Establishment of a thermal compensation model for a turning–milling compound machine

Abstract Thermal errors are considered the most significant errors in precision machining. The heat sources of a turning– milling compound machine (TMCM) are diverse and complex. This study proposes a new thermal compensation model in a TMCM to compensate for thermal errors. First, the relationship between the tool tip deformation direction and the temperature increase trend is considered. The temperature points that have a pronounced influence on the tool tip deformation are selected from 76 measured temperature points. Temperature sensitive points are then identified through F-test feature engineering. The temperature increases and thermal deformation data of the temperature sensitive points under the characteristic and training conditions are used as the input and output in training a recurrent neural network model. When the recurrent neural network model was trained, the automatic parameter adjustment method of the RandomSearch in the Keras tuner toolbox was used. The minimum mean-square error (MSE) of the verified operating conditions was taken as the evaluation index, and a thermal compensation model was established. Compared with the average root-mean-square errors (RMSE) of the validation and test conditions along the six axes, the maximum difference in the average RMSE (only 2.12 μm) occurred in the direction of the spindle Y-axis. Noticeably, the thermal compensation model established in this study has high robustness and good universality. In addition, the average accuracy of the validation and testing conditions along the six axes improved by 84.3%. The proposed thermal compensation model is highly reliable and can effectively improve the thermal compensation accuracy.