Measuring Reliability of Spot-Weld Nondestructive Tests

An improved metric for the reliability of spot-weld nondestructive tests is proposed. The traditional metric uses rates of false calls and misses obtained by comparing results of the nondestructive test with subsequent destructive determinations of button size. Such error rates, however, are not determined by the characteristics of the nondestructive test alone but are confounded by the distribution of weld quality of the welds input into the test. The only meaningful way to characterize the reliability of the nondestructive test, independent of the weld distribution, is by quantifying the uncertainty in the test’s predictive power.