Prediction and analysis of corrosion rate of 3C steel using interpretable machine learning methods

Abstract

Machine learning has become increasingly important in materials science, particularly for predicting material properties as artificial intelligence and big data continue to converge rapidly. Studying the corrosion rate of 3 C steel in various seawater environments is the purpose of this study. As a first step, correlation coefficients were used to establish the significance of the features considered. Following the training of 46 sample sets, machine learning methods were utilized to determine the most accurate machine learning model based on errors in the test set. In the end, the SHAP interpretation technique was employed to clarify the model and facilitate feature visualization. Through local interpretability, the effect of distinct seawater environments on 3 C steel corrosion rates was examined using SHAP contribution values for single and multiple feature interactions. According to the results, the oxidation-reduction potential (ORP) had the greatest impact on corrosion rate, showing a positive correlation. As dissolved oxygen (DO), pH value, and temperature (T) declined in magnitude, salinity (Sal) had a comparatively minor effect on corrosion.