In the distributed large-scale system, the behavior of any subsystem is not only influenced by variables belonging to it (local variables), but also by the variables in other subsystems during its interaction with neighboring subsystems. The effect of the fault in one subsystem will be quickly propagated to other subsystems due to their interconnections. Currently, most of the fault detection and diagnosis schemes are focused on centralized system which do not consider the interaction terms and can not efficiently detect the faults.
In this thesis, a distributed fault detection scheme is developed for a class of large-scale nonlinear uncertain systems with unstructured modeling uncertainty. For each subsystem in the large-scale system, a fault detection estimator (FDE) is designed by utilizing local measurements and certain communicated information from neighboring FDEs associated with subsystems that are directly interconnected to the particular subsystem under consideration. Under certain assumptions, adaptive threshold for fault detection in each local subsystem is derived, and its robustness property with respect to modeling uncertainty and interactions among interconnected subsystems is also investigated. Also, the fault detectability conditions characterizing the class of faults in each subsystem that can be detected by this approach is analyzed. A simulation example of automated highway systems is used to illustrate the effectiveness of the proposed method.