SENSOR DATA VALIDATION FOR NUCLEAR-POWER-PLANTS THROUGH BAYESIAN CONDITIONING AND DEMPSTERS RULE OF COMBINATION

Sensor data fusion and interpretation, sensor failure detection, isola tion and identification are extremely important activities for the saf ety of a nuclear power plant. In particular, they become critical in c ase of conflicts among the data. If the monitored system's description model is correct and its components work properly, then incompatibili ties among data may only be attributed to temporary deterioration or p ermanent breakage of one or more sensors. This paper introduces and di scusses three simple ideas: 1. classical ''model-based diagnosis'' can be extended straightforwardly to encompass the sensors' models into t he system's description in order to diagnose even their own faults 2. from the ''log-file'' of the diagnosed minimal conflicts among the sen sors, one can draw interesting conclusion regarding their relative rel iability (e.g., through Bayesian conditioning) 3. the estimated reliab ility of the sensors is useful when assessing (e.g., through Dempster' s Rule of Combination) the actual state of the monitored physical syst em, even in cases of conflicting data. These ideas lead to the concept ion of a distributed monitoring system able to attach to each sensor a statistically evaluated relative reliability, which is especially use ful for devices situated in dangerous zones or areas, difficult to rea ch inside huge and complex power plants.