Modeling Continuous Cooling Transformations for HSLA Steels With Physical Metallurgy Guided Hereditary Machine Learning

Phase transformations during continuous cooling play a vital role in controlling final microstructure and mechanical properties of hot-rolled high-strength low-alloy (HSLA) steels. Therefore, accurate prediction of continuous cooling transformation (CCT) diagrams is the key to optimizing hot-rolling processes. But, because phase transformation behaviors are complex and the accumulated data are insufficient, it is of great difficulty to accurately model CCT diagrams. In this paper, a hereditary modeling method based on the combination theories of physical metallurgy (TPM) and machine learning (ML) is proposed. Through thermodynamic and kinetic analyses, the key factors affecting behaviors of continuous cooling transformation are clarified. Combined with the existed data, the feature parameters in direct correlations with phase transformation temperatures are obtained by theoretical calculations. By using the algorithm of support vector machine (SVM), the model for predicting CCT diagrams has been developed, demonstrating superior prediction accuracy over the traditional data-driven ML models, especially in predicting the temperatures for pearlite and bainite transformations. By applying the established ML models to industrial production of HSLA steel plates, their CCT diagrams were predicted and verified through metallographic observations of final microstructures formed under different cooling paths.