銑削加工振動訊號應用於刀具磨耗監控之研究

Tool Wear Monitoring System Using Vibration Signals in Milling

蔡乙陞(CAI, YI-SHENG)

指導教授 ： 李貫銘

國立臺灣大學/工學院/機械工程學研究所/碩士(2019年)

https://doi.org/10.6342/NTU201901980

摘要

近年來製造業對於預測機械系統故障或刀具壽命之模型開發愈趨重視。過往刀具壽命預測模型以物理為基礎之方式建立，然而若欲推演出該系統解析解之數學模型，需要對目標系統進行深入的探討。但對複雜的製造系統而言，以物理為基礎建立適合的數學模型相當困難，因此有越來越多基於數據分析的預測模型被開發。
過往基於數據分析之刀具磨耗監測系統的研究中，是根據其所制定的切削條件建立刀具磨耗監控模型並進行辨識，對於此切削條件下之數據具有良好的辨識能力，但對於切削條件外的數據辨識能力卻鮮少著墨，如此一來監測模型的泛化性(generalization)便無法被驗證。本研究嘗試將特定切削條件下的所有數據排除於訓練數據集中，並做為測試樣本，測試模型之辨識率，藉此檢驗刀具磨耗監測模型之泛化性。
本研究利用主軸電流以及置於虎鉗上的加速規擷取之振動訊號，做為刀具磨耗狀態的對應特徵。主軸電流為機台本身的訊號來源，不需要額外的感測器，該訊號與刀具磨耗有高度正相關。而加速規安裝空間需求小、成本低，且有較高的信噪比(Signal-to-Noise ratio)，比切削力能更即時的監測刀具狀態的變化。本研究利用上述特徵建立機器學習模型以辨識刀具磨耗之狀態。當特定轉速數據排除於訓練數據集，對於該轉速數據的辨識準確率在類神經網路上很差。而經由前處理的改善並搭配隨機森林模型後，模型辨識準確率得以提升10%-20%。

關鍵字

振動訊號 ； 加速規 ； 刀具磨耗 ； 機器學習 ； 銑削

並列摘要

In recent years, the manufacturing industry has paid more attention to the development of models for predicting mechanical system failure or tool life. In past tool life prediction model in milling was established in a physics-based manner. However, for complex manufacturing systems, it is quite difficult to build a suitable mathematical model so predictive models based on data analysis are being developed.
In this study, the spindle current and the vibration signal captured by the controller and the accelerometers placed on the vise are used as the corresponding features of the tool wear state. The spindle current is the signal source of the machine itself, so no additional sensors are needed. This signal is highly positively correlated with tool wear. Accelerometers are small, low cost, and a high signal-to-noise ratio, which can monitor the change of tool status more instantaneously than the cutting force. This study utilizes the above features to establish a machine learning model to identify the state of tool wear, then compare the identification results of the artificial neural network with the random forest model, and discuss the physical meaning behind the important features. The recognition rate of the random forest model is 10%-20% higher than that of the neural network.

並列關鍵字

vibration signal ； accelerometers ； tool wear ； machine learning ； milling

參考文獻

[1] J. Corbett, R. A. McKeown, G. N. Peggs, and R. Whatmore, “Nanotechnology: International developments and emerging products,” CIRP Annals - Manufacturing Technology, pp. 523-545, 2000.

[2] L. Monostori, and H. Van Brussel, “Machine Learning Approaches to Manufacturing,” CIRP Annals, pp. 675-712, 1996.

[3] M.C.Shaw, “Metal cutting principles,” Oxford University Press, New York, 2005.

[4] J. S. Lemaster, “The use of process monitoring techniques on a CNC wood router. Part 1. Sensor selection,” Forest Products Journal, pp. pp.31-38, 2000.

[5] J. K. Tae-Yong Kim, “Adaptive cutting force control for a machining center by using indirect cutting force measurements,” International Journal of Machine Tools and Manufacture, pp. pp.925-937, 1996.