Corrosion area detection and depth prediction using machine learning

Abstract

Corrosion reduces the thickness of a structure, making it less safe and reducing its lifespan. In particular, ships are vulnerable to corrosion because they are always submerged in seawater. This corrosion is identified through regular inspections of the ship structure, and gradually increases in scope if no action is taken at an early stage. In this study, we developed a model to detect the corrosion areas and predict the depth of corrosion in the detected areas. The corrosion area detection model used a machine learning model based on Mask R-CNN. The 35,753 images were used to map corrosion images and measured corrosion depths. Four different color maps and regression algorithm were used to predict corrosion depths and their performance was compared. The new attempt to predict the corrosion depth from images in this study will contribute to improving existing corrosion control methods by providing information for corrosion prevention and maintenance.