Evaluation of Supercritical Carbon Dioxide Corrosion by High Temperature Oxidation Experiments and Machine Learning Models

Corrosion behaviors of ferritic-martensitic, austenitic steels, and Ni-based alloys were examined in high-temperature CO₂ environments. Aside from the conventional evaluation, we further used machine learning based on experimental data and an existing SCO₂ database, from which contributing factors influencing the SCO₂ corrosion were quantified and the correlation between the CO₂ and SCO₂ corrosion was revealed. Among the tested alloys, Ni-based alloys revealed most exceptional corrosion resistance at 500 °C to 800 °C. The random forest model learning the SCO₂ database suggested that the most important factor was the material type, followed in sequence by temperature, exposure time, Cr content, flow rate, and pressure. In the ferritic-martensitic steel, we observed a strong correlation between the CO₂ evaluation data and the linear regression model. The machine learning models (linear regression, decision tree, and random forest) revealed a relatively weak correlation in the austenitic steel, but a relatively strong correlation in the Ni-based alloy.