A Study of Diagnosing Reciprocating Compressor Faults using EMD-entropy of the Airborne Acoustic Signals

For more effectively and less-costly monitoring of reciprocating compressors, this paper focuses on the developing of a new monitoring method based on airborne sounds which can be easily obtained in and remote way and contains richer information of an entire machine. Possible sound sources of the compressor have been examined according to the working process of mechanical motions and fluid dynamics in order to understand the sound characteristics under different operating conditions. Especially, the sound signal from the compressor is highly non-stationary due to the periodic excitation resulted by the combined effect of gas dynamics and the mechanical forces acting on the compressor associated with the random disturbances of valve motions and flow turbulence and the variations of discharge pressures. In addition, the acoustic signals are also veritably influenced by background noises which often are of unsteady. To characterise such signals for fault detection and diagnosis, Empirical Mode Decomposition (EMD), an effective tool for non-stationary signal analysis, is used to find and enhance the inherent information that correlates more to the various acoustic events involved in compressor operations. Experimental studies, carried out based on a two-stage reciprocating compressor, have shown that Intrinsic Mode Functions (IFM) from EMD can depict more of the signals to indicate the conditions of the machine. In particular, using EMD- entropy as a diagnostic parameter allows common faults such as inter-cooler leakage (ICL) and discharge valve leakage (DVL) to be discriminated and separated from the baseline operation over a wider range of discharge pressures, demonstrating that the proposed EMD acoustic signatures can be an effective approach for monitoring reciprocating machines.