Abstract
When using stochastic data in a probabilistic engineering assessment it is common practice to fit the central portion of the data, using for example the Normal distribution. This generally works for predicting expected behaviour; however, it does not necessarily describe extreme behaviour very well.
Extreme value theory and more specifically the peaks-over-threshold method [1] is adopted to assess the extreme behaviour represented by the distribution’s tail. A generalized Pareto distribution is used to fit all samples exceeding a certain threshold value. Estimation of the distribution’s shape parameter provides valuable information on whether the upper tail of the fitted distribution has “finite” (having an endpoint) or “infinite support”.
Line pipe out-of-roundness affects weldability, fatigue performance and collapse resistance. Especially for deep water pipelines it is important to meet tight tolerances. Out-of-roundness data were evaluated and show finite support of the upper tail; however, the evidence is not as strong as the evidence that was found for the line pipe material strength [7]. Hence it is important to measure and confirm the out-of-roundness of every pipe joint during manufacture.
The degree of tail support can have a major effect on the calculated return period of failure events, particularly when related to low occurrence probabilities.