Model assisted calculations for NDE reliability

There are multiple techniques to assess the reliability of non-destructive evaluation (NDE). Probability of detection (POD) curves can be used to find out the smallest flaw that will be detected with a certain probability and receiver operating characteristic (ROC) curves can be used to determine the performance of the amplitude-based classification systems. However, the problem with these techniques is that they require a large number of measurements for good results. In this study, ultrasound simulations were used to generate data for POD and ROC analysis. In total 463 cracks with height from 0.5 mm to 10 mm were simulated. Skew and tilt angles of the cracks were normally distributed random variables with deviation of 5°. From height vs. amplitude data, POD and ROC curves were extracted. Results indicate that due to varying tilt and skew angles, both detectability and classification accuracy decreases due to high variation of the amplitude signal. Moreover, the results show also the capabilities of model assisted probability calculations as a tool to improve NDE methods.