Statistical analysis of a simulator based anomaly detector
In a nuclear power plant the enormous amount of information arriving at the control room from the sensors placed throughout the plant is drastically reduced to a few indicators, easily attended by the operators. In the course of this data reduction it may happen that the statistical fluctuations combine so unfavourably as to simulate falsely an ongoing accident. This problem is considered with reference to a plant whose main components are equipped with physical simulators yielding the expected values of measured and/or state variables. By comparing the measured signals with the simulated ones, three successive anomaly levels are activated by fault indicators, relating the first to each single detector, the second to the totality of them and the last one to the temporal permanence of the first two indicators in the alarm state. An alarm is presented to the operator only when the third level is activated. A statistical analysis of the various probabilities involved is performed and the results indicate that, at least in the numerical examples worked out here, the proposed procedure yields fewer false alarms than a standard one based on the chi-square calculation.
Bibliographic Reference: Paper presented: 22nd Informal Meeting on Reactor Noise (IMORN), Delft (NL), June 13-15, 1990
Availability: Available from (1) as Paper EN 35573 ORA
Record Number: 199011562 / Last updated on: 1994-12-02
Original language: en
Available languages: en