This paper presents an Artificial Neural Network (ANN) approach that classifies a cavitation signal into 3 distinct classes viz., no cavitation, Incipient and developed cavitation signal. In this paper an Elman recurrent network have been used for the classification of cavitation signals in a pressure drop devices which are used for flow zoning in a Prototype Fast Breeder Reactor (PFBR). Classification process can be divided into stages, pre-processing, range fixing stage and classification stage. In pre-processing and range fixing stage, various types of cavitation signal from different zones are fed to recurrent network as input to get a simulated output. Initial processing of signal is carried out on neural network and through vigorous analysis of various cavitation signals, the classification range has been obtained from output of recurrent network based on the magnitude of RMS value of signal acquired from an accelerometers installed downstream of various flow zones. In classification stage, an Elman recurrent network have been used to evaluate the classification results and the optimum network architecture is evolved through an elaborate trial and error procedure. The classification results shows that recurrent network employing resilient back propagation algorithm was effective to distinct between the classes based on the good selection of both network and algorithm parameters. The proposed Elman recurrent model with resilient algorithm gives better performance, classification rate and only requires less computation time. The classification rate was 84.21% for the training sets and 92.89% for test data sets. It is concluded that the performance of the neural network is carried out zone wise and it is optimum, and the errors are very less. The paper also discusses the future research directions.

Key words:Classification of cavitation Signal, ANN model, Elman Recurrent Network, Resilient BPN Algorithm.