Application of Data Mining Techniques for Sensor Drift Analysis to Optimize Nuclear Power Plant Performance

The Power Plants are engineered and instrumented to ensure safety in all modes of operation. Hence they should be continuously monitored and maintained with necessary Instrumentation to identify performance degradation and the root causes to avoid calling for frequent maintenance. The degraded performance of Instrumentation & Control systems may also lead to plant outages. Different studies have suggested that a well maintained instrumentation with errors and response times within the permissible limits may increase the availability minimizing outages. The I&C systems are designed for monitoring, control and safety actions in case of an event in a power plant. The sensors used are single, redundant, triplicated or diverse based on the type of application. Where safety is of prime concern, triplicated and 2/3 voting logic is employed for initiating safety actions. Diverse instruments are provided for protecting the plant from any single abnormal event. Redundant sensors are used to improve plant availability. Wherever 2/3 logics are used, the sensors shall uniformly behave and the drifts across the sensor may lead to crossing the threshold, initiating a protective action. Instead of waiting for the regular preventive maintenance schedule for recalibrating the sensors, the drift in the sensors are analyzed by developing a combined overall online monitoring parameter which will give an early warning to the operator the need for recalibration of the redundant sensors. This paper deals with development of one such parameter through data mining techniques for a representative process in a nuclear power plant.