New insight in predicting martensite start temperature in steels

Martensite start temperature (MS) plays an important role in the chemical composition and heat treatment designs of advanced high-strength steels. To improve the performance of MS prediction model trained using machine learning methods, dataset augmentation strategy and the addition of new features play a key role. In the present work, the effects of new features related to atomic parameters on the performance of the prediction model trained using an machine learning algorithm were studied and discussed by using the datasets obtained from the literature and industrial companies. In addition, the effects of the interaction between carbon and alloying elements on martensite start temperature are studied. The results revealed that the new features related to electronegativity difference between the alloying elements and C, the number of covalent electrons and the lattice constant of austenite grains considerably improved the prediction performance of the trained model. Moreover, Si-C interaction intensified the role of C in reducing martensite start temperature, whereas V-C, Ni-C, N-C, Mo-C, Cr-C and Al-C interactions weakened the role of C. [GRAPHICS] .