臺灣博碩士論文加值系統

工具機在切削過程中，刀具磨耗程度為影響加工精度的主要因子，當加工件精度超過允收規格時，常代表該刀具需予以更換。然而在更換刀具前，已生產之超規工件與所歷經的加工資源將造成浪費。因此，若能有效推估刀具可能磨耗程度，預測刀具更換時機，以適時更換刀具，將可提升工具機之整體加工效益。
本研究整合加工特性與智慧型演算法，提出一刀具磨耗推估機制，以預測刀具可用壽命。在特徵分析上，首先收集與彙整在不同加工動作時的刀具振動訊號與主軸電流變化，以將其轉換成時域、頻域與時頻域的候選特徵值。在特徵選取上，由於候選特徵眾多，藉由非支配排序多目標基因演算法(NSGA-II)，可就最少化特徵數，最小化平均絕對誤差與最大誤差的目標下，選出刀具磨耗的關鍵特徵。在磨耗推估上，利用關鍵特徵藉由倒傳遞類神經網路(BPNN)模型，以估計刀具磨耗量。最後，在剩餘壽命預測上，將所推估的刀具磨耗量應用線性迴歸，以預測刀具可能剩餘壽命範圍。
在研究成果上，於三軸CNC銑床經6組不同刀具條件下(包含加工材料、主軸轉速倍率，進給倍率、與預留量等)，共72個標準件實驗結果得知，刀具磨耗之預測平均誤差為0.011 mm。此外，在刀具剩餘壽命推估上，尚可獲特定加工條件下得合理的預測範圍。

During machining of a machining tool, the wearing degrees of tools dominate machining precision. A tool needs to be replaced when precision of machined workpiece indicating out of tolerances of acceptable quality level; however, it has wasted the rejected workpieces and the machining resource before the broken tool being replaced. Hence, if the wearing degree of a tool can be effectively estimated and the broken tool can be predicted and replaced in time, the overall machining efficiency of machine tools can be improved.
This work integrates machining features and intelligent algorithms to propose a tool wearing estimation scheme for predicting tool residual life. In feature analysis, the candidate features covering time, frequency, and time-frequency domains are derived and transformed from the signals of the tool vibration and spindle current during various operations. For feature selection, we use the non-dominated sorting Genetic Algorithm II (NSGA-II) to select the key features of tool wearing while minimizing the three objectives key feature numbers, mean absolute errors, and the maximum error. For wearing estimation, the back propagation neural network (BPNN) is utilized to estimate the wearing degree of tools. Finally, the estimated wearing degree is applied to a linear regression for estimating the residual life of the tool.
In the results, the mean average error of tool wearing estimation is 0.0106 mm under six different machining conditions (including workpiece material, ratio of spindle speed, ratio of feed rate, and cutting depth) with 72 workpieces in a three-axis CNC machine. In addition, the feasible range of the tool residual life can be derived based on the specific conditions.

摘要 i
Abstract ii
圖目錄 v
表目錄 vii
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 4
第 2 章 理論方法 5
2.1 文獻探討 5
2.2 特徵計算方法 6
2.2.1 時域分析 6
2.2.2 頻域分析 7
2.2.3 資料正規化方法 9
2.3 特徵選取法 9
2.3.1 遺傳基因演算法 9
2.3.2 基於精英策略之非支配排序遺傳演算法 13
2.3.3 K-means演算法 16
2.4 預測方法 18
2.4.1 類神經網路 18
2.4.2 剩餘壽命預測 23
第 3 章 機制分析與設計 26
3.1 機制分析 26
3.2 機制設計 28
第 4 章 實驗分析 30
4.1 案例設計 30
4.1.1 實驗設計與架構 31
4.1.2 案例假設與限制 36
4.2 實驗結果分析 38
4.2.1 實驗說明 38
4.2.2 實驗步驟 38
4.3 實驗結果討論 48
第 5 章 結論與未來工作 50
5.1 結論 50
5.2 未來工作 50
參考文獻 52

