Modeling the relationship between tempering process-microstructure-property of steel using deep learning

• StyleGAN was introduced to model the tempering process-microstructure relationship. • The generated images showed high similarity to real microstructures. • Resnet101 was applied to model the microstructure-property relationship. • The predicted properties showed good agreement with experimental properties. The slewing bearing module is extensively used in large-scale machinery and equipment, requiring high strength. For manufacturing slewing bearings, 42CrMo4 steel is preferred due to its excellent mechanical properties, achieved through various manufacturing processes. Among these, tempering critically influences both microstructure and mechanical characteristics. Conventionally, evaluating the relationships among processing conditions, microstructure, and properties relies on repetitive experiments, which are time-consuming, costly, and limit exploration of large design spaces. To address this, our study introduces a deep learning-based approach to capture the process–structure–property (PSP) relationships of 42CrMo4 steel under various tempering conditions. Steel samples were tempered at multiple temperatures, and the resulting microstructures were characterized by optical microscopy (OM), while tensile strength, yield strength, and elongation were measured through tensile testing. A conditional StyleGAN model was employed to model the link between tempering temperature and microstructure, and a ResNet-101 model was used to predict mechanical properties from microstructural images. The StyleGAN-generated microstructures showed strong agreement with experimental observations, and the ResNet-101 predictions closely matched measured properties. Both models also demonstrated robust generalization to previously unseen tempering conditions. These results highlight the potential of deep learning to reduce experimental workload and reliably predict material behavior under new processing conditions.