Effect of Welding Parameters and Artificial Intelligence-Based Prediction of Maximum Temperature in Friction Stir Welding of AA3003 Alloy

The present study highlights the influence of feed rate, rotational speed, and tool inclination angle on the evolution of the maximum temperature generated during friction stir welding (FSW) of aluminum AA3003. To predict these thermal variations, six machine learning models were developed and trained using a dataset composed of 64 experimental trials covering a wide range of process parameters. The models include three artificial neural networks (ANNs) optimized using the Levenberg-Marquardt (LM), Scaled Conjugate Gradient (SCG), and Bayesian Regularization (BR) algorithms; one support vector machine (SVM) with a quadratic kernel; and two Gaussian Process Regression (GPR) models with Matérn 5/2 and exponential kernels. The models were evaluated using standard statistical indicators (RMSE, MAE, R²). The results demonstrate the superiority of the GPR model with a Matérn 5/2 kernel, with an RMSE of less than 0.02°C and an R² coefficient of determination close to unity. This model also stood out for its robustness on unprecedented configurations, with a relative error of less than 1.6%. The proposed approach demonstrates the potential of machine learning techniques to model the complex thermal phenomena of FSW accurately and represents a step towards intelligent predictive control of welding processes.