Automating the Diagnosis of Valve Faults in Reciprocating Compressors

In the previous issue of Maintenance the authors showed that analysis of the Smoothed Pseudo Wigner Ville Distribution (SPWVD) can provide an interpretation of compressor valve vibration response and thus form a basis for valve fault diagnosis. The procedure is far from being routine, however, and is not without its difficulties. In this paper, a technique is presented for compressing SPWVD results, using a pattern similarity analysis, and hence deriving diagnostic parameters by means of which valve faults may be more readily assessed and diagnosed. Also important is the ability to correlate valve faults to deterioration in compressor performance. A 'trained' neural network method for accomplishing this has been developed.