Coating Damage Detection of Vessels Using Machine Learning-Based Underwater Electric Field

There has been extensive research on corrosion electric fields because of their special characteristics. There are apparent distribution characteristics corresponding to different coating damage regions of a vessel. To accurately detect where the coating damage region is located on vessels using underwater electric field signature, in this study, a method called "refined composite multivariate multiscale fluctuation reverse dispersion entropy (RCMMFRDE)" was proposed. This method effectively extracted the multichannel signal in a complex environment of poor stability and high noise. By combining RCMMFRDE and hierarchical prototype (HP), a novel coating damage detection method was developed. First, the accumulated differential value was calculated to determine the side on which the vessel had the damaged coating. Second, RCMMFRDE was used to extract the features of electric field signals. Then, feature vector dimensionality reduction was carried out using pair-wise feature proximity (PWFP). The low-dimensional features were given as input to the HP classifier to detect the damaged region. Finally, a series of numerical and physical scale experiments were conducted to validate the feasibility of the proposed method. The damaged region was efficiently predicted, achieving satisfactory accuracy of 96.67% and 92.00% in numerical and measurement data respectively; thus, it is a suitable alternative to conventional methods, especially when there is a lack of previous environmental information.