|
The paper provides a procedure for validating flaw detection size using smaller number of signal response data points or number of flaws. MIL-HDBK-1823 probability of detection (POD) analysis of inspection test data results in a flaw size estimate, denoted as a90/95. The a90/95. flaw size has 90% POD and minimum 95% confidence. Here, a case is considered where the inspection test data points are not sufficient to run the POD analysis but are sufficient for limited validation approach provided here. The procedure is based on an approach developed by the author. The approach is conservative and it validates a flaw size that is greater than a theoretical a90 used in the simulation data. By using simulated signal response data, the paper shows that conditions based on limits on ratios, called merit ratios, such as contrast-to-noise ratio (CNR), and net decision threshold-to-noise ratio (TNR), and contrast to threshold ratio (CTR) can be used in validating a reliably detectable flaw size. Monte Carlo random data samples from a population of data of a signal response characteristics are used. Decision threshold for each sample and corresponding merit ratios are computed. Using the decision threshold computed from each sample and the signal to flaw size characteristics, the theoretical ath90/95 is computed for each sample. The merit ratios and theoretical ath90/95 for each sample are plotted as distributions. These distributions are analyzed to calculate confidence in the meeting the merit ratio conditions. The target flaw size is compared with the distribution of the theoretical ath90/95 to calculate confidence in validating the target size. The procedure requires measurement of noise. Results of the simulated data prove validity of the approach. Minimum six data points are recommended. Although, small number of flaws are sufficient in this validation procedure, these flaw sizes are carefully selected based on experimental work to find the smallest flaw size that can meet the merit ratio conditions. Therefore, some trial-and-error effort is needed to choose correct target flaw size so that the merit conditions can be met.
|
