Optimization of friction stir welding parameters for steel joints in shipbuilding using machine learning and finite element analysis

Friction Stir Welding (FSW) has emerged as a promising technique for joining steel in shipbuilding, offering enhanced mechanical performance and environmental sustainability compared to conventional welding methods. However, the application of FSW to steel remains constrained by challenges associated with high processing temperatures and forces, requiring advanced tool materials and precise parameter optimization. This study addresses these limitations by employing machine learning algorithms, including Gradient Boosting Regressor (R2 = 0.69) and Artificial Neural Networks (R2 = 0.70), alongside finite element (FE) simulations, to optimize key process parameters, such as tool rotational speed and welding speed, for steel joints commonly used in shipbuilding. Experimental data, collected in collaboration with Tringali Shipyard, were integrated with an extensive database from the literature, comprising approximately 200 records, to develop and validate predictive models. The results identify optimal process windows and correlations between parameters and joint performance, while FEA further enhance the database and refine the predictive accuracy. These findings provide actionable guidelines for the implementation of FSW in shipbuilding, demonstrating the potential of data-driven methodologies to accelerate the adoption of advanced joining technologies in industrial contexts.