Machine learning aided process design of Fe-Cr-Ni-Al/Ti multi-principal element alloys for excellent mechanical properties

Abstract

Fe-Cr-Ni-Al/Ti multi-principal element alloys (MPEAs) with the good mechanical properties were designed by utilizing machine learning (ML) methods in our previous work. It can be noted that the post-processing (thermo-mechanical treatments, TMT) can tailor microstructures of the designed MPEAs with an aim to further obtain excellent mechanical properties. However, for a novel MPEA, determining the optimal post-processing parameters often requires time-consuming and laborious experiments due to the absence of existing precedents. In this work, we collected approximately 400 sets of mechanical properties data of MPEAs. The various feature selection methods were employed to train ML models for the process optimization of MPEAs. In addition, two input strategies were compared to offer a simplified approach for model inputs: one considering all process features and the other focusing solely on TMT process. The results showed that the difference in model accuracy between the two input strategies was minimal, indicating a strong possibility to simplify the ML model by disregarding specific processing features. Based on the trained models, we conducted optimization design for the TMTed parameters of Fe-Cr-Ni-Al/Ti MPEAs. The as-TMTed (Fe10Cr35Ni55)95Al5 and (Fe10Cr35Ni55)97Al2Ti1 samples demonstrated a balance between strength and toughness, with yield strength (YS) and ultimate tensile strength (UTS) values of approximately 600 MPa and 900 MPa, respectively, and a fracture elongation (FE) exceeding 30 %. As compared to samples without the TMT process in our previous work, the as-TMTed samples demonstrated a significant increase in YS and UTS, while only a certain loss in FE. This work confirmed the viability of ML-assisted efficient design for post-processing treatment in novel MPEAs.