Isolating errors in models of complex systems

One of the steps in creating a mathematical model of a system is to test the model after it has been fully specified, to see whether it is performing adequately. Often, it is found that the model is not performing acceptably (e.g. the model is not giving accurate predictions of the performance of the actual system). The same lack of fidelity can also be observed in established models that had been performing well, indicating a change in the actual system. At this point, it is necessary to diagnose where the problem in the model lies; a process called error isolation. An error isolation technique for detecting the misspecified parameter (or set of parameters) is described. This technique is especially designed for use on state-space models of large-scale systems. The authors report on an example of an application of the methodology to localizing errors in the model of an inertial navigation system