[1]台灣區機械同業公會，台灣工具機產銷統計，http://www.tami.org.tw/statistics/week2.htm
[2]V. J. Johnson, A. Moosman, and P. Cotter, “A Hierarchical Model for Estimating the Early Reliability of Complex Systems,” IEEE Trans. on Reliability, vol. 54, no. 2, pp. 224-231, 2005.
[3]D. Wang and K. S. Trivedi, “Reliability Analysis of Phased-Mission System with Independent Component Repairs,” IEEE Trans. on Reliability, vol. 56, no. 3, pp. 540-551, 2007.
[4]F.-T. Cheng, H.-C. Huang, and C.-A. Kao, “Dual-Phase Virtual Metrology Scheme,” IEEE Transactions on Semiconductor Manufacturing, vol. 20, no. 4, pp. 566-571, 2007.
[5]MORI SEIKI, http://www.moriseiki.com/chinese/index.html
[6]MAZAK, http://www.mazak.jp/
[7]OKUMA, http://www.okuma.co.jp/chinese/
[8]應用振動訊號與模糊演算法偵測刀具狀態之研究，國立中興大學機械工程學系所，林敬偉，2008。
[9]應用類神經網路與振動訊號之微銑刀具狀態偵測系統開發，國立中興大學機械工程學系所，謝萬澔，2008。
[10]T. Doguer and J. Strackeljan, “Vibration Analysis using Time Domain Methods for the Detection of small Roller Bearing Defects,” SIRM 2009 - 8th International Conference on Vibrations in Rotating Machines, Vienna, Austria, Paper ID-16, February 23–25, 2009.
[11]B. Li, M.-Y. Chow, Y. Tipsuwan, and James C. Hung, “Neural-Network-Based Motor Rolling Bearing Fault Diagnosis,” IEEE Transactions on Semiconductor Manufacturing, vol. 47, no. 5, pp. 1060-1069, 2000.
[12]R. G. Silva, R. L. Reuben, K. J. Baker, and S. J. Wilcox, “Tool wear monitoring of turning operations by neural network and expert system classification of a feature set generated from multiple sensors,” Mechanical Systems and Signal Processing, vol. 12, issue 2, pp. 319-332, 1998.
[13]Rosas-Orea, M.C.E., Hernandez-Diaz, M., Alarcon-Aquino, V., and Guerrero-Ojeda, L.G., “A Comparative Simulation Study of Wavelet Based Denoising Algorithms,” in Proc. of Electronics, Communications and Computers, 2005. CONIELECOMP 2005. Proceedings. 15th International Conference. pp. 125-130, 2005.
[14]P. D. Welch, “The use of fast Fourier transform for the estimation of power spectra: A method based on time averaging over short, modified periodograms,” IEEE Transactions on Audio and Electroacoustics, vol. 15, no. 2, pp. 70-73, 1967.
[15]J. W. Cooley and J. W. Tukey, "An algorithm for the machine calculation of complex Fourier series," Mathematics of Computation, vol. 19, no. 90, pp. 297–301, 1965.
[16]I. Daubechies, ”Orthonormal bases of compactly supported wavelets,” Communications on Pure and Applied Mathematics, vol. 41, no. 4, pp. 909-996, 1988.
[17]J. H. Holland, Adaptation in Natural and Artificial Systems, Ann Arbor, MI: Univ. Mich. Press, 1975.
[18]K. Deb, A. Pratap, S. Agarwal, and T. Meyarivan, “A fast and elitist multiobjective genetic algorithm: NSGA-II,” IEEE Transactions on In Evolutionary Computation, vol. 6, no. 2, pp. 187-197, 2002.
[19]J. B. MacQueen, "Some Methods for classification and Analysis of Multivariate Observations," in the Proceedings of 5th Berkeley Symposium on Mathematical Statistics and Probability, University of California Press. pp. 281–297, 1967.
[20]機械工業雜誌，291期，民國96年6月號，"刀具破壞之探討"，金屬工業研究發展中心，精微成形研發處，馬寧元、李新中。