Coupling physics in machine learning to predict interlamellar spacing and mechanical properties of high carbon pearlitic steel

Abstract

The present study proposed an improved composition-structure–property model which incorporates physical features in the machine learning(ML) process. Herein, the physical parameters, the volume fraction of ferrite, cementite and carbides and the transformation temperature of pearlite, were included to the dataset as extra input variables to guide the ML process. As a result, the dataset mixing physical features greatly improved the generalization ability and prediction accuracy of the generalized regression neural network(GRNN) model, which clearly demonstrates the practicability of the present physics coupled ML approach. Furthermore, several optimization algorithms are applied to improve the GRNN model and the fruit fly optimization algorithm(FOA) is demonstrated more effective than particle swarm optimization(PSO) algorithm. Thus then, a novel high-strength pearlitic steel was synthetized and experimentally validated with outperformed microstructural and mechanical characteristics.