Acoustic Diagnosis for Compressor Using Neural Networks

Abstract

The application of hybrid neural networks consisted of an acoustic feature extraction network and a fault discrimination network to an acoustic diagnosis for compressor is described. The acoustic feature extraction network uses a five-layered feed-forward neural network which is able to extract the nonlinear features from the input information. The fault discrimination network uses a Gaussian potential network which is able to adjust the number of hidden units based on the learning algorithm. The experiment is performed under the various experimental conditions in order to examine the performance of the hybrid neural network for a practical use in the plant. The obtained result shows that the network has the superior performance in the discrimination accuracy, the learning speed and the adjustment of the number of hidden units.