Nondestructive Testing - Reliability and Statistics

Reliability and Statistics

Probability of detection (POD) tests are a standard way to evaluate a nondestructive testing technique in a given set of circumstances, for example "What is the POD of lack of fusion flaws in pipe welds using manual ultrasonic testing?" The POD will usually increase with flaw size. A common error in POD tests is to assume that the percentage of flaws detected is the POD, whereas the percentage of flaws detected is merely the first step in the analysis. Since the number of flaws tested is necessarily a limited number (non-infinite), statistical methods must be used to determine the POD for all possible defects, beyond the limited number tested. Another common error in POD tests is to define the statistical sampling units (test items) as flaws, whereas a true sampling unit is an item that may or may not contain a flaw. Guidelines for correct application of statistical methods to POD tests can be found in ASTM E2862 Standard Practice for Probability of Detection Analysis for Hit/Miss Data and MIL-HDBK-1823A Nondestructive Evaluation System Reliability Assessment, from the U.S. Department of Defense Handbook.