LINK TẢI LUẬN VĂN MIỄN PHÍ CHO AE KET-NOI
Integrate ANN-GA into improve particle classification algorithm in automotive production moving forward to green manufacturing
ABSTRACT
In pursuit of promoting Green Manufacturing in the automotive industry, this
research integrates Artificial Neural Network (ANN) Genetic Algorithm (GA), called ANN-GA, to enhance the optimization of Particle Classification Algorithm (PCA) was previously developed by Taguchi method, then upgraded by Programming C++ and Matlab optimization function. This innovation comes from the need to solve the problem of automotive machinery small chip lodged and complex particles created in various engine processing processes such as casting, anodizing, machining, High-pressure water jet, anodizing, sandblasting, grinding, polishing, and assembly. These processes will induce the rist of creation of burr, cast and chip,... These particles lodge inside the transmission, engine and crankshaft, and then damage the functions of these components, posing risks to drivers, such as: suddenly increase speed until losing control, mal- function, suddenly stop engine, negative impact to valve control systems called TC21,
TC35, TC 29... which mainly working based on the electro-magnetic oil sensors.
Based on the partical experiment in research of Dr. Phan Quoc Bao, and Professor
Sung Lim không from the high-tech manufacturing laboratory at Hyundai Motor Korea, South Korea, and the Precison Machining Lab, Konkuk University, South Korea, from 2009 - 2015, has developed a particle expert system (PES) that proved the possibility to classify burrs, blanks and chips, by using very basic image processing techniques. An advanced model of PCA, consisting of 12 distinct particle types, was found by using ANN-GA. This new approach does not only trains PCA parameters with higher accuracy but also ensures a stable success rate, thereby improving the reliability of particle source identification processing. Even though the whole project result was statiscal evaluation, it showed up the trend of creating particle from different sources, processes of metal fabrication. These decisions may support the top leader CEO, Managers to make suitable
decision to invest into suitable production technology related to improving Cleanability
This integration of ANN-GA into PCA is a forward-thinking approach to solving
cleanability challenges in automotive manufacturing. It aligns with the principles of Green Manufacturing by providing a methodologically sound, environmentally friendly solution to improve product quality and safety. Here is the discussion from Professor Dornfield at ISGMA 2013 Sheraton Hawaii Waikiki Green Manufacturing International
Conference.
Ho Chi Minh City, Ferbruary 19, 2024
Tran Minh Thuan
TABLE OF CONTENTS
CHAPTER 1: OVERVIEW.............................................................................................................2
1.1. Overview of the research...................................................................................................2
1.2. The need for research ........................................................................................................3
1.3. The goal of the research ....................................................................................................4
1.4. The object and scope of research ......................................................................................4
1.4.1. The object of the research ........................................................................................4
1.4.1. The scope of the research.........................................................................................4 1.5. Approach and research method .........................................................................................4
1.5.1. Approach..................................................................................................................4
1.5.2. Research methods ....................................................................................................4
1.6. Project results ....................................................................................................................4
1.7. Efficiency specification.....................................................................................................4
CHAPTER 2: BASIC THEORIES..................................................................................................5 2.1. Cleanability problems and technology. .............................................................................5
2.1.1. Introduction..............................................................................................................5
2.1.2. Advanced cleaning methods ....................................................................................6
2.1.3. Deburring Processes and Cleanability...............................................................................7
CHAPTER 3: CHARACTERIZATION AND RECOGNITION OF PARTICLES FOR IMPROVING CLEANABILITY IN AUTOMOTIVE PRODUCTION ........................................9
3.1. Mechanism of particles generation....................................................................................9
3.2. Standard for particle classification ..................................................................................13
3.3. Algorithm of classification..............................................................................................15
CHAPTER 4: INTEGRATING ANN-GA INTO THE CLASSIFICATION ALGORITHM ......18 4.1. Introduce ANN-GA.........................................................................................................18
4.1.1. Artificial Neural Network (ANN)..........................................................................18
4.1.2. Genetic Algorithm (GA) ........................................................................................20
4.2. ANN training...................................................................................................................22
4.2.1. Graphical representation of ANN model ...............................................................22
4.2.2. Result and Discussion............................................................................................24
4.3. Use GA from ANN data to optimize parameters ............................................................27
4.3.1. Function sim(x) to use ANN network for optimization.........................................27
4.3.2. Optimization results using ANN-GA.....................................................................29
4.3.3. New PCA base on ANN-GA .................................................................................32
