Experiments on resistance spot welding of three layers of unequal thickness steel based on deep learning and multi-objective optimization

Resistance spot welding (RSW) of three-layer steel sheets with unequal thicknesses presents significant challenges in accurately simulating weld nugget formation and process signal behavior. This paper proposes a hybrid approach that combines deep learning and multi-objective optimization to improve simulation accuracy. A 1D convolutional neural network (1DCNN), bidirectional long short-term memory (BiLSTM), and attention mechanism are integrated to predict dynamic resistance curves from process parameters. These predicted curves are then used as benchmarks in an ANSGA-II and Bayesian optimization framework to calibrate thermal-electrical contact parameters in a finite element model. Experimental results demonstrate that the optimized simulations closely match measured data, achieving a mean absolute error (MAE) of 0.132, a root mean square error (RMSE) of 0.156, and an $R^2$ value of 0.91. The calibrated model reduces resistance prediction error by over 30% and improves nugget diameter and weld depth prediction accuracy across multiple thickness configurations. This integrated framework offers a practical and data-efficient solution for enhancing RSW simulations in complex multi-layer welding scenarios.