RT Reliability Analysis

Non-destructive testing (NDT) has become indispensable in areas of our lives where damage can result in high follow-up costs or pose a threat to human life (examples: transportation, power generation, the chemical industry). In practice, an inspection system may reach its limits, for example, when dealing with small defects. Defects of a critical size may not be detected. Therefore, probabilistic evaluation methods must be used to characterize the inspection system. The goal is to identify an objective quality metric that can be used to define the applicability of the inspection method. The probability of detection (POD) meets this requirement. Based on the relationship and dispersion of the data, the POD indicates whether the method can be accepted for the inspection task or needs further improvement. The original POD method was developed for quasi-one-dimensional defects in thin aerospace components. In industrial practice, this evaluation is a balancing act between statistics and feasibility: The inspection must be evaluated using real defect data from the component’s subsequent production (or periodic maintenance inspections). However, the necessary comparison between micrograph data—used to determine the true defect size of spatially defined defects—and the signal from a NDT system proves to be a challenging and cost-intensive task. Both the establishment of a common coordinate system and the description and alignment of the data constitute necessary preliminary work. This thesis develops a possible approach that can be applied in the future. While the literature on POD has frequently recognized the limitations of using a one-dimensional POD (POD with a single defect parameter) for real defects, this thesis also aims to comprehensively extend the method on the signal side to enable the inclusion of real defects in the POD evaluation.

To this end, this thesis introduces two significant innovations in POD evaluation:

1. The display area is introduced into the evaluation as an important indicator for detection. The Observer-POD approach, which describes the detectability of a defect, offers one possibility for extending the evaluation. However, the amount of data required for an Observer-POD is rarely achieved in experiments. Therefore, we propose the introduction of a smoothing algorithm to capture the area dependence even based on a small amount of data. The algorithm’s functionality is verified using simulated data before it is applied to real defects. At the same time, the simulated data help facilitate a comparison with previous approaches.
2. Furthermore, data from real defects are often insufficient to meet the statistical requirements, making it necessary to include artificial defects. Therefore, the available artificial defects should be included as reference defects to strengthen the statistical basis. However, important influencing factors (e.g., surface roughness) are not available for the testing of reference defects. Due to the varying informative value of the data and to avoid an overly optimistic estimate, a simple mixture of the data is ruled out. To ensure that the characteristics of real defects can appropriately influence the results of the method’s evaluation, a weighted combination of defect data is proposed for the evaluation. The approach is demonstrated using the example of radiographic inspection of an electron-beam-welded joint. The weld joint connects the lid to the outer wall of a copper container designed for the subsequent final disposal of spent fuel rods from nuclear power plants. The measurement results were provided by Posiva Oy, the company responsible for the final disposal of spent fuel rods from nuclear power plants in Finland. In this context, the POD assessment represents an important element in the overall risk assessment for the final disposal system.