Non-intrusive Internal Corrosion Characterization using the Potential Drop Technique for Electrical Mapping and Machine Learning

This paper describes a non-intrusive method for collecting data about internal corrosion damages in AISI-304 stainless steel plates and classifying them according to severity. The mapping of the electric potential gradient is derived using the potential drop technique, which is then analyzed using image processing techniques including edge enhancement and segmentation. Simulations were run using finite element modeling to produce examples of damaged plates, with four types of defects that can be considered part of pitting corrosion. The image processing stage plays the role of an extractor of features that, when employed as inputs of machine learning algorithms, make it possible to determine the damage severity. With the Gradient Boosting regressor, the maximum absolute error of 0.879 mm was obtained in the estimate of the depth of the defects. Additionally, with the application of a Convolutional Neural Network, an accuracy of 94.84% was achieved to classify of the severity of the damages.