Coating damage localization of naval vessels using artificial neural networks

Abstract

For the localization of coating damages of naval vessels numerical simulation using the FEM software COMSOL Multiphysics were carried out to calculate the corresponding underwater electric potential (UEP) signature. Therefore, we defined said damages at random hull surface positions and used the information provided by the impressed current cathodic protection (ICCP) system, more exactly the cathodic current itself, and the calculated UEP signatures as input parameters to train an artificial neural network (ANN) for predicting the coating damage location. With this deep learning approach, more than 90% of all coating damages are predicted correctly, considering a generic ship model with 50m length, whose hull is divided into 12 different sectors. Even the mere use of ICCP currents as highly aggregated input parameters for the ANN lead to a satisfactory prediction rate over 80% within the predefined sectors, thus providing quite accurate results using minimal amount of data.