Elucidating precipitation in FeCrAl alloys through explainable AI: A case study

Abstract

A primary challenge of using FeCrAl in high temperature industrial settings is the formation of α′-precipitates that causes brittleness in the alloy, resulting in failure through fracture. The precipitation causes hardness change during thermal aging which is sensitive to both alloy composition and experimental condition (i.e., temperature and time of heat treatment). A Gaussian Process Regression (GPR) model is built on the hardness data collected at GE Research. Subsequently, for the first time, SHapley Additive exPlanations (SHAP) built upon the GPR is used as an Explainable Artificial Intelligence (XAI) tool to understand the effect of feature values in driving the hardness change. SHAP analysis has confirmed that the primary chemical driver for α′ age hardening in the FeCrAl system is Cr as expected. However, the analysis also indicated that Al does not have a clear trend of only suppressing the formation of α′ which contradicts current literature. This lack of a trend on the effect of Al on age hardening may be due to Al ability to both suppress thermodynamically and enhance kinetically the formation of α′. Similarly, SHAP analysis points towards Mo having no clear trend towards either enhancing or suppressing α′. This study indicates that more in depth studies focusing on both the chemistry and different aging temperatures (to study kinetics) should be performed to better understand the aging of this system.