Fault Diagnosis of Steam Generator Using Signed Directed Graph and Artificial Neural Networks
- Nuclear Eng. Department, Fac. of Eng., Alex. Univ., Alex. (Egypt)
- Nuclear Power Plants Authority, Cairo (Egypt)
Diagnosis is a very complex and important task for finding the root cause of faults in nuclear power plants. The objective of this paper is to investigate the feasibility of using the combination of signed directed graph (SDG) and artificial neural networks for fault diagnosis in nuclear power plants especially in U-Tube steam generator. Signed directed graph has been the most widely used form of qualitative based model methods for process fault diagnosis. It is constructed to represent the cause-effect relations among the dynamic process variables. Signed directed graph consists of nodes represent the process variables and branches. The branch represents the qualitative influence of a process variable on the related variable. The main problem in fault diagnosis using the signed directed graph is the unmeasured variables. Therefore, neural networks are used to estimate the values of unmeasured nodes. In this work, different four cases of faults in the steam generator ( SG) have been diagnosed, three of them are single fault and the fourth is multiple fault. The first three faults are by pass valve leakage (Vbp(+)), main feed water valve opening increase (Vfw(+)), main feed water valve opening decrease (Vfw (-)). The fourth fault is a multiple fault where by-pass valve leakage and main feed water valve opening decrease (Vbp(+) and Vfw (-)) in the same time. The used data are collected from a basic principle simulator of pressurized water reactor 925 Mwe. The signed directed graph of the steam generator is constructed to represent the cause-effect relations among SG variables. It consists of 26 nodes represent the SG variables, and 48 branches represent the cause effect relations among this variables. For each fault the values of measured nodes are coming from sensors and the values of unmeasured nodes are coming from the trained neural networks. These values of the nodes are compared by normal values to get the sign of the nodes. The cause-effect graph for each fault is constructed from the steam generator signed directed graph by removing the invalid (normal) nodes and inconsistent branches. Then in the cause-effect graph we search about the node which does not have an input branch. This node is the fault origin node. The result of this work demonstrated that this method can be used in nuclear power plant fault diagnosis. The advantages of this method are, it enables us to diagnose a multi fault, it is not restricted by pre-defined faults, and it is fast method. (authors)
- Research Organization:
- American Nuclear Society, 555 North Kensington Avenue, La Grange Park, IL 60526 (United States)
- OSTI ID:
- 21021051
- Resource Relation:
- Conference: 2006 International congress on advances in nuclear power plants - ICAPP'06, Reno - Nevada (United States), 4-8 Jun 2006; Other Information: Country of input: France; 14 refs; Related Information: In: Proceedings of the 2006 international congress on advances in nuclear power plants - ICAPP'06, 2734 pages.
- Country of Publication:
- United States
- Language:
- English
Similar Records
Incipient multiple fault diagnosis in real time with applications to large-scale systems
Complexity of system-level fault diagnosis and diagnosability