A Predictive Model of Spindle Thermal Error Based on DCGAN

Thermal error is one of the main error sources to affect machining accuracy, accounted for 40%-70% of the total error for machining. The compensation of thermal error is an efficient way to improve machining quality. Tradition data-driven modeling methods of thermal error usually relied on the selection of key temperature measurement points and training based on the shallow neural network with these selected temperature measurement points. Furthermore, the different working-conditions of CNC machine tools have different key temperature measurement points. In this paper, a new modeling method based on deep conditional generative adversarial network (DCGAN) is proposed. This method can automatically extract the features. The experiment results show that the method could reach an accuracy of 91% for thermal error prediction.