Combining with the local wave method, image processing and CPN classification pattern, a new intelligent diagnosis method on fault information extraction was proposed in this paper. Firstly using the local wave method, the vibration signal from reciprocating compressor was disposed to local wave time-frequency spectra, which was further processed by the image partition method, while the result was set as the input of CPN classification mode. In CPN mode, the terminal was associated with the typical fault character of reciprocating compressor. Application of the method in fault diagnosis on a cylinder of the reciprocating compressor showed it a practicability.