Design the metastable austenite-containing advanced steels based on machine learning

Zhenli Mi^{1, 2*}, Yonggang Yang^{1, 2}

¹ Beijing Advanced Innovation Center of Materials Genome Engineering, Beijing, 100083, China

² Institute of Engineering Technology, University of Science and Technology Beijing, Beijing, 100083, China

EXTENDED ABSTRACT: The development of advanced steels goes from the traditional trial-and-error approach towards the prediction-and-validation approach. With the increasing requirements of the performance, the effect of the metastable austenite on mechanical properties draws the attentions of researchers. Hence, austenite stability have been taken into consideration in order to tailor the mechanical properties of advanced steels. Considering the reform of the development approach mentioned above, the combination of the tuning austenite stability and the prediction-and-validation approach can significantly benefit the design of advanced steels. It is known that austenite stability is associated with the stacking fault energy (SFE). The composition-SFE-property relationship predicted base on machine learning and validated by experiments method will definitely accelerate the development of metastable austenite-containing advanced steels. In the current study, three alloys systems, Fe-Mn, Fe-Mn-C, and Fe-Mn-Si-C, were taken into consideration; the data of SFE originated from the first-principles calculation, thermodynamic calculation, and experimental measurements were used; and the five machine learning regressions models, i.e. Linear Regression (LR), Regression Tree (RT), Gaussian Process Regression (GPR), Support Vector Regression (SVR), and Ensemble Learning Methods (ELM) were chosen. The correlation of 15 elements and the effect of the regression model were analyzed. Based on the related result, the composition-SFE-property relationship was revealed, which can benefit the development of advanced steels.