Deep learning accelerated the discovery of multi-principal element alloys with various strength-toughness trade-offs

Machine learning has significantly enhanced the efficiency of multi-principal element alloys (MPEAs) development. Nonetheless, despite its potential, the rapid discovery of MPEAs with various strength-toughness trade-offs remains a largely unexplored area. This challenge lies in the inherent trade-off between strength and toughness, the complexity and scarcity of existing MPEAs data, and the absence of efficient strategies for Pareto front optimization in high-dimensional and sparse composition design spaces. Here, we present an alloy design framework that integrates multiple deep learning models and Pareto optimization algorithms to address these challenges. Remarkably, through merely three iterations, the framework yields eight MPEAs that notably surpassed the original dataset benchmarks, showing varied strength-toughness trade-offs. Microstructural analysis further confirmed the framework's ability to influence phase formation and microstructure through precise alloy composition adjustments, achieving outstanding and various strength-toughness combinations. Given its effectiveness, it holds substantial application potential in accelerating the design of materials tailored to meet a wide range of strength and toughness requirements.