Progress of Machine Learning Applications in Steel Research
Author of the article:CHEN Wenhui1,2,3,4, YEXinlan1,2,3,4, ZHANG Jiaxue1,2,3,4, ZHANG Menghao1,2,3,4, ZHANG Yan5, WU Hongxing1,2, WANG Weijun6,7, ZHI Huihui1,2,3,4, WANG Haifeng1,2
Author's Workplace:1. State Key Laboratory of Solidification Processing, Northwestern Polytechnical University, Xi'an 710072, China ; 2. Center of Advanced Lubrication and Seal Materials, Northwestern Polytechnical University, Xi'an 710072, China; 3. Research & Development Institute of Northwestern Polytechnical University in Shenzhen, Shenzhen 518063, China; 4. Innovation Center NPU Chongqing, Chongqing 401135, China; 5. Northwest Institute for Non-ferrous Metal Research, Xi'an 710016, China; 6. Hefei Institutes of Physical Science, Chinese Academy of Sciences, Hefei 230031, China; 7. Institute of Energy, Hefei Comprehensive National Science Center, Hefei 230031, China
Key Words:steel materials; machine learning; predictive modelling; material design; microstructure and properties
Abstract:
Steel materials have attracted sustained attention because of their excellent mechanical properties and broad compositional design space; however, the intrinsic complexity of the composition-processing-microstructure-property relationships makes them difficult to efficiently resolve using traditional approaches. Machine learning offers a promising alternative, yet its application remains constrained by issues such as data quality, generalizability, and physical consistency. In this work, machine learning methods in steel research from the perspective of representative data modalities and task types were systematically reviewed. Focusing on key problems, including microstructure segmentation, property mapping, and stacking fault energy prediction, the applicability and limitations of different modelling paradigms were summarized.Furthermore, fundamental challenges were critically examined, such as data heterogeneity, inconsistent evaluation protocols, and limited extrapolation capability. This review indicates that while machine learning holds significant promise for accelerating property prediction and material design in steels, its effective deployment relies on high-quality data paradigms, rigorous validation strategies, and the integration of physical knowledge. Future efforts should emphasize cross-modal data integration, model interpretability, and engineering transferability, thereby advancing toward a "data-knowledge co-driven" paradigm for materials design.