An Automated Electrochemical Noise Analysis for Corrosion Type Identification Using Random Forest: Features Selection and Cross-Material Performance

Electrochemical noise has been used to monitor four materials undergoing uniform and localised corrosion and in passive conditions. The data are used to train a random forest (RF) model to classify the corrosion behaviours, achieving an accuracy of 96%, when training and testing the model with all materials and conditions. Key features, including current standard deviation and peak to peak difference, were identified as critical to obtain an accurate classification. The RF model was also trained on single metals and cross-validated on the others. The accuracy in the classification was generally maintained for similar materials, but decreased slightly for training and validation on significantly different materials.