CHAPTER 5: TESTING AND EVALUATION ..........................................................................33 5.1. Particle Classification System .........................................................................................33 5.1.1. Analysis process.....................................................................................................33
5.1.2. Image processing ...................................................................................................33
5.1.3. Feature extraction...................................................................................................35
5.1.4. Decision conditions................................................................................................36
5.1.5. Particle Classification System................................................................................37
5.1.6. Result particle classification system ......................................................................38
5.2. Evaluate the results of the new algorithm based on ANN-GA .......................................39
CHAPTER 6: NEW METHOD FOR PARTICLE CLASSIFICATION - BOUNDARY MATRIX .......................................................................................................................................................43
6.1. Introduction .....................................................................................................................43
6.2. Analysis process ..............................................................................................................43
6.3. Image process ..................................................................................................................45
6.4. Results .............................................................................................................................46
CHAPTER 7: CONCLUSION AND FUTURE SCOPE ..............................................................48 10.1. Conclusion.......................................................................................................................48 10.2. Futurescope....................................................................................................................48
REFERENCE ................................................................................................................................50
LIST OF FIGURES
Fig. 1.1 Shows particles extracted from transmission by grid filter ................................... 2 Fig. 1.2 Intersection between casting surface and drilling holes of Automotive Transmission at control valve ............................................................................................. 2 Fig. 1.3 SEM analysis of particles with chemical composition and surface structure. ............... 3 Fig. 2.1 Cleanability in the design-to manufacturing cycle and main influences[1] ......... 5 Fig. 2.2 Design of intersection holes considering cleanability[3]...................................... 5 Fig. 2.3 Chip morphology, formed by drilling & milling.................................................... 6 Fig. 2.4 Real industrial cleanability problem in automotive .............................................. 7 Fig. 2.5 Morphology of surface after different deburring processes..................................7 Fig. 2.6 Burrs at different locations after HPWJ & Brushing ............................................ 8 Fig. 2.7 Burr at window after HPWJ and brushing ............................................................ 8 Fig. 3.1 Samples with 4 areas: casting surface, machined surfaces (drilling and milling) ......... 9 Fig. 3.2 Stable burrs formed along the edge of a workpiece even after deburring process.......10 Fig. 3.3. Unstable burr after machining or deburring process ........................................ 10 Fig. 3.4. Generation of cast debris and cast surface ........................................................ 12 Fig. 3.5. V ariety of chip shape .......................................................................................... 13 Fig. 3.6. Standard of particle classification after image processing to get clear shape/boundary of particles .............................................................................................. 14 Fig. 3.7. Definition of shape parameters in (a) burr, (b) cast, (c) chip and (d) filament of brush .................................................................................................................................. 14 Fig 3.8. Schematic illustration of Particle Classification Algorithm (PCA)[4] ............... 17 Fig. 4.1. Basic working mechanism of neural networks ................................................... 19 Fig. 4.2. MATLAB's Neural Network Toolbox .................................................................. 19 Fig. 4.3. Simple diagram explaining the GA algorithm .................................................... 20 Fig. 4.4. Global Optimization Toolbox with GA in Matlab .............................................. 21 Fig. 4.5. Graphical representation of ANN model............................................................22 Fig. 4.6. Network achitecture ANN ................................................................................... 22 Fig. 4.7. Use the Levenberg-Marquardt algorithm to train the ANN ............................... 23 Fig. 4.8. Validation and test data......................................................................................23 Fig. 4.9. ANN training results for group 1 and group 2 ................................................... 24 Fig. 4.10. Comparison of experimental SR% with ANN(Group 1) and ANN(Group 2) and Taguchi predicted SR%. .................................................................................................... 25 Fig. 4.11. Contour Plot of SR% vs L2 and L3...................................................................26 Fig. 4.12. Contour Plot of SR% vs R1 and L3. ................................................................. 26 Fig. 4.13. Contour Plot of SR% vs L2 and R1 .................................................................. 26 Fig. 4.14. How to transform when you need to find the maximum value using the minimum value optimization algorithm.............................................................................28 Fig. 4.15. New PCA base on ANN-GA..............................................................................32 Fig. 5.1. Flow diagram of Particle Classification System (PCS) ..................................... 33 Fig. 5.2. Examples of the process of image processing in each particle .......................... 34 Fig. 5.3. Feature extraction with length (L) and width (W) for burr, cast, chip, and filament of brush................................................................................................................36 Fig. 5.4. Programming decision-making structure in Matlab. ......................................... 36 Fig. 5.5. GUI (Graphical User Interface) of Particle Classification System ................... 37
