programmable Logic and
Applications (FPL), pp. 1-8, 2017.
[3] S. Xi, Y. Yao, K. Bhardwaj, P. Whatmough, G.-Y. Wei, and D. Brooks,
“SMAUG: End-to-end full-stack simulation infrastructure for deep
learning workloads,” ACM Transactions on Architecture and Code
Optimization (TACO), Vol. 17, No. 4, pp. 1-26, 2020.
[4] V, Subramanian, “Deep Learning with PyTorch: A practical approach
to building neural network models using PyTorch,” Packt Publishing
Ltd, 2018.
[5] Shukla, N, “Machine Learning with TensorFlow,” Manning
Publications, Shelter Island, NY, USA, 2018.
[6] X. Zhang et al., "DNNBuilder: an Automated Tool for Building High-
Performance DNN Hardware Accelerators for FPGAs," 2018
IEEE/ACM International Conference on Computer-Aided Design
(ICCAD), pp. 1-8, 2018.
[7] R. Venkatesan et al., "MAGNet: A Modular Accelerator Generator for
Neural Networks," 2019 IEEE/ACM International Conference on
Computer-Aided Design (ICCAD), pp. 1-8, 2019.
[8] S. Xi, Y. Yao, K. Bhardwaj, P. Whatmough, G.-Y. Wei, and D. Brooks,
“SMAUG: End-to-end full-stack simulation infrastructure for deep
learning workloads,” ACM Transactions on Architecture and Code
Optimization (TACO), Vol. 17, No. 4, pp. 1-26, 2020.
[9] J, Choi. “Implementing Full-Stack Simulator for Deep Learning
System using Non-Volatile Memory,” KIISE Transactions on
Computing Practices, Vol. 29, No. 2, pp. 61-67, 2023.
[10] S. Mittal and J. Vetter, “A survey of techniques for modeling and
improving reliability of computing systems,” IEEE Transactions on
Parallel and Distributed Systems, vol. 27, no. 4, pp. 1226-1238, 2015.
[11] EN, B. “EN 50129: 2018 Railway applications-Communication,
signalling and processing systems-Safety related electronic systems for
signalling,” BSI Standards Publication, 2018.
- 475 -
A Study on the Optimal Traffic Signal Timing
Transition Algorithms for Diverse Demand
Fluctuations on Corridors
Yong-Bin, Cho 1, Jin-Tae, Kim 1*, Thomas JW Kim 2
1
Korea National University of Transportation, Korea, Republic of
2
UI networks, Korea, Republic of
Email: [email protected]
Abstract— This study aims to develop and validate timing immediately, and incorporate it into traffic signal operations.
transition techniques for real-time traffic signal operation, The ultimate goal is to enhance traffic flow at intersections.
departing from conventional methods based on past commute In fact, smart signal operation systems are being implemented
traffic patterns. In the past, traffic signal timing plans were to create efficient traffic environments, leveraging collected
primarily designed for peak commuting hours, neglecting the traffic information and AI technologies to generate and
remaining periods with lower congestion levels due to budget implement optimal signal timings. The real-time variability of
constraints. However, advancements in technologies such as traffic signals implies frequent adjustments of signal
video-based vehicle detection, radar, and lidar have enabled operation variables. This is in contrast to the past fixed-time
real-time vehicle detection and analysis using artificial
signal operations, which focused on peak commuting hours,
intelligence and big data techniques, providing cost-effective
access to on-site traffic information. The flexibility to select and
as illustrated in Fig. 1. Consequently, the number of traffic
change target time periods for traffic signal timing design has signal transitions occurring within a day is expected to differ.
increased. With the increasing number of traffic signal This phenomenon can be interpreted as an expansion in the
transitions resulting from these changes, there is a need to select temporal range of traffic signal transitions, with their
and apply optimal transition techniques for different transition proportion increasing compared to the entire 24-hour day.
scenarios. This study proposes two transition techniques: Furthermore, it is important to note that future traffic
Method 1 and Method 2. Method 1 involves unconditional operations can significantly impact the effectiveness of
transitions within one cycle and distributes offset changes even network signal operations, depending on the method
during non-peak hours, while Method 2 transitions within one- employed for traffic signal transition operations.
or two-cycles based on the offset change rate, also considering
non-peak hours. To validate the proposed techniques, various
traffic signal transition scenarios were analyzed and compared
in terms of delay. The results indicate that Method 1 is most
suitable for situations with increased traffic volume leading to
congestion, while the Subtract method is more appropriate for
situations with the decreased traffic volume and the reduced
congestion. These findings emphasize the importance of
selecting and applying appropriate transition techniques
tailored to individual traffic scenarios. The proposed transition
techniques provide valuable insights for improving real-time (a) Traffic signal transition under TOD signal operation
traffic signal operation and contribute to the overall efficiency
and effectiveness of traffic management on highway corridors.
Keywords—Traffic signal timing, Transition algorithm,
Simulation, Corridor
I. INTRODUCTION
Recent advancements in vehicle detection technologies
such as video, radar, and LiDAR, combined with cutting-edge
technologies like Artificial Intelligence (AI) and Big Data, (b) Traffic signal transition under TRC signal operation
have enabled the real-time collection, processing, and
analysis of traffic information in urban road networks. As a Fig. 1. Aspects of traffic signal transition according to traffic signal operation method
result, the previously fragmented system for collecting real-
time traffic information has been restored, facilitating smooth Traffic signal transition also needs to be changed to a
traffic data collection. This, in turn, has allowed for various method that applies techniques suitable for various situations,
attempts and innovations in traffic signal operations. and for this, it is necessary to be accompanied by the
development of new traffic signal transition techniques that
In this context, both central government agencies and can respond to rapid signal switching.
local authorities have shown significant interest and
investment in establishing smart intersections to improve This study aims to develop and validate timing transition
urban traffic environments. Departing from traditional fixed- techniques for real-time traffic signal operation, departing
time signal operations, they aim to collect real-time traffic from conventional methods based on past commute traffic
information through new detection systems, process it patterns.
- 476 -
II. DEVELOPMENT OF SIGNAL TRANSITION METHOD TABLE. 2. REVIEW OF TRANSITION CYCLE LENGTH
The transition technique proposed in this paper was Existing transition Proposed transition
Division Remark
designed in consideration of functional improvements to method method
supplement the limitations of traditional transition techniques. -+/- direction offset difference - Fixed the phase distribution to Design of
In the subsections below, the limitations of the traditional Tansition variable the + direction(if the offset
two
difference is negative, transition
transition method are reviewed and the transition method is direction by adding the cycle of Plan B to proposed
designed considering the derived functional improvements. the offset difference) methods
- offset difference distribution on - Apply fixed ratio considering according
A. Reviewing the limitations of traditional transition techniques phase coordinated phase the coordinated and non- to the
- Apply variable ratio coordinated phase
distribution transition
In this study, traditional transition techniques were considering the coordinated
phase length
examined, and five representative transition techniques (Add,
Subtract, Immediate, Two-cycle, Three-cycle) were evaluated.
For the Add and Subtract techniques the offset changes are To achieve the second functional improvement of
distributed to each time slot according to the current phase achieving short-term signal transitions within 1-2 cycles, two
allocation, and the transitions are performed accordingly. transition techniques have been designed based on the
However, in the case of the Dwell technique, the offset operational principles of traditional transition techniques, as
changes are uniformly applied to the coordinated time slot, shown in Table 3.
resulting in an elongation of the coordinated time slot. This
elongation can lead to congestion in the adjacent roads, The Proposed-1 technique is a modification of the
affecting the traffic situation even after transition to Plan B. Immediate technique. In the existing Immediate technique,
offset changes are allocated only to the coordinated time slots.
The duration of the transition is determined by the In the Proposed-1 technique, offset changes can also be
magnitude of the offset changes. When the offset changes are allocated to non-coordinated time slots, including the adjacent
small, the number of transition cycles and the length of each road time slots. Similarly, the Proposed-2 technique allows
cycle are reduced, resulting in a short duration of the for the allocation of offset changes to both coordinated and
transition. However, as shown in Table 1, when the offset non-coordinated time slots, similar to the Proposed-1
changes are large, the number of transition cycles and the technique. Additionally, if the magnitude of the offset change
length of each cycle increase, leading to a longer transition exceeds the transition threshold value (50% of the Plan B
period. With a threshold ratio of 20% for the Add and Subtract cycle length), the transition can be extended up to 2 cycles.
techniques, the transition is performed over a maximum of Otherwise, if the magnitude is smaller, the transition can be
five cycles. On the other hand, the Immediate technique completed within 1 cycle.
assigns all offset changes to the coordinated time slot within
one cycle, resulting in a relatively shorter transition duration. TABLE. 3. DETERMINETHETRANSITION LENGTHOFTHEPROPOSEDTRANSITIONMETHOD
TABLE. 1. REVIEW OF TRANSITION CYCLE LENGTH Number of transition cycles
Division
Transition 1 cycle 2 cycles
Division Strengths Weaknesses Develop motive
cycle
Proposed-1 ○ ×
1 cycle -Quick Plan B -Concern about - Quick Plan B transition
(Immediate)
transition adverse effect on through short transition ○ ○
Short -Optimum signal traffic flow through Proposed-2
time availablehasty transition process (offset difference ＜ threshold1)) (offset difference ≥ threshold1))
transition 2 cycles - There is no statistical
quickly
1)
(Two-cycle) difference when applying Threshold is calculated based on 50% of the Plan B cycle length. At this time, 50% s the transition
-Expected to - Slow Plan B 1-2 cycles, but it has not threshold rate
reduce adverse transition been specifically
effect on traffic -Prolonged confirmed
Long
3~5 cycles flow operation time of III. TRANSITION TECHNIQUE VALIDATION
(Add, inefficient traffic
transition signals
Subtract) -Concern about
A. Validation method
changes in traffic
flow patterns
In order to confirm the function of the proposed method,
the proposed method was applied to the simulation and a
B. Design of proposal transition technique verification process was performed through comparative
analysis with the existing traditional transfer method. To this
Taking into consideration the limitations of traditional
end, verification conditions and comparative analysis
transition techniques, this paper proposes enhancements to
methods for each transfer technique were established in the
the transition technique in order to address two functional
detailed paragraphs below.
improvements: (1) alleviating congestion on adjacent roads
and (2) achieving short-term signal transitions within 1-2 To compare and analyze the performance of the seven
cycles. transition techniques (including five traditional transition
techniques and two proposed transition techniques), a
To address the first functional improvement, which is
simulation environment was constructed, and the analysis was
alleviating congestion on adjacent roads, the proposed
conducted. In order to create signal transition scenarios, a
transition technique has been supplemented to allow for the
reference analysis scenario was established, along with
allocation of offset changes not only to the coordinated time
extended scenarios based on the reference scenario. The
slots but also to non-coordinated time slots, as shown in Table
reference analysis scenario included geometric structure
2.
conditions, traffic conditions, and control conditions. The
extended scenarios involved adjusting the traffic conditions
- 477 -
from the reference analysis scenario and deriving modified v/c ≒ 0.85 level, according to the extended analysis
control conditions to create traffic signal transition scenarios. scenario, a new scenario was created by adjusting the traffic
To define the geometric structure conditions in the volume and left-turn ratio. The Control conditions are
reference analysis scenario, an interconnected intersection presented in Table 7.
was constructed. The geometric structure conditions were
TABLE. 7. SIGNAL CONTROL CONDITION OF SCENARIO1) (UNIT : SEC)
defined by specifying the elements and their corresponding
detailed conditions, as presented in Table 4. Case
Cycle Major Major Minor Minor
LT Intersection length Offset Thru LT Thru LT
TABLE. 4. BASIC GEOMETRIC CONDITION OF TRANSITION SIMULATION Volume
ratio
Division Condition Remark #1 0 44 12 30 14
5% 100
#2 33 44 11 30 15
Signal group offset design possible
# of intersection 2 intersections
by intersection combination #1 0 40 18 28 14
80% 10% 100
Distance between #2 34 39 18 29 14
400m -
intersections #1 0 36 23 27 14
15% 100
Major 3 straight lanes, 1 right turn shared #2 34 35 22 29 14
4 lanes
# of arterial lane, 1 left turn bay
#1 0 95 20 67 28
lanes Minor 2 straight lanes, 1 right turn shared 5% 210
3 lanes #2 20 95 20 67 28
arterial lane, 1 left turn bay
#1 0 95 37 69 29
120% 10% 230
Saturated traffic flow rate : 1,800vph #2 16 93 37 71 29
#1 0 90 53 69 28
The traffic condition element items applied to the set 15% 240
geometric structure conditions and the detailed conditions #2 3 86 51 73 30
accordingly were set at the level of v/c ≒ 0.8 5, and Same as 1)
Amber time：3, All red : 1
Table 5.
In order to analyze the traffic signal transition scenario,
TABLE. 5. BASIC TRAFFIC CONDITION OF TRANSITION SIMULATION the extended analysis scenario was set as Plan A and operated
the signals from 0 to 3,600 seconds. At 3,600 seconds, the
Division Condition Remark
transition process started, and after the completion of the
Major arterial approach balance EB:WB = 60:40 - transition process, the extended scenario was set as Plan B and
Major arterial 10% Minor arterial outbound volume the signals were operated until 7,200 seconds. Fig. 2 depicts
Left turn
ratio the schematic diagram of the individual traffic signal
Minor arterial 15% Major arterial outbound volume
transition simulation analysis conducted over a total duration
East bound 2,100 60% of approach balance of 7,200 seconds.
