Modelling and prediction of thermal error in rotary axis of machine tools under variable working conditions based on TPE-IndRNN

Thermal error of rotary axis is one of the key factors affecting the machining accuracy of 5-axis machine tools. Currently, traditional methods struggle to model thermal error sequences over varying working conditions and extended periods. A spectral clustering method is utilized to automatically select temperature-sensitive points. Furthermore, a lightweight and efficient independent recurrent neural network (IndRNN) model is developed to accurately model and predict the thermal positioning errors of the rotary axis. The proposed model is capable of processing longer time series data. Training and predictive validation were conducted under variable working conditions. This lightweight model not only captures the time lag and thermal errors in temperature data but also exhibits strong generalization capabilities under varying working conditions. Experimental results indicate that the model's predictive accuracy exceeds 95%, with a maximum residual of less than 0.78μrad. It outperforms comparative models such as LSTM, GRU, and BiLSTM in terms of training speed and accuracy.