Corrosion pit depth extreme value prediction from limited inspection data

Abstract

Passive alloys like stainless steels are prone to localized corrosion in chlorides containing environments. The greater the depth of the localized corrosion phenomenon, the more dramatic the related damage that can lead to a structure weakening by fast perforation. In practical situations, because measurements are time consuming and expensive, the challenge is usually to predict the maximum pit depth that could be found in a large scale installation from the processing of a limited inspection data. As far as the parent distribution of pit depths is assumed to be of exponential type, the most successful method was found in the application of the statistical extreme-value analysis developed by Gumbel. This study aims to present a new and alternative methodology to the Gumbel approach with a view towards accurately estimating the maximum pit depth observed on a ferritic stainless steel AISI 409 subjected to an accelerated corrosion test (ECC1) used in automotive industry. This methodology consists in characterising and modelling both the morphology of pits and the statistical distribution of their depths from a limited inspection dataset. The heart of the data processing is based on the combination of two recent statistical methods that avoid making any choice about the type of the theoretical underlying parent distribution of pit depths: the Generalized Lambda Distribution (GLD) is used to model the distribution of pit depths and the Bootstrap technique to determine a confidence interval on the maximum pit depth.