臺灣博碩士論文加值系統

近年來因少子化的關係導致勞動人口下降，以至於出現人力成本上升的現象，因此許多工具機大廠紛紛朝向智能化工具機的方向發展，以此來提高製程效率與加工品質，並藉由開發無人化智慧工廠來因應目前人力短缺之情況，然而追求高精密加工的情況下，國內外針對智能化工具機之研究還是以車床與銑床作為主要主題，然而大多精密零件於最後一道製程皆為研磨加工以此來獲得較高精度的表面品質，因此如何保持砂輪的銳利與工件品質為重要之議題，故本研究針對CNC平面磨床開發一套智能化加工品質與砂輪壽命系統，能夠提供操作者能即時得知砂輪壽命與工件表面精度之情形，可將其分成三大部分進行探討，第一部分為使用霍爾感測器與加速規擷取研磨加工時的主軸負載值與振動之時域訊號，將時域訊號透過均方根、核心尋峰演算法與無因次化處理得到其加工時的關鍵數據，並於加工後經由非接觸式表面粗糙度量測儀進行工件表面粗糙度的量測；第二部分為使用倒傳遞類神經網路將異質感測器訊號與研磨參數作為神經網路的輸入層，工件表面粗糙度值為輸出層進行加工品質預測模型的建置，並透過試誤法找出最佳神經網路結構、學習效率以及學習次數，最後在探討最陡坡降法、Adagrad、RMSProp、Momentum與Adam五種疊代演算法之計算原理和學習效果之差異；第三部分為砂輪壽命預測系統，由無因次化負載訊號與表面粗糙度之關係找出其最佳的修砂時機，再將此無因次化負載值作為砂輪壽命終了之百分比，於加工的過程中即時量測主軸負載並預測砂輪壽命，以保持砂輪與工件為最佳狀態。由無因次化主軸負載值與工件表面粗糙度分析結果，觀察當高於穩定負載值9.4%後工件表面粗糙度度急遽上升，故將其作為砂輪壽命終了之條件，而在最陡坡降法、Adagrad、RMSProp、Momentum與Adam五種疊代演算法之加工品質預測系統中，其工件表面粗糙度預測誤差百分比分別3.7%、3.1%、4.6%、3.7%、3.7%。

In recent years, the labor forces are very shortages leaded to higher labor costs due to the low birth rates. Therefore, many machine tool manufacturers have developed the intelligent machine tools and unmanned smart factories to resolve the labor shortage problems which can also improve the efficiency and the quality of manufacturing process. In general, the research reports in intelligent machine tools are ususally focused on the lathes and milling machines for high-precision manufacturing process. However, the grinding machine is one of the most important manufacturing process for high precision parts by low material removal rate and high surface finish. Therefore, the issuees of how to maintain the sharpness of the grinding wheel and the quality of the workpiece is important during the grinding manufacturing process. This study investages the preditions of grinding wheel life-time and workpiece surface roughness during manufacturing process using backpropagation neural network. This study can be divided into three parts for discussion. In the first part, we used the Hall sensor, G-sensor, and roughness meter to measure the spindle load, spindle vibration, and roughness of the workpiece. The time-domain signals of spindle load and vibration are extracted by root mean square, findpeaks algorithm, and dimensionless processed from raw data. In the second part, the quality prediction of workpiece surface roughness is built through the backpropagation neural network based on different heterogeneous sensing signals and grinding parameters. Then, the study used the trial and error methods to find optimally the numbers of hidden layer neurons and the times of neural network architecture iterations. The different iterative parameters are then used to analyze the different algorithm convergences for predicting workpiece surface roughness, including Gradient descent, Adagrad, RMSProp, Momentum, and Adam algorithms. In the third part, in order to effectively analyze the end of life-time conditions of the grinding wheel, the preditions of grinding wheel life-time is based on the relationahip between the spindle loading and the workpiece roughness. The dimensionless ratio of grinding wheel life-time is defined between the spindle stable loading and current loading. Experimental results showed that the surface roughness of the workpiece sharply increases after the dimensionless ratio is higher than 109.4% that is defined the end of life time in this study. The predictive percentage errors between learning samples and predictive models for Adam, Adagrad, RMSProp, Gradient descent, and Momentum are 2.4%, 3.1%, 3.5%, 4%, and 4.2%, respectively. The predictive percentage errors between verificating samples and predictive models for Adagrad and RMSProp are 3.1% and 4.6%, respectively, others errors of Gradient descent, and Momentum algorithms are all 3.7%. Therefore, the quality prediction model of workpiece roughness is estimated by Adagrad algorithm in this study.