#1
intersection North bound 1,050 -
Input South bound 1,050 -
volume West bound 1,401 40% of approach balance
#2
intersection North bound 1,050 -
South bound 1,050 -
Free flow speed : 50 kph
Fig. 2. Delay diagram analysis for each transition method according to signal operation change
When a transition request occurs, it leads to different
To facilitate signal coordination, a dual-ring scheme traffic flow patterns depending on the transition technique.
consisting of eight signal phases, with separate phases for These patterns can vary in terms of transition length,
through movements and left turns, was implemented for the transition methods, and other factors. To compare and analyze
intersection signal display combination. Furthermore, based the delays among different transition techniques, it was
on the previously defined geometric structure conditions, necessary to establish analysis approaches for each technique.
traffic conditions, and signal phase configuration, optimal During the 7,200-second simulation, a comparison was made
signal operation parameters were derived to achieve a v/c on the segments where delays occurred after the transition
ratio of approximately 0.85. This was accomplished using request. Individual transition segments and the delay
Transyt-7F, developed by the University of Florida's comparison of Plan B's one cycle were analyzed to perform
McTrans, and the results are presented in Table 6. the comparative analysis.
TABLE. 6. BASIC SIGNAL CONTROL CONDITION OF V/C ≒ 0.851) (UNIT : SEC) B. Validation results
A simulation analysis was performed for analysis scenario
Cycle Major Major Minor Minor
Division Offset and the extension scenario set previously. The analysis results
length Thru LT Thru LT
are divided into two types: a situation of switching to non-
v/c ≒ 0.85 #1 intersection 0 43 19 32 16 peak hours and a situation of switching to peak hours.
110
condition #2 intersection 20 42 19 33 16
1)
1) Non-peak time transition situation
Amber time：3, All red : 1 A simulation analysis was performed by setting the
analysis scenario as Plan A and the off-peak situation
- 478 -
expansion scenario as Plan B. Comparative analysis was disruptions within a short period. The Proposed-1 technique
performed as in table 8. involves transitioning unconditionally within one cycle and
allows for the allocation of offset changes to both the
TABLE. 8. NON-PEAK TIME TRANSITION SIMULATION RESULTS (UNIT : SEC/VEH) coordinated and non-coordinated phases. The Proposed-2
technique performs transitions within 1-2 cycles based on the
Case Proposed offset change rate and also considers the non-coordinated
LT Add Sub Imme Two Three phase for allocating offset changes. Furthermore, the
Volume 1 2
ratio functional improvements of the proposed techniques were
5% 41.94 44.93 59.57 54.66 56.33 40.14 40.14 validated. To conduct the validation, simulated traffic signal
transition scenarios were created, and a comparative analysis
80% 10% 45.22 46.98 65.15 53.20 57.71 43.34 43.34
of the transition techniques was performed based on selected
15% 48.22 49.24 66.08 62.20 60.37 43.30 43.30 analysis approaches. The results showed that the Proposed-1
technique exhibited the lowest delay during peak hour
transitions, while the Subtract technique showed the lowest
2) Peak time transition situation delay during non-peak hour transitions. These results indicate
A simulation analysis was performed by setting the that there are suitable transition techniques for different
analysis scenario as Plan A and the peak situation expansion situations, highlighting the importance of understanding the
scenario as Plan B. Comparative analysis was performed as in characteristics of each situation and applying the appropriate
Table 9. transition technique accordingly. However, since this study
focused on functional validation through three specific
TABLE. 9. PEAK TIME TRANSITION SIMULATION RESULTS (UNIT : SEC/VEH) analysis scenarios, further research is needed to analyze the
characteristics of different transition techniques under various
Case Proposed
conditions.
LT Add Sub Imme Two Three
Volume 1 2 REFERENCES
ratio
5% 75.93 60.86 86.60 84.74 85.58 59.67 59.67 [1] Kim, J., Park, J. and Chang, M., A Comparison Study of Different
Offset Transition Lengths in Simulation Environment, J. Korean Soc.
120% 10% 87.24 74.90 93.01 98.79 99.85 81.17 90.75 Transp., Vol.23, No.8, pp.43-52, 2005
15% 89.91 87.12 105.28 105.35 108.73 86.52 91.89 [2] Shelby, S., D. Bullock, and D. Gettman, Transition Methods in Traffic
Signal Control, Transportation Research Record Journal of the
Transportation Research Board No. 1978(1), pp.130~140, 2006
[3] Obenberger, J. and J. Collura, Methodology to Assess Traffic Signal
IV. CONCLUSION Transition Strategies for Exit Preemption Control, Transportation
Due to limited past traffic data collection, the Research Record Journal of the Transportation Research Board
No.2035, pp.155~168, 2007
advancement of traffic signal operations has been constrained.
[4] Cohen, D., Head, L., and Shelby, S. G., Performance analysis of
However, with the emergence of cutting-edge detection coordinated traffic signals during transition, Transportation Research
technology and artificial intelligence, the field is now Record: Journal of the Transportation Research Board, Vol.2035, No.1,
progressing to a higher level. Real-time signal operations pp.19~31, 2007
based on continuously collected traffic information are [5] Federal Highway Administration, Traffic Signal Timing Manual - 6.0
expected to play a crucial role in the development of urban COORDINATION, pp.24~28, 2008
traffic management. Additionally, traffic signal transitions [6] Tobias, P., Bernhard, F., Traffic Signal Transition in Coordinated
that adapt to changing traffic conditions are considered Meshed Networks, Transportation Research Record: Journal of the
Transportation Research Board, Issue 2192, pp 97-107, 2010
essential and influential factors.
[7] Peñabaena, N. R., Cantillo, V., Moura, J., Impact of transition between
In this study, we proposed two traffic signal transition signal timing plans in social cost based in delay, fuel consumption and
techniques that can perform transitions while minimizing air emissions, Transportation Research Part D: Transport and
Environment, Volume.41, pp. 445~456, 2015
- 479 -
Investigation of the output parameters of a
definable adaptive gear variator
Konstantin Samsonovich Ivanov 1, Gulmira Nurlankyzy Abdreshova 2
1
AUES, Almaty, Kazakhstan
Email: [email protected]
2
Al-Farabi Kazakh National University, Almaty, Kazakhstan
Email: [email protected]
Abstract – A gear variator is a gear mechanism with a variable gear ratio. The gear variator is the prospect of creating gearboxes.
Currently, two types of units with variable gear ratios are used as automatic transmissions of the car: automatic and variator.
The automatic machine is a multi–stage gear mechanism with a stage switching device and a hydraulic clutch. The variator is
a friction mechanism with stepless regulation of the gear ratio. Currently, there are only friction variators (Vbelt, toroidal). A
gear variator, unlike a friction variator, is capable of providing reliable force transmission and high efficiency. In existing
aggregates, a variable gear ratio is created using differential coupling. Differential coupling is an equation of the relationship
of parameters, which includes, in addition to geometric parameters, relative speeds of movement. In the automaton, differential
coupling is provided by hydraulics, in the variator - by friction. This connection leads to energy losses and a decrease in
efficiency. A gear variator with a constant engagement of gears in the usual representation cannot have a differential connection.
To create a gear variator, it is necessary to use a fundamentally new approach to the design of the mechanism. A fundamentally
new approach to the design of the variator was found intuitively when analyzing the gear differential mechanism. As is known,
the differential mechanism can decompose the movement of the input link into two independent output movements of two
output links. If the output links of the two differentials are combined, then you get a mechanism with one input link and one
output link, which should move independently. But such a mechanism will have two degrees of freedom, and according to the
laws of mechanics it will be inoperable. The adaptive gear variator contains an input link, an output link and a movable closed
loop placed between them. The variator has a variable gear ratio, which is automatically adjusted to the load without a control
system. Mechanical adaptation is the ability of an adaptive mechanism to transfer motion from a constant–power motor to an
output working organ at a speed inversely proportional to the technological load. Mechanical adaptation provides continuous
automatic stepless self-regulation. Adaptive gear variator is a continuously variable gear self-regulating transmission, fully
adequate to variable load. Mechanical adaptation opens up fundamentally new possibilities of gear mechanisms - it allows you
to create the simplest automatic continuously variable transmissions without any control systems. Most technological machines,
transport machines, wind turbine installations, robots, manipulators, drives of space technology operate in a variable power load
mode. This mode requires the use of variable gear ratio drives for efficient use of engine power. The existing controlled drives
are complex, and in some cases simply unacceptable (for example, controlled drives of manipulators). Adaptive gear variators
are able to solve the problem of creating highly efficient drives with variable gear ratios in all branches of technology. The
developed theory is a tool for the analysis and synthesis of a fundamentally new type of mechanisms – stepless self-regulating
gear variators that adapt to variable technological load.
.
- 480 -
DEVELOPMENT OF AN INTEGRATED MULTIFUNCTIONAL COMPLEX FOR
LIGHTING THE SPACE SITUATION
Boranbayeva A.T., Tulekenova D.T.
Email: [email protected], [email protected]
Annotation. In this article, the analysis of possible methods for estimating motion
parameters based on radar measurements under conditions of a priori uncertainty of the dynamics
of the target movement is carried out and a method for calculating the quality indicator of the
evaluation based on radar measurements by mathematical modeling is developed.
Keywords: radar, target detection range at flight altitude, azimuth, elevation, altitude,
digital phased antenna, multibeam klystron, transmitter power
Introduction
The creation of air and space control systems, remote sensing of the Earth, short-range
navigation systems and space global positioning systems, as well as communication and television
systems became possible thanks to the development of radio engineering systems for various
purposes. A radio engineering system (RTS) is an organized set of devices designed to receive,
process, accumulate and transmit information using radio waves. Radio waves, reflecting from the
inhomogeneities of the surrounding space, for example, airplanes, ships, make it possible to detect
these objects and measure their position in space. Such a property of radio waves as a different
nature of reflection from surfaces of different structures, for example, the sea, land, mountains, is
the basis of remote sensing – a scientific direction that collects information about the Earth's
surface without actual contact with it. Remote methods are divided into active and passive. When
using active methods, the satellite sends a signal from its own energy source, for example, a radar
transmitter, to the Earth and registers its reflection from the surface.
The purpose of the article is to create an integrated multifunctional complex for lighting
the space situation in the sector of the Caspian Sea.
The main part.
Any active RTS includes devices for generating and emitting radio waves, as well as
receiving and processing radio signals. The parameters of the radio signal carry certain information
about the properties of objects and changes in their state. For example, in radio navigation systems,
at the stage of primary processing of radio signals, signal parameters are measured, and at the stage
of secondary processing, estimates of the state vector of the consumer are calculated using the
estimates of signal parameters obtained at the first stage.
Figure 1.1 defines the size of the controlled area using the GIS project “Google Earth".
Figure 1.1 – Dimensions of the controlled area of the Caspian Sea
- 481 -
Analyze the tactical and technical characteristics of modern samples of the Ukrainian
mobile radar of circular view [2] and choose the most suitable for solving the task.
79K6 is a Ukrainian mobile three-coordinate all—round radar station with a phased array
antenna (80K6 is an export version of 79K6). The first prototype was manufactured in 2006. In
the same year, state tests were conducted, and in the summer of 2007, the radar was officially
adopted.
The maximum limits of the radar 79K6:
• Target detection range at flight altitude:
o 100 m — 40 km,
o 1000 m — 110 km,
o 10-30 km — 300-350 km.
• Maximum limits of the 80K6 radar operation:
o 100 m — 40 km,
o 1000 m — 110 km,
o 10000m — 400 km,
o azimuth, deg: 360
o angle of location, deg: 0...35, 55
o in height, km: 40
• Antenna Type: Digital phased array antenna
• Transmitter Type: Multipath klystron
• Peak power of the transmitter, kW: 130
Resolution:
- by range, m 400 - 500
- by azimuth, degree 2,5 – 3,5
Root-mean-square errors of coordinate
measurement
- by range, m 100
- in height, m не более 300 - 600
- azimuth, min. 15 - 20
Using the Google Earth GIS project, select the location of the radar (see Figure 1.3) on the
ground, taking into account the necessary overlap of the viewing areas of neighboring stations and
the closing angles of the radar, taking into account the terrain, for a target altitude of 10,000 m.
Figure 1.2 – Terrain from the radar to the water surface of the Caspian Sea
- 482 -
Figure 1.3 – Radar locations
The scheme of the air situation control system with missing radar anchored on the ground,
taking into account the necessary overlap of the viewing areas of neighboring stations and the
angles of radar closure, taking into account the terrain, to ensure control of the target flight at an
altitude of 100 m is shown in Figure 1.4. Also at an altitude of 1000 m is shown in Figure 1.5 and
at an altitude of 10000 m is shown in Figure 1.6.
Figure – 1.4 Ensuring the control of the target's flight at an altitude of 100 m
Figure – 1.5 Ensuring the control of the target's flight at an altitude of 1000 m
The scheme of the radar air situation monitoring system for monitoring the flight of a target
at an altitude of 1000 m is shown in Figure 1.5.
- 483 -
Figure – 1.6 Ensuring the control of the target's flight at an altitude of 10000 m
Practical acquisition of radar measurements using a simulator of the operation of a circular
view radar (CO radar).
Install a simulator of the operation of the radar CO on the computer using the module
Setup_RTS302-187.exe . After the simulator installation is completed, a window of the "Micro"
subroutine will appear on the desktop. Run the "Micro" routine for execution. After receiving the
image of the program's working window, which is shown in Fig. 1, select the "File" submenu and
upload a file named "Test". Select the operating modes of the simulator shown in Fig.1, namely,
the rotation speed of the antenna is 2 rpm, the range scale is 200 km. Click on the "Start" button.
In total, the simulator generates the trajectories of four radar targets RS1-RS4, which are located
at different distances from the radar and move from different azimuthal directions (see Fig. 1.7
and Fig. 1.8).
Figure -1.7 Trajectory of RC1, RC2
Using the radar simulator, get an array of radar measurements (range - Dc, azimuth - C )
to plot the trajectory of the target RC1-RC4 in a spherical coordinate system for 20 minutes with
an interval of 1 min for a given height of the target movement.
