An intelligent diagnosing method to compressors' faults based on fuzzy RBFNN

Abstract

In order to diagnose the compressors' fault on line, an intelligent checking method is presented in the paper. A vibration sensor was put on the compressors that should be detected. The vibration signals obtained by the sensor contain a great deal information, which reflects the compressors' qualities and their type of faults. It has been proved that, the vibration signals obtained from compressors with different faults have different time domain features and frequency domain features. We extract those features, and then get a feature vector which is sent to an intelligent information processor. In order to improve the generalization and robustness of the processor, we adopt a fuzzy clustering radial basis function (RBF) neural networks as the information processor. A method of fuzzy C-means clustering based on minimized mean square error rule is used to determine the RBF layer, and the shape factors of RBF neurons are determined by the grades of membership. The experimental results show that, fuzzy clustering RBF neural networks neural networks have powerful ability of pattern recognition, and the faults diagnosis method is feasible to diagnose the fault of the revolving machinery.