This research deals with control and fault diagnosis of large scale complex reactive safety critical systems like air crafts, power plants and highspeed trains. Research in this field is a high priority research area in Australia, Sri Lanka and in many other countries of the world. Failure to do fault diagnosis and fault accomadation in these systems can result in severe disasters killing a lot of people who get their service. Being the approach of this thesis a model based approach, it

more »

... tes correctby-design real-time fault accommodation. When the complexity of man-made physical systems increases, the possibilities of using centralised mechanisms for control and fault diagnosis may become severely limited. This happens because of the many disadvantages associated with centralised systems; high spatial complexity, weak robustness, and poor scalability. Hence, the idea of distributed systems is mooted as the most promising paradigm. In order to develop the necessary algorithms to achieve this requirement, it appeared sensible to carry out a corresponding decomposition of the control architecture in order to continue with the distributed approach. Therefore, the main goal of this thesis is to systematically develop a distributed fault diagnosis system that is consistent with the control architecture of complex, tightly coupled safety-critical systems. The system must guarantee that all known types of the faults in the system are detected (completeness) without producing false alarms (soundness) and while providing targeted control behaviour. The research recognised the dominant hybrid nature of the target systems, comprising discrete-event dynamics and continuous variable dynamics, and the extensive research cultures developed separately for the two types of dynamics, with the more recent developments in hybrid approaches. The literature review revealed that a distinct gap in the existing knowledge is the unavailability of compositional (hence, modular) approaches for the integrated development of fault diagnosis and hybrid control. Bond graphs were seen as the most promising choice for continuous variable dynamics, whereas the automata theory that supports both logic design and supervisory control theory was seen as the choice for discrete-event dynamics. iv Fault