臺灣博碩士論文加值系統

CNC在現今的製造業中，扮演著或不可缺的角色，然而製程排配及交貨期限的安排，導致切削時間的預測相當重要，而在大部分的CAM 軟體中，如:NX，時間的預測是透過切削路徑長除上進給率來求得，但由於機台真實切削時會隨著運動型態的不同而有不同的加減速再加上每台機台的加減速性能都不同，使得切削時間無法很準確的預測。伴隨者大數據及人工智慧的崛起，類神經網路被應用於各個領域，而在CNC這塊，被用來預測表面粗糙度、表面形貌，及預測切削力、加工溫度等，本論文則利用類神經網路中的倒傳遞類神經網路來預測高進給時的加工時間，利用實驗所得到的數據進行模型的訓練，其中考慮到了各個運動型式所造成的影響，最後利用訓練好的模型來預測加工時間，研究顯示，利偉擺頭式小五軸機台採用Levenberg-Marquardt演算法配合一層隱藏層60個神經元數目訓練出來的模型，預測誤差率可以達到0.3%～8.3%，然而NX的預測誤差率為6.7%~18.5%，接著依照同樣的方法擷取QUASER UX300五軸機台的時間參數進行模型的訓練，模型預測誤差率可以達到0.37%～6%，研究證實，透過此實驗的方法，能夠快速的訓練出適合各個機台的時間預測模型。

In the manufacture industry, CNC become a part of indispensable roles. Because of the process scheduling and delivery deadline, predict the machining time become very important. All of the CAM software likes the NX, used the length of cutting roots divided
by the feedrate to predict the machining time, but in reality, machine have different
acceleration, deceleration followed by different motion type and different machine have its own machine performance, so the time predict will not very accuracy. Neural network is
applied in every field accompanied by the big data and Artificial Intelligence. It is used to predict the surface roughness, surface topography, cutting force and machining temperature. In this thesis, we are used backpropagation neural network to predict the
machining time. Between that we consider the effect factor which is caused by the motion
type, linear and circular. Finally, we can use this model to predict the machining time.
According to the research, using LM, one hidden layer and sixty hidden layer neurons to
train the little 5-axis machine model can reach the error rate between 0.3 to 8.3% ; however, the NX error rate is between 6.7 to 18.5%. Then we used the same method to get the QUASER UX300 CNC machining time and trained it, the model error rate can reach 0.37
to 6%, this thesis confirmed that used this experiment method can train the time model
rapidly and suitable for every machine.

摘要 I
Abstract II
致謝 III
圖索引 VI
表索引 VIII
第1章 緒論 1
1.1 研究背景 1
1.2 研究目的 2
1.3 論文架構 2
第2章 文獻回顧與理論基礎 3
2.1 切削加工時間預測文獻回顧 3
2.2 類神經網路 7
2.2.1 類神經網路簡介 7
2.2.2 影響神經網路效能的因素 11
2.2.3 倒傳遞類神經網路 13
2.2.4 隱藏層數、神經元數及訓練函數之選用 14
2.2.5 梯度下降法 16
2.2.6 Levenberg Marquardt倒傳遞法 19
2.2.7 參數正規化 21
第3章 實驗方法與設備 22
3.1 實驗方法 22
3.1.1 訓練方法一 28
3.1.2 訓練方法二 30
3.1.3 訓練方法三 32
3.2 切削設備、刀具與工件材料 36
3.2.1 切削機台 36
3.2.2 切削刀具 38
3.2.3 工件材料 38
3.3 開發工具 38
3.3.1 Skymars Express 38
3.3.2 Visual Studio 2012 39
3.3.3 Matlab R2013a 39
第4章 結果與討論 40
4.1 擺頭式小五軸切削中心機訓練方法一預測結果 40
4.2 擺頭式小五軸切削中心機訓練方法二預測結果 42
4.2.1 驗證一 42
4.2.2 驗證二 44
4.2.3 驗證三 44
4.3 擺頭式小五軸切削中心機訓練方法三預測結果 45
4.4 QUASER UX300五軸機台預測模型訓練 56
第5章 結論與未來展望 58
5.1 結論 58
5.2 未來展望 58
參考文獻 59
附錄A 61
使用者操作說明 61

