Machine learning assistance for electrochemical curve simulation of corrosion and its application

In this paper, we used machine learning algorithms such as k-nearest neighbour, decision tree, gradient boosting decision tree, random forest, and support vector machine in a scikit-learn module of Python to construct polarization curves and electrochemical impedance spectra. After application of this method to a high-level nuclear waste disposal tank material of pure copper, the polarization curves, and electrochemical impedance spectra of pure copper under different chloride ion concentrations, sulfide concentrations, and temperature environments were used as training sets. The combination of the cross-validation and experimental validation results revealed that the RF algorithm had the best effect on predicting the polarization curve and EIS. Through the input weight analysis, it was found that the sulfide concentration had the greatest influence on the polarization curve, followed by the chloride ion concentration, and the temperature influence was the smallest. For electrochemical impedance spectroscopy, sulfide and temperature had a large effect, while chloride ions had little effect. The result of the weight analysis was consistent with traditional electrochemical results.