To obtain the RC marks, for each moment of measuring the DC range, it is necessary to
calculate the value of the angle of the RC position according to the method that is explained in
Figure 2.5, according to the expression [3] according to the specified values of the height of the
NC:
𝐷𝑐2
𝐻𝑐 = 𝐷𝑐 ∗ 𝑠𝑖𝑛𝜀 +
2𝑅𝑒
Create an array of received RC marks (t, Dc, c, c) in a spherical coordinate system and
plot the time variation of the received values.
- 484 -
Figure – 1.8 Obtaining arrays of radar measurements
Figure – 1.9 Determination of the coordinates of the targets
To map, select the option of the trajectory of the target 03 relative to the radar using the
AZBook software, which is shown in Figure 1.10. and the data were entered in Table 2.
Figure 1.10 – Trajectory of the target
The radar was selected on the scheme of the air situation control system, which observed
the target.
Radar No. 19 was selected as a radar in the scheme of the air situation control system,
which observed the target shown in Figure 1.11.
- 485 -
RADAR No. 19, with coordinates:
N.L. - 46°48'37.89"N.
E.L. - 50°12'55.46"W.
The change in the coordinates of the target in time is plotted in the form of marks of latitude
Vc and longitude Lc on the scheme of the air situation control system using the GIS project
“Google Earth”.
Figure 1.11 – Trajectory of the target relative to radar No. 19
Determine the latitude of the radar radar and the longitude L of the radar in the geographical
coordinate system (GSK) (Fig.1.11) [3].
Figure 1.12- Geographical Coordinate system (GSK)
Coordinates of radar No. 19, in the geographical system:
Brs = 46.878°
Lrs = 50.177°
To describe the movement of an object relative to a point on the Earth's surface, a
topocentric local rectangular coordinate system (MPSC) and a spherical SC are used (Fig.5).
- 486 -
Figure 1.13 - Local rectangular coordinate system (MSK)
The center of the topocentric MPSC is located on the Earth's surface. The ОtYt axis is
perpendicular to the plane of the local horizon (PLH) and passes through the center of the Earth.
The Otitis ОtXt and ОtZt Otitis ОtZt axes are located in the PLH so that the right triple of vectors
is formed. Let the axis from ОtZt coincide with the direction to the North Pole of the Earth N. Then
the angle between the axis ОtZt and the projection of the radius vector ОtC on the PLH, counted
clockwise, is called the azimuth βc.The position of the target in this SC is characterized by the
coordinates Xt, Yt, Zt [4].
Recalculated for each moment of time, the resulting array of measurements of the
coordinates of the target from the spherical SC in the MSK radar [4]:
𝑋𝑡 = −𝑅𝑐 𝑐𝑜𝑠 𝜀𝑐 𝑠𝑖𝑛 𝛽𝑐
𝑌𝑡 = 𝑅𝑐 𝑠𝑖𝑛 𝜀𝑐 }.
𝑍𝑡 = 𝑅𝑐 𝑐𝑜𝑠 𝜀𝑐 𝑐𝑜𝑠 𝛽𝑐.
Determine the position of the radar in the GCC. The position of the radar on the Earth's
surface in the geographical coordinate system is determined by the latitude of the radar and the
longitude L of the radar, so the position of the radar in the GCC can be determined as follows [6]:
𝑋1 = 𝑅𝐸 𝑐𝑜𝑠 В𝑟𝑙𝑠 𝑐𝑜𝑠 𝐿𝑟𝑙𝑠
𝑌1 = 𝑅𝐸 𝑐𝑜𝑠 В𝑟𝑙𝑠 𝑠𝑖𝑛 𝐿𝑟𝑙𝑠 }.
𝑍1 = 𝑅𝐸 𝑠𝑖𝑛 В𝑟𝑙𝑠
Recalculate the array of coordinates of the target from MSK to GSK [7]:
𝑋𝑐 𝑠𝑖𝑛 𝐿𝑟𝑙𝑠 𝑐𝑜𝑠 В𝑟𝑙𝑠 𝑐𝑜𝑠 𝐿𝑟𝑙𝑠 − 𝑐𝑜𝑠 𝐿𝑟𝑙𝑠 𝑠𝑖𝑛 В𝑟𝑙𝑠 Х𝑡 𝑋1
[𝑌𝑐 ] = [− 𝑐𝑜𝑠 𝐿𝑟𝑙𝑠 𝑐𝑜𝑠 В𝑟𝑙𝑠 𝑠𝑖𝑛 𝐿𝑟𝑙𝑠 − 𝑠𝑖𝑛 В𝑟𝑙𝑠 𝑠𝑖𝑛 𝐿𝑟𝑙𝑠 ] [𝑌𝑡 ] + [𝑌1 ]
𝑍𝑐 0 𝑠𝑖𝑛 В𝑟𝑙𝑠 𝑐𝑜𝑠 В𝑟𝑙𝑠 𝑍𝑡 𝑍1
The movement of the target in time in the geographical coordinate system is determined
by calculating the latitude of the Bц and the longitude of the Lц, which can be determined by
knowing the position of the target in the GCC, as follows [7]:
- 487 -
𝑅𝑐 = √𝑋𝑐2 + 𝑌𝑐2 + 𝑍𝑐2
𝐿𝑐 = 𝑎𝑟𝑐𝑡𝑔(𝑌𝑐 /𝑋𝑐 ) }.
В𝑐 = 𝑎𝑟𝑐𝑠𝑖𝑛( 𝑍𝑐 /𝑅𝑐 )
Analyze the possibility of controlling the surface situation in the Kazakh sector of the
Caspian Sea with the involvement of the radar of the developed air situation control system.
Several radar posts installed at a distance from each other allow you to create a continuous
control zone. The range of the radar is up to 42 km, which allows you to fully monitor the situation.
The distance of 42 km is taken from the consideration that the visible horizon, from a height
of 100 meters, is approximately 42 km from the observation point.
The creation of a network of coastal radars will allow us to take under reliable control the
situation along key sections of the coast of the Republic of Kazakhstan.
Conclusions
To date, one of the main sources of information about the air situation in the controlled
territory is radar. To obtain reliable data on the parameters of the target movement in these systems,
the radar image processing system plays an important role. The main problem of the radar image
processing system is the choice of a statistical processing method in case of errors in the
measurement of object motion parameters.
The accuracy of the estimation of the object motion parameters based on the results of the
radar images obtained depends on a priori information about the dynamics of the observed object.
Therefore, the development of a methodology for evaluating the effectiveness of methods for
estimating the parameters of the object's motion in conditions of a priori uncertainty of the
dynamics of the observed object's motion is an urgent problem considered in the qualification
examination.
Thus, the radio navigation parameters of the target were detected using the Google-Earth
software. A variant of building an air situation control system in the Kazakh sector of the Caspian
Sea has been developed, which will allow monitoring the surface situation and the flight of air
targets at an altitude of 1000 to 10000 m.
A sequential recalculation was made according to spatial coordinate systems.
The skills of creating an integrated multifunctional complex for lighting the space situation
in the Caspian Sea were acquired. The trajectory of the target is relatively directed from north to
east.
References
1. Babak V.P. Suputnikova radionavigation/ V.P. Babak, V.V. Konin, V.P. Kharchenko −
K.: Technika, 2004. − 328 p.
2. Bakulev P.A. Radar systems: textbook for universities / P.A. Bakulev. – M.: Radio
Engineering, 2004. – 320 p.
3. Berdyshev V.P. Radar systems: textbook / V.P. Berdyshev, E.N. Garin, etc.; under the
general editorship of V.P. Berdyshev. – Krasnoyarsk: SFU, 2011. – 400 p.
4. Friz S.P. Theory and technique of radar and radio navigation: textbook. manual / S.P.
Friz, A.V. Andreev, A.R. Rykhalsky, A.F. Dubina – Zhytomyr: Zhytomyr Polytechnic State
University, 2020. – 180 p.
5. Nikolaev A.I. Radar systems: textbook. manual/ A.I. Nikolaev, V.V. Akhiyarov, S.I.
Nefedov. − M.: Bauman Moscow State Technical University, 2018. – 349 p.
6. Shirokov Yu.F. Fundamentals of the theory of radar systems: electron. studies. manual
/ Yu.F. Shirokov. – Samara: SSAU named after S.P. Korolev, 2012.
7. Shirman Ya.D. Theory and technique of radar information processing against
interference/ Ya.D. Shirman, V. N. Manzhos. – M.: Radio and Communications, 1981. – 416 p.
- 488 -
Safety of aircraft operation in modern
conditions
Gulnara Nauryzbayeva, Guldana Aldzhambekova, A Tuktibayev
Almaty University Power Engineering and Telecommunications after G. Daukeyev, Kazakhstan
Email: [email protected]
Abstract— High-speed descent through the atmosphere information for future missions and applications. The
generates a significant amount of heat, which, if not properly knowledge gained from thermal computing can be applied to
managed, can lead to structural failure, thermal stress, and a wide range of engineering disciplines, including aerospace,
other problems. Accurate thermal calculations are essential to automotive, and power generation. The development and
mission safety and success. They play a key role in predicting research of methods for the thermal calculation of descent
the thermal loads and temperature distribution of vehicles, vehicles is necessary to ensure the safety, success and
allowing engineers to design and build vehicles that can efficiency of spacecraft flights, as well as for the development
withstand extreme entry, exit and landing conditions. By of modern technologies of thermal engineering. Accurate
understanding the thermal environment during descent, thermal calculations help optimize vehicle thermal design,
engineers can also design and develop systems that can reduce costs and increase productivity, making them an
operate more efficiently and effectively. Finally, research and important aspect of spacecraft design and development.
development of thermal computing techniques will help
improve the state of the art in thermal engineering and
provide valuable information for future missions and Keywords— safety, aircraft, operation, space, method
applications. The knowledge gained from thermal computing Introduction
can be applied to a wide range of engineering disciplines, Today, it is important to develop and research lander
including aerospace, automotive, and power generation. thermal calculation methods to ensure the safety, success, and
Relevance of the topic. The development and research of efficiency of spacecraft flight, as well as to advance state-of-
methods for the thermal calculation of the descent vehicle is the-art thermal engineering. The development and research of
very relevant for several reasons: Safety: Accurate thermal thermal calculation methods for the descending vehicle are
calculations are essential to ensure the safe and successful very relevant for several reasons:
landing of a vehicle and its crew or payload. The heat Safety: Accurate thermal calculations are essential
generated during atmospheric descent can lead to structural to ensure a safe and successful landing of the vehicle and its
failure, thermal stress, and other problems if not properly crew or payload. The heat generated during descent through
managed. By understanding the mechanisms of heat transfer the atmosphere can lead to structural failure, thermal stress,
and temperature distribution in a vehicle, engineers can and other problems if not managed properly. By
design and build vehicles that can withstand the extreme understanding the mechanisms of heat transfer and
conditions of entry, descent and landing. temperature distribution in a vehicle, engineers can design
Mission success: The thermal environment during the descent and build vehicles that can withstand the extremes of entry,
phase can significantly affect the success of a mission. The descent, and landing.
ability to accurately predict thermal loads and temperature Mission Success: The thermal environment during
distribution within a vehicle helps achieve mission objectives the descent phase of a mission can significantly affect
and avoid damage to the vehicle and its payload during the mission success. The ability to accurately predict heat loads
landing phase. Cost Savings: Accurate thermal calculations and temperature distribution in the vehicle helps to achieve
help optimize vehicle thermal design by reducing the amount mission objectives and avoid damage to the vehicle and its
of materials and thermal management systems needed. This payload during the landing phase.
can result in significant cost savings in terms of vehicle Cost savings: Accurate thermal calculations help
development and operating costs. Improved Performance: By optimize vehicle thermal design, reducing the amount of
understanding the thermal environment during descent, thermal management materials and systems needed. This can
engineers can design and develop systems that can operate result in significant cost savings in terms of vehicle
more efficiently and effectively. For example, by predicting development and operating costs.
the temperature distribution in a car, engineers can design Improved performance: By understanding the
performance-optimized insulation and cooling systems, thermal environment during descent, engineers can design
reducing overall vehicle weight and power consumption. and develop systems that can operate more efficiently and
Technology Development: Research and development of effectively. For example, by predicting temperature
thermal computing techniques can help advance the state of distribution in a vehicle, engineers can design performance-
the art in thermal engineering and provide valuable
- 489 -
optimized insulation and cooling systems, reducing overall thermal characteristics of the vehicle and its environment.
vehicle weight and power requirements. Models can range from simple geometric models to complex
Technology Advancement: Research and models that include detailed information about the vehicle's
development of thermal computing techniques can help materials, systems, and other components. After the thermal
advance the state of the art in thermal engineering and models are created, input data is generated to represent the
provide valuable insights for future missions and applications. conditions encountered during reintroduction. This data may
Knowledge gained from thermal computing can be applied to include information on atmospheric conditions such as
a wide range of engineering disciplines, including aerospace, density, pressure, and temperature, as well as heating rates
transportation, and energy production. and heat transfer coefficients. This data is used to determine
The development and research of spacecraft thermal the thermal environment in which the vehicle will operate.
calculation methods is important to ensure the safety, success, Thermal calculations are then performed using
and efficiency of spacecraft missions, as well as to advance numerical methods such as computational fluid dynamics and
the state of the art in thermal engineering. Accurate thermal finite element analysis. These methods solve the equations
calculations help optimize vehicle thermal design, reduce describing the heat transfer through the vehicle and its
costs, and improve performance, making them an important environment and allow predicting the heat loads and
aspect of spacecraft design and development. temperature distribution in the vehicle.
