Multi-objective optimization of T-shaped bilateral laser welding parameters based on NSGA-II and MOPSO

In the realm of welding production, optimizing parameters for enhanced quality and efficiency remains a challenge. This study introduces a novel approach using ABAQUS for simulating the thermal-mechanical behavior in T-shaped double-sided welded joints of Q345 steel. Initial simulations were experimentally validated. A range of welding parameters was then established, and the "Unifrnd" function helped sample within this range. This study considers eight welding parameters as input variables for an artificial neural network (ANN), trained to predict residual stress and deformation. Optimization algorithms NSGA-II (non-dominated sorting genetic algorithm-II) and MOPSO (multi-objective particle swarm optimization) algorithm were employed to refine the ANN's outputs. Notably, the Pareto front revealed an optimal balance between minimizing residual stress and deformation. While single optimization achieved up to 5.12% reduction in residual stress and over 50% in deformations, the tri-objective optimization resulted in a more balanced reduction (about 1.88% in residual stress and over 20% in deformations). This highlights the need for parameters weighting in decision-making. The findings demonstrate the effectiveness of integrating ANN with optimization algorithms for welding parameter optimization, with implications for improved welding practices.