Machine learning and experimental study on a novel Cr–Mo–V–Ti high manganese steel: Microstructure, mechanical properties and abrasive wear behavior

Abstract

Machine learning combined with traditional experimental methods can promote the efficient research and development of materials. In this work, five kinds of algorithm models combined with a genetic algorithm (GA) were used to optimize the compositions of the alloyed high manganese steel. And then the effect of solid solution temperatures on the microstructure, mechanical properties, and impact wear properties of the steel were systematically investigated. The results showed that Categorical Boosting (CB) model exhibited the high validation accuracy (R2 > 0.95, RMSE<4.11, MAE<2.44). Based on the trained CB model and GA, the optimal steel compositions were obtained. The as-cast microstructure of the steel contained coarse austenite, little pearlite and networks and irregular carbides. After water toughening treatment, the pearlite completely dissolved into the matrix and the number of carbides gradually decreased. Compared to as-cast alloy, the austenite grain size significantly decreased, and it decreased first and then increased with the increase of solid solution temperatures. The average impact energy, ultimate tensile strength (UTS) and elongation (EL) of the steel increased and then decreased with the increase of solid solution temperatures. However, the wear loss showed an opposite trend. The steel of water quenching at 1100 °C with an average impact energy of 185.1 J, hardness of 242.4 HBW, UTS of 742 MPa, yield strength (YS) of 458 MPa, and EL of 37.4%, exerted best impact toughness, mechanical properties and wear resistance. It was attributed to the interactions among dispersed small sized second phases, high density dislocations, fine austenite grains and twins.

Keywords

• High manganese steel
• Machine learning
• Microstructures
• Mechanical properties
• Impact wear