Artificial Intelligence for the Prediction of Tensile Properties by Using Microstructural Parameters in High Strength Steels

Abstract

Artificial intelligence is widely employed in metallurgy for its ability to solve complex phenomena, which are associated with the learning process of previously obtained experimental data. Although numerous physical modeling techniques have been implemented for the prediction of mechanical strength using equations, several empirical efforts are necessitated to evaluate specific constants for different models. Microstructure is a good feature to understand mechanical properties because it incorporates the effects of material composition and process conditions. The complex combination of material composition and process parameters determine the microstructure of steel. In this study, the information on microstructural volume fraction is utilized for the prediction of tensile strength, yield strength, and yield ratio via artificial neural networking. Various combinations of PF (polygonal ferrite), AF (acicular ferrite), GB (granular bainite), BF (bainitic ferrite), and M (martensite) are investigated for the prediction of yield strength, ultimate tensile strength, and yield of high strength steel via back-propagation linear regression and neural network based algorithm. The effects of each microstructure on the three mechanical properties were successfully predicted by employing back-propagation linear regression. A deep learning algorithm with hyper-parameter tuning and cross-validation enabled high accuracy in predicting experimental data with mean absolute percentage errors of 6.59% and 10.78% for the validation and test sets, respectively. These studies can open a new avenue for applying the microstructural design effects to find optimum yield strength, tensile strength, and yield ratio of high strength steel.