Application of Wavelet Packet Transform and Envelope Analysis to Non-stationary Vibration Signals For Fault Diagnosis of a Reciprocating Compressor

Reciprocating compressors play a major role in manufacturing industries such as oil and gas refineries, petrochemical industrial plants, etc. Therefore, it is necessary to implement online condition monitoring for early and accurate detection of faults which if not controlled can lead to machine inefficiency, damage, or total system shutdown. This paper presents the application of wavelet packet transform (WPT) and envelope analysis to non-stationary vibration signal from a two-stage reciprocating compressor for fault diagnostics. Vibration signal measured on the reciprocating compressor consist of a series of impulsive events with non-stationary random characteristics, which result mostly from mechanical impacts of the valves and impulsive fluid excitation of high-pressure turbulent flows. To characterize such vibration signal, WPT is employed to decompose the measured signal for the extraction of time-frequency information. With the help of statistical based analysis, the most optimal terminal node of the wavelet packet is selected for further study. Envelope spectrum of the optimal terminal node is processed and used for the classification of three common faults including intercooler leakage, second-stage discharge valve leakage and a combined fault at five critical tank discharge ranges (0.55, 0.62, 0.69, 0.76, and 0.83 MPa) for condition monitoring of a reciprocating compressor.