Fig. 5.6. Graph of classification for particles inside a transmission at the final values of parameters. ........................................................................................................................ 38 Fig. 5.7. Formula to calculate SR%..................................................................................39 Fig. 5.8. Confusion matrix of Group 1..............................................................................40 Fig. 5.9. Confusion matrix of Group 2..............................................................................40 Fig. 6.1. Simple simulation of the particle boundary matrix ............................................ 43 Fig. 6.2. Diagram to extract classification using boundary matrix..................................44 Fig. 6.3. Result LargestObject images. ............................................................................. 45 Fig. 6.4. Some results of the particle processed(The PDF file shows more than 1100 particle tests) ..................................................................................................................... 46 Fig. 6.5. Results when drawing the grain distribution chart of Eigenvalue with more than 1100 particle...................................................................................................................... 47 Fig. 7.1. Robot Arm Programming to point out positions. ............................................... 48 Fig. 7.2. Particles randomly collected at Huyndai Motor Company, Ulsan, S.Korea. .... 49
LIST OF TABLES
Table 3.1. Classes of particles in PCA[4] ...........................................................................13 Table 3.2. Automatically measured shape parameters (L, W, A) of sample particles[4].15 Table 3.3. L-8 experiments, with 2 trial test results from group 1, 2[3] ..........................16 Table 3.4. Proper value from ANOVA[3] ......................................................................... 16 Table 3.5. Final values of variables R1, L2, L3[3]...........................................................17 Table 3.6. Initial values of algorithm parameters[4] .......................................................17 Table 4.1. Optimization results ......................................................................................... 30 Table 4.2. Determination of all parameters with success rates 99% (group 1) and 98%(group 2)..................................................................................................................... 31 Table 5.1. Compare the SR% results of the PCAs ............................................................ 41 Table 5.2. Overlapping brushes lead to image processing errors....................................42 Table 6.1. Examples of the process of image processing in each particle .......................46
ANN : GA : GUI : MDM : HPWJ : PCA :
PCS : PES :
Graphical User Interface Multilayer Detection Model High-Pressure Water Jets Particle Classification Algorithm
Particle Classification System Particle Expert System
LIST OF ABBREVIATIONS
Artificial Neural Network
Genetic Algorithm
CHAPTER 1: OVERVIEW 1.1. Overview of the research
In the field of precision machining or advanced manufacturing, especially in automobile manufacturing, the problem of burr is causing headaches for many experts. These are burrs left on the part surface after machining, appearing in processes such as turning, milling, drilling, and casting. These issues do not only reduces the accuracy of the part, but also burr removal process is often very much time-consuming and costly. Fig. 1.1 Shows particles extracted from transmission using 120 um grid filter.
Fig. 1.1 Shows particles extracted from transmission by grid filter.
Burr is the root cause of many serious problems inside the engine, transmission, and crankshaft. The inability to completely remove burrs has led to an increase in the number of auto recalls in recent times. These parts have common designs, including very small oil line systems created from drilling intersecting or inclined cross-section holes, cast surfaces, drilled holes combined with milled surfaces, and combined milled surfaces and the casting surface. In addition, in the engine and transmission, there are various types of sensors and solenoid valves to control oil pressure and motion signals. Burrs that are not cleaned thoroughly can detach out during engine operation, follow the lubricating oil, and easily stuck in locations such as oil lines, solenoid valves, and sensors, causing serious errors. When mixing with lubricating oil that is circulated through all parts in the engine, burrs/chip/cast particles created direct contact or strong impact into metal surfaces, creating negative effects on the surface or connection between parts. In particularly, when the car is running at high speed, the consequences can become more serious, affecting the working surface and the function of key components. Fig. 1.2 Cross-section of Intersection between casting surface and drilling holes of Automotive Transmission.
Fig. 1.2 Intersection between casting surface and drilling holes of Automotive Transmission at control valve.
2
D. Dornfeld [1,2] defined the concept of Cleanability, which totally change the concept of DFM (Design for Manufacturing) to become DFC (Design for Cleanability), with the meaning to care much more about issues related to the final processes at manufacturing: deburring, cleaning, ... then call back these designs, so that can change basically the cutting conditions, parameters of robot arm handling, controlling system, anodizing, and much more importance is the changing in the Design of original parts.
Particle Classification Algorithm, and Particle Expert System were generated by Phan Quoc Bao, không Sung Lim [3,4] to predict the main effects into the system were burrs, cast, chip, then propose the solution to improve the cleanliness. The results allowed users to identify particle origins after PCA parameter training and post-image processing. From there, they determine which particle is created by which machining step, which helps support the direction to solve that Particle. Especially, that can be possible to upload into web-based expert system to support worldwide users to upload their issues with particles similarity to these research. However, there were need a better database processing with ANN
In this research paper, we apply Artificial Neural Network combined with Genetic Algorithm (ANN_GA) to improve parameter optimization for the Particle Expert System (PES) classification algorithm. Therefore improving the accuracy of particle classification. Our researches showed promising to create a self-learning particle expert system that can recognize different particle types with at high speed and accuracy.
