Computer-Based Sensor and Process Fault Diagnosis in Real Time

Abstract

Computer-based, on-line fault detection and identification in nuclear power plants (NPP’s) by monitoring the time behavior of measurement signals can be performed in three levels. These are the level of hardware redundant signals, the level of processes where methods of analytical redundancy can be applied and the level of closed systems where a new technique using Petri nets has been derived. Methods of the 1st level have already been applied in NPP’s, while a software module related to the 2nd level was tested on-line on a living plant monitoring a steam generator. A Petri net module (3rd level) has been implemented for leak detection and was tested off-line using plant data. Modules of these three levels have been combined to a system for the earLY sensor and process fault detection and DIAgnosis, called LYDIA, for the monitoring of ~ 250 measurement signals of a NPP. LYDIA has been implemented on a parallel processing system. The testing data are generated by the GRS plant simulator ATLAS. The diagnosis based on the modules’ output will be performed on a hybrid AI-system consisting of an expert system and a neural net.