New methods for leaks detection and localisation using acoustic emission

Real time monitoring of Pressurized Water nuclear Reactor secondary coolant system tends to integrate digital processing machines. In this context, the method of acoustic emission seems to exhibit good performances. Its principle is based on passive listening of noises emitted by local micro-displacements inside a material under stress which propagate as elastic waves. The lack of a priori knowledge on leak signals leads us to go deeper into understanding flow induced noise generation. Our studies are conducted using a simple leak model depending on the geometry and the king of flow inside the slit. Detection and localization problems are formulated according to the maximum likelihood principle. For detection, the methods using a indicator of similarity (correlation, higher order correlation) seems to give better results than classical ones (rms value, envelope, filter banks). For leaks location, a large panel of classical (generalized inter-correlation) and innovative (convolution, adaptative, higher order statistics) methods of time delay estimation are presented. The last part deals with the applications of higher order statistics. The analysis of higher order estimators of a non linear non Gaussian stochastic process family, the improvement of non linear prediction performances and the optimal-order choice problem are addressed in simple analytic cases. At last, possible applications to leak signals analysis are pointed out. (authors).264 refs., 7 annexes