1.2. The need for research
The current automobile manufacturing industry, when using 6-speed engines, always finds hidden risks: sudden acceleration, loss of throttle, lack of throttle, breakage, damaged sensors, damaged control valves because particles are always present. are separated from the transmission system when we accelerate, under the compression pressure of gasoline and thrust when starting the engine, these particles are easily separated from the casting made from ADC12 aluminum alloy.
Fig. 1.3 SEM analysis of particles with chemical composition and surface structure.
The high level of noise in the sharpness of images taken of Particles with the Burr Measurement System H.Phuong[5] led to the construction of input data for particles by Bao, không Sung Lim [3,4] limited to the range of 100-400 μm, in these areas it is very difficult for us to classify, calculate, and build a standard database for classification, so we use ANN artificial intelligence to advance Prediction combined with GA optimization algorithm as a development platform at the micro-scale level.
3
1.3. The goal of the research
− Improve the reliability of the classification algorithm on the platform developed with Taguchi from input from experts.
− Develop identification software for hardware of metal scrap sorting equipment, applied at the micro to nano size level.
− Build intelligent classification algorithms to integrate into production design software from the early stages of the production chain.
1.4. The object and scope of research 1.4.1. The object of the research
− Image of particles taken from inside car engines.
− Algorithm for classifying particles.
− Optimized parameters for particle classification algorithm.
− ANN artificial neural network and reasonable optimization method.
1.4.1. The scope of the research
− Analyzing, calculating, simulating and optimizing parameters for the particle classification algorithm.
1.5. Approach and research method 1.5.1. Approach
− A classification algorithm with new parameters and a particle identification system created by inheriting the old classification algorithm have been enhanced ability to detect particles in sizes from micro to nano.
− Optimize parameters for classification algorithm using Artificial Neural Network ANN and genetic algorithm GA.
1.5.2. Research methods
− Theoretical reference method.
− The method of inheriting the research has been made.
− Methods of analysis and evaluation.
− Computational research methods.
− Programming and simulation using Matlab and Python software.
1.6. Project results
− Description:
• Classification algorithm with optimized parameters using ANN-GA.
• The particle classification system is based on an optimized classification
algorithm
CHƯƠNG 7: KẾT LUẬN VÀ PHẠM VI TƯƠNG LAI 10.1. Phần kết luận
Việc tích hợp Mạng thần kinh nhân tạo (ANN) với Thuật toán di truyền (GA) đã tạo ra sự tối ưu hóa tham số để nâng cấp thuật toán phân loại trước đó. Cách tiếp cận tích hợp này khai thác khả năng điều chỉnh khả năng học sâu của mạng lưới thần kinh.
Ngoài ra, chúng tui đã đề xuất một cách mới để giải quyết khó khăn của các hạt không thể đoán được ở ranh giới của các lớp, bằng cách sử dụng phân tích Ma trận ranh giới. Một bức ảnh của hạt có thể được trích xuất để lấy pixel ma trận ranh giới và giá trị riêng là đặc tính của từng ma trận, trong trường hợp này chúng tui sử dụng kích thước ma trận nhị phân vuông [N x N]. Cách tiếp cận này có thể gợi ý giải pháp tốt hơn cho hạt micro đến nano hay có thể khắc phục một số điểm nghẽn về tiếng ồn trong quá trình chụp ảnh.
Hơn nữa, một hệ thống phần mềm phân loại đã được phát triển, đặt nền tảng cho các ứng dụng phần mềm trong tương lai về phần cứng phân loại phế liệu kim loại.
10.2. Phạm vi tương lai
Trong tương lai, nghiên cứu được thực hiện trong luận án này cung cấp nền tảng vững chắc cho nhiều hướng đi đầy hứa hẹn:
- Nghiên cứu chuyên sâu về phân loại hạt bằng phương pháp ma trận biên sẽ được ưu tiên. Việc khám phá này nhằm mục đích cải thiện sự hiểu biết về mối tương quan giữa các đặc điểm ranh giới với các tính chất của hạt, có khả năng cách mạng hóa độ chính xác của các thuật toán phân loại đối với các hình dạng và thành phần phức tạp.
- Nghiên cứu việc tích hợp Deep-Learning ANN, hay còn gọi là Kohonen mờ ANN.
- Tích hợp phân loại các hạt có kích thước nhỏ hơn từ lưới lọc 20 micromet, ở kích thước từ 100 – 250 micromet.
- Phát triển hạt có kích thước siêu nhỏ bằng quy trình in 3D SLA. Mục đích là tạo ra mẫu hạt mà sau này sử dụng để Nhận dạng Cánh tay Robot. Dự án này được phát triển bởi Cleanability Lab của Công ty TNHH Phần cứng Kim loại Việt Nam.
Do Drive thay đổi chính sách, nên một số link cũ yêu cầu duyệt download. các bạn chỉ cần làm theo hướng dẫn.
