Rolling Rotor Compressor Fault Diagnosis Method Combined With Wavelet Packet And Neural Network

Rotary compressor has many features such as small size, light weight, high efficiency and high reliability and is widely used in residential equipments such as refrigerator, air conditioner and so on. In this paper, a fault diagnosis method is researched for this type of compressor based on modern signal processing theory. The method can be used in automatic detecting system on the production line.Firstly, a brief introduction is gived about the structure and working mechanism of the rotary compressor. The causes of fault and position are also analyzed. After this, a great deal of vibration and noise data is acquired on the production line. Secondly, traditional signal processing method is applied to determine positions and direction that include relatively abundant fault information. Lastly, the vibration signals from the top and flank of the compressor crust are decomposed and reconstructed by wavelet packet method. The energy of different frequency ranges is put into back-propagation neural network to judge normal and fault compressors. The effectiveness of this approach has been proved by the diagnosis examples of many rotary compressors and the precision rate reaches 97%. In addition, we also try to investigate noise fault analysis using this method.