Finally, the results of the thermal calculations are
MODERN HEAT CALCULATION TECHNOLOGY validated and verified to ensure their accuracy and reliability.
The study of thermal calculations for landing This can be done through a variety of methods, including
spacecraft has been an active area of research and comparisons with flight test data, ground test data, and other
development for many decades. There is a substantial body experimental data, as well as sensitivity analysis and error
of literature on this topic, ranging from theoretical studies to propagation studies.
practical design and analysis tools. The methodology used to develop and investigate
The current state of thermal engineering for spacecraft lander thermal calculations typically involves
spacecraft launch vehicles includes analytical and numerical creating thermal models, generating input data, performing
methods for predicting thermal loads and temperature thermal calculations, and validating and verifying the results.
distributions. Analytical methods based on thermal This multi-step process helps ensure that the results of
conductivity and radiation are useful for predicting steady- thermal calculations are accurate and reliable, providing
state thermal conditions. However, these methods have valuable information for the design and development of
limitations when dealing with the complex and dynamic spacecraft landing gear.
thermal environment encountered during reentry. Spacecraft lander thermal calculations include a
Numerical methods, such as computational fluid series of equations and numerical models to accurately
dynamics and finite element analysis, are becoming predict heat transfer and temperature distribution on the
increasingly popular for thermal calculations due to their surface of the spacecraft. These equations and models depend
ability to predict dynamic and transient thermal conditions on various factors such as speed, altitude, angle of attack, and
that occur during re-entrainment. These techniques allow thermal properties of the surrounding atmosphere. Below are
engineers to model the flow of heat through the vehicle and the thermal calculation equations that can be used for the
its surrounding environment, providing a detailed planter:
understanding of the temperature distribution and heat loads Convective heat transfer equation:
experienced by the vehicle. 𝑄 = ℎ ∗ 𝐴 ∗ (𝑇 − 𝑇 )
In addition to numerical methods, there are also where 𝑄 − is the convective heat transfer, ℎ −
various design and analysis tools for thermal calculations. is the convective heat transfer coefficient, 𝐴 − is the vehicle
There is also a growing body of literature on the use of surface area, 𝑇 − is the temperature of the vehicle surface,
advanced materials and systems to manage heat generated 𝑇 − is the ambient temperature.
during reentry. This includes the use of thermal protection Radiation heat transfer equation:
systems such as thermal blankets and tiles, and the 𝑄 = ℇ ∗ ℴ ∗ 𝐴 ∗ (𝑇 − 𝑇 )
development of advanced cooling systems for electronics and where 𝑄 − radiative heat transfer, ℇ − surface
other critical components. radiation coefficient, ℴ − Stefan-Boltzmann constant, 𝐴 −
A review of the literature reveals that there is a vehicle surface area, 𝑇 − vehicle surface temperature, 𝑇 −
wealth of information on the subject of thermal calculations ambient temperature of the surrounding atmosphere.
for spacecraft landing vehicles. The current state of the art Equation of conductive heat transfer:
includes analytical and numerical methods, as well as a 𝑇 −𝑇
variety of design and analysis tools, and growing research 𝑄 =𝑘∗𝐴∗
𝑑
into the use of advanced materials and systems to manage the where 𝑄 − conductive heat transfer, 𝑘 −
heat generated during reintroduction. thermal conductivity of the material, 𝐴 − vehicle surface area,
The methodology used to develop and study 𝑇 − vehicle surface temperature, 𝑇 − ambient temperature,
spacecraft lander thermal calculations typically involves 𝑑 − material thickness.
several steps, including generating thermal models, Energy balance equation:
generating input data, performing thermal calculations, and 𝑄 +𝑄 +𝑄 =𝑄
validating and verifying the results. where 𝑄 − is the heat energy absorbed by the
The first step in the methodology is to create thermal vehicle.
models. These models are used to represent the geometry and
- 490 -
Using these equations, as well as numerical generation during flight while ensuring the safety of the crew
techniques such as finite element analysis and computational and equipment. Here is a brief explanation of where each of
hydrodynamic modeling, engineers can accurately predict the the values used in the calculation of radiative and convective
thermal behavior of a vehicle during a descent and design heat transfer in the union comes from.
effective thermal protection systems. Radiation coefficient (ε): The radiation coefficient
Radiation heat transfer for "Soyuz-16" of the surfaces of the Soyuz-16 spacecraft was assumed to be
Before entering the Earth's atmosphere, the Soyuz- equal to 0.2. This is a characteristic of spacecraft materials.
16 spacecraft will be exposed to sun radiation and deep space Surface Area (A): The surface area of Soyuz-16 is calculated
radiation. Radiant heat transfer from these sources can be based on its dimensions of approximately 7.48 meters in
calculated using the following equation: length and 2.72 meters in diameter. The total area was 96.4
𝑞 = 𝜎𝜀𝐴 (𝑇 − 𝑇 ) square meters. temperature (T): the temperature used in the
q- radiative heat transfer (W) calculations is based on conditions during the Union-16 flight.
σ- Stefan-Boltzmann constant (5,67 ∙ 10 − 8 W/ The temperature of the spacecraft in space was estimated to
m K ) be 2.7 Kelvin, that is, the temperature of deep space. The
ε- radiation coefficient of the spacecraft surface temperature of the earth in the atmosphere during the flight
(assumed to be equal to 0.2 for Soyuz-16 spacecraft). was equal to 800 Kelvin, and during the descent it was equal
A- spaceship surface area exposed to radiation to 1200 Kelvin. Stefan-Boltzmann constant (σ): This is a
(m ). basic physical constant used in radiation heat transfer
calculations and its value is 5,6710-8 W/(m K ). Convective
T- spacecraft surface temperature (K). heat transfer coefficient (h): The convective heat transfer
T - deep space temperature (should be 2.7 K) coefficient used in the calculation of the atmospheric flight of
As the Soyuz-16 spacecraft enters the Earth's the earth is taken as equal to 50 W / (m K), which is a typical
atmosphere, it collides with atmospheric gases, causing value of the heat transfer in the air. Velocity (v): The velocity
convective heat transfer. Convective heat transfer can be used in the calculation of convective heat transfer for the
calculated using the following equation: flight of the earth in the atmosphere was equal to 255.5 m/s,
𝑞 = ℎ𝐴(𝑇 − 𝑇∞ ) which is the maximum speed of the Soyuz-16 spacecraft
q - convective heat transfer (W) during this period.
h - convective heat transfer coefficient (W/m K) Convective heat transfer for Soyuz-16 spacecraft
A- surface area of the spacecraft in contact with the The convective heat transfer for the Soyuz-16
atmosphere (m ). spacecraft can be calculated using the following equation:
T- spacecraft surface temperature (K). 𝑞 = ℎ ∗ 𝐴 ∗ (𝑇 − 𝑇 )
T∞ — the temperature of the atmosphere (taken as where 𝑞 − is the heat transfer rate, h is the
equal to 255.5 K at sea level). convective heat transfer coefficient, 𝐴 − is the surface area,
After landing, the Soyuz-16 spacecraft is still 𝑇 is the temperature of the surface, and 𝑇 is the
exposed to radiation from the sun and the Earth's surface. ambient temperature. During the entry into the atmosphere,
Radiant heat transfer can be calculated using the same Soyuz-16 was exposed to high temperatures due to friction
equation as in step 1. The heat output from the Soyuz-16 with the air. This resulted in a significant increase in the
equipment can be calculated using the following equation: surface temperature of the spacecraft. The heat produced by
𝑃
𝑞 = the equipment inside the spacecraft is dissipated to the
𝜀𝐴 environment by convective heat transfer. The convective heat
q-heat output (watts) transfer coefficient depends on the speed, density and thermal
Radiation was the dominant factor influencing the conductivity of the air. To calculate the convective heat
thermal characteristics of Soyuz-16 during the spaceflight transfer, we need to estimate the convective heat transfer
and reentry phases. However, during the descent phase, both coefficient ℎ . For this purpose we can use the following
radiation and convective heat transfer played an important equation:
role in dissipating the heat produced by the equipment. 𝑣 , 𝜌 ,
During descent, the Soyuz-16 spacecraft had to withstand ℎ = 0,029 ∗ ∗ ∗ 𝜆,
high temperatures due to a combination of radiative and 1000 𝜌
convective heat transfer. A thermal protection system where 𝑣 is the speed of the spacecraft, 𝜌 is the air
consisting of ablative heat shields and insulating materials density, 𝜌 is the air density at sea level, λ is the thermal
provided protection from extreme heat. The calculated conductivity of the air. Using the known physical
convective heat transfer was 1,06107 watts. This was mainly characteristics of the Soyuz-16 spacecraft, it is possible to
due to the high speed of the spacecraft through the Earth's estimate the convective heat transfer coefficient at different
atmosphere. stages of the flight. During descent, Soyuz-16 had a speed of
As the spacecraft descends, air molecules collide about 230 m/s, air density and thermal conductivity at an
with the surface of the heat shield, transferring heat through altitude of 120 km were 1,3610-4 kg/m3 and 2,610-2 W/mK,
convection. The estimated radiation heat transfer during respectively. Using these values, we can estimate the
descent was 1,58107 watts. This was the dominant factor convective heat transfer coefficient as follows:
,
affecting the thermal performance of the 10 spacecraft during 230 , 10
ℎ = 0,029 ∗ ∗ 1,36 𝑥
this period. The heat produced by the equipment is radiated 1000 1,23
into the surrounding atmosphere. Overall, the thermal design ∗ (2,6 𝑥 10 ) ,
= 50 W / (m K)
of the Soyuz-16 spacecraft successfully managed heat
- 491 -
During the entry phase, the speed of the Soyuz-16 used for spacecraft equipment and structures, as well as the
spacecraft was about 7900 m/s, the air density and thermal thickness of the materials. Below are the conduction heat
conductivity at an altitude of 80 km, respectively 3,4710-3 transfer estimates for Soyuz-16: Conduction heat transfer
kg/m3 and 2,210-2 W/mK. Using these values, we can between spacecraft equipment and structures during
estimate the convective heat transfer coefficient as follows: spaceflight: Assume an average temperature of 20°C for the
7900 , 10
, equipment and 5°C for the spacecraft design. Suppose that
ℎ = 0,029 ∗ ∗ 3,47 𝑥 the thermal conductivity of materials is 0.2 W/mK. Assume
1000 1,23 that the average distance between the hardware and the
(2,2 ) ,
∗ 𝑥 10 = 5100 W / (m K) spacecraft design is 1 cm. Estimate the contact surface area
Note that the convective heat transfer coefficient as 10 m
may vary depending on the shape and orientation of the 293 − 278
spacecraft, as well as the composition and properties of the 𝑞 = 0,2 ∗ 10 ∗ = 29,0 𝑊
0,01
surrounding atmosphere. These estimates provide a general Conductive heat transfer between spacecraft
idea of the amount of convective heat transfer during the equipment and structures during atmospheric entry: Assume
Soyuz-16 flight. an average temperature of 150°C for the equipment and
The convective heat transfer coefficient for the 1200°C for the spacecraft design. Suppose that the thermal
Soyuz-16 spacecraft in the atmospheric phase on Earth is conductivity of materials is 0.2 W/mK. Assume that the
estimated to be 50 W/(m K). The surface temperature was average distance between the hardware and the spacecraft
estimated at 255.5K based on radiation heat exchange design is 1 cm. Estimate the contact surface area as 10 m
equations. The ambient temperature at an altitude of 30 km 423 − 1473
was equal to 255K. Using these values, we can calculate the 𝑞 = 0,2 ∗ 10 ∗ = −5,5 𝑘𝑊
0,01
convective heat transfer for the atmospheric phase of the
(note the negative sign indicating that heat is transferred from
Soyuz-16 spacecraft.
the spacecraft design to the equipment) Conductive heat
During descent, the convective heat transfer
transfer between the equipment and the structure during
coefficient may increase due to high density and air velocity.
takeoff: Assume an average temperature of 80°C for the
The temperature of the Earth's surface also increases due to
equipment and 1000°C for the spacecraft design.
the compressional heat generated by the shock wave in front
Suppose that the thermal conductivity of materials is
of the spacecraft. Values of convective heat transfer
0.2 W/mK. Assume that the average distance between the
coefficient and descent phase surface temperature are not
hardware and the spacecraft design is 1 cm. Estimate the
available and depend on several factors such as angle of
contact surface area to be 10 m
attack, speed, altitude and atmospheric properties. 353 − 1273
The values of the convective heat transfer q = 0,2 ∗ 10 ∗ = −3,6 k𝑊
coefficient (h) used in the calculations were estimated based 0,01
(Note the negative sign indicating that heat is transferred
on typical values of the spacecraft inlet. The values of the
from the spacecraft design to the equipment) These estimates
Earth's surface (A) were taken from the physical
give a general idea of the amount of conductive heat transfer
characteristics of the Soyuz-16 spacecraft, a spherical shape
during the flight of the Soyuz-16 spacecraft. For conductive
with a diameter of 2.7 meters. To calculate the convective
heat transfer, we used the thermal conductivity of the
heat exchange of the Earth's atmosphere, the ambient
materials and the temperature gradient between the interior
temperature is 255.5 K
and exterior of the spacecraft to calculate the heat transfer rate.
