Extreme Value Statistics for Pitting Corrosion of Steel Pipelines

Abstract

Data for maximum pit depths obtained from intelligent pigging runs for several older operational water injection pipelines have been used extensively for prediction of likely future pipe wall penetration. Theoretically pit depth data should be consistent with the Gumbel extreme value (EV) distribution, but most data for longer-term exposures in pipelines do not show the expected linear trend on Gumbel EV paper. The usual approach for dealing with this is to invoke the concept of the ‘domain of attraction’ in extreme value theory. This may be attractive from a statistical point of view but it provides little or no insight about the reasons for the departure from the Gumbel EV distribution or about the pitting corrosion mechanisms involved or the possible influencing factors. It is more satisfactory to consider the development of pitting corrosion with time and then examine the resulting effects on pit depth uncertainty. It is shown that this provides a way to interpret the pigging pit depth data. In particular, it is shown that long-term pit depth data are from a different statistical population than the data from shorter exposures. This differentiation appears not to have been made previously but has important implications in practice.