Password giải nén nếu cần: ket-noi.com | Bấm trực tiếp vào Link để tải:
Integrate ANN-GA into improve particle classification algorithm in automotive production moving forward to green manufacturing
ABSTRACT
In pursuit of promoting Green Manufacturing in the automotive industry, this
research integrates Artificial Neural Network (ANN) Genetic Algorithm (GA), called ANN-GA, to enhance the optimization of Particle Classification Algorithm (PCA) was previously developed by Taguchi method, then upgraded by Programming C++ and Matlab optimization function. This innovation comes from the need to solve the problem of automotive machinery small chip lodged and complex particles created in various engine processing processes such as casting, anodizing, machining, High-pressure water jet, anodizing, sandblasting, grinding, polishing, and assembly. These processes will induce the rist of creation of burr, cast and chip,... These particles lodge inside the transmission, engine and crankshaft, and then damage the functions of these components, posing risks to drivers, such as: suddenly increase speed until losing control, mal- function, suddenly stop engine, negative impact to valve control systems called TC21,
TC35, TC 29... which mainly working based on the electro-magnetic oil sensors.
Based on the partical experiment in research of Dr. Phan Quoc Bao, and Professor
Sung Lim không from the high-tech manufacturing laboratory at Hyundai Motor Korea, South Korea, and the Precison Machining Lab, Konkuk University, South Korea, from 2009 - 2015, has developed a particle expert system (PES) that proved the possibility to classify burrs, blanks and chips, by using very basic image processing techniques. An advanced model of PCA, consisting of 12 distinct particle types, was found by using ANN-GA. This new approach does not only trains PCA parameters with higher accuracy but also ensures a stable success rate, thereby improving the reliability of particle source identification processing. Even though the whole project result was statiscal evaluation, it showed up the trend of creating particle from different sources, processes of metal fabrication. These decisions may support the top leader CEO, Managers to make suitable
decision to invest into suitable production technology related to improving Cleanability
This integration of ANN-GA into PCA is a forward-thinking approach to solving
cleanability challenges in automotive manufacturing. It aligns with the principles of Green Manufacturing by providing a methodologically sound, environmentally friendly solution to improve product quality and safety. Here is the discussion from Professor Dornfield at ISGMA 2013 Sheraton Hawaii Waikiki Green Manufacturing International
Conference.
Ho Chi Minh City, Ferbruary 19, 2024
Tran Minh Thuan
TABLE OF CONTENTS
CHAPTER 1: OVERVIEW.............................................................................................................2
1.1. Overview of the research...................................................................................................2
1.2. The need for research ........................................................................................................3
1.3. The goal of the research ....................................................................................................4
1.4. The object and scope of research ......................................................................................4
1.4.1. The object of the research ........................................................................................4
1.4.1. The scope of the research.........................................................................................4 1.5. Approach and research method .........................................................................................4
1.5.1. Approach..................................................................................................................4
1.5.2. Research methods ....................................................................................................4
1.6. Project results ....................................................................................................................4
1.7. Efficiency specification.....................................................................................................4
CHAPTER 2: BASIC THEORIES..................................................................................................5 2.1. Cleanability problems and technology. .............................................................................5
2.1.1. Introduction..............................................................................................................5
2.1.2. Advanced cleaning methods ....................................................................................6
2.1.3. Deburring Processes and Cleanability...............................................................................7
CHAPTER 3: CHARACTERIZATION AND RECOGNITION OF PARTICLES FOR IMPROVING CLEANABILITY IN AUTOMOTIVE PRODUCTION ........................................9
3.1. Mechanism of particles generation....................................................................................9
3.2. Standard for particle classification ..................................................................................13
3.3. Algorithm of classification..............................................................................................15
CHAPTER 4: INTEGRATING ANN-GA INTO THE CLASSIFICATION ALGORITHM ......18 4.1. Introduce ANN-GA.........................................................................................................18
4.1.1. Artificial Neural Network (ANN)..........................................................................18
4.1.2. Genetic Algorithm (GA) ........................................................................................20
4.2. ANN training...................................................................................................................22
4.2.1. Graphical representation of ANN model ...............................................................22
4.2.2. Result and Discussion............................................................................................24
4.3. Use GA from ANN data to optimize parameters ............................................................27
4.3.1. Function sim(x) to use ANN network for optimization.........................................27
4.3.2. Optimization results using ANN-GA.....................................................................29
4.3.3. New PCA base on ANN-GA .................................................................................32
CHAPTER 5: TESTING AND EVALUATION ..........................................................................33 5.1. Particle Classification System .........................................................................................33 5.1.1. Analysis process.....................................................................................................33
