Model-Assisted Probability of Detection (MAPOD) Initiative – Center for Nondestructive Evaluation

The overall objective of this initiative is to establish a comprehensive methodology for model-assisted POD evaluation with reduced empirical testing. The first aim is to investigate the use of simulation and statistical models to help assess the statistical implications of only having a limited sample set. The second aim is to explore challenging transfer function examples, considering interaction effects between flaw type and part geometry state. Model validation will be performed to assess model accuracy and viability to support limited set POD and transfer function approaches. This initiative is expected to revitalize the core capability at CNDE in NDE simulation and reliability evaluation.

The MAPOD initiative has led to the development of an improved process for the consideration of models in POD evaluation (see Figure). For any POD study, the scope of the evaluation must be initially assessed, and all critical factors for the NDT technique, part material, part geometry, and discontinuity characteristics that control signal and noise must be identified. As well, the amount of variability, ideally as distribution functions, must be assessed. If the effect of and variability within key factors is not well understood, design of experiments (DOE) studies and NDT inspection surveys should be performed. The use of Bayesian estimation and model-based inversion techniques can provide quantitative assessment of variability with uncertainty bounds. Three POD design considerations must be made on whether to pursue a MAPOD evaluation versus a fully experimental-based study. First, the state of physics based models for NDT simulations must be considered, in terms of accuracy and speed, for the problem of interest. Physics-based models must also be mature in order to leverage model-assisted based approaches. Second, the cost to manufacture specimens with discontinuities is a critical consideration in POD design. If the cost to build representative samples with real discontinuities is deemed excessive, MAPOD becomes a viable alternative. Lastly, there is value in researching whether any existing, related POD study data sets or specimen sets have already been created. Based on these three POD design consideration, decisions can be made on whether to perform fully empirical, transfer function (TF) or full-model assisted (FMA) approaches.