Application of minimum entropy deconvolution in diagnosis of reciprocating compressor faults based on airborne acoustic analysis

The airborne acoustic signals from reciprocating compressors (RCs) exhibit an impulsive periodic transient response and are modulated due to several reasons, including structural and acoustic resonance. The occurrence of faults such as intercooler leakage, filter blockage and compound faults such as a combination of intercooler and discharge valve leakage can enhance the feature characteristics of the signal. As a result, the randomised periodic impulse and the presence of non-linearity due to valve fluttering can contribute to the series of harmonic components in the acquired signal. Thus, common methods have limitations in terms of identifying the characteristic features from signals submerged in high levels of background noise. In this paper, a deconvolution technique named minimum entropy deconvolution (MED) has been adopted to extract the features of the impulses, filtering out the non-transient components from the signal and providing a filtered output that only contains the periodic and transient components of the signal. The filtered signals are then analysed by estimating the root mean square (RMS) and entropy values under various operating pressures with the presence of different faults. The analysis result from the entropy of the filtered signal performs adequately to diagnose the conditions of the reciprocating compressor and hence finds suitable application of the method in diagnosis of compound faults using airborne acoustic signals, making it a remote and cost-effective condition monitoring technique.