Qualitative and quantitative approaches to the diagnosis of plant faults

The architecturally relevant issues which arise in the development of both qualitative and quantitative fault-diagnostic systems are discussed, and the technical issues of acquiring the knowledge and using it effectively are addressed. A direct comparison is drawn between the performances of both fault-detection systems while detecting the same fault on a laboratory test rig. It is concluded that fault-diagnostic systems must be sensitive to the appearance of faults but robust enough not to produce false alarms. It is also concluded that the fault signatures associated with plant-transgression-based methods are stronger, and therefore easier to detect, than model-based transgressions. However, the model-based approach can be extremely effective at detecting faults which manifest themselves as changes in internal plant dynamics