The critical information in condition assessment often is an estimate of the localized maximum extent of damage (e.g., pitting in pipe wall).  NDT sampling data may include a large number of points, but do not necessarily contain the maximum or minimum values present in the asset.  Physical processes (like corrosion), typically can be described by a large collection of random observations, and typically generate probability distributions of one of several characteristic shapes.  All of these characteristic probability distributions have long (“fat”) tails, that can be analyzed to make reasonable estimates of their extreme values.  For example, a set of NDT in-line corrosion inspection data can be analyzed to predict the maximum depth of pitting or the maximum density of pitting within the piping in question.  This is called extreme value analysis (EVA).

Knowledge-based services

• Our team of expert engineers will perform “tail fitting,” of inspection data using one of several extreme value functions (e.g.,  General EV, Pareto, Weibull, Zipf) to define a distribution that best accounts for the observed points, and then use this distribution to predict the maximum value of the damage process in question.
• Alternately, sequential analysis may be used to analyze the sampling data in increments until a sufficient statistical basis for estimating the extreme (maximum) value of the damage process is reached. This approach can be conditioned by a customer’s risk tolerance and budget.
• In general, any low probability high consequence condition that is the result of a continuous physical process is likely to be amenable to extreme value analysis, as an alternative to more extensive inspection or testing.
• Extreme value analysis is a proven method to make optimum use of limited sampling data, and often is the best choice when time and resources are tight

Software

• Several commercial packages (e.g., Matlab) support extreme value analysis.
• We use calculated extreme values as bounding conditions in conventional FEA analysis (using, for example, ABAQUS) or in our proprietary FEA and remaining life assessment products.

Applications

• Inspection of pitted plant – universal inspection of piping is not generally feasible, so EVA is used on partial (e.g., 10%) inspection data to predict maximum pit depth or maximum pit density.
• Limited data sets from different inspection episodes can be analyzed as a time series in EVA, and used in remaining life predictions.
• Difficult, inaccessible or dangerous to inspect:  EVA can use relatively small data sets to estimate the limits of damage processes within a mechanical system.

For further information about Extreme Value Analysis, please contact us.