Understanding oxidation of Fe-Cr-Al alloys through Explainable Artificial Intelligence

AbstractThe effect of alloy composition and oxidation condition on specific mass gain of FeCrAl alloys was studied and analyzed using a combination of experimental and AI approaches. A Neural Network (NN) classification model was used on the experimental FeCrAl dataset produced at GE Research from steam oxidation studies at both high (~ 1000°C) and low temperature (~ 400°C). Furthermore, using the Shapley Additive exPlanations (SHAP) explainable Artificial Intelligence (XAI) tool, we explore how the NN can identify an alloy at specific oxidation condition to form a protective oxide or not. We found high Al and Cr concentration increases the chances of forming protective oxide layer, which is consistent with literature studies. Contrary to Al and Cr, the presence of Mo in FeCrAl creates thick unprotective oxide scale that results in high mass gain per unit area.