Machine learning-assisted prediction of mechanical properties in WC-based composites with multicomponent alloy binders

Abstract

The development of WC-based composites for harsh environment applications has been impeded by trial-and-error approaches, which are inherently time-consuming and costly. In this study, a machine learning (ML) framework was developed to rapidly predict the hardness and fracture toughness of WC-based composites, focusing on alternatives to conventional Co binder that was susceptible to corrosion in marine environments. Experimental data were collected from published literature and used to train three ML models, i.e., backpropagation neural networks (BPNN), gradient boosting decision tree (GBDT), and support vector regression (SVR). The results showed that the BPNN algorithm demonstrated good predictive performance, achieving R² values of 0.913 and 0.906 for hardness and fracture toughness, respectively. The predictive accuracy was experimentally validated using samples prepared with binders composed of Co, Ni, Fe, or their alloys. SHAP (SHapley Additive exPlanations) analysis revealed that grain size significantly impacted the hardness model of WC-based composites, and electronegativity was the most influential chemical descriptor affecting the hardness and fracture toughness models. This proposed framework shows the effectiveness of the ML approach for the development of multicomponent alloy binders in WC-based composites, with superior mechanical properties and enhanced applicability in harsh environments.