Estimating uncertainty in maximum pit depth from limited observational data

The expected maximum depth of corrosion pits in structural steel exposed to marine environments is important for estimating the reliability of engineered systems such as pipelines, tanks, ships and other containment structures. The conventional approach is to use observed data to determine the best fit extreme value (EV) distribution and then to estimate its parameters. In turn these may be used to estimate the maximum pit depth for other situations. However, in many cases there is insufficient data for this classical approach. Progress can be made by invoking a Bayesian statistics approach when a prior EV distribution can be invoked based on previous experience with similar data. The procedure is described herein for estimating the coefficients of variation for the maximum depth of pits on steel surfaces exposed to immersion, tidal, coastal atmospheric and marine inland atmospheric exposure conditions. For these situations previous experience for longer-term exposures has shown that Frechet is the most appropriate prior distribution.