Study on time-temperature-transformation diagrams of stainless steel using machine-learning approach

Abstract

Time-temperature-transformation (TTT) diagram is an essential tool for studying the microstructure of stainless steels to enhance properties. Therefore, it is of great practical importance to predict TTT diagram accurately and rapidly. In the present work, a combination of machine learning (ML) algorithms, including BP artificial neural network, Random Committee, Random Forest and Bagging, is developed for prediction of TTT diagram with relevant descriptors containing the alloying elements, austenitizing temperature and holding time. The results show that, such combination can achieve high predictive accuracy on the TTT diagrams of stainless steel with a high correlation coefficient value and low root mean squared error value. Comparison among the results of ML and commonly used empirical formulas illustrate the obvious advantage of machine learning approaches. Also, the ML techniques have advantages over the commercial JMatPro software which can predict TTT diagrams for only a part of stainless steels.