Multidimensional thermo-mechanical simulation of laser welding dynamics in magnesium–steel joints via hybrid finite element–machine learning frameworks

Lightweight structural welding of dissimilar magnesium-steel joints Laser welding is essential but faces challenging complex thermo-mechanical behaviour because of material mismatch. This paper introduces a welding dynamics optimization and modeling hybrid finite element (FE) and machine learning system. To model transient temperature field, melt pool development and residual stress development under different laser parameters, a multidimensional thermo-mechanical FE model was developed. Artificial neural networks (ANN) were trained on high-fidelity FE data to make scalar predictory forecasts and deep neural networks (DNN) to make high-dimensional spatiotemporal field predictions. ANN was able to learn nonlinear correlations between process parameters and weld quality measures, whereas DNN was able to reproduce temperature and stress distributions. The thermocouples and infrared thermography Assessed with thermal experiments showed a high degree of agreement with the predictors of FE with a difference of less than 6%. The hybrid framework obtained a coefficient of determination, which was greater than 0.94 and prediction errors of less than 5%. Furthermore, the calculation time was cut about 80% over so-called full FE simulations. It was observed as a result of sensitivity analysis that laser power and scanning speed have a significant impact on thermal gradients, melt pool geometry, and residual stress distribution. The suggested method allows quick prediction, optimization of the process and better quality of the weld. This framework offers a scalable and computationally efficient platform of intelligent laser welding of dissimilar magnesium steel joints and offers future application with real-time monitoring and digital twin systems.