Thermal deformation prediction method for feed systems incorporating residual compensation and memory factors

Traditional temperature prediction mechanistic models are constrained by idealized boundary conditions during the prediction process, making it difficult to characterize actual operating conditions. Meanwhile, although data-driven models offer flexibility, they suffer from strong data dependency and insufficient robustness. To overcome these limitations, this study proposes an interpretable mechanistic-data-driven compensation term modeling method to achieve high-precision predictions of equipment temperature and thermal deformation. First, a dynamic compensation term for the temperature prediction mechanism model is constructed based on the error between actual and simulated temperatures; then, a WOA-optimized LSTM-XGBoost neural network fusion model is established to enable intelligent prediction of the compensation term; second, a memory factor is proposed to describe the accumulation of thermal deformation effects. The model’s validity was verified through temperature and thermal deformation experiments. The results indicate that: (1) component temperature prediction errors can be controlled within 2%; (2) the axial deformation of the lead screw reached 100.9 µm, and the peak radial deformation of the bearing was 12.35 µm; (3) the memory factor effectively characterizes the non-steady-state hysteresis effect of thermal deformation. This study provides a new modeling method and theoretical foundation for high-precision prediction of thermal deformation in precision equipment.