(-17.6°C), which is the standard temperature of the
Values are obtained from various sources such as NASA
atmosphere at an altitude of about 11 km. The temperature of
technical reports, material specifications, and scientific
the Earth's surface was estimated to be 800 K based on
literature. The dimensions and composition of the spacecraft
radiation heat transfer calculations. To calculate the
were also considered to estimate the surface layer and
convective heat transfer of the lunar atmosphere, the ambient
thickness of the various layers. The values used in the
temperature was assumed to be 120 K (-153.15°C), which is
calculations are based on the physical characteristics of the
the average temperature of the lunar surface. The temperature
Soyuz-16 spacecraft, including Temperature gradient
of the Earth's surface was estimated at 100K based on the
between interior and exterior space: based on assumptions
calculation of radiation heat transfer.
regarding the mission profile and internal heat distribution
Conductive heat transfer occurs when two
and distribution. The equations used are based on Fourier's
substances at different temperatures are in direct contact,
Law of heat conduction and assuming stationary conditions.
when heat is transferred from a hot substance to a cold
The heat transfer rate was calculated as the product of surface
substance. In the case of Soyuz-16, conductive heat transfer
area, temperature gradient, and thermal conductivity.
mainly occurred between the on-board equipment and the
Conclusion
spacecraft design. To calculate conductive heat transfer, we
For the above calculation, a thermal conductivity
use the following equation:
value of 10 W/mK and a thickness of 0.01 m were assumed
𝑞 = 𝑘 ∗ 𝐴 ∗ (𝑇 − 𝑇 )/𝑑
for the spacecraft design. The used surface area was 0.2
where q is the speed of heat flow, k is the thermal
square meters, and the hot and cold temperatures were
conductivity of the material, A is the area of the contact
estimated to be 353 K and 1273 K, respectively, "Soyuz- 16"
surface between two objects, 𝑇 is the temperature of the
based on equipment and spacecraft design temperature ranges
hot object, 𝑇 is the temperature of the cold object, and d
during flight.
is the distance between the two objects. The conductive heat
transfer coefficient k can vary depending on the materials
- 492 -
The main type of heat transfer during this period was 12 Allen, G. A. (2012). Fundamentals of space systems.
a combination of radiative and convective heat transfer. On Oxford University Press.
board "Soyuz-16" there was a variety of equipment that
produced heat during operation, such as the guidance and
navigation system, communication equipment and life
support systems. The heat generated by this equipment had to
be dissipated through the spacecraft's cooling system to
prevent overheating and damage to the equipment. The
amount of heat coming out of the equipment can be estimated
from the information about the power consumption and
efficiency of the equipment.
Similarly, the heat generated by other onboard
equipment can be estimated using the same equation. Once
the heat is released, it must be dissipated by the spacecraft's
cooling system. The cooling system of the Soyuz-16
consisted of two separate circuits: one for the equipment and
the other for the crew cabin. The equipment circuit used a
liquid coolant that circulated through the equipment to absorb
heat, which was then sent to a heat exchanger, where the heat
was released to space through a radiator.
Overall, the heat distribution and separation process
for Soyuz-16 was a complex and critical aspect of the
spacecraft's thermoregulation system, requiring precise
calculations and design to ensure mission safety and success.
As it entered the atmosphere, Soyuz 16 experienced strong
atmospheric drag, which caused frictional heating of the outer
surfaces of the spacecraft. This heat had to be dissipated to
prevent damage to the spacecraft and its passengers. The
main types of heat transfer during this period were radiative
and convective heat transfer.
List of references
1 Smith, J. M., & Johnson, R. V. (2018). Spacecraft
Thermoregulation Manual: Basic Technologies. Cambridge
University Press.
2 Anderson, Jr., J. D. (2016). Familiarity with flying.
McGraw Hill Education.
3 Incropera Devitt, DP, Bergman, T. L., and Lavigne, A. S.
(2017). Fundamentals of heat and mass transfer. John Wiley
& Sons.
4 NASA. (2019). Thermal protection systems: Materials
and design. Retrieved from
https://www.nasa.gov/directorates/spacetech/tdm/thermal_pr
otection_systems/materials.html
5 CFD Online. (2021). Computational hydrodynamics.
Retrieved from https://www.cfd-online.com
6 SOLIDWORKS. (2021). Modeling in SOLIDWORKS.
Retrieved from
https://www.solidworks.com/products/simulation/
7 Menter, F. R. (1994). Eddy Turbulence Models with Two
Viscosity Equations for Engineering Applications. AIAA
Journal, 32(8), 1598-1605.
8 Ferstig and Malalasekera, 2007). Introduction to
Computational Hydrodynamics: The Finite Volume Method.
Pearson Education Limited.
9 Rossow, K. S. (2014). Thermal analysis and design of
passive systems of thermoregulation of space vehicles. John
Wiley & Sons.
10 Gilmore, D. G., & Gilmore, T.F. (2018). Handbook of
Spacecraft Thermoregulation: Cryogenics. CRC Press.
11 Cheng, K. Y., and Louis, R. Yu. (2016).
Thermoregulation systems of space vehicles. Jumper.
- 493 -
A Molecular Dynamics Simulation Study on
Superlubric Carbon-based Nanodevice
Tae Hoon Kim 1, Jeong Won Kang 1,2*, Haram Kim 1, and Ki-Sub Kim 3*
1
Department of SMART Railway System, Korea National University of Transportation, Republic of Korea
2
Department of Transportation System Engineering, Korea National University of Transportation, Republic of Korea
3
School of Chemical and Energy Engineering, Korea National University of Transportation, Republic of Korea
Email: [email protected] (KSK), [email protected] (JWK)
Abstract— Superlubric motions of graphene nanoflakes on
graphene have opened up more applications of graphene for
micromachines and nanomachines. Here, we investigate the
dynamic behavior of a graphene nanoflake shuttle on a
graphene nanoribbon with carbon nanotube blocks via
molecular dynamics simulations, and present a design of a
nanoscale inertial measurement unit or a data archive using a
graphene-nanoflake sandwiched between crossed graphene-
nanoribbon junctions. The graphene nanoflake moves on a
graphene nanoribbon superlubrically, and the carbon
nanotubes as building blocks induce bistable potential wells so
that the graphene nanoflake is stabilized. Molecular dynamics
simulation results indicate that when a graphene nanoflake
shuttle approaches the carbon nanotubes, a potential well is
created by an increase in the attractive van der Waals energy
between the graphene nanoflake and carbon nanotubes, and
bistability at the local energy minima positions can be achieved
near the carbon nanotubes. In order for the graphene nanoflake
shuttle to escape the local energy minima positions, a high
external force must be applied to overcome the potential energy
barrier. However, after the graphene nanoflake shuttle escapes
from one of the bistable positions, only a low external force is
required to stabilize the graphene nanoflake shuttle. When an
external force applied is below the retracting force, the inertial
force exerted on the movable graphene nanoflake can telescope
it. Then, the self-restoring force increases as the attractive van
der Waals force between the graphene nanoflake and the
graphene nanoribbons, which enables the graphene nanoflake
to automatically and fully retract back into the sandwich
position immediately after the externally applied force is
released. When the external force exceeds the retracting force,
the graphene nanoflake escapes from the crossed a graphene
nanoribbon junction, which enables the device to be used as non-
volatile memory. This work explicitly demonstrates that a
graphene nanoflake-graphene nanoribbon / carbon nanotubes
system could be applied to alternative nonvolatile memory and
high-speed mass storage by using graphene nanoribbon - carbon
nanotubes arrays.
Keywords—Carbon Nanotube, Graphene, Nanoribbon,
Nanoflake, Molecular Dynamics
ACKNOWLEDGMENT
This work was also supported by the Korea Agency for
Infrastructure Technology Advancement (KAIA) grant
funded by the Ministry of Land, Infrastructure and Transport
(Grant No. 22QPWO-C158106-04).
- 494 -
Extracted Cellulose with Deep Eutectic Solvent
and Ionic Liquid: Changes in Structural and
Thermal Characteristics
Yeung Chan Kim 1, Ho Hyun Han 2, Jeong Won Kang 2,3*, and Ki-Sub Kim 1*
1
School of Chemical and Energy Engineering, Korea National University of Transportation, Republic of Korea
2
Department of SMART Railway System, Korea National University of Transportation, Republic of Korea
3
Department of Transportation System Engineering, Korea National University of Transportation, Republic of Korea
Email: [email protected] (KSK), [email protected] (JWK)
Abstract— In this study, replacing toxic organic solvents, we
use deep eutectic solvents(DESs) and ionic liquids(ILs) to
extract cellulose. The extracted cellulose with DESs and ILs is
characterized by scanning electron microscopy(SEM), X-ray
diffraction(XRD) and thermogravimetric analysis(TGA). when
we used DESs and ILs to extract the cellulose, the crystallinity
of extracted cellulose by DESs was higher than extracted
cellulose by ILs. In the result of TGA, it was confirmed that the
pyrolysis temperature of the extracted cellulose with the ILs and
DESs decreased. The result can be interpreted it was confirmed
that when cellulose was extracted with DESs and ILs, structural
characteristics of cellulose changed depending on the type of
solvents and thermal properties of regenerated cellulose was
decreased the pyrolysis temperature with DESs and ILs. The
experimental results from this study would be helpful for a
better understanding of the various solvents for cellulose and
increase the applicability of biomass to various fields.
Keywords—cellulose, deep eutectic solvent, ionic Liquid,
regeneration
- 495 -
Implementation of carbon neutrality through
the RE100 verification system in the agricultural
sector
Minseong Kang 1, Sangho Park 1* and Kihyun Kim2, Kyungwan Koo2
1
Korea Testing Laboratory, Korea
2
Hoseo University, Korea
Email: [email protected]
Abstract—Various global campaigns were born according 409 members have participated in the campaign so far, and
to the global carbon neutral stance. Among them, RE100 most of them are well known companies. Thanks to the
moved global companies through the union of the two non- trend of the carbon neutral, that activate renewable energy
profit organizations, and the global companies implement market and as shown below, it can be seen that the current
RE100 and they requesting vendors to join the campaign. members of RE100 are on the rise and the corresponding
RE100 in the Industrial sector has been progressing at a rapid RE achievement rate, is continuously increasing.
pace. And solving the climate crisis, which started from the
Kyoto protocol, was continuously worked at the adoption of
the Paris Climate Agreement in which all 195 UN Parties
participate in 2015. Following carbon neutrality in the
industrial sector, the agricultural sector is no exception. We
studied ways to implement and verificate RE100 in order to
expand and apply carbon-neutral steps in the industrial sector
to the agricultural sector.
Keywords—Carbon Neutral, Global RE100, K-RE100,
Agricultural RE100, RE100 Verification
I. INTRODUCTION
Fig. 1. RE100 Members and energy consuption
The Kyoto Protocol began in 1997 to reduce emissions
of six types of greenhouse gases, including carbon dioxide B. K-RE100
and methane, to prevent global warming. In 2015, 195
countries participated in the plenary session of the 21st Match with the global RE100, Korea renewable energy
United Nations Framework Convention on Climate market also gained momentum and was born as the K-
Change(COP21) in Paris. These efforts has led to several RE100. This campaign established for Korea companies to
climate change campaigns, and so far, the RE100 been respond the Global RE100. The biggest difference thing is
recognized as a representative global company-led carbon participant condition. More about that will talk about next
neutral effort. This flow in the industrial sector begin to be subject. As a result, various companies such as public
required from global companies to the subcontractors, and institutions and companies joined and so far 157 members
have to join the RE100 campaign to contract with them. And on it. This is about 3.8 times more than the 32 currently
now, efforts are needed to implement carbon neutrality in participating in the Global RE100. This phenomenon
the agricultural sector following carbon neutrality in the suggests that Korea, as an export-based manufacturing
industrial sector. Typically, for carbon neutral in country, is sensitive to issues that effect GVD, such as the
agricultural and livestock sector, which using a lot of power RE100 campaign(except for government-related
to dry and maintain temperature, we selected a pig breeding organizations), and that many global companies are putting
house test bed and conducted a study for the application of pressure on domestic suppliers to join RE100.
RE100. C. Difference of Global RE100 vs K-RE100
II. GLOBAL RE100 AND K-RE100 K-RE100 is little bit different with Global RE100. The
biggest point is participants, that most of the RE100
A. Global RE100 members are based on the power consumption more than
The RE100 campaign was launched in 2014 as a 0.1 TWh or selected as a Fortune 1,000 company. On the
collaboration of the Carbon Disclosure Project(CDP) and other hand, K-RE100 has no restriction. Also, in the case of
the multinational non-profit organization The Climate former, implementation goal are set and recommended for
Group. Starting with 14 companies such as Ingka Group and 2030, 2040 and 2050. Latter, only needs to be satisfied by
Nestle in 2014, a large number of global companies such as 2050 of 100% of renewable energy. The middle step leave
Google, Adobe and Microsoft participated in 2015. The it to the company by autonomous. When the company reach
influence of global companies participation has been the target point they have different recognition system that
boosted by the movement toward global carbon neutrality, use some platform in case of K-RE100. In addition, the
and some contracts concluded only when vendors planning scope is different that former is for global business sites and
or join the RE100 campaign at least. Through there100.org, latter limited to domestic only. See the table below.
- 496 -
TABLE I. COMPARISON RE100 AND K-RE100 international standards, it is necessary to establish an
independent quality system that reflects the requirements of
Global RE100 K-RE100 ISO/IEC 17029, which is configured as follows according
Influential companiess, to KTL regulations.
such as companies that All including industrial,
consume more than 0.1 general and KTL's quality documents are classified and linked in
Participants vertical and horizontal systems. Vertically, it is composed
TWh of electricity per agricultural(partly)
year or Fortune 1,000 electricity consumers of a total 5 steps: ‘company manual’ > ‘manual by sector’ >
companies ‘procedure’ > ‘guideline’ > ‘work standard’. It is divided
Green Premium System Green Tariff into 'common documents' that define the overall operational
REC Purchase REC Purchase
Method Third-party PPA Third-party PPA
direction and common functions of the horizontally
Captial Investment Direct PPA structured institution, and 'individual documents' that
Self-generation Self-generation describe differentiated contents by field. In the case of
Geothermal, solat(heat), Sane as left renewable energy, in addition to common documents, ‘1
Energy tidal, hydro, wind, Biomass + fuel cell manual for verification field / 1 procedure (individual) / 4
Source biomass(gas), green *GHG reduction recognition:
guidelines (individual)’ were separately prepared and
hydrogen fuel cells solar(heat), wind, hydro
Fulfillment 2050 100% RE100 2050 100% RE100 constituted, and a total of 23 types of quality documents are
Goal 2040 90% / 2030 60% Middle step autonomous used.
