Prediction of weld morphology in laser-welded 316L stainless steel using a multilayer feedforward neural network

This study aims to predict the melt pool depth and width of 316L stainless steel welds during laser welding using a multilayer feed-forward neural network (MLFNN). The Taguchi method was employed to design the laser welding parameters and generate experimental data on melt depth and width. This allowed for an in-depth investigation of the effects of these parameters on the melt pool characteristics of 316L stainless steel. The results demonstrate that the MLFNN model, with a 3–10-10–10-2 structure, exhibits high accuracy and stability across training, validation, and testing phases. The correlation coefficient R-value between predicted and experimental results is 0.99995, indicating an excellent fit to the experimental data. The model’s predictions can effectively reduce defects in 316L stainless steel during laser welding, significantly enhancing weld quality.