臺灣博碩士論文加值系統

近年來，由於少子化的關係進而影響了勞動人口，以至於出現人力成本上升的現象，而現今工具機廠大多依靠擁有加工經驗的技師來操作機台，使企業紛紛朝向智能化工具機發展來提升製程效率以及品質，且礙於人力不足，多數工具機廠紛紛藉由開發無人化智慧工廠來因應目前人力短缺之情況，以效地解決人力短缺以及人為操作失誤所導致的品質控管問題，然而追求高精密加工的情況下，無論是車床或銑床，對於刀具的銳利以及其更換時機都非常的重視，因此如何隨時監視刀具使用情形並預估刀具壽命終了時機，以保持工件品質為重要之議題。故本研究針對CNC銑床開發了一套刀具線上監測磨耗與預測刀具使用壽命終了時機之智能化系統，能夠提供操作者即時得知刀具當前磨耗狀況與剩餘使用壽命。本論文主要可分成三大部分進行探討，第一部分為線上視覺檢測刀具磨耗系統，透過兩台工業相機搭配兩支遠心鏡頭，同時拍攝刀具側刃(Flank)與底刃(Land)部位的當前磨耗情形，再加上LED燈具投射白色光源照亮刀具，探討各種光源照明方式對於刀具影像清晰度之影響，使刀具磨耗特徵更加明顯；第二部分為利用微型光頻譜特性分析儀檢測環境光線特性，以利控制光源強度並保持系統穩定性。透過自行撰寫之程式計算刀具磨耗值，並將其數值與工具顯微鏡量測之刀具磨耗值相比較以觀察兩者之間的誤差，由實驗結果得知，於穩定光源情況下，刀具全刃磨耗使用程式與工具顯微鏡之平均誤差在0.005mm以內；第三部分為透過神經網路檢測刀具壽命並與影像處理方式進行比較，首先使用更快的區域卷積神經網路(Faster Region-Convolutional Neural Networks, Faster R-CNN)，將工業相機搭配遠心鏡頭所擷取之刀具影像作為神經網路的輸入條件，刀刃磨耗特徵部位為輸出條件進行刀具ROI預測模型的建置，並找出最佳學習效率以及疊代次數。接著再建立刀具壽命預測模型，基於刀具ROI預測模型辨識刀刃磨耗特徵區域，再次由Faster R-CNN演算法進行刀具壽命之預測，神經網路的輸入條件為程式所計算之磨耗值，輸出條件為刀具剩餘使用壽命，整體實驗得知Faster R-CNN預測模型的平均誤差為2~3%；而影像處理則是使用K-means聚類分割演算法進行特徵提取，最後再計算刀具剩餘使用壽命其平均誤差為3~5%，依結果可看出神經網路準確性優於影像處理。本系統可於每次加工後線上拍攝刀具影像並預測刀具壽命，使操作人員獲取即時刀具加工狀況，使生產線保持穩定運作以及節省生產成本，提升機台生產率、產品精度與加工品質以獲得事半功倍的效果。

In recent years, due to the relationship of fewer children, the labor force has been affected, and the labor cost has increased. Nowadays, most tool and machine factories rely on technicians with processing experience to operate the machines, making enterprises develop towards intelligent chemical tools to improve the process efficiency and quality. Due to the shortage of manpower, most tool and machine factories have developed unmanned intelligent factories to cope with the current shortage of manpower, In order to effectively solve the quality control problems caused by manpower shortage and human operation errors, however, in the pursuit of high-precision machining, both lathes and milling machines attach great importance to the sharpness of the tools and the timing of their replacement. Therefore, how to monitor the use of the tools at any time and estimate the timing of the end of the tool life is an important issue to maintain the quality of the workpiece. Therefore, this research develops an intelligent system for monitoring wear and predicting the end of tool life on the CNC milling machine, which can provide the operator with the current wear status and the remaining service life of the tool. This paper can be mainly divided into three parts for discussion. The first part is the online tool wear machine vision detection system. Through two industrial cameras and two telecentric lenses, the current wear situation of the side edge and the land edge of the tool is photographed at the same time.In addition, the white light source projected by the LED lamp illuminates the tool, and the influence of various light source illumination methods on the image clarity of the tool is discussed, so that the tool wear characteristics are more obvious; the second part is to use the micro optical spectrum analyzer to detect the characteristics of the ambient light, so as to control the intensity of the light source and maintain the stability of the system. The tool wear value is calculated by the self-made program, and the value is compared with the tool wear value measured by the tool microscope to observe the error between the two. The experimental results show that the average error between the tool wear program and the tool microscope is within 0.005mm under the condition of stable light source; the third part is to detect the tool life through the neural network and compare it with the image processing method. First, the Faster Regional Convolutional Neural Networks (Faster R-CNN) is used to build the tool ROI prediction model by using the tool image captured by the industrial camera and the telecentric lens as the input condition of the neural network and the characteristic part of the blade wear as the output condition, and find out the best learning efficiency and iteration times. Then, a tool life prediction model is established. Based on the tool ROI prediction model, the characteristic area of tool edge wear is identified, and then the Faster R-CNN algorithm is used to predict the tool life. The input condition of the neural network is the wear value calculated by the program, and the output condition is the remaining service life of the tool. The overall experiment shows that the average error of the Faster R-CNN prediction model is 2 ~ 3%;in image processing, K-means clustering algorithm is used for feature extraction, and finally the remaining service life of the tool is calculated. The average error is 3-5%. It can be seen from the results that the accuracy of neural network is better than that of image processing. The system can take the tool image online after each machining and predict the tool life, so that the operator can obtain the real-time tool processing status, keep the production line running stably and save the production cost, improve the productivity, product accuracy and processing quality of the machine to achieve twice the result with half the effort.

