Research on High-Speed Electric Spindle Thermal Error Modeling Based on Stacking Ensemble Learning Method

Spindle thermal error is a significant factor affecting the machining accuracy of CNC machine tools. This study addresses the limitations of single thermal error prediction models, which often show large prediction errors and poor generalization in complex working conditions. Based on the VF150-X01 electric spindle manufactured by China General Technology Dalian Machine Tool, relevant thermal error data was collected to establish four candidate base learner models: BP, BiLSTM, GRNN, and GAM. The best-performing models under both constant speed and complex working conditions were selected for stacking ensemble learning modeling research. Comparative analysis shows that the ensemble learning model improved prediction accuracy by 0.189µm under constant speed conditions and by 0.178µm under complex conditions compared to the best single model. These results demonstrate that the stacking-based ensemble learning model offers higher prediction accuracy and better generalization capability.