A neural network approach for the automated detection of faulty electromagnetic probes in a nuclear fusion experiment

RFX (Reverse Field Experiment) is one of the large nuclear fusion experiments within the framework of the co-ordinated nuclear fusion research program of the European Community. Its configuration requires precise knowledge of the magnetic quantities for the understanding of the plasma behaviour. Due to the large number of signals acquired from the electromagnetic probes, an automated test procedure is required to monitor their functionality. We report the results of a novel approach for the automatic detection of faulty signals, based on Neural Network techniques. The Adaptive Resonance Theory (ART) network architecture proved to be best suited for this kind of application.<<ETX>>