B0931
Title: How to evaluate the probability of detection based on data from undamaged structures
Authors: Alexander Mendler - Technical University of Munich (Germany) [presenting]
Abstract: Probability of detection (POD) curves are standard tools to quantify the performance of non-destructive testing methods and, depending on the method employed, they require a minimum of 30 data sets from damaged specimens or destructive tests. Statistical methods are presented that construct POD curves based on data from undamaged specimens. This is relevant for structural health monitoring and other applications, where no data from the damaged state is available at a reasonable cost. The methods can be applied for damage detection and localization, and multiple data-driven features can be evaluated simultaneously (e.g. multiple modal parameters) for the evaluation of a single material parameter. The method explicitly quantifies the uncertainties in the data-driven features (e.g. due to measurement errors) and requires an analytical model for the computation of sensitivity vectors (e.g. finite element models or wave propagation equations), but no simulations in the damaged state are necessary. For proof of concept, various case studies from structural health monitoring and non-destructive testing are presented, e.g. based on modal parameter and ultrasonic testing, and ultimately, all limitations and assumptions are critically discussed and juxtaposed with the ones of existing POD methods.