5.1.2. Image processing ...................................................................................................33
5.1.3. Feature extraction...................................................................................................35
5.1.4. Decision conditions................................................................................................36
5.1.5. Particle Classification System................................................................................37
5.1.6. Result particle classification system ......................................................................38
5.2. Evaluate the results of the new algorithm based on ANN-GA .......................................39
CHAPTER 6: NEW METHOD FOR PARTICLE CLASSIFICATION - BOUNDARY MATRIX .......................................................................................................................................................43
6.1. Introduction .....................................................................................................................43
6.2. Analysis process ..............................................................................................................43
6.3. Image process ..................................................................................................................45
6.4. Results .............................................................................................................................46
CHAPTER 7: CONCLUSION AND FUTURE SCOPE ..............................................................48 10.1. Conclusion.......................................................................................................................48 10.2. Futurescope....................................................................................................................48
REFERENCE ................................................................................................................................50
LIST OF FIGURES
Fig. 1.1 Shows particles extracted from transmission by grid filter ................................... 2 Fig. 1.2 Intersection between casting surface and drilling holes of Automotive Transmission at control valve ............................................................................................. 2 Fig. 1.3 SEM analysis of particles with chemical composition and surface structure. ............... 3 Fig. 2.1 Cleanability in the design-to manufacturing cycle and main influences[1] ......... 5 Fig. 2.2 Design of intersection holes considering cleanability[3]...................................... 5 Fig. 2.3 Chip morphology, formed by drilling & milling.................................................... 6 Fig. 2.4 Real industrial cleanability problem in automotive .............................................. 7 Fig. 2.5 Morphology of surface after different deburring processes..................................7 Fig. 2.6 Burrs at different locations after HPWJ & Brushing ............................................ 8 Fig. 2.7 Burr at window after HPWJ and brushing ............................................................ 8 Fig. 3.1 Samples with 4 areas: casting surface, machined surfaces (drilling and milling) ......... 9 Fig. 3.2 Stable burrs formed along the edge of a workpiece even after deburring process.......10 Fig. 3.3. Unstable burr after machining or deburring process ........................................ 10 Fig. 3.4. Generation of cast debris and cast surface ........................................................ 12 Fig. 3.5. V ariety of chip shape .......................................................................................... 13 Fig. 3.6. Standard of particle classification after image processing to get clear shape/boundary of particles .............................................................................................. 14 Fig. 3.7. Definition of shape parameters in (a) burr, (b) cast, (c) chip and (d) filament of brush .................................................................................................................................. 14 Fig 3.8. Schematic illustration of Particle Classification Algorithm (PCA)[4] ............... 17 Fig. 4.1. Basic working mechanism of neural networks ................................................... 19 Fig. 4.2. MATLAB's Neural Network Toolbox .................................................................. 19 Fig. 4.3. Simple diagram explaining the GA algorithm .................................................... 20 Fig. 4.4. Global Optimization Toolbox with GA in Matlab .............................................. 21 Fig. 4.5. Graphical representation of ANN model............................................................22 Fig. 4.6. Network achitecture ANN ................................................................................... 22 Fig. 4.7. Use the Levenberg-Marquardt algorithm to train the ANN ............................... 23 Fig. 4.8. Validation and test data......................................................................................23 Fig. 4.9. ANN training results for group 1 and group 2 ................................................... 24 Fig. 4.10. Comparison of experimental SR% with ANN(Group 1) and ANN(Group 2) and Taguchi predicted SR%. .................................................................................................... 25 Fig. 4.11. Contour Plot of SR% vs L2 and L3...................................................................26 Fig. 4.12. Contour Plot of SR% vs R1 and L3. ................................................................. 26 Fig. 4.13. Contour Plot of SR% vs L2 and R1 .................................................................. 26 Fig. 4.14. How to transform when you need to find the maximum value using the minimum value optimization algorithm.............................................................................28 Fig. 4.15. New PCA base on ANN-GA..............................................................................32 Fig. 5.1. Flow diagram of Particle Classification System (PCS) ..................................... 33 Fig. 5.2. Examples of the process of image processing in each particle .......................... 34 Fig. 5.3. Feature extraction with length (L) and width (W) for burr, cast, chip, and filament of brush................................................................................................................36 Fig. 5.4. Programming decision-making structure in Matlab. ......................................... 36 Fig. 5.5. GUI (Graphical User Interface) of Particle Classification System ................... 37
