Abstract
This study proposes a procedure for the prediction of the bottom conditions of storage tanks; it is based on the observations of the evolution of the pit corrosion in some carbon steel samples that were exposed over one year to a corrosive environment simulating a real case scenario. The pitting corrosion on the samples was monitored by laser-scanning their surface, then, the maximum corrosion depth was calculated for each sample. A probabilistic model, based on the extreme values theory, was developed to model the distribution of pits. The model aims to predict the probability distribution of the corrosion depths at a different time, building a simple time dependence for the distribution parameters and allowing to determine the expected bottom perforation probability. By comparing with previous studies, Bayesian inference was used to adjust the prediction of both the distribution parameters, without losing their interdependence. The extreme values theory, combined with Bayesian inference, allows the model to give an accurate prediction of failure probability; consequently, the residual useful life of the tank can be precisely estimated, mitigating the risk of release of dangerous substances while optimizing management costs.
