Hybrid modeling with finite element—analysis—neural network for predicting residual stress in orthogonal cutting of H13

Residual stress is an important surface integrity index to evaluate the crack initiation and failure of die surface. The efficient prediction of cutting residual stress can guide the high-quality machining of die and improve its service life. The existing cutting residual stress prediction models are complex, time-consuming and inefficient. In this paper, a hybrid prediction method of cutting residual stress based on finite element-analytical-neural network is proposed. Firstly, the stress, strain and temperature of the cutting surface are obtained based on the orthogonal cutting finite element model. Then, the stress relaxation analytical algorithm considering the elastic-plastic state of the material is used to replace the stress release process of the finite element, and the residual stress distribution data were obtained based on the joint model of finite element and analytical algorithm. Secondly, the surrogate model of BP neural network (SSA-BP) is improved based on SSA algorithm to realize the rapid prediction of characteristic value of residual stress. The effectiveness of the finite element-analytical-neural network hybrid model was verified by the cutting residual stress test of H13 steel. Finally, the effects of tool structure parameters and cutting parameters on the residual stress distribution and the maximum compressive stress and maximum tensile stress of H13 steel were studied. This method can provide a flexible and efficient basic model for obtaining the optimal cutting conditions for controlling the residual stress of H13 steel and other metals.