[1] Jung, J. Y. (2002). Manufacturing cost estimation for machined parts based on manufacturing features. Journal of Intelligent Manufacturing, 13(4), 227-238.
[2] Hbaieb, M., Othmani, R., & Bouzid, W. (2011). Time modeling in high-speed machining of mold pocket. The International Journal of Advanced Manufacturing Technology, 53(1), 113-120.
[3] Yan, X., Shirase, K., Hirao, M., & Yasui, T. (1999). NC program evaluator for higher machining productivity. International Journal of Machine Tools and Manufacture, 39(10), 1563-1573.
[4] Heo, E. Y., Kim, D. W., Kim, B. H., & Chen, F. F. (2006). Estimation of NC machining time using NC block distribution for sculptured surface machining. Robotics and Computer-Integrated Manufacturing, 22(5), 437-446.
[5] Coelho, R. T., de Souza, A. F., Roger, A. R., Rigatti, A. M. Y., & de Lima Ribeiro, A. A. (2010). Mechanistic approach to predict real machining time for milling free-form geometries applying high feed rate. The International Journal of Advanced Manufacturing Technology, 46(9), 1103-1111.
[6] Maropoulos, P. G., Baker, R. P., & Paramor, K. Y. G. (2000). Integration of tool selection with design: Part 2: Aggregate machining time estimation. Journal of Materials Processing Technology, 107(1), 135-142.
[7] So, B. S., Jung, Y. H., Park, J. W., & Lee, D. W. (2007). Five-axis machining time estimation algorithm based on machine characteristics. Journal of materials processing technology, 187, 37-40.
[8] Liu, C., Li, Y., Wang, W., & Shen, W. (2013). A feature-based method for NC machining time estimation. Robotics and Computer-Integrated Manufacturing, 29(4), 8-14.
[9] 胡其寬，數值控制加工中心機程式加工時間估算與鑽孔路徑最佳化研究，碩士論文，國立台北科技大學製造科技研究所，台北，台灣，2006。
[10] Sanjay, C., & Jyothi, C. (2006). A study of surface roughness in drilling using mathematical analysis and neural networks. The International Journal of Advanced Manufacturing Technology, 29(9), 846-852.
[11] Gelmereanu, C., Morar, L., & Bogdan, S. (2014). Productivity and Cycle Time Prediction Using Artificial Neural Network. Procedia Economics and Finance, 15, 1563-1569.
[12] Xu, P., & Lee, R. S. (2016). Feedrate optimization based on hybrid forward-reverse mappings of artificial neural networks for five-axis milling. The International Journal of Advanced Manufacturing Technology, 87(9-12), 3033-3049.
[13] 陳永昌，應用類神經網路與遺傳演算法對端銑削加工參數之最佳化，博士論文，國立高雄第一科技大學機械與自動化工程研究所，高雄，台灣，2002。
[14] 張斐章、張麗秋，類神經網路，臺灣東華書局股份有限公司，台北，台灣 2005。
[15] 葉怡成，類神經網路模式應用與實作，儒林圖書股份有限公司，台北，台灣，2009。
[16]梁淑芳，導入 LM 法之平行倒傳遞演算法，碩士論文，東吳大學資訊科學系，台北，台灣，2006。
[17] Lourakis, M. I. (2005). A brief description of the Levenberg-Marquardt algorithm implemented by levmar. Foundation of Research and Technology, 4(1).
[18] Hagan, M. T., & Menhaj, M. B. (1994). Training feedforward networks with the Marquardt algorithm. IEEE transactions on Neural Networks, 5(6), 989-993.
[19] Sharma, B., & Venugopalan, K. (2014). Comparison of neural network training functions for hematoma classification in brain CT images. IOSR-JCE, 16(1), 31-5.
[20] 神經網絡學習算法matlab應用分析，取自：https://read01.com/KDM776.html#.Wbdr8LIjHIV
[21] 不同的神經網絡訓練函數training function的比較，取自： https://read01.com/8Pk3AO.html#.WbeSQ7IjHIU