CNN-LSTM-Attention based Predictive Model for Thermal Error in CNC Machine Tool Spindles

Aiming at the problem that most thermal error prediction models for machine tools adopt a single neural network architecture, which is difficult to comprehensively and deeply learn the multidimensional features of thermal error signals, a thermal error prediction model and compensation method based on a CNN-LSTM multi-attention mechanism are proposed. A CNN-LSTM-Attention series structure model and a parallel structure model are designed. Compared with traditional machine learning models and single neural network models, the proposed CNN-LSTM multi-attention model shows significantly higher precision and generalization ability in machine tool thermal error prediction, with the parallel CNN-LSTM-Attention model achieving the best performance. The parallel CNN-LSTM-Attention model is a multi-branch collaborative learning parallel neural network architecture that includes a CNN branch, an LSTM branch, and a direct input branch. It integrates multidimensional information through a feature fusion strategy. Experimental results show that the parallel CNN-LSTM-Attention model outperforms the series CNN-LSTM-Attention, LSTM, and GRU models in terms of model evaluation indicators such as MAE, RMSE, and R². The precision and reliability of the spindle thermal error prediction model are higher.