Speaker
Description
Artificial intelligence is increasingly reshaping metallic materials design by accelerating exploration beyond conventional trial-and-error approaches. However, purely data-driven models often suffer from limited interpretability, excessive data requirements, and poor generalization across different alloy systems/processing routes. This study presents a physically guided AI framework for intelligent metallic materials design and processing, with a focus on steels and other high-performance structural alloys. By systematically embedding physical metallurgy knowledge into machine learning models, a multi-level strategy is established, spanning thermodynamics-informed learning, microstructure-centered deep learning, and mechanics-guided transfer learning. Thermodynamic and solidification-related descriptors are first incorporated to capture processing-dependent phase evolution, significantly improving prediction accuracy and alloy design rationality under limited data conditions. For complex, processing-induced microstructures, deep learning models guided by SEM/EBSD knowledge and multimodal imaging are developed to enable robust classification, quantification, and property prediction directly from microstructural images. Furthermore, physics-guided transfer learning frameworks integrating fatigue and creep mechanisms allow reliable prediction of property curves and long-term performance while reducing experimental cost. Representative applications are demonstrated in advanced steels, nickel-based superalloys, and aerospace and energy alloys. Overall, this work highlights how coupling physical metallurgy and processing physics with AI enhances accuracy, interpretability, and transferability, paving the way toward reliable, industry-ready metallic materials design.