关键词: 有限元整体应力修匀, BP神经网络, 两种训练模型, Mises应力

Abstract: This paper takes the quadrilateral plane stresses isoparametric element as an example and Gaussian integral points as the sample points, based on the BP neural network, there are two training models investigated：The coordinates of the Gaussian integration points are input and the Mises stress is output. The coordinates and displacements of the Gaussian integral points are input at the same time, and the Mises stress is output. Compared with the traditional finite element global stress smoothing, the Mises stress of the nodes is studied by BP neural network, whether it is the coordinate as the input or the coordinate and the displacement as the input at the same time, the accuracy of the Mises stress is higher than that of the traditional finite element global stress smoothing, and the second type of BP neural network training model has the higher accuracy than the first one.

Key words: finite element integral stress smoothing, BP neural network, two training models, Mises stress