Intelligent prediction of mechanical properties in metallic materials based on machine learning: A review & perspective

The rapid advancement of machine learning is transforming the research paradigm of mechanical properties of metallic materials from experience-driven to data-driven. This review systematically summarizes recent progress and challenges in machine learning based prediction of mechanical properties in metallic materials. We first outline commonly used ML algorithms and workflows, with an emphasis on cutting-edge methods such as explainable AI and physics-informed machine learning. We then review typical applications and predictive performance of ML models across three scales: micro/mesoscopic properties (e.g., microstructural evolution, fracture behavior), macroscopic properties (e.g., hardness, stress response, fatigue life), and cross-scale coupling properties (e.g., flow stress, yield strength, constitutive parameter inversion), highlighting their advantages in high-throughput computation and multi-scale modeling. Finally, we identify persistent challenges such as data scarcity, heterogeneity, and insufficient accuracy under wide temperature/strain-rate ranges, and propose potential solutions including transfer learning, large language models, and multi-modal fusion. Looking forward, we outline a technical pathway integrating multi-modal data and physical mechanisms for accurate prediction of mechanical behavior under extreme conditions, aiming to advance materials mechanics toward digitalization and intelligence.