Fig. 5.6. Graph of classification for particles inside a transmission at the final values of parameters. ........................................................................................................................ 38 Fig. 5.7. Formula to calculate SR%..................................................................................39 Fig. 5.8. Confusion matrix of Group 1..............................................................................40 Fig. 5.9. Confusion matrix of Group 2..............................................................................40 Fig. 6.1. Simple simulation of the particle boundary matrix ............................................ 43 Fig. 6.2. Diagram to extract classification using boundary matrix..................................44 Fig. 6.3. Result LargestObject images. ............................................................................. 45 Fig. 6.4. Some results of the particle processed(The PDF file shows more than 1100 particle tests) ..................................................................................................................... 46 Fig. 6.5. Results when drawing the grain distribution chart of Eigenvalue with more than 1100 particle...................................................................................................................... 47 Fig. 7.1. Robot Arm Programming to point out positions. ............................................... 48 Fig. 7.2. Particles randomly collected at Huyndai Motor Company, Ulsan, S.Korea. .... 49
LIST OF TABLES
Table 3.1. Classes of particles in PCA[4] ...........................................................................13 Table 3.2. Automatically measured shape parameters (L, W, A) of sample particles[4].15 Table 3.3. L-8 experiments, with 2 trial test results from group 1, 2[3] ..........................16 Table 3.4. Proper value from ANOVA[3] ......................................................................... 16 Table 3.5. Final values of variables R1, L2, L3[3]...........................................................17 Table 3.6. Initial values of algorithm parameters[4] .......................................................17 Table 4.1. Optimization results ......................................................................................... 30 Table 4.2. Determination of all parameters with success rates 99% (group 1) and 98%(group 2)..................................................................................................................... 31 Table 5.1. Compare the SR% results of the PCAs ............................................................ 41 Table 5.2. Overlapping brushes lead to image processing errors....................................42 Table 6.1. Examples of the process of image processing in each particle .......................46
ANN : GA : GUI : MDM : HPWJ : PCA :
PCS : PES :
Graphical User Interface Multilayer Detection Model High-Pressure Water Jets Particle Classification Algorithm
Particle Classification System Particle Expert System
LIST OF ABBREVIATIONS
Artificial Neural Network
Genetic Algorithm
CHAPTER 1: OVERVIEW 1.1. Overview of the research
In the field of precision machining or advanced manufacturing, especially in automobile manufacturing, the problem of burr is causing headaches for many experts. These are burrs left on the part surface after machining, appearing in processes such as turning, milling, drilling, and casting. These issues do not only reduces the accuracy of the part, but also burr removal process is often very much time-consuming and costly. Fig. 1.1 Shows particles extracted from transmission using 120 um grid filter.
Fig. 1.1 Shows particles extracted from transmission by grid filter.
Burr is the root cause of many serious problems inside the engine, transmission, and crankshaft. The inability to completely remove burrs has led to an increase in the number of auto recalls in recent times. These parts have common designs, including very small oil line systems created from drilling intersecting or inclined cross-section holes, cast surfaces, drilled holes combined with milled surfaces, and combined milled surfaces and the casting surface. In addition, in the engine and transmission, there are various types of sensors and solenoid valves to control oil pressure and motion signals. Burrs that are not cleaned thoroughly can detach out during engine operation, follow the lubricating oil, and easily stuck in locations such as oil lines, solenoid valves, and sensors, causing serious errors. When mixing with lubricating oil that is circulated through all parts in the engine, burrs/chip/cast particles created direct contact or strong impact into metal surfaces, creating negative effects on the surface or connection between parts. In particularly, when the car is running at high speed, the consequences can become more serious, affecting the working surface and the function of key components. Fig. 1.2 Cross-section of Intersection between casting surface and drilling holes of Automotive Transmission.
Fig. 1.2 Intersection between casting surface and drilling holes of Automotive Transmission at control valve.
2
D. Dornfeld [1,2] defined the concept of Cleanability, which totally change the concept of DFM (Design for Manufacturing) to become DFC (Design for Cleanability), with the meaning to care much more about issues related to the final processes at manufacturing: deburring, cleaning, ... then call back these designs, so that can change basically the cutting conditions, parameters of robot arm handling, controlling system, anodizing, and much more importance is the changing in the Design of original parts.
Particle Classification Algorithm, and Particle Expert System were generated by Phan Quoc Bao, không Sung Lim [3,4] to predict the main effects into the system were burrs, cast, chip, then propose the solution to improve the cleanliness. The results allowed users to identify particle origins after PCA parameter training and post-image processing. From there, they determine which particle is created by which machining step, which helps support the direction to solve that Particle. Especially, that can be possible to upload into web-based expert system to support worldwide users to upload their issues with particles similarity to these research. However, there were need a better database processing with ANN
In this research paper, we apply Artificial Neural Network combined with Genetic Algorithm (ANN_GA) to improve parameter optimization for the Particle Expert System (PES) classification algorithm. Therefore improving the accuracy of particle classification. Our researches showed promising to create a self-learning particle expert system that can recognize different particle types with at high speed and accuracy.
1.2. The need for research
The current automobile manufacturing industry, when using 6-speed engines, always finds hidden risks: sudden acceleration, loss of throttle, lack of throttle, breakage, damaged sensors, damaged control valves because particles are always present. are separated from the transmission system when we accelerate, under the compression pressure of gasoline and thrust when starting the engine, these particles are easily separated from the casting made from ADC12 aluminum alloy.
