Spindle thermal error modeling method considering the operating condition based on Long Short-Term Memory

Abstract

Thermal error has a significant effect on machine tools’ precision. Traditional methods usually establish the thermal error model by temperature. As the temperature only describes the instant state of the machine tool, it is inadequate to clarify the generation process of thermal error. To improve the description completeness of thermal error generation process, the operating condition data such as spindle speed, rotational number and power are employed besides the temperature. Furthermore, Long Short-Term Memory (LSTM) based thermal error model was proposed which combined the operating condition with temperature. This model not only considers the dynamic characteristics of the thermal error, but also selects and remembers the previous thermal error data information. To improve the accuracy for the whole speed range, the K-Means clustering analysis was adopted to obtain the optimal speed categorization, then different LSTM sub-models were established separately. Experiments demonstrated that the predicting accuracy of the proposed K-Means segmental LSTM method (K-S-LSTM) is over 85.0% under different operating conditions. Moreover, comparison experiments indicated that the predicting performance of K-S-LSTM is better than Elman and IRNN models. Finally, the compensation experiment showed that the thermal error can decrease from 49 μm to 12 μm after compensation in actual numerical control system.