摘要...i
Abstract...ii
誌謝...iv
目錄...v
表目錄...viii
圖目錄...ix
符號說明...xii
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 檢測砂輪壽命狀態的相關研究...3
2.2 機器學習預測應用於工具機之收斂疊代演算法相關研究...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 異質感測器-主軸負載與振動量測設備...14
3.2.2 非接觸式表面粗糙度儀...15
3.2.3 實驗機台介紹...15
3.2.4 實驗刀具與材料性質...16
3.3 主軸負載與振動訊號擷取系統...17
3.3.1 霍爾效應(Hall effect)...17
3.3.2 訊號均方根(Root Mean Square)處理...18
3.3.3 核心尋峰演算法...19
3.3.4 訊號無因次化(Dimensionless)...20
3.4 倒傳遞類神經網路...20
3.4.1 倒傳遞類神經網路基本概論...20
3.4.2 資料正規化處理(Normalize)...21
3.4.3 活化函數(Activation function)...22
3.4.4 倒傳遞類神經網路學習與預測流程...22
3.4.5 疊代演算法...26
3.5 智能化砂輪壽命與加工品質預測系統...32
3.5.1 新代控制器通訊與資料傳遞...32
3.5.2 平面研磨-新代MACRO程式設計與撰寫...33
第 四 章 實驗結果...35
4.2 實驗架設...35
4.3 磨削實驗...36
4.3.1 前置作業...36
4.3.2 磨削實驗1...37
4.3.3 磨削實驗2-空修...39
4.3.4 建立加工品質預測模型之學習與驗證樣本...39
4.3.5 砂輪壽命分析...41
4.4 加工品質預測系統...42
4.4.1 網路結構與學習次數分析...42
4.4.2 疊代演算法分析...43
4.4.3 倒傳遞神經網路之預測結果...50
4.5 智能化砂輪壽命與加工品質預測系統...53
4.5.1 PC-Based量測與分析軟體...54
4.5.2 新代控制器人機介面設計...54
第 五 章 結論與未來展望...57
參考文獻...58
Extended Abstract...60

[1]陳泓衛 , “砂輪在削銳區間磨削力變化之探究,國立高雄第一科技大學”, 機械與自動化工程系, 碩士論文, 第1頁, 2014。
[2]羅佐良 , “智慧機械趨勢與應用”,工業技術研究院,智慧機械科技中心, 2018。
[3]Kalpana Kannan, N. Arunachalam, “ Grinding wheel redress life estimation using force and surface texture analysis”,Procedia CIRP ,Volume 72, Pages 1439-1444, 2018.
[4]Kalpana Kannan, N. Arunachalam, Aakash Chawla, Sundararajan Natarajan, “Multi-Sensor Data Analytics for Grinding Wheel Redress Life Estimation – An Approach towards Industry 4.0” ,Procedia Manufacturing,Volume 26, Pages 1230-1241, 2018.
[5]J.L.Lanzagorta, L. Urgoiti , P. Ruiz Vazquez , D. Barrenetxea ,J.A. Sanchez, “Experimental approach for a grinding burn in-process inspection system based on Eddy Current”, Procedia CIRP, Volume 87, Pages 391-396, 2020.
[6]Kevin Tanaka, Philip Koshy, “A pneumatic sensor for grinding wheel condition monitoring”,Precision Engineering ,Volume56 , Pages 62-68, March 2019.
[7]Jae-Seob Kwak, Man-Kyung Ha, “Detection of dressing time using the grinding force signal based on the discrete wavelet decomposition”, Int J Adv Manuf Technol 23, Pages 87-92, 2004.
[8]Amrut Mulay, B. Satish, Syed Ismail, Andrzej Kocanda, “Prediction of average surface roughness and formability in single point incremental forming using artificial neural network”,Archives of Civil and Mechanical Engineering, Volume 19, Issue 4, Pages 1135-1149, August 2019.
[9]Muhammad Rizal, Jaharah A. Ghani, Mohd Zaki Nuawi, Che Hassan Che Haron, “Online tool wear prediction system in the turning process using an adaptive neuro-fuzzy inference system”,Applied Soft Computing, Volume 13, Issue 4, Pages 1960-1968, April 2013.
[10]RavinderKumar, SantramChauhan, “ Study on surface roughness measurement for turning of Al 7075/10/SiCp and Al 7075 hybrid composites by using response surface methodology (RSM) and artificial neural networking (ANN)”,Measurement, Volume 65, Pages 166-180, April 2015.
[11]林伯尚, “基於倒傳遞類神經網路之刀具壽命與表面粗糙度研究”,國立虎尾科技大學 ,碩士學位論文 ,2018。
[12]İlhan Asiltürk, Mehmet Çunkaş, “ Modeling and prediction of surface roughness in turning operations using artificial neural network and multiple regression method”,Expert System with Application, Volume 38, Issue 5, Pages 5826-5832, May 2011.
[13]Ronny, “圖像分析：投影曲線的波峰查找” , 2014。
[14]吳少澤, “基於類神經演算法之疊代收斂性預測與補償螺桿熱變形研究”, 國立虎尾科技大學, 碩士學位論文, 2019。
[15]葉怡成，類神經網路模式應用與實作，儒林圖書有限公司，1993 年 1 月初版，2000 年 4 月七版。
[16]吳聲治, “工具機無人化生產及遠端監視系統建置”, 國立虎尾科技大學, 碩士學位論文, 2016。
[17]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, 碩士學位論文, 2019。
[18]許少霖, “基於漸開線自動生成齒輪3D輪廓磨削加工路徑開發”, 國立虎尾科技大學, 碩士學位論文, 2020。