FDI utilizes system models to predict operating values for a chosen set of system variables in a given mode of operation (Fig. 1) [6, 7]. This set of variables, called observations, is continuously monitored during normal operation.
The diagnosis system maps
observations, y, that deviate from predicted normal behavior,
,
onto a system model (Fig. 1).
Analysis of descrepancies, r in the context of
the model helps to generate one or more hypothesized
root-causes, f, that explain the observed deviations.
Hypothesized faults suggest modifications to the system models which are then employed to predict future system behavior. Continued monitoring and comparison with these predictions helps refine the initial fault set, f. Faults whose predictions remain consistent with the observations determine the root-causes for the observed problems. From a computational viewpoint, the better the prediction, the easier it becomes for FDI algorithms to quickly prune the search space through continued monitoring and comparison with predictions. The goal is to continue the monitoring, comparison, and refinement process till the exact set of faults occurring in the system are isolated. The overall process of monitoring, generating hypothetical faults, prediction, and fault isolation using system models as the primary basis is referred to as model based diagnosis.
Figure 1: Diagnosis of dynamic systems.