Part machining deformation prediction based on spatial-temporal correlation learning of geometry and cutting loads

The machining deformation prediction and control is the key issue to ensure the dimensional accuracy of parts and guarantee the quality of products. However, with the material removal and residual stress release, the part geometry and cutting loads keep changing during the machining process, and the complex superposition of the two eventually affects the machining deformation, making it difficult to establish a theoretical prediction model. Moreover, due to the limitation of the theoretical approach, the model cannot be continuously modified with the change of the cutting loads affecting the machining deformation during the actual machining process, which leads to low prediction accuracy. In this paper, we propose a prediction method for machining deformation based on spatial-temporal correlation learning of geometry and cutting loads. Firstly, the process of geometrical state change of workpiece considering material removal was modeled, and then the time-varying process model of external loads including cutting force and clamping force was established. Next, according to the tool trajectory during machining, the theoretical geometric variation and the actual cutting loads variation are fused in tem-poral and spatial terms, and the fused information also reflects the deformation state and internal stress variation of the workpiece. Finally, a deep spatio-temporal learning network is proposed to model the complex geometry-loads-deformation relationship for deformation prediction. The proposed method was validated in the actual part machining process, and compared with the end-to-end model as well as the simulation model, results proved that the method can accurately predict the machining deformation. The proposed method combines the theoretical variation of the machining process with the actual physical quantity variation data, and in terms of the loads variation predicted by the model and the actual collected loads, the model is more consistent with the actual machining state.