Extreme value analysis of ultrasonic thickness measurements

Abstract

Modern infrastructure and industrial plants have a finite design life. Their effective and profitable operation depends on well organised maintenance and condition assessments. Non-destructive evaluation and inspection is a key tool for condition assessment. However, despite best efforts, full inspection coverage of a plant is not always possible because of access problems, time constraints and limited budgets.

Many inspection companies are beginning to use partial coverage inspection (PCI) techniques to solve this problem. PCI describes the use of inspection data collected from a small area of the component to extrapolate to the condition an the rest of the component. Extreme value analysis (EVA) is a technique of particular interest for this application as it allows an inspector to construct a statistical model of the smallest thickness measurements across the component. This model can then be used to extrapolate to the most likely minimum thickness.

In this thesis, an analysis of the uncertainties that can arise when using EVA for extrapolation is performed. A clear outline of the uncertainties expected to result from EVA extrapolation is presented and it is made usable for inspectors. In addition, a simple test algorithm to analyse when EVA is suitable for a set of inspection data is described. It is hoped that the work described in this thesis will enhance confidence in the practical use of the technique in the field.

Furthermore, the effect of surface roughness on ultrasonic thickness measurements is investigated with joint experimental and computational studies. It is shown that the thickness measurement distribution can differ significantly from the actual thickness distribution, particularly for the smallest values of thickness and with rougher surface conditions. Consequently, extrapolations from extreme value models using ultrasonic thickness data are shown to be conservative compared to the true condition of the component.