Real-time diagnosis of incipient multiple faults with application for large-scale systems

Abstract

In a complex and large-scale systems such as nuclear power plants which are strongly nonlinear systems, signal validation is very important for proper plant surveillance, control, and safe operation. Especially, it is very difficult to locate the fault(s) origin when there is influx of abnormal signals because abnormal states in the plant are often detected at devices other than at the failed device (fault origin). A number of useful techniques for the automated fault diagnosis have been suggested[1-38], but there are some problems for the real application which are limited detection capability in case of model-based fault diagnosis[28,35], required large historical data treatment for the exact diagnosis in artificial neural networks(ANNs)-based diagnosis[6,8-10,12,15], and exact rule-derivation problem for signed directed graphs(SDGs)-based fault diagnosis[7,11,13,17]. In this study, by employing the merits of the existing ANN- and SDG- based fault diagnosis and adding a new concept, a method is proposed to overcome the typical drawbacks as described above. The SDG used here involves the feedback and feedforward control loops, adopts newly fault propagation time on each branch for faults diagnosis under a transient case, and was also modified to make possible the diagnosis of the troubled pipes such as broken, leaking, or throttled pipes by treating the troubled pipes as new source nodes. Also, the large-scale system is divided into several subsystems and a neural network is incorporated on each subsystem basis for the on-line fault diagnosis. A method for extracting data for the subsystem diagnosis from the SDG is also presented. The value of each node of the SDG changes automatically according to various operation conditions through the system structure identifier which classifies the characteristics of the system into several operating patterns. The method in this study also applies for diagnosis in the transient cases as well as in the steady-state cases u...