Prediction of Thermal Deformation and Real-Time Error Compensation of a CNC Milling Machine in Cutting Processes

Abstract

This study constructs a prediction model of thermal deformation with an artificial neural network and implements the real-time error compensation for a three-axis vertical CNC milling machine in cutting processes to improve the thermal error of the workpiece. There are 32 PT-100 thermal sensors installed in key parts of the machine in order to measure the temperature of key machine parts in actual cutting conditions. Pearson’s correlation coefficients are used to select crucial temperature sensors for building the prediction model of thermal deformation. The reduced number of crucial temperature sensors in model construction can simplify the model complexity and speed up the response time of prediction. This study constructs a long short-term memory (LSTM) neural network model to predict the thermal error of the machine in cutting processes. This prediction model of thermal deformation can be further used in real-time error compensation of the workpiece in cutting processes. In an 8 h cutting experiment, the dimensions of the workpiece show that, with real-time error compensation, the thermal error in X-axis decreases from 7 µm to 3 µm, the thermal error in Y-axis decreases from 74 µm to 21 µm, and the thermal error in Z-axis decreases from −64 µm to −20 µm. The results indicate that the prediction model of thermal deformation and the real-time error compensation can significantly reduce the thermal error and improve the dimensional accuracy of the workpiece.

Keywords

• thermal deformation
• long short-term memory (LSTM)
• prediction model
• real-time error compensation