Machine learning-driven optimization of hardness and toughness in high-entropy alloy coatings based on composition and descriptor

Abstract

Optimizing the mechanical properties of high-entropy alloy coatings, a class of high-performance materials, is critical for advanced engineering applications. Here, we present an innovative strategy that integrates generative adversarial networks with interpretable machine learning to overcome two key challenges in machine learning-assisted material design: the limitation of small sample data and the complexities of mapping component properties. We develop a de-heterogeneous conditional generative adversarial network that effectively addresses the modeling challenges posed by data scarcity and distribution imbalance, establishing a mapping that connects composition, process, and descriptor to performance. The model achieves high predictive accuracy, with R2 values of 0.93 and 0.88 for hardness and elastic modulus, respectively, on the test set. Our modeling analysis further revealed that the elemental heat of evaporation is a key factor influencing hardness variations. In particular, the high heat of evaporation of W (824.0 kJ/mol) contributes to the enhanced hardness of coatings. By incorporating the multi-objective optimization algorithm, we obtain the Pareto-optimal solution set, identifying the Al–Fe–V–Co–Nb–Ti–Mo–W–Ta alloy as a promising candidate with superior mechanical properties (hardness >20 GPa, elastic modulus <115 GPa). This machine learning framework provides a new paradigm for addressing the strength–toughness trade-off in material property optimization.