2021-03-29 13:30 Estimating the Thermal-Error Model of a Machine-Tool through Machine Learning Spindle
演 講 者：胡敏忠 國立宜蘭大學機械與機電工程學系教授
組 別：全體組別
演講時間：2021-03-29 13：30
演講地點：MEA401演講廳
演講摘要：
The thermal error of spindle is a primary factor that affects the cutting accuracy of a machine tool. During machining, increased temperatures due to friction between bearings and joints cause the thermal deformation of the spindle, thereby affecting the cutting accuracy of the machine tool. This paper used three machine learning algorithms to establish an estimation model for the thermal error of a high-speed spindle by using some of its characteristic temperature points and its rotation speed. The performances of the three machine learning algorithms, namely feedforward neural network (FFNN), gated recurrent unit (GRU), and extreme gradient boosting (XGBoost), were also discussed. The FFNN was simple and adaptive but could not reflect the dynamic temporal behavior of the system. GRU comprises gating mechanisms in recurrent neural networks and can more completely describe the temporal dynamic behavior of the system. XGBoost uses an ensemble learning algorithm to greatly improve the performance of the tree model, and it is fast, accurate, and has high reliability. The experiments demonstrated that in thermal error estimation for the spindle, GRU and XGBoost models outperformed FFNN model. Specifically, the error between the prediction result of XGBoost model and the experimental result was within ±3 μm.