Abstract: The accuracy characteristics and robustness properties of mathematical model are significant, and its implementation impact of thermal error compensation of computer numerical control (CNC) machine tool technology in the practical cannot be ignored. The temperature of sensitive points and thermal error of the spindle in z direction are measured. Using some algorithms to establish prediction models, and fitting accuracy of different algorithms are analyzed. The temperature values of sensitive points and the thermal error in the experimental conditions of different ambient temperatures and different spindle speeds are measured. By comparing the prediction accuracy of various prediction models under different experimental conditions verify the robustness of the models. Experimental results show that the prediction accuracy of multiple linear regressions of least squares and least absolute estimation methods and distributed lag model is declined and their robustness are poor while support vector regression model has good prediction accuracy and its robustness remains strong when changing the experimental conditions. The robustness analysis of different models provides a useful reference for the thermal error compensation model selection of CNC machine tools and has good engineering applications.

Key words: distributed lag model, machine tools, multiple linear regression model, robustness, support-vector regression model, thermal error