數控加工中心機之刀具熱誤差建模

Abstract

現今大多加工廠都採用濕式加工，但是切削液的排放會造成環境汙染，加上近年來環保意識抬頭，因此乾式加工成為了永續加工的趨勢。然而，乾式加工會面臨熱誤差的問題，但是目前針對熱誤差補償的研究將重心放在工具機主軸，因為主軸的溫度與熱變形量較好量測。因此，我選擇建立面銑刀具的熱誤差預測模型，提供一個預測模組，其輸入為切削參數，輸出為切削溫升與刀具熱變形量之關係，達到透過實時的監控切削溫度來預測即時的刀具熱誤差值得效果。為了建立刀具熱誤差模型，使用無線溫度感測模組量測切削溫度；使用雷射位移計量測刀具熱變形量。蒐集了不同切削參數下的切削溫升和刀具熱變形量之關係，並使用了以下五種機器學習模型: Random Forest、Extre Tree、Support Vector Regression、Gradient Boostiong Decision Tree和Extreme Gradient Boosting。切削溫度模型的特徵為每刃進給的切削槽體積、切削速度和切削時間；標籤為切削溫度。刀具熱變形量模型的特徵為切削溫度和切削時間；標籤為刀具熱變形量。根據模型的訓練結果，最佳的切削溫度模型為Extre Tree，RMSE值僅有1.05℃；而性能最佳的刀具熱變形量模型為Random Forest，RMSE值僅有1.06微米。

Nowadays, most manufacturing plants use wet machining, but the discharge of cutting fluids causes environmental pollution. With the rise of environmental awareness in recent years, dry machining has become a trend for sustainable manufacturing. However, dry machining faces the challenge of thermal error. Currently, research on thermal error compensation focuses on the machine tool spindle because the temperature and thermal deformation of the spindle are easier to measure. Therefore, I chose to establish a thermal error prediction model for face milling cutters. This model provides a predictive module, where the input is cutting parameters, and the output is the relationship between cutting temperature rise and tool thermal deformation. The goal is to achieve real-time prediction of tool thermal error through real-time monitoring of the cutting temperature. To build the tool thermal error model, I used a wireless temperature sensing module to measure the cutting temperature and a laser displacement sensor to measure tool thermal deformation. I collected the relationship between cutting temperature rise and tool thermal deformation under different cutting parameters and utilized the following five machine learning models: Random Forest, Extra Tree, Support Vector Regression, Gradient Boosting Decision Tree, and Extreme Gradient Boosting. The features of the cutting temperature model include the cutting slot volume per tooth feed, cutting speed, and cutting time, with the label being the cutting temperature. The features of the tool thermal deformation model include cutting temperature and cutting time, with the label being the tool thermal deformation. According to the training results of the models, the best cutting temperature model is the Extra Tree model, with an RMSE value of only 1.05℃. The best-performing tool thermal deformation model is the Random Forest model, with an RMSE value of only 1.06 microns.