摘要...i
Abstract...ii
誌謝...iv
目錄...v
表目錄...vii
圖目錄...viii
符號說明...xi
第 一 章 緒論...1
1.1 研究背景與動機...1
1.2 研究目的...2
1.3 論文架構...2
第 二 章 文獻回顧...3
2.1 國內外影像檢測刀具磨耗的相關研究...3
2.2 國內外使用智能化演算法預測刀具壽命的相關研究...8
第 三 章 研究架構與方法...13
3.1 研究架構與流程...13
3.2 硬體及系統架構...14
3.2.1 實驗機台介紹...14
3.2.2 刀具磨耗影像擷取設備...14
3.2.3 輔助照明設備...16
3.2.4 微型光頻譜特性分析儀...17
3.2.5 實驗刀具與材料...18
3.2.6 相機與鏡頭成像...18
3.2.7 照明設備之光源投射原理...20
3.2.8 光譜儀量測原理與光色參數介紹...21
3.3 端銑刀常見磨耗種類與刀具壽命...26
3.4 Faster R-CNN演算法...27
3.5 K-means聚類分割演算法...36
第 四 章 實驗結果...38
4.1 線上視覺檢測刀具磨耗系統建置...38
4.1.1 實驗架設...38
4.1.2 相機之水平與垂直度校正...39
4.1.3 各種輔助照明設備對於刀具影像清晰度影響...41
4.2 刀具磨耗銑削實驗...42
4.2.1 刀具加工參數條件及銑削實驗方式...42
4.2.2 刀具磨耗量測結果...44
4.3 智能化搜索刀具ROI影像及壽命預測系統...50
4.3.1 建立刀刃特徵之Faster R-CNN辨識模型...50
4.3.2 學習效率對於誤差函數收斂性影響...52
4.3.3 神經網路之疊代次數分析...55
4.3.4 建立刀具壽命預測模型與驗證結果...58
4.3.5 刀具銑削部位影像處理...59
4.3.6 智能化刀具壽命監控系統...60
第 五 章 結論與未來展望...62
參考文獻...63
Extended Abstract...65
1 . Introduction...67
2 . Research of tool life prediction technique...68
3 . Methodology...69
3.1 Experimental tools and materials...69
3.2 Experimental parameters...70
3.3 Faster R-CNN Algorithm...71
3.4 K-means clustering algorithm...80
4 . Result and Discussion...81
4.1 Experimental setup...81
4.2 Results of various auxiliary lighting...82
4.3 Tool wear results...83
4.4 Analysis on learning rate of Faster R-CNN ROI model...84
4.5 Change the learning rate or iterations to predict the ROI...85
4.6 Tool life prediction results...86
4.7 Image processing results of K-means...87
4.8 Comparison of K-means and Faster R-CNN...88
4.9 Intelligent tool life monitoring system...89
5 . Conclusion...89
6 . References...90

[1]Yiquan Dai, Kunpeng Zhu, “A machine vision system for micro-milling tool condition monitoring”, Precision Engineering, Volume 52, Pages 183-191, 2018
[2]吳廷凱 , “開發線上刀具磨耗之量測系統”, 國立高雄第一科技大學, 電機工程研究所, 碩士學位論文, 2017
[3]Jitin Malhotra, Sunil Jha, “Fuzzy c-means clustering based colour image segmentation for tool wear monitoring in micro-milling”, Precision Engineering, Volume 72, Pages 690-705, 2021
[4]Junjie Zhou, Jianbo Yu, “Chisel edge wear measurement of high-speed steel twist drills based on machine vision”, Computers in Industry, Volume 128, Pages 103436, 2021
[5]Jianbo Yu, Xun Cheng, Liang Lu, Bin Wu, “A machine vision method for measurement of machining tool wear”, Measurement, Volume 182, Pages 109683, 2021
[6]Xuewei Zhang, Tianbiao Yu, Pengfei Xu, Ji Zhao, “In-process stochastic tool wear identification and its application to the improved cutting force modeling of micro milling”, Mechanical Systems and Signal Processing, Volume 164, Pages 108233, 2022
