Abstract: Conventional inherent strain methods are mostly derived by the thermoelastic-plastic finite element method in order to ensure the accuracy of the acquired inherent strain. However, using the thermoelastic-plastic finite element method to obtain the inherent strain of T-joints will consume a lot of time. In order to improve the efficiency and accuracy of obtaining the inherent strains of T-joints, the inherent strain values of 44 groups of T-joint welding experiments were obtained, and based on the data of the experiments, the transverse and longitudinal inherent strain values of T-joints were predicted by using the multilayer perceptual machine and the support vector machine regression algorithms, respectively, and the accuracies of the prediction results of the two algorithms were compared. For the prediction of transverse and longitudinal inherent strains of T-joints, the coefficients of determination (R²) of the support vector machine regression algorithm are 0.998 and 0.974 for the data test set, and the coefficients of determination (R²) of the multilayer perceptron algorithm are 0.985 and 0.945 for the data test set, respectively, The results show that the support vector machine regression algorithm is more accurate in predicting the inherent strains of T-joints compared with the multilayer perceptron algorithm. joint inherent strain values more accurately than the multilayer perceptron algorithm.

Key words: welding deformation, inherent strain method, MLP, SVR, T-joint