A NOISE INTERFERENCE MODEL FOR ESTIMATINGPROBABILITY OF DETECTION FOR NONDESTRUCTIVEEVALUATIONS

摘要

A traditional way to estimate probability of detection (POD) from quantitativeinspection data involves estimating the relationship between signal response and flaw size using alinear regression model of the (possibly) transformed data. Noise response data, when available, areused to estimate the detection threshold to control the probability of a false alarm. One of the directresults from the traditional POD method is that the POD will be close to zero where there is no flaw.However, the POD for the limiting case of a very small flaw (or no flaw) is actually the probability ofa false alarm. In this paper, we will use both the noise data and the measured signal data to estimatethe parameters of a combined model for signal and noise. The resulting POD estimate from the noiseinterference model provides a correct estimate of the positive probability of detection for small andzero flaw sizes.