Machine learning modeling of time-dependent corrosion rates of carbon steel in presence of corrosion inhibitors

We have employed supervised machine learning methods to model measurements of corrosion rates of carbon steel as a function of time when corrosion inhibitors are added in different dosage and dose-schedules. The experiments show that the time-profile of corrosion rates depend on the dose schedule, while the final rates depend mainly on the environment severity. We find that Random Forest was the best algorithm that predicted the entire time-profile of corrosion rates with the mean squared error ranging from 0.005 to 0.093. Sensitivity of corrosion rates to changes in the environmental variables are well-predicted by the trained Random Forest model.