Machine learning-enhanced laser cladding process for high-entropy alloy coatings with concurrent strength and ductility optimization

Abstract

This study proposes a machine learning framework integrated with laser cladding additive manufacturing to enable rapid development of high-entropy alloy (HEA) coatings with balanced strength and ductility. We demonstrate a closed-loop manufacturing methodology combining composition-driven predictive modeling with process-aware experimental validation. Six machine learning models (RF, ANN, SVM_rbf, GBM, LightGBM, XGBoost) were implemented to establish direct composition-property correlations from a curated HEA database, with LightGBM achieving superior prediction accuracy (EL: R² = 0.938, RMSE = 4.76 %) and GBM excelling in strength modeling (YS: R² = 0.858, RMSE = 184.82 MPa). SHAP analysis quantitatively identified Al and Nb as dominant elements governing ductility-strength trade-offs. A multi-objective genetic algorithm then generated Pareto-optimal compositions, which were manufactured through laser cladding – a high-precision additive technique. Experimental characterization revealed that process-induced microstructural evolution (grain coarsening, residual stress distribution) mediates the translation of computational predictions to actual mechanical performance. The hybrid approach demonstrates significant reductions in development cycles and costs compared to conventional trial-and-error methods, validated through Vickers hardness testing and tensile characterization. This work establishes a replicable paradigm for AI-enhanced smart manufacturing systems, where computational alloy design and laser-based processing are co-optimized to achieve target component specifications while maintaining production efficiency.