Data-driven predictive corrosion failure model for maintenance planning of process systems

Abstract

Extreme value analysis (EVA) is occasionally used to predict corrosion progress. This paper adopts EVA to predict the depth of extreme pits to prioritize inspection and maintenance. It considers the peaks over threshold (POT) method to illustrate the predictive capacity of this method in assessing degradation progress based on consecutive inspection reports. The proposed approach uses distribution parameters to establish stochastic corrosion models. Four consecutive inline inspections of a pipeline are used to validate the model. As the block maxima (BM) method is often used in extreme value analysis of corrosion damage depths, the POT approach is compared to the BM's predictive results. The POT approach is considerably more capable (33%) of assessing failures in individual sections than the same workflow implemented with BM. With the downside of increased falsely categorized failures (10.6%). The method's performance in assessing failures makes it most useful for data-driven maintenance of process systems.