Effects of nonmodel errors on model-based testing

In previous work, model-based methods have been developed for efficient testing of components and instruments that allow for their full behavior to be predicted from a small set of test measurements. While such methods can significantly reduce the testing cost of such units, these methods are valid only if the model accurately represents the behavior of the units. Previous papers on this subject described many methods for developing accurate models and using them to develop efficient test methods. However, they gave little consideration to the problem of testing units which change their behavior after the model has been developed, for example, as a result of changes in the manufacturing process. Such changed behavior is referred to as nonmodel behavior or nonmodel error. When units with this new behavior are tested with these more efficient methods, their predicted behavior can show significant deviations from their true behavior. This paper describes how to analyze the data taken at the reduced set of measurements to estimate the uncertainty in the model predictions, even when the device has significant nonmodel error. Results of simulation are used to verify the accuracy of the estimates and to show the expected variation in the results for many modeling variables