Air compressor fault diagnosis based on lifting wavelet transform and probabilistic neural network

Abstract

Due to the wide variety of faults, the components of air compressor fault condition vibration signal are complex. Fault features extracted by conventional time-frequency analysis and multi-resolution analysis are not comprehensive enough to reflect the fault condition of air compressor. To overcome this deficiency, this paper proposes an air compressor fault diagnosis algorithm based on lifting wavelet. Firstly, the vibration signal of the air compressor is decomposed by lifting wavelet; then the statistic such as the peak value and Kurtosis of the decomposition layer is calculated as the fault features. Finally, Probabilistic Neural Network is utilized to classify the fault state. In the experiment, the fault extraction methods such as Wavelet Packet Decomposition and Continuous Wavelet Transform are compared. The results indicate that the fault features extracted by lifting wavelet are more comprehensively to reflect the fault condition of air compressor while other fault character extraction methods need more fault characteristics; the fault diagnosis accuracy by utilizing proposed method is high and the training time is short, which is more suitable for the online fault diagnosis of the actual air compressor.