Epistemic Uncertainties in Minimum Thickness Estimates Based on Extreme Value Analysis

Ultrasonic Thickness (UT) measurement is typically used at sample locations in piping system inspection to monitor wall thickness loss. Statistical methods based on Extreme Value Theory (EVT) extrapolate from the measured part of the thickness distribution into the unmeasured lower tail to estimate minimum thickness bounded by a confidence interval. This paper assesses how two sources of epistemic uncertainty affect estimates based on Extreme Value Analysis: measurement uncertainty and uncertainty surrounding the threshold in the sample distribution separating samples taken from corroded and non-corroded locations. Data from an in-service piping system inspection is analyzed to demonstrate that quantification of these additional uncertainties provides useful information when assessing asset integrity. When only sample uncertainty is considered, the lower confidence bound on the expected minimum thickness is 3.7 mm, well above the minimum required thickness of 2.8 mm. However, it is found that when the additional two uncertainties are considered, the lower confidence limit on the lower interval bound of the expected minimum thickness drops to 2.7 mm. A further useful finding is that the minimum thickness estimate does not decrease monotonically with the number of block minima values from the lower distribution tail fitted to the extreme value distribution model.