Probabilistic ship corrosion wastage model with Bayesian inference

Corrosion wastage is one of the critical problems for the ship structures and prediction of corrosion depth is essential to monitor and maintain the ageing parts. This study targets to propose a probabilistic method to predict the corrosion depth considering the uncertainties, potentially induced by measurement. To achieve this goal, the probabilistic distributions of parameters were employed to the conventional corrosion wastage model. Then Bayesian inference was introduced to update the obtained probabilistic model using inspection data. Firstly, one of the nonlinear corrosion wastage models was selected for a fundamental model and the parameters of the model and variance of error term were assumed as random variables. Hence, the number of corrosion wastage models corresponding to sets of random variables and their prior joint probabilities were obtained. At the second stage, likelihoods of each corrosion model were calculated using the corrosion field data and the error distributions which were originated from the variance of error term. Bayesian inference was then applied and the updated joint probability, called posterior joint probability, was obtained. Finally, the corrosion depth distribution over time was calculated based on the posterior joint probability and the reliability of the corrosion depth was evaluated.