Unsupervised process monitoring of corrosion based on electrochemical noise and multivariate image analysis

Electrochemical noise (EN) is a crucial technique in the monitoring of corrosion systems due to its ability to provide real-time, non-intrusive insights into the corrosion process. By measuring the spontaneous fluctuations in voltage and current that occur naturally in a corroding system, EN allows for the detection of localised corrosion events, such as pitting, without the need for external perturbation. In this investigation, a multivariate statistical process monitoring framework (MSPC) based on the use of deep learning models and principal component analysis (PCA) is proposed. Electrochemical noise associated with uniform corrosion is segmented with a sliding window, with the segments converted to images from which features are extracted with deep learning models. Finally, these features are used to construct a principal component model that can be used to detect deviations from uniform corrosion.