Supervised machine learning for corrosion assessment of multi-principal element alloys using experimental and generative datasets

Multi principal element alloys (MPEAs) have emerged as a novel class of materials with complex compositions and tuneable properties, making them promising candidates for corrosion-resistant applications. However, their vast compositional space complicates the establishment of clear composition-property relationships using conventional experimental and computational methods, which typically yield insufficient data for effectively mapping causal relationships. This study investigates MPEA corrosion behaviour via supervised machine learning (ML). A dataset of alloy compositions, phases, electrolyte characteristics, and corrosion properties was used to train ML models, including random forest, neural network, kernel ridge regression, and LassoIC. To address data scarcity and enhance prediction accuracy, a generative adversarial network (GAN) was employed to generate synthetic data. Retraining the ML models with augmented data led to improvements in predictive accuracy for corrosion properties. The results highlight the effectiveness of an integrated GAN-ML approach in exploring MPEA compositional complexity and accelerating the design of corrosion-resistant alloys.