Research on TTT Curve of Alloy structural Steel based on Machine Learning

Abstract

Austenite isothermal transformation curve (IT) of steel, also known as time-temperature-transformation curve (TTT) is an very important basic data for the heat treatment process design of steel. Traditionally, obtaining TTT information on metal mainly depends on experiments, so there are problems such as data dispersion, large errors, and inconvenient access in use. Using artificial intelligence and machine learning technology, the TTT curve of steel can be predicted with limited experimental data. Taking the authoritative data collected as training samples, the TTT curve of alloy structural steel was predicted based on a variety of machine learning algorithms.. Alloying element category, austenitizing temperature, phase transformation time are taken as input characteristics, and 10 kinds of transformation characteristics are taken as output targets. Correlation coefficient (R), and error analysis (RMSE, MAE) are used to evaluate and finalize the model, and the best algorithm is selected to form a combined-machine-learning algorithm (CML), and predict the TTT curve. Take the application of CML multi-model prediction method in 40Cr, 38CrMoAl, 35SiMn, and 20Mn2, and the predicted results reflect that the CML model has high prediction ability and good generalization.