Artificial Intelligence Approach to Predict Elevated Temperature Cyclic Oxidation of Fe-Cr and Fe-Cr-Ni Alloys

Predictive modelling of cyclic oxidation (CO) behavior is challenging. CO is dependent on several factors including time, temperature, atmosphere (composition, pressure) and alloy composition. The machine learning algorithm CatBoost is used to model, for the first time, CO of binary Fe-(10-20)% Cr alloys and ternary Fe-(16-20) %Cr-(10-30)%Ni alloys using published data. The CO conditions were 650 degrees C and 800 degrees C under air + 10% water vapor atmosphere. The CatBoost model was successfully trained and tested using 80:20% of the data with the composition, temperature and cycle time as input variables and mass change as the output. The five-fold cross-validation showed that the model had an average accuracy of 0.98 (R-2). The CatBoost algorithm was also used as a classifier that, given a composition, predicts if the alloy will form protective oxide scale, oxidize rapidly or undergo spallation after 100 h of CO, accurately.

Automated summary

In this study, predictive modeling of cyclic oxidation behavior of different alloys was conducted using the machine learning algorithm CatBoost, achieving high accuracy. The results showed that the CatBoost algorithm can not only predict the mass change of alloys after cyclic oxidation, but also accurately predict whether they will form protective oxide scale, oxidize rapidly, or undergo spallation.