Uniaxial and multiaxial cyclic deformation behavior prediction of Z2CN18.10 austenitic stainless steel based on Transformer deep learning method

Abstract

In this study, the experimental study of cyclic deformation behavior of Z2CN18.10 austenitic stainless steel under uniaxial, cross, rhombus and circular paths is investigated. Based on Transformer deep learning method, three models are built to predict the cyclic deformation behavior of the material. One is the single input–single output model, which is used to predict the uniaxial cyclic deformation behavior. The second is the double input–single output model, which is used to predict the biaxial cyclic deformation behavior. The third is the time series multi-step model, which is used to predict the stress evolution under all paths. The Cross-Entropy loss function is adopted to train the models and the rationality of its application in the regression model is deduced from the theoretical point of view. Transformer deep learning method is based on attention mechanism, which can focus on the historical effect of Z2CN18.10 stainless steel cyclic deformation and accurately describe the uniaxial and multiaxial cyclic deformation behavior of the material.