Machine Learning Approach for the Prediction of Mechanical Properties Based on the Steel Microstructure Volume Fraction Ratio

Abstract

Artificial intelligence in material science has attracted many attention in 4.0 industrial revolution contemporary. The process optimization for desired mechanical property is one of the important issue with machine learning algorithm to verify the optimal combination of parameters. Cooling conditions are known to affect the volume fractions of AF, GB, BF and MA of bainitic steel. With increasing cooling rate, grains are generally refined in conventional steels as the grain growth is constrained, and the volume fraction of MA increases as the martensite start temperature (Ms) is rapidly reached. In the bainitic steels, the volume fractions of GB and BF formed at lower temperatures than AF increase under fast cooling rates, but the low-temperature toughness decreases as GB and BF have larger effective grain size than AF. In the bainitic steel specimens, AF and GB are mostly formed during cooling, and the grain growth occurs without forming new phases after cooling. These formed microstructure affects to the yield stress, ultimate tensile strength and other mechanical properties. There have been several studies to predict the effect those volume fraction composition of microstructure of AF, GB, BF and MA with a few number of specimens. However, the overall tendency of the microstructural volume fraction effect has barely been studied. In this study, the AF, GB, BF and MA volume fraction effect onto mechanical property have been characterized with linear regression model by tensorflow (with learning rate of 1e-8 and step size 1e+7). Learning rate and The overall tendency of each effect onto YS, UTS, YR, USE and ETT has been predicted.