Axis-Space Synchronized Motion Error Estimation for Five-Axis Contour Error Control

This study addresses the modeling and prediction of tracking error in CNC machine tool feed systems by developing a neural network prediction model based on an LSTM-Transformer architecture. To overcome the limitation of conventional data in adequately exciting the dynamic characteristics of servo systems, a featured trajectory construction method is proposed. By incorporating excitation trajectories that encompass various operating conditions—such as startup stopping sequences, direction reversals, and speed variations—the distribution of motion states in the training dataset is systematically enriched, thereby enhancing the model's ability to characterize nonlinear dynamic behaviors. Comparative experiments conducted under the same dataset with conventional LSTM and GRU networks demonstrate that the proposed modeling and featured trajectory design approach significantly reduces the prediction error of the feed system output while maintaining strong generalization performance under complex trajectories. This provides high-precision prior information for subsequent contour error analysis and compensation in CNC machine tools.