INCIPIENT MULTIPLE-FAULT DIAGNOSIS IN REAL-TIME WITH APPLICATION TO LARGE-SCALE SYSTEMS

By using a modified signed directed graph (SDG) together with the dist ributed artificial neural networks and a knowledge-based system, a met hod of incipient multi-fault diagnosis is presented for large-scale ph ysical systems with complex pipes and instrumentations such as valves, actuators, sensors, and controllers. The proposed method is designed so as to (1) make a real-time incipient fault diagnosis possible for l arge-scale systems, (2) perform the fault diagnosis not only in the st eady-state case but also in the transient case as well by using a conc ept of fault propagation time, which is newly adopted in the SDG model , (3) provide with highly reliable diagnosis results and explanation c apability of faults diagnosed as in an expert system, and (4) diagnose the pipe damage such as leaking, break, or throttling. This method is applied for diagnosis of a pressurizer in the Kori Nuclear Power Plan t (NPP) unit 2 in Korea under a transient condition, and its result is reported to show satisfactory performance of the method for the incip ient multi-fault diagnosis of such a large-scale system in a real-time manner.