[7]Mohamed Marei, Shirine El Zaatari, Weidong Li, “Transfer learning enabled convolutional neural networks for estimating health state of cutting tools”, Robotics and Computer-Integrated Manufacturing, Volume 71, Pages 102145, 2021
[8]Sumant Bagri, Ashish Manwar, Alwin Varghese, Soham Mujumdar, Suhas S. Joshi, “Tool wear and remaining useful life prediction in micro-milling along complex tool paths using neural networks”, Journal of Manufacturing Processes, Volume 71, Pages 679-698, 2021
[9]T. Mikołajczyk, K. Nowicki, A. Bustillo, D. Yu Pimenov, “Predicting tool life in turning operations using neural networks and image processing”, Mechanical Systems and Signal Processing, Volume 104, Pages 503-513, 2018
[10]P.J. Bagga, K.S. Bajaj, M.A. Makhesana, K.M. Patel, “An online tool life prediction system for CNC turning using computer vision techniques”, Materials Today: Proceedings, issn 2214-7853, 2021
[11]方彥文, “基於工具機主軸電流智能化預測刀具壽命”, 國立虎尾科技大學, 機械與電腦輔助工程系, 碩士學位論文, 2019
[12]吳俊賢, “神經網路疊代收斂性分析與智能化預測加工品質及砂輪壽命”, 國立虎尾科技大學, 機械與電腦輔助工程系, 碩士學位論文, 2021
[13]陳彥叡, “基於視覺檢測與智能監控電流預估刀具磨耗與壽命”, 國立虎尾科技大學, 機械與電腦輔助工程系, 碩士學位論文, 2021
[14]光線與色溫的變化(https://read01.com/6GMyddj.html#.Ypcd_6hBxPY)
[15]更換臻光彩Ra95高演色性濾藍光護眼燈泡，還你一個清晰明亮的未來(https://www.mobile01.com/topicdetail.php?f=168&t=5429643)
[16]ISO 8688-2:1989(en) Tool life testing in milling — Part 2: End milling
[17]Shaoqing Ren, Kaiming He, Ross Girshick, Jian Sun, “Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks”, in IEEE Transactions on Pattern Analysis and Machine Intelligence, Volume 39, Pages 1137-1149, 2017
[18]陳建宏, “以基於區域的卷積神經網路實現空拍影片之魟魚偵測和辨識”, 國立中山大學, 機械與機電工程學系, 碩士學位論文, 2017
[19]Mengqi Chen, Lingjie Yu, Chao Zhi, Runjun Sun, Shuangwu Zhu, Zhongyuan Gao, Zhenxia Ke, Mengqiu Zhu, Yuming Zhang, “Improved faster R-CNN for fabric defect detection based on Gabor filter with Genetic Algorithm optimization”, Computers in Industry, Volume 134, Pages 103551, 2022
[20]Tangbo Bai, Jianwei Yang, Guiyang Xu, Dechen Yao, “An optimized railway fastener detection method based on modified Faster R-CNN”, Measurement, Volume 182, Pages 109742, 2021
[21]Jihao Shi, Yuanjiang Chang, Changhang Xu, Faisal Khan, Guoming Chen, Chuangkun Li, “Real-time leak detection using an infrared camera and Faster R-CNN technique”, Computers & Chemical Engineering, Volume 135, Pages 106780, 2020
[22]Yuanbin Wang, Minggao Liu, Pai Zheng, Huayong Yang, Jun Zou, “A smart surface inspection system using faster R-CNN in cloud-edge computing environment”, Advanced Engineering Informatics, Volume 43, Pages 101037, 2020