A Bayesian probabilistic approach for ship corrosion prediction using the Adaptive Metropolis-Hastings algorithm

초록

Accurate estimation of corrosion depth is essential for evaluating the structural integrity of vessels operating in harsh ocean environments and for establishing effective maintenance strategies. This study proposes a probabilistic nonlinear model for corrosion loss. The corrosion depth is assumed to follow a lognormal distribution, and both the model parameters and the standard deviation in the log-space are treated as random variables. Bayesian inference is employed to estimate the posterior distributions of these parameters using observed inspection data. To improve computational performance and convergence, the Adaptive Metropolis-Hastings algorithm is implemented, allowing dynamic adjustment of the proposal covariance matrix throughout the sampling process. The likelihood function is formulated by incorporating observation uncertainty derived from the lognormal distribution, enabling probabilistic estimation of the corrosion depth distribution and its associated uncertainty. The proposed methodology is validated using full-scale inspection data collected from commercial ships with distinct operational characteristics. The model is applied individually to multiple structural members, allowing probabilistic estimation of corrosion depth distributions at each location. Furthermore, additional validation is conducted to provide a preliminary assessment of the model's capability to forecast future corrosion progression.

키워드