Thermal Error Prediction of Five-axis Machine Tools Based on Collaborative Mapping of Multi-physics Field Simulation and Spatio-temporal Network

The original data constructed in this study is derived from a high-fidelity "thermal-structural" bidirectional coupled finite element simulation of a five-axis high-speed machining center. Using the ANSYS Workbench platform, time-series data of the temperature field and thermal deformation field during four hours of dynamic operation were obtained. The temperature data cover the temperature rise curves of ten key thermally sensitive areas, including the spindle front bearing, guide rail pairs, ball screw nut pairs, and the rotary table, with temperatures ranging from an initial 22°C to a maximum of 47.68°C. The thermal deformation data synchronously record the displacements of the tool tip in the X, Y, and Z directions, exhibiting significant anisotropic characteristics. Notably, the Z-axis shows a non-monotonic variation pattern of "decreasing first and then increasing." Finally, the continuous time-series data are constructed into a physics-informed dataset containing 100 standardized samples using the sliding window method. Each sample consists of the temperature time series from ten measurement points and the corresponding three-axis thermal error values, forming a "temperature-deformation" mapping pair with clear physical significance. It can be seen that the high-fidelity simulation replaces physical experiments, which are costly and limited in operating condition coverage, providing the deep learning model with physically consistent and comprehensively covered training samples, which holds certain practical significance.