Fig. 1.3 SEM analysis of particles with chemical composition and surface structure.
The high level of noise in the sharpness of images taken of Particles with the Burr Measurement System H.Phuong[5] led to the construction of input data for particles by Bao, không Sung Lim [3,4] limited to the range of 100-400 μm, in these areas it is very difficult for us to classify, calculate, and build a standard database for classification, so we use ANN artificial intelligence to advance Prediction combined with GA optimization algorithm as a development platform at the micro-scale level.
3
1.3. The goal of the research
− Improve the reliability of the classification algorithm on the platform developed with Taguchi from input from experts.
− Develop identification software for hardware of metal scrap sorting equipment, applied at the micro to nano size level.
− Build intelligent classification algorithms to integrate into production design software from the early stages of the production chain.
1.4. The object and scope of research 1.4.1. The object of the research
− Image of particles taken from inside car engines.
− Algorithm for classifying particles.
− Optimized parameters for particle classification algorithm.
− ANN artificial neural network and reasonable optimization method.
1.4.1. The scope of the research
− Analyzing, calculating, simulating and optimizing parameters for the particle classification algorithm.
1.5. Approach and research method 1.5.1. Approach
− A classification algorithm with new parameters and a particle identification system created by inheriting the old classification algorithm have been enhanced ability to detect particles in sizes from micro to nano.
− Optimize parameters for classification algorithm using Artificial Neural Network ANN and genetic algorithm GA.
1.5.2. Research methods
− Theoretical reference method.
− The method of inheriting the research has been made.
− Methods of analysis and evaluation.
− Computational research methods.
− Programming and simulation using Matlab and Python software.
1.6. Project results
− Description:
• Classification algorithm with optimized parameters using ANN-GA.
• The particle classification system is based on an optimized classification
algorithm
CHƯƠNG 7: KẾT LUẬN VÀ PHẠM VI TƯƠNG LAI 10.1. Phần kết luận
Việc tích hợp Mạng thần kinh nhân tạo (ANN) với Thuật toán di truyền (GA) đã tạo ra sự tối ưu hóa tham số để nâng cấp thuật toán phân loại trước đó. Cách tiếp cận tích hợp này khai thác khả năng điều chỉnh khả năng học sâu của mạng lưới thần kinh.
Ngoài ra, chúng tui đã đề xuất một cách mới để giải quyết khó khăn của các hạt không thể đoán được ở ranh giới của các lớp, bằng cách sử dụng phân tích Ma trận ranh giới. Một bức ảnh của hạt có thể được trích xuất để lấy pixel ma trận ranh giới và giá trị riêng là đặc tính của từng ma trận, trong trường hợp này chúng tui sử dụng kích thước ma trận nhị phân vuông [N x N]. Cách tiếp cận này có thể gợi ý giải pháp tốt hơn cho hạt micro đến nano hay có thể khắc phục một số điểm nghẽn về tiếng ồn trong quá trình chụp ảnh.
Hơn nữa, một hệ thống phần mềm phân loại đã được phát triển, đặt nền tảng cho các ứng dụng phần mềm trong tương lai về phần cứng phân loại phế liệu kim loại.
10.2. Phạm vi tương lai
Trong tương lai, nghiên cứu được thực hiện trong luận án này cung cấp nền tảng vững chắc cho nhiều hướng đi đầy hứa hẹn:
- Nghiên cứu chuyên sâu về phân loại hạt bằng phương pháp ma trận biên sẽ được ưu tiên. Việc khám phá này nhằm mục đích cải thiện sự hiểu biết về mối tương quan giữa các đặc điểm ranh giới với các tính chất của hạt, có khả năng cách mạng hóa độ chính xác của các thuật toán phân loại đối với các hình dạng và thành phần phức tạp.
- Nghiên cứu việc tích hợp Deep-Learning ANN, hay còn gọi là Kohonen mờ ANN.
- Tích hợp phân loại các hạt có kích thước nhỏ hơn từ lưới lọc 20 micromet, ở kích thước từ 100 – 250 micromet.
- Phát triển hạt có kích thước siêu nhỏ bằng quy trình in 3D SLA. Mục đích là tạo ra mẫu hạt mà sau này sử dụng để Nhận dạng Cánh tay Robot. Dự án này được phát triển bởi Cleanability Lab của Công ty TNHH Phần cứng Kim loại Việt Nam.
Do Drive thay đổi chính sách, nên một số link cũ yêu cầu duyệt download. các bạn chỉ cần làm theo hướng dẫn.
Password giải nén nếu cần: ket-noi.com | Bấm trực tiếp vào Link để tải:
You must be registered for see links