Scoope Wolrdwide offices Domestic only
After registering the
Report renewable
usage in the K-RE100
energy usage to CDP
Report management system,
once a year
receiving a certificate of
(submit spreadsheet, etc)
renewable energy use
Domestic RE100
declaration and external
publicity, use of
Global RE100 declation
Utilization greenhouse gas reduction
and publicity
performance and global
RE100 performance
(planned)
One thing that K-RE100, which was originally available
only to industrial and general electricity consumers, has
amended that can be implemented by who currently use
agricultural electricity. Those kind of happened because of
the Korea electricity consumption structure. Fig. 3. Mamnual and Procedure for validation and verification
III. K-RE100 VERIFICATION AND CERTIFICATION SYSTEM Based on the internationally used carbon footprint
verification system of the KOLAS verification system, the
Carbon neutrality and energy-related verification fields
K-RE100 verification system will be established as follows
have been announced as new introductions in Korea. As
to meet the main requirements of the international mutual
follows, according to National Agency for Technology and
recognition system.
Standards notice No.2023-0702, the field and scope of
accreditation of environmental information and energy
management accredited institutions have been expanded by
the introduction of a validation and verification system.
Fig. 4. Operating system of RE100 Verification
Fig. 2. MLA Conformity Assessment In accordance with this system operation, the K-RE100
verification system will be managed by the Korea Energy
As a result, ISO 14065, the existing V&V institution Agency as the Scheme Owner, and KTL will proceed with
operation standard, was downgraded from Level 3 to Level the verification. After that, the Scheme Owner for the
4, and the contents were expanded to include greenhouse Global RE100 connection is also the Korea Energy Agency,
gas and other environmental information. In Level 3, and KTL is in charge of verification, so that the system
ISO/IEC 17029, the standard for all V&V agency operations operation and verification system will be equipped.
including environmental information, was enacted and
placed. Therefore, in order to provide conformity
assessment services in the verification field that conform to
- 497 -
IV. TEST-BED FOR RE100 EXEMPLICATION The hydrogen fuel cell system has a maximum monthly
power generation of 72,000 kWh and a gas consumption of
A. General of Test-bed 68 Nm3/h.
We selected the test bed of Mundong Farm, located at
610-19 Geumak-ri, Hallim-eup, Jeju-si, measured the power TABLE II. H2 AND POWER GENERATION COMPARED TO INPUT
consumption for RE100, and conducted demonstration
Input of pig H2 Day Year
using this site. This test-bed is equipped with a pig breeding
manure Generation Generation Generation
facility with a capacity of 5,000 heads and a manure
(ton) (Nm3/h) (kWh) (kWh)
treatment facility.
1 34.8 1,781 427,440
2 69.54 3,562 854,880
5 173.88 8,906 2,137,440
In the case of a single drying facility in the farm, as
measured by a power meter, it consumed about 25,692 kWh
of electricity per year. Total electricity consumption
according to KEPCO's electricity bill was about 980,925
kWh per year in 2022. This is a value corresponding to
RE87 of the total power when 2 ton is input, and in the case
of a single drying facility, a value that is satisfied with 1 ton
Fig. 5. Demonstration farm breeding facility
comes out. Based on this, when the power conversion factor
0.4747 tco2/MWh is applied according to the input of 2 tons
The Test-bed is equipped with infrastructure that can of pig manure, it will have an effect of reducing CO2 by
conduct demonstrations for eco-friendly treatment of about 405 tons per year.
livestock manure and realization of RE100 by establishing
V. CONCLUSION
livestock manure energy resource infrastructure also
greenhouse gas and energy environment integrated New policies and regulations are being established for
management monitoring system as shown in the following each country to achieve the NDC goal, and international
figure. efforts are being made to keep the global average
temperature increase below 2°C compared to pre-industrial
levels, and further limit the temperature increase to 1.5°C or
less. RE100, which is concentrated in the industrial sector,
as we demonstrate the effect of reducing carbon in the
electric power sector and acceleration carbon neutrality in
the agriculture and forestry sector, a verification and
certification system will effectively achieve the transition to
renewable energy. And then in the entire energy sector will
be expanded.
Based on the research tasks achieved so far, the
applicability of RE100 for agriculture was verified, and
based on the results, research and demonstration of energy
resource utilization using livestock manure will be
continued. In addition, based on this, it is necessary to
Fig. 6. Structure of Monitoring System promote the expansion of supply so that RE100 certification
can be operated in the agricultural and livestock sector.
The compost and liquid manure treatment system goes
through the following process. The compost and liquid ACKNOWLEDGMENT
manure are separated through a centrifugal separator, and This work was supported by Korea Institute of Planning
the liquid manure is reprocessed and reused in the farm. The and Evaluation for Technology in Food, Agriculture and
separated pig manure is directly transported to the fermenter Forestry(IPET) through Livestock Industrialization
storage area, spread evenly, and then agitated and mixed Technology Development Program(or Project), funded by
using an agitator to ferment and dry the pig manure for a Ministry of Agriculture, Food and Rural Affairs(MAFRA).
certain period of time. After then it transport to a public (322102-03)
treatment facility.
B. RE100 demonstration of livestock manure resoure REFERENCES
[1] Park Sang-ho, “A study on the Domestic and Foreign RE100 Trend
In the demonstration, a 100 kW PEMFC hydrogen fuel and Verification System,” Proceedings of the 2022 Fall Conference
cell is operated using H2 generated by utilizing the livestock of the Electrical Equipment Division of the Korean Electrical
manure energy resource system as an energy source. At that Society, 2022.
time, the representative chemical formula for gasification of [2] RE100 ANNUAL REPORT, 2022.
pig manure is as shown in (1). [3] Lee Min-ah, “Preliminary Study on the RE100 International
Recognition Plan,” Proceedings of the 2022 Fall Conference of the
Cx, Hy, Oz + mH2O = xCO + nH2 (1) Electrical Equipment Division of the Korean Electrical Society,
2022.
- 498 -
[4] Lee Jung-Hoo, “Preliminary Study on Establishment of the RE100
Certification System,” Proceedings of the 2022 Fall conference of
the Electrical Equipment Division of the Korean Electrical Society,
2022.
[5] Kang Min-Seong, “A Study that Certification RE100 of Pig
Breeding House by Utilization Livestock Manure”, The
Transactions of the Korean Institute of Electrical Engineers, vol. 72,
no. 3, pp. 453~458, 2023
- 499 -
Study on the collapsing of artificial retaining
wall facility using acoustic emission
Kwang Bok Kim 1, Bong Ki Kim 1, Woon Kyung Kim 1, Tae Hun Kim 1,
Sung Hun Park 1, Saurabh Kumar 1, Jun Gil Kang 1 , Sang-Tae Han2 and Kyung-Wan Koo 2*
1
Integrity Diagnostic Korea Co. Ltd., Republic of KOREA
2
HoSeo University, Republic of KOREA
Email: [email protected]
Abstract—This paper is a study of the process of retaining pressure of the soil. After soil compaction, nine 1m-long
wall collapse using the AET technique. Pile type AE Sensor pile-type sensors were installed at 1m distance from each
specially designed to collect AE data during the collapse of the other. These pile-type sensors were designed to detect micro-
retaining wall was manufactured. The artificial reinforced displacement-induced elastic waves by attaching
retaining wall was made to be 27 square meter and height 3 piezoelectric ceramics to the sensor line, which resonates at a
meter. It consists of reinforced soil and concrete blocks that are maximum resonance at 60kHz.
most commonly used for reinforced soil retaining wall. The pile
type sensor was installed vertically in retaining wall at a depth a) AE Data collection
of 1 m and connected to the AET system using a cable. AE data The acoustic emission testing collection device (IDK-
was collected and AE parameters were converted by collapsing PU16) developed by IDK was waterproofed and placed in a
into forklifts. The signal strength in AE parameters showed the
waterproof box for real-time data collection. The data
best AE characteristics during the collapse process. After
applying the 2D CNN model of artificial intelligence, 85% of collected from the file-type sensors were amplified through a
the five-step collapsing process could be confirmed. (Abstract) preamplifier and amplified with a gain value of 40dB. The
AET system, equipped with two PCIe slot-mounted DAQ
Keywords— Acoustic emission testing, retaining wall, data boards (8-channel type), measured and automatically saved
collection, prediction (key words) the signals flowing from the nine AE sensors to a PC. The
data were collected at a maximum speed of 10Msps for each
sensor. The collected AE data were analyzed using the AE
I. INTRODUCTION Studio software developed by IDK, after removing
The application of inclinometers, subsurface environmental noise. The timing control function values
inclinometers, displacement gauges, drones, and imaging during data collection were set according to Table 1.
techniques in the prediction of wall and slope collapse has
been increasing. Real-time digital data conversion is
necessary to enable the transition to early warning systems
by detecting small ground movements. To develop a
prediction technique for wall collapse using acoustic
emission testing (AET), a pile-type AE sensor was developed
[1]. This sensor is a real-time system that is able to detect
ground movements by drilling and inserting it deep into the
ground and transmitting micro-elastic waves to provide
warnings of the collapse point. In this study, a reinforced soil [Fig. 1] Retaining wall testbed
wall(retaining wall) was constructed in an outdoor test bed as
a preliminary step of empirical research, and AE data related
to the collapse process were collected to establish a collapse
prediction based on AE feature values. Table 1: AE Data collection parameters
Parameter Value Parameter Value
II. EXPERIMENT AND RESULTS P/A gain (dB) 40 HDT (us) 400
A. Reinforced retaining wall sampling rate (MHz) 10 PDT (us) 200
Reinforced soil walls were constructed in a designated Pre-trigger 400 MHD (us) 10,000
area, with two sets of 2m x 2m and one set of 3m x 3m as
shown in Fig. 1. The exterior was made of concrete blocks, Threshold (dB) 30 HLT (us) 4,000
and three types of reinforcing soil were compacted inside.
HPF (kHz) 50 Noise AVG TC (us) 100,000
Pressure gauges were installed to measure the overall
- 500 -
LPF (kHz) 300 Signal AVG TC (us) 100,000
[Fig. 3] Retaining wall collapse procedure
B. characteristics during wall collapse During the wall collapse, the signal strength data from
The AET system was installed in the outdoor test bed and AE showed significant increases each time the forklift crane
observed using a real-time web monitoring system. CCTV made contact with the wall and exerted force during the five
cameras were installed for safety reasons, considering stages of collapse, corresponding to the movement of the
potential hazards, and environmental data were collected reinforced soil. Figure 4 depicts the relationship between
during the summer season, including typhoons, rainfall, wind, slope displacement and AE data by installing an
etc. Subsequently, during the winter season and thawing inclinometer sensor. The lower part of Figure 4 shows that
period, the wall was artificially collapsed, and AE data were when stress is applied to the wall due to forklift crane
collected and analyzed in March 2023. contact, the initiation of horizontal loading, and the
completion of horizontal loading, the AE data increase
vertically. This becomes essential data for developing
a) Environmental data and AE integrity indicators for wall collapse.
The relationship between AE signal strength values
and precipitation, wind and temperature were investigated.
The AE signal strength values increased with precipitation, c) HS Index
as the top part of the pile-type sensors did not submerge The zonal intensity method evaluates the structural
entirely into the reinforced soil. However, the AE values severity of defects by determining the intersection of two
showed little variation with wind and temperature. independent parameters: the history index (HI) and severity
index (SI) [2].
N
N ∑ Soⅈ
t= k +1
H (t )= N
N− K ∑ Soi
i= 1
N= 10
1
Sav=
10
∑ Soi (1)
i= 1
Where N is the hit count up to time t, Soi is the signal
strength of the i-th hit, and K is defined as follows:
If N > 1000, K = N - 200
If J ≤ N ≤ 1000, K = 0.8N
[Fig. 2] AE data variation according to precipitation, wind If N < J, historical index not applied
J is a specific event count, which may vary depending on the
and temperature test material.
b) Wall collapse process
Figure 3 illustrates the images of the wall collapse
process. The wall was artificially collapsed using a forklift
crane with an 800kg load. AE data were collected
throughout this process.
[Fig. 4] Angle sensor and AE data relation
- 501 -
Figure 5 shows the trend of hit count intervals for
generating the H index every 5 seconds during the collapse
process. The collected hit rates for each of the ten intervals
were used to calculate the HS index using Equation (1).
[Fig. 5] HS index for retaining wall collapse
As shown on the right side of Figure 5, the HS index
provides a comprehensive view of the wall collapse process,
allowing prediction of the collapse process into six stages,
labeled A to F.
III. CONCLUSION
The entire process of wall collapse was performed in an
outdoor test bed using acoustic emission technology.
Despite the influence of precipitation, wind, and temperature,
environmental factors were easily eliminated during AET
data collection. Clean AE data collected during the collapse
process were secured by utilizing the AET system,
commercially developed by IDK, from collection to signal
processing. The collapse algorithm was completed using AE
data from different collapse stages. Additionally, an
integrity index was proposed and developed, which is
expected to contribute significantly to the future
development of slope collapse prediction technologies.
ACKNOWLEDGEMENT
This paper presents the research findings (Project Unique
Number 1414183064) conducted with the support of the
Ministry of Industry, Trade, and Resources and the Korea
Institute for the Advancement of Technology under the
'Chungnam Dangjin Industrial-Academic Convergence Zone
Industrial-Academic Convergence Promotion Support
Project (No. P0016967).
