Electrochemical impedance spectroscopy analyses based on interpretable machine learning and global optimization algorithms
Ying Jin*, Zhaoyang Zhao, Yang Zou, Peng Liu, Zhaogui Lai, Lei Wen
National Center for Materials Service Safety, University of Science and Technology Beijing, 102206, China

EXTENDED ABSTRACT: Electrochemical impedance spectroscopy (EIS) is an important method for studying the mechanism of electrochemical corrosion process and evaluating the corrosion resistance of metals. While, it is difficult to analyze the EIS data due to the same test result can be fitted into different equivalent circuits (ECM), and its interpretation is not unique. Traditional experimental data analysis mostly relies on artificial identification is obviously not suitable for analysis of large quantities of EIS data. In this work, 7 classic MLs (k-nearest neighbor, logistic regression classifier, naïve bayes classifier, support vector machine, decision tree, random forest and AdaBoost) were applied to predict the equivalent circuit model of the given EIS instead of the traditional manual fitting, which shows Adaboost taking optimized hyperparameters found by grid search achieved the known highest prediction accuracy and had a prediction basis that was consistent with chemical knowledge. The performances of 20 GOAs on identifying parameters for 9 ECMs were assessed on simulated and experimental impedance spectra. The feasibility of GOA fitting parameters was further verified on the experimental EIS, and compared with the results of manual fitting. By comparison, it is found that GOA can obtain similar results with manual fitting accuracy without manual assistance (setting initial values and repeated attempts) when EIS is not disturbed by obvious outliers. Therefore, the prediction of the equivalent circuit of EIS by interpretable machine learning and subsequent automatic fitting of circuit parameters by GOA make it possible for high-throughput, automatic and intelligent EIS processing.