Prediction of residual stresses in welded structures based on neural network: a review

The wide application of welding manufacturing across key significant industries has aroused an increasing concern on weldments’ reliability. Weld residual stress, as a critical factor, is closely related to structural and material-level failures of weldments. Hence, the prediction of residual stress is necessary for ensuring weldments’ reliability. Neural network models have emerged as promising methods to achieve this prediction. In this review, the prediction of residual stress based on neural network models is reviewed to summarize current state and uncover their limitations. In terms of these limitations, this review gives substantial suggestions for future research on neural network modeling.