Mechanism-data hybrid-driven model for spatial thermal error prediction of horizontal machining center

Thermal error is one of the main factors affecting the machining accuracy of machine tools. Existing thermal error prediction methods mainly focus on local components and have limited ability to characterize spatial thermal errors (STE) across the machining space. To address this problem, a mechanism-data hybrid-driven model, named thermal perception physics-constrained xRFM (TPC-xRFM), is proposed for spatial thermal error prediction of machine tools. A transient finite element model (FEM) is first established to obtain the temperature field and thermal deformation field of the machine tool, and the simulated temperature data are fused with measured data to provide mechanism-based prior information for model training. A temperature-weighting strategy based on the average gradient outer product (AGOP) is then introduced to identify and enhance thermally sensitive variables, while thermal error consistency and delay constraints are incorporated to improve physical consistency and temporal stability. In addition, a rapid STE measurement scheme is designed to acquire multi-position thermal errors in the machining space. Experimental results show that TPC-xRFM significantly outperforms the baseline models in different directions, providing an effective and interpretable solution for spatial thermal error prediction of machine tools.