Blind identification methods applied to Electricite de France's civil works and power plants monitoring

In this article, we would like to present results obtained on industrial data with source separation techniques in instantaneous mix. One introduces three applications developed to perform the monitoring of Electricite' de France civil works and power plants. The first application concerns the monitoring of nuclear power plants. Each internal component generates specific vibration modes and ''neutron noise'' is a combinaison of all modes generated. The aim of this study is to separate such independent vibration modes. The second application concerns the dams supervision : it consists in separating the various types of motion of a dam according to their physical origin. The third application concerns non destructive resting on steam generators in nuclear power plants. The aim is to reduce the flattening noise. The classical methods operate only when a noise reference is available. We propose to use the multi-sensor approach with the blind separation methods (the noise reference is not necessary). Considering the specifications of the signals, we get better performance using a two-order statistics algorithm than a higher-order statistics algorithm.