In previous work, methods have been developed for efficienttesting of components and instruments that are based on models of theseunits. These methods allow for the full behavior of these units to bepredicted from a small but efficient set of test measurements. Suchmethods can significantly reduce the testing cost of such units byreducing the amount of testing required. But these methods are validonly as long as the model accurately represents the behavior of theunits. Previous papers on this subject described many methods fordeveloping accurate models and using them to develop efficient testmethods. However, they gave little consideration to the problem oftesting units which change their behavior after the model has beendeveloped, for example, as a result of changes in the manufacturingprocess. Such changed behavior is referred to as nonmodel behavior ornonmodel error. When units with this new behavior are tested with thesemore efficient methods, their predicted behavior can show significantdeviations from their true behavior. This paper describes how to analyzethe data taken at the reduced set of measurements to estimate theuncertainty in the model predictions, even when the device hassignificant nonmodel error. Results of simulation are used to verify theaccuracy of the estimates and to show the expected variation in theresults for many modeling variables
展开▼
