Development of Surrogate Models for Predicting Thermal and Mechanical Responses in Friction Stir Welding of Aluminum Plates

This paper presents a computationally efficient framework for modeling Friction Stir Welding (FSW) by integrating thermomechanical Finite Element Analysis (FEA) with data-driven surrogate modeling. We validate an analytical heat-input model against Coupled Eulerian–Lagrangian (CEL) simulations and develop surrogate models for the prediction of thermal and mechanical responses, such as temperature, stress, and distortion. Experimental measurements of thermal history and distortion were conducted. The temperature data are used to validate both the CEL simulations and the analytical heat-input model implemented in an explicit FEA framework, while the distortion measurements are used to validate the stress analysis model under cooling conditions in an unclamped setup. Although CEL captured detailed material flow behavior, the analytical approach provides broader parametric analysis with significantly reduced computational cost. Finite element simulation data with 30 different combinations of transverse speeds (100–500 mm/min) and rotational speeds (500–1750 rpm) are used to train surrogate models with high predictive precision (R2 = 0.99). This provides a reliable tool for optimizing the FSW process.