Joint Research Conference
June 24-26, 2014
When assessing a binary measurement system using a gold standard to verify the true conforming status of a part can be expensive. In these situations it is important to be intelligent and selective in what parts are verified. Complete verification is too expensive. Verification of a random subset of parts ends up finding very few non-conforming parts. A common approach is to verify only parts that failed the binary measurement system. None of these approaches are particularly effective or well thought out. In this talk a more efficient plan for assessing binary measurement systems is introduced. This plan measures parts with the measurement system of interest multiple times and then uses the information gained to determine which parts to verify. This talk will show that verifying parts that had approximately equal numbers of passes and failures gives immensely more information than verifying parts that either failed or passed most of the time. By selecting parts using this knowledge one can, practically speaking, match estimate precision of the complete verification plan while measuring only a small fraction of the parts with the gold standard.