Abstract Machine learning (ML) has emerged as a powerful tool for accelerating the computational design and production of materials. In materials science, ML has primarily supported large-scale discovery of novel compounds using first-principles data, as well as digital twin applications aimed at optimizing specific manufacturing processes. However, developing and applying general-purpose ML frameworks to complex industrial materials, such as steel, remains a significant challenge. A major obstacle is accurately capturing the intricate relationship between chemical composition, processing parameters, and the resulting microstructure and properties. To address this gap, we introduce a computational framework that combines physical insights with ML to develop a physics-informed continuous cooling transformation (CCT) model for steels. Our model, developed using a dataset of unprecedented scale (4100 diagrams), is validated against both literature data and experimental results. The model demonstrates high computational efficiency, generating full CCT diagrams with 100 cooling curves in less than 5 seconds (Using a Lenovo ThinkPad with i7-8665U CPU), and shows strong generalizability across low-alloy steels and cooling conditions. It achieves phase formation classification F1-scores > 88 pct for all phases and cooling rates. For phase transition temperature regression, it attains mean absolute errors (MAE) < 20 °C across all phases except bainite, which exhibits slightly higher MAE of 27 °C. By extending the framework with additional generic and customized ML models, a platform can be established that serves as a universal digital twin for heat treatment and more. Integration with complementary simulation tools and targeted experiments will further support accelerated workflows.
Hedström et al. (Mon,) studied this question.