Molten pool characterisation in multi-track laser cladding: A data-driven study on steel/superalloy coatings on C45 steel

Abstract

This study investigates the use of machine learning (ML) to predict and analyse the thermal and geometric characteristics of molten pools formed during laser cladding of steels and superalloys onto C45 steel. Three ML models – random forest (RF), XGBoost, and support vector regression (SVR) – were trained using a comprehensive dataset and evaluated for predictive accuracy. XGBoost outperformed the others, achieving R² values of .951 for maximum temperature, .932 for effective temperature gradient, .913 for dilution rate, and .945 for shape factor. However, this higher accuracy required longer training times due to the model's complexity. RF showed greater robustness to input feature variation, while XGBoost and SVR were more sensitive to feature selection, occasionally overlooking important parameters. Finally, case studies involving SS316 cladding on C45 steel validated the models' practical relevance and demonstrated the potential to optimize cladding parameters for improved control over molten pool behaviour and cladding quality