REFERENCES
[1] ASNT, NDT Handbook Vol 6, 315p
[2] ASNT, NDT Handbook Vol 6, 121p
- 502 -
Estimation of AIS-based ship GHGs emission
using PAQman System : ⓒ
focusing on Pyeongtaek·Dangjin Ports
Yongchan Lee*, Heekwan Lee*, CheolGyo Yim**, KyungWan Koo***
Incheon National University*,JIN Electrical Engineering Co.,Ltd**, Hoseo University***
South Korea
Abstract – This research estimates the GHGs emission from
ocean-going vessels approaching into Pyeongtaek·Dangjin Ports
for October 2021, while applying the AIS-based algorithm. Based
on the estimated ship GHGs emission inventory, the high-
resolution ship GHGs emission map is produced for the waters
near Pyeongtaek·Dangjin Ports. The results of this study could be
utilized to strengthen of the regulations on the international
shipping GHGs emission. Moreover, the results of this study
could provide a great deal of potential to achieve sustainable port
city management worldwide. (Abstract)
1. INTRODUCTION
Ocean-going vessels have been taking a long-lasting role
in international logistics, and as maritime trade market is Figure 1. Flow chart for marine vessel emission by PAQman ⓒ
rapidly expanding, its economic and social influence is also
increasing. However, as the environmental burden on various
types of air pollutants and greenhouse gases such as carbon 2.2 Port-city Air Quality management system (PAQmanⓒ)
dioxides, sulfur oxides, and fine dust from vessels attracts Port-city Air Quality Management Systemⓒ (PAQmanⓒ)
international attention, the effective management of ship air is the AIS based ship air and GHGs emission estimation
emission and improvement of air quality in the port city is system developed by Air / Climate Group of Incheon
also an emerging issue as of strong social concerns. In
National University, and PAQmanⓒ is the technology that
particular, the European Commission (EC) is planning to
enables the accurate ship air pollutants and GHGs emission
expand the scope of the EU Emission Trade Scheme (EU-
inventory in a form of high-resolution space emission
ETS) to international shipping in order to manage and reduce
mapping.
greenhouse gas emissions from ships. In order to implement
the emission trading scheme, it is very important to identify PAQmanⓒ collects AIS data for estimating ship
the exact emission of each company or country, and it is emissions by using PAQman-watcherⓒ, AIS data receiver,
essential to secure a reliable spatial emissions inventory and creates the database for emission inventory. In this study,
considering the characteristics of ships traveling across PAQmanⓒ is used to estimate the ship GHGs emissions in
multiple waters. Pyeongtaek·Dangjin Ports.
2. BODY
2.1 AIS based ship GHGs emission estimation methodology
The ship emission estimation methodology using AIS
(Automatic Identification System) data, unlike the existing
fuel consumption-based method, estimates the ship emission
by considering individual movements and specifications of
each vessel, so it could be used as a useful means for the
regional GHGs emission management in port cities. In
particular, it is possible to produce a high-resolution emission
map through accurate spatial allocation of emissions based
on AIS data.
ⓒ
Figure 2. Systematic concept of PAQman System
- 503 -
2.3 Study area and data collection) In addition, it was found that most of the emissions in the
We directly collected AIS data (Oct. 1 to 30, 2021), area near Pyeongtaek Port, where a large amount of CO2
which is ship activity data, to estimate ship air emissions, and was emitted, were reduced by the preparation of emission
the AIS receiver was installed at Hoseo University campus. maps according to the AMP application.
This study conducted the analysis of ship GHGs emission
from ocean-going vessels entering and departing Pyeongtaek
and Dangjin ports for one month. Among them, the CO2
emissions were estimated and analyzed for ocean-going
vessel operated within the emission control areas (ECAs) of
Pyeongtaek and Dangjin ports stipulated by the Vessel
Traffic Service System being operated by the Korea Coast
Guard.
(a) CO2 emssion map (BAU)
(b) Control area of
(a) Visualization of AIS data
Pyeongtaek·Dangjin ports
Figure 3. Visualization of AIS data and the emission control area
(ECA) of Pyeongtaek·Dangjin ports
2.4 Result of ship GHGs emission inventory by PAQmanⓒ
The CO2 emissions from the ocean-going vessels in
Pyeongtaek·Dangjin Ports during October 2021 are as
follows in Table 1. Currently, Pyeongtaek·Dangjin Ports are
desiginated and employed for the emission control areas
(ECAs), so the emission was estimated based on the ECAs,
and the expected emission due to the implementation of
Alternative Marine Power Supply System (AMP) scheduled
for additional introduction was estimated. (b) CO2 emission map
(AMP)
Table 1. CO2 emission from ocean-going vessels in Figure 5. High-resolution ship CO2 emission map by reduction
Pyeongtaek·Dangjin ports by PAQmanⓒ policies in Pyeongtaek·Dangjin ports
(unit : kg/month)
BAU* AMP
Region
CO2 CO2
Pyeongtaek 10,095 3,593 3. CONCLUSION
Dangjin 5,548 5,025
Total 15,643 8,618 In this study, the AIS-based algorithm was used to
* BAU : Business As Usual
estimate the CO2 emission and produce the high-resolution
ship emission map at Pyeongtaek·Dangjin ports. Furthermore,
As the result of estimating CO2 emissions from ships based on the emission map, the reduction potential of the
within the emission control areas of Pyeongtaek·Dangjin ocean-going vessel entering and departing from
Pyeongtaek·Dangjin ports was evaluated based on the
Port, it was estimated that the CO2 before AMP application
implementation of the reduction policy. As the result of the
wwas emitted of 15,643 kg. It was also analyzed that the
evaluation, it was analyzed that approximately 55% of CO2
CO2 was reduced by 55% compared to BAU under AMP emitted from ocean-going vessel operating
application. Pyeongtaek·Dangjin ports was reduced. In addition, by
creating a high-resolution ship air emission map based on
AIS data, AIS-based algorithm was potential to accurately
2.5 High-resolution ship air emission map by PAQman allocate spatial mapping. Through this, it was possible to
The emission in 2.3 was allocated based on the location evaluate the amount of reduction expected from the
coordinates according to the ship operation included in the introduction of a more accurate reduction policy. Therefore,
AIS data. The spatial domain was set to Pyeongtaek·Dangjin the results of this study could be expected to contribute to the
Ports emission control areas disclosed by the Korea Coast effective reduction of greenhouse gas emissions in port cities
Guard. As the result, most of the emissions were emitted on and the establishment of policies to respond to climate
designated routes, especially near Pyeongtaek Port, change.
approximately 50% of CO2 emission was discharged.
- 504 -
ACKNOWLEDGMENT
This work was supported by Korea Institute of Planning
and Evaluation for Technology in Food, Agriculture and
Forestry(IPET) through Livestock Industrialization
Technology Development Program(or Project), funded by
Ministry of Agriculture, Food and Rural Affairs(MAFRA).
(322102-03)
REFERENCES.
[1] Johansson L, Jalkanen J-P, Kukkonen J., “Global assessment of
shipping emissions in 2015 on a high spatial and temporal resolution”,
Atmospheric Environment, 167:403-15, 2015
[2] EEA, “EMEP/EEA Air Pollutant Emission Inventory Guidebook
2016: technical guidance to prepare national emission inventories”,
EEA-Report:21, 2016
[3] Kwon Y, Lim H, Lim Y, Lee H., “Implication of activity-based vessel
emission to improve regional air inventory in a port area”,
Atmospheric Environment, 203:262-70, 2019
[4] Kwon Y., “Study on the Air Quality Impact of Shipping Fine Dust
Emission and Its Inventory Methodology in a Port-City”, Graduate
School, Incheon National University, 2019
[5] IMO, “Fourth IMO Green House Gas Study, 2020”, International
Maritime Organization, 2020
[6] Lee Y., “Study on the applicability of AIS-based ship air emission
inventory methodology”, Graduate School, Incheon National
University, 2021
[7] 국가미세먼지정보센터, “2019 년 국가 대기오염물질 배출량”,
2022
[8] 해양경찰청 전국 VTS 센터, “평택항 VTS 관제구역도”
[9] 이희관, 임효지, 이용찬, “항만 대기오염 배출 관리 장치 및
방법”, 등록 특허 10-2300919, 2021
¹⁾AIS-based algorithm: the bottom-up estimation approach using
the ship activity data collected from Automatic Identification
System (AIS) (Tier 3 approach) by PAQman, (4th IMO GHGs
study, 2020)
- 505 -
Development of a Waste Plastic Energy
Resource System by Integrating Pyrolysis and
High-Temperature Reduction Processes
Song Jea Jun¹, Kyung Wan KOO²
¹HOWON CO., LTD PRESIDENT. ²Hoseo University, Korea
Abstract— As plastic consumption increases worldwide, the continuous waste plastic pyrolysis process were investigated,
amount of waste plastic generated tends to increase, and (2) the possibility of hydrogen production was confirmed by
accordingly, appropriate recycling methods are urgently high-temperature reduction treatment of black carbon, which
needed. Currently, more than half of waste plastics are is a residue among the produced materials, (3) Finally, the
landfilled or incinerated, but diversification of recycling carbon reduction amount was calculated during the pyrolysis
methods is being carried out due to environmental problems. treatment compared to the incineration of the existing waste
Therefore, in this study, performance evaluation and by- plastics
product utilization methods were studied for the continuous
pyrolysis method, which is a chemical recycling method among
waste plastic recycling methods. For the study, a 100 kg/d
continuous waste plastic pyrolysis device was constructed and
operated 24 hours a day, 5 times. As a result of the study,
2. MATERIAL AND METHODS
59.3±1.2% of pyrolysis oil, 16.8±0.8% of black carbon,
16.3±1.5% of non-condensable gas, and 7.6±0.7% of 2.1 Waste plastic
wastewater were generated during the pyrolysis of waste
plastic. In addition, as a result of performing calorific value
The raw material of waste plastic used in this study was 5
and hydrogenation for resource recycling of generated black mm size waste plastic generated from crushing and crushing
carbon, the high calorific value was 4,391.3 Kcal/kg and the facilities. The table 1 below shows the raw material
low calorific value was 4,063.3 Kcal/kg. was found to be able to composition of waste plastics in the relevant crushing and
produce 13.20 L/min (19.01 Nm3/d). In addition, it is thought crushing facilities.
that it is possible to reduce about 16.0% of carbon emissions
per year when the pyrolysis method is applied compared to
incineration. (Abstract)
TABLE1. COMPOSITION OF WASTE PLASTIC
Keywords— Waste plastic, Continuous pyrolysis, Pyrolysed polyethylene
oil, Black carbon, Hydrogen production (key words) polyethylene polypropylene polystyrene
terephthalate Others
(PE) (PP) (PS)
(PET)
1. INTRODUCTION
Due to the increase in plastic consumption worldwide, 39.6% 27.5% 18.9% 9.7% 4.3%
more than 300 million tons of waste plastics are generated
every year, but despite efforts for recycling, more than half of
them are not collected and are landfilled or incinerated in the
form of waste [1]. Currently, the most discharged plastics can 2.2 Configuration of continuous pyrolysis experiment
be classified into polyethylene (PE), polypropylene (PP), equipment
polyvinyl-chloride (PVC), polyethylene terephthalate (PET),
and polystyrene (PS) [2]. Recycling methods for these waste The main components of the continuous pyrolysis
plastics include (1) a method of mechanically recycling process for waste plastics are divided into the input part, the
recovered waste plastics by sorting them, (2) manufacturing continuous pyrolysis process, the specific gravity separation
them in the form of R.P.F (refuse plastic fuel) or R.D.F process, and the condensation process. The processing
(refuse derived fuel) and thermally processing them in steel capacity per hour of the continuous pyrolysis process was
mills or incinerators. (3) chemical methods that produce fuels 4.17 kg/hr, and it was manufactured in a lab-pilot scale
and chemicals through processes such as pyrolysis or capable of processing 100 kg per day and used for the
gasification [3]. Pyrolysis, a chemical recycling method, is a experiment. The input of raw materials to the continuous
method of producing liquid oil that can be used as fuel from pyrolysis furnace was fed in a fixed amount through a screw,
plastic, that is, pyrolysis oil, through a chemical reaction in and the heating of the pyrolysis device was maintained at a
which waste plastic is reduced and converted into low- constant temperature around 350 ~ 400 ℃ using thermal oil.
molecular compounds under medium-temperature anoxic In addition, a specific gravity screening process was used to
conditions in the range of 400 to 600 °C [4]. . In this study, remove the wax component contained in the pyrolysis oil
(1) the composition and yield of products generated in the vapor, and the wax component was put into the pyrolysis
- 506 -
process again and the apparatus was configured with a 3. RESULT
structure in which it was discharged as black carbon. The
condensation process is a device for recovering liquid oil by 3.1 Composition and content of pyrolysis products
cooling the vapor to below 60 ° C. A water bath is used to
circulate the cooling water. Figure 1 below shows a
schematic diagram of the experimental device, Table. 2 3.1.1 Pyrolysis oil production and wastewater generation
summarizes the operating conditions of the experimental In order to confirm the production of pyrolysis oil
device. generated in the continuous pyrolysis process of waste
plastics, 5 repeated experiments were conducted. The input
amount of waste plastic is 100 kg/d, and about 66.9±0.9 kg/d
of liquid oil was produced during a total of 5 experiments.
The amount of separated wastewater was confirmed to be 7.6
kg/d, and about 11.4% of the liquid oil produced was
generated as wastewater. The average amount of pyrolysis oil
in liquid oil was 59.3 ± 1.2 kg/d, and about 60% of pyrolysis
oil was produced compared to the input amount of 100 kg/d.
picture below. Figure 3 shows the amount of liquid oil,
wastewater, and pyrolysis oil produced in the condensation
process according to repeated experiments.
