運用主成份田口類神經模組建立端銑表面粗糙度可調式控制系統

A PCA-Taguchi-Neural Adaptive Control of surface finishing in end milling operations

賴柏辰(Bo-Chen Lai)

指導教授 ： 黃博滄

中原大學/電機資訊學院/工業與系統工程研究所/碩士(2010年)

https://doi.org/10.6840/cycu201000893

摘要

表面粗糙度是端銑加工中重要的品質指標之一，為了達到良好的表面粗糙度通常需要額外的生產成本或降低生產力。過去的許多研究為了改善此問題，提出在加工前透過輸入的機器加工參數而事先預測出表面粗糙度，但在預測出的表面粗糙度不符合時，還是會產生不良品或停工的問題，進而本研究提出即時可調雙向式全面品質量測系統的概念。此概念為運用主成份田口類神經模組建立端銑表面粗糙度控制系統，其量測方式能在加工同時即時調整參數，且將表面粗糙度控制在一個品質範圍內，以降低量測和調整所耗費的時間與物件成本。

在過去文獻中大多只考量端銑製程參數是不夠的，本研究為了達到調整的精準度，將切削力量、預測與需求表面粗糙度差值和實際的表面粗糙度值也列入考量。為此，力量感測器(dynamometer)被使用在監控這不易控制的切削系統中，以增加表面粗糙度控制的準確度。從過去文獻發現，一個適當的切削力量信號是在XY平面的平均端點力量(Fap_xy)和Z軸力量絕對值(Fa_z)的交互作用。這兩種力量表現在表面粗糙度中關係密切，證明了切削刀具對粗糙度的影響和切削力與粗糙度之間的關係。

而在主成份田口類神經模組建立端銑表面粗糙度控制系統中，主要應用的決策系統為類神經網路，類神經網路已被廣泛應用在預測、診斷和調整上，所以本研究選擇以類神經網路作為決策系統。再者，為了提升模組的穩健性，本研究以端銑加工之實驗數據，透過主成份分析法轉換成為類神經網路的訓練來源，且訓練過程結合田口方法去優化類神經網路的建置。透過實驗之結果去驗證本研究之績效，結果發現主成份分析效果顯著。另外發現，製程參數中轉軸速度(S)相較於實際表面粗糙度(Ra)對於進給速度的調整關係較為顯著，且本研究最終實驗的調整誤差函數(Mean Square Error, MSE)已經可達準確之績效結果。

關鍵字

表面粗糙度 ； 田口方法 ； 類神經網路 ； 端銑 ； 主成份分析法

並列摘要

Surface roughness is an important quality index of machined parts. It usually needs additional production costs or to reduce productivity to result in a good surface roughness. To improve this problem, many researches suggested that predict the surface roughness through input parameters before processing; nevertheless, when its prediction does not meet quality specification, it still brings problems such as defective products or stop work. Therefore, this study proposed a real-time and two-way adjustable concept of quality measurement systems for surface roughness. The study used PCA-Taguchi-Neural Model to establish a control system of surface roughness in end milling operations, which can adjust parameters during machining processes, makes the surface roughness control in the range of quality specification, and reduces the cost of time of measurement and adjustment.

In the past, the research, which considered only the setting of processing parameters, is not enough. To be more accurate, other factors such as division of prediction and requirement, and the real surface roughness should also be considered, and therefore dynamometer was used in this uncontrolled processing system to increase the precision of surface roughness control. From the research, an empirical approach was applied to discover the proper cutting force signals, the average resultant peak force in XY plane (Fap) and the absolute average force in the Z direction (Faz). These two forces were employed to represent the uncontrollable cutting tool conditions for surface roughness control.

In a PCA-Taguchi-Neural Model that established a control system of surface roughness in end milling operations, the major decision-making system is neural networks. Neural networks have been widely used in prediction, diagnosis and adjustment. Therefore, this study used neural network as a decision-making system. Furthermore, to enhance the robustness of the module, the study used the experimental data of end milling operations, through PCA to convert the training source of the neural network, and the training process combined with Taguchi method to optimize the training parameters of the neural network. Four different models were used to verify the performance of this study, the results showed a significant effect that applied the principal component analysis in NN with spindle speed as one of input parameters, provided a better result than others.

並列關鍵字

the principal component analysis ； surface roughness ； end-milling ； Neural Network ； Taguchi

參考文獻

[18] Lu, H.S., Hwang, N.C., Chang, C.K. (2007). Optimal parameter design of high-speed end milling using Taguchi -principle component analysis approach, Journal of Technology, 22 (4), 325-333

[2] Huang, B. P., Chen, J. C. (2003). An in-process neural network-based surface

roughness prediction (INN-SRP) system using a dynamometer in end milling operations, Int. J. Adv. Manuf. Technol., 21, 339–347.

[3] Singh, S.K., Srinivasan, K.,Chakraborty, D.(2004). Acoustic characterization and prediction of surface roughness, Journal of Materials Processing Technology, 152, 127-130.

[4] Chen, J. C., Lou, M. S. (2000). Fuzzy-nets based approach to using an accelerometer for an in-process surface roughness prediction system in milling operations, Int. J. Comput. Integr. Manuf., 13 (4), 358–368.