Analysis of friction stir welding in SS304L using transient coupled thermomechanical finite element and machine learning-based Gaussian Process Regression framework

Abstract

Friction stir welding of austenitic stainless steel 304L was numerically investigated using a three-dimensional fully coupled transient thermomechanical finite element model at different rotational (angular) speeds (300, 400 and 500 revolutions per minute (rpm)), respectively. The computational model integrates temperature-dependent material properties and verified mesh independence constraints to predict temperature distribution, heat flux, stress, strain and deformation behaviour. The results indicate a maximum temperature of approximately $600\circ \mathrm{C}$ at 500 rpm, accompanied by a reduction in equivalent stress and elastic strain due to thermal softening. Total deformation and material flow were observed to increase with rotational speed, indicating enhanced plasticization at higher heat input conditions. Gaussian Process Regression models are trained using finite element model (FEM) data, generated from the computation methods to establish relationships between process parameters and thermomechanical responses. The model achieved coefficients of determination $(R^2)$ of 0.945, 0.783, 0.998 and 0.996 for temperature, stress, strain and deformation, respectively, with low prediction error and quantified uncertainty. The integrated FEM-Gaussian Process Regression framework demonstrates high predictive capability and computational efficiency, offering a reliable approach for process optimization and performance prediction.