Acoustic Leak Detection at Complicated Topologies Using Fuzzy Classifiers and Neural Networks

A method for detecting and localizing leaks at complicated three-dimensional topologies by measuring the leak induced structure-borne and airborne sound and by applying pattern recognition procedures is being developed. The sound patterns necessary to train fuzzy logic classifiers and neural networks are generated with simulated leaks at the original structure. As features for characterizing the occurrence and the location of a leak, coherence values between high-frequency microphone signals as well RMS-values of acoustic emission sensors are used. The method is even applicable when localization based on propagation time differences or sound attenuation differences fail. The method is prototypically developed for a soviet-type pressurized VVER-reactor.