[Fig. 1] Schematic diagrams of continuous pyrolysis system
TABLE2. OPERATING CONDITION OF CONTINUOUS PYROLYSIS PROCESS
Division Unit Contents
Processing Capacity kg/hr (kg/d) 4.17 (100)
Operating Time hr 24
Pyrolysis Temperature ℃ 350~400
[Fig. 3] Production of Liquid Oil, Wastewater and Pyrolysed Oil
Operating Pressure kgf/cm2 0.2~0.3
Rotation speed rpm 0.5 3.1.2 Black carbon production
As a result of checking the amount of black carbon
generated in the continuous waste plastic pyrolysis process
during the five experimental periods, an average of 16.8±0.8
2.3 Composition of high-temperature reduction treatment kg/d of black carbon, a solid residue, was produced when
system test equipment 100 kg/d of waste plastic was input. The figure shows the
amount of black carbon produced according to the
experiment. 4. Elemental analysis was conducted to
determine the possibility of using black carbon as a resource
as energy, and the results of elemental analysis are shown in
the table below. It is summarized and shown in 3.
[Fig. 2] Process diagram of black carbon energy resource system
[Fig. 4] Production of Black Carbon
- 507 -
TABLE3. ELEMENTAL COMPOSITION OF BLACK CARBON TABLE4. COMPOSITION OF NON-CONDENSIBLE GAS
구분 C H O N S Ash Moisture Division Propene Propane Ethane 1-Butene Butane Others
% 40.77 6.07 23.28 1.63 0.58 27.63 0.04 Content
60.5 9.3 8.2 3.1 1.7 17.2
(%)
In order to determine whether it can be used as a fuel
through the analysis results of black carbon elements 3.2 Black carbon utilization for energy
specified in 4, the high heating value and the low heating
In order to check the amount of hydrogen that can be
value were calculated based on the Durong formula. The
used as energy after high-temperature and reduction
calculation formulas are shown in Equations (1) and (2)
treatment of black carbon generated in the continuous
below. As a result of calculating the high and low heating
pyrolysis process of waste plastics, a high-temperature and
values based on the Durong formula, the high heating value
reduction treatment system experimental device was
was 4,391.3 Kcal/kg and the low heating value was 4,063.3
constructed and used for the experiment. The gasification
Kcal/kg, confirming that it can be used as an auxiliary fuel in
reaction in the high-temperature reduction treatment system
incinerators and calcination furnaces.
proceeds based on the following formulas (3) and (4). The
gasification reaction by water vapor proceeds as an
endothermic reaction as shown in Equation (3) below at
1,000℃ or higher.
식(1)
식(2)
C + H2O = CO + H2 ; △H = 31.4 Kcal/mol 식(3)
In the gasification reaction by CO2, the higher the
3.1.3 Amount of non-condensable gas production temperature and the lower the pressure, the more the reaction
The amount of non-condensable gas generated in the proceeds.
continuous waste plastic pyrolysis process was estimated
using the difference between the amount of liquid oil and
black carbon in the total amount of input. It was found that C+ CO2 = 2CO ; △H = 38.2 Kcal/ mol 식(4)
non-condensable gas was produced. The figure shows the
generation of non-condensable gas. 4. In order to confirm
whether non-condensable gas can be used as an alternative Black carbon generated when 100 kg/d of waste plastic is
fuel such as LPG or LNG, non-condensable gas was injected is about 16.8 kg/d, and when the result of elemental
collected and component analysis was performed. As a result analysis and the above reaction formula are followed, the
of component analysis, the main components of the non- syngas generated after high-temperature reduction resource
condensable gas were C3 ~ C4 propane, propylene, butane, conversion is 10.26 L/min for CO and 7.95 L for H2. /min,
butylene, etc. It was confirmed that the content was 74% or CO can be converted to H2 through water gas reaction, and
more. The table below shows the results of component when the conversion rate is 98.0%, CO 10.26 L/min is
analysis of non-condensable gases. converted to H2 10.06 L/min. Therefore, the total hydrogen
gas produced is 18.01 L/min. However, the temperature of
the high-temperature reduction reactor must be maintained at
1,300 ° C and the amount of heat required for the reaction
must be supplied. If this is supplied as hydrogen gas, the
required amount of hydrogen gas is 4.81 L / min. Therefore,
it was confirmed that the final hydrogen gas can be produced
at 13.20 L/min (19.01 Nm3/d).
3.3 Carbon reduction
Greenhouse gas emissions were calculated based on the
IPCC guidelines in order to confirm CO2 emissions during
pyrolysis compared to incineration, which is the most widely
used method of treating waste plastics. GHG emissions were
calculated assuming that 100 tons of waste plastics are
[Fig. 5] Production of non-condensibla gas processed per day, and the amount of GHGs generated when
using fuel (LPG) and the amount of GHGs generated by
converting waste plastics into substances were added
together. . In addition, additionally generated CH4 and N2O
were calculated by converting them into tCO2 by applying
global warming potentials of 21 and 310, respectively. table
below. Figure 6 summarizes the calculation results of CO2
- 508 -
emissions from incineration and pyrolysis based on 100 tons
of waste plastic. It is considered that if 100 tons of waste
plastic is treated by pyrolysis, greenhouse gas emissions can REFERENCE
be reduced by about 16.0% compared to incineration. [1] OECD, Improving Plastics Management: Trends, Policy Responses,
and The Role of International Co-operation and trade, Environ. Policy,
20, 2018.
[2] S. Abalansa, B. El Mahrad, G. K. Vondolia, J. Icely, and A. Newton,
TABLE5. COMPARISON OF GREENHOUSE GAS REDUCTION INCINERATION The marine plastic litter issue: A social-economic analysis,
AND PYROLYSIS
Sustainablility, 12, 1-27, 2020.
Globla Warming [3] PWMI, An introduction to plastic recycling, Plastic Waste
Division Incineration Pyrolysis Management Institute, 1-33, 2019.
Potential
[4] D. Czajczynska, T. Nannou, L. Anguilano, R. Krzyzynska H. Ghazal,
N. Spencer, and H. Jouhara, Potentials of pyrolysis processes in the
1. By the waste plastic recycling methods
waste management sector, Energy Procedia, 123, 387-394, 2017.
CO2 emission 15,382.4 tCO2/yr 12,936.6 tCO2/yr -
N2O emission
445.8 tCO2/yr 374.9 tCO2/yr 310
(Exchange the CO2)
2. By the Fuel(LPG) using
CO2 emission 4,200.1 tCO2/yr 3,540.7 tCO2/yr -
CH4 emission
86.7 tCO2/yr 73.1 tCO2/yr 21
(Exchange the CO2)
N2O emission
4.13 tCO2/yr 3.5 tCO2/yr 310
(Exchange the CO2)
Total CO2 emission 20,119.1 tCO2/yr 16,928.7 tCO2/yr -
4. CONCLUSION
The conclusions obtained through the experiment of the
100 kg/d continuous waste plastic pyrolysis device are as
follows.
(1) The materials and contents produced when waste
plastic is pyrolyzed in a continuous manner were confirmed
as pyrolysis oil 59.3±1.2%, black carbon 16.8±0.8%, non-
condensable gas 16.3±1.5%, and wastewater 7.6±0.7%.
(2) Black carbon generated in the pyrolysis process was
treated at high temperature and reduced to produce hydrogen,
which is energy, to be used as a resource. It is considered that
the final hydrogen gas can be produced at 13.20 L/min
(19.01 Nm3/d) when 16.8 kg/d of black carbon is subjected
to high-temperature reduction treatment.
(3) As a result of calculating carbon emissions in
accordance with the IPCC guidelines in order to confirm the
carbon reduction during pyrolysis compared to incineration,
which is the main treatment method for waste plastics, it is
estimated that about 20,119.1 tCO2 of CO2 is emitted
annually during incineration, and pyrolysis It is expected to
reduce carbon emissions by about 16.0% per year compared
to incineration treatment with about 16,928.7 tCO2 per year.
ACKNOWLEDGMENT
This work was supported by Korea Institute of Planning
and Evaluation for Technology in Food, Agriculture and
Forestry(IPET) through Livestock Industrialization
Technology Development Program(or Project), funded by
Ministry of Agriculture, Food and Rural Affairs(MAFRA).
(322102-03)
- 509 -
You might also like
- Proceedings of The 2nd International Conference On Electronic Engineering and Renewable Energy Systems: ICEERE 2020, 13-15 April 2020, Saidia, Morocco Bekkay Hajji Instant Download100% (3)Proceedings of The 2nd International Conference On Electronic Engineering and Renewable Energy Systems: ICEERE 2020, 13-15 April 2020, Saidia, Morocco Bekkay Hajji Instant Download59 pages
- Proceedings of The 4th International Conference On Electrical Engineering and Control Applications: ICEECA 2019, 17-19 December 2019, Constantine, Algeria Sofiane Bououden Instant Download100% (3)Proceedings of The 4th International Conference On Electrical Engineering and Control Applications: ICEECA 2019, 17-19 December 2019, Constantine, Algeria Sofiane Bououden Instant Download37 pages
- CPM Pert by Sandeep Jyani Sir PDF Notes Civil Engineering PDF FreeNo ratings yetCPM Pert by Sandeep Jyani Sir PDF Notes Civil Engineering PDF Free222 pages
- M. Sc. Mathematics With Computer ScienceNo ratings yetM. Sc. Mathematics With Computer Science95 pages
- Artificial Intelligence in Renewable Energetic Systems 1st Ed Mustapha Hatti PDF DownloadNo ratings yetArtificial Intelligence in Renewable Energetic Systems 1st Ed Mustapha Hatti PDF Download86 pages
- Digital Service Supply Chain Management - Current Realities and Prospective Visions by Badr Bentalha PDFNo ratings yetDigital Service Supply Chain Management - Current Realities and Prospective Visions by Badr Bentalha PDF1,284 pages
- User Guide (Manual) For Exp - COPYLOT EnglishNo ratings yetUser Guide (Manual) For Exp - COPYLOT English51 pages
- Intelligent Communication and Computational Technologies 2017 PDFNo ratings yetIntelligent Communication and Computational Technologies 2017 PDF390 pages
- ICMIMT 24 X ICII 24 Conference Program-5.8No ratings yetICMIMT 24 X ICII 24 Conference Program-5.838 pages
- Beautiful Yellow Flower PowerPoint TemplatesNo ratings yetBeautiful Yellow Flower PowerPoint Templates61 pages
- Youcef Zennir - Artificial Intelligence in RoboticsNo ratings yetYoucef Zennir - Artificial Intelligence in Robotics44 pages
- File Explorer - Window Environment: Submitted by - Ritik Bhadoria BCA (2018-21) - CNo ratings yetFile Explorer - Window Environment: Submitted by - Ritik Bhadoria BCA (2018-21) - C17 pages
- Ntdaswebinar 2024 V 2 Onk 1715260904580No ratings yetNtdaswebinar 2024 V 2 Onk 171526090458036 pages
- FINAL Detailed Conference Program 7th Pak-Turk ESTE 2024No ratings yetFINAL Detailed Conference Program 7th Pak-Turk ESTE 20249 pages
- Call For Papers ETFG 2025 A4 Flyer 20250403No ratings yetCall For Papers ETFG 2025 A4 Flyer 202504032 pages
- Battery Model Parameter Estimation Using A Layered Technique: An Example Using A Lithium Iron Phosphate CellNo ratings yetBattery Model Parameter Estimation Using A Layered Technique: An Example Using A Lithium Iron Phosphate Cell15 pages
- Conference Program: 4 International Conference On Electrical Engineering and Control Applications, ICEECA2019No ratings yetConference Program: 4 International Conference On Electrical Engineering and Control Applications, ICEECA201914 pages
- GHT-200030 How To Start The ControlST Documentation Global SearchNo ratings yetGHT-200030 How To Start The ControlST Documentation Global Search6 pages
- Tatmeen WKI-0062 Technical Guide For Dispensers v2.0No ratings yetTatmeen WKI-0062 Technical Guide For Dispensers v2.027 pages
- 1986 - Diário de Bitita - Carolina Maria de Jesus PDFNo ratings yet1986 - Diário de Bitita - Carolina Maria de Jesus PDF4 pages
- ICEES 2023 - Paper Presentation ScheduleNo ratings yetICEES 2023 - Paper Presentation Schedule15 pages
- Product Data Sheet: 4G LTE, Data Logger, 3 Port PremiumNo ratings yetProduct Data Sheet: 4G LTE, Data Logger, 3 Port Premium3 pages
- Training For SAP ERP in For ApplicationsNo ratings yetTraining For SAP ERP in For Applications4 pages
- Acef Week Events: 4 June 5 June 8 June Monday Tuesday Friday Thursday 7 June 6 June WednesdayNo ratings yetAcef Week Events: 4 June 5 June 8 June Monday Tuesday Friday Thursday 7 June 6 June Wednesday1 page
- SAP HR-Set Up Personnel Action - Configuration Steps100% (2)SAP HR-Set Up Personnel Action - Configuration Steps16 pages
- Emerging Trends in Electric - Mobility & Sustainable DevelopmentNo ratings yetEmerging Trends in Electric - Mobility & Sustainable Development1 page
- Detecting Lateral Movement Through Tracking Event Logs - Version2100% (1)Detecting Lateral Movement Through Tracking Event Logs - Version217 pages
- 7th International Conference On Engineering and Emerging TechnologiesNo ratings yet7th International Conference On Engineering and Emerging Technologies1 page
- First virtual Bilateral Conference on Functional Materials (BiC-FM)From EverandFirst virtual Bilateral Conference on Functional Materials (BiC-FM)No ratings yet
