Condition monitoring of reciprocating compressor valve health via a statistical time-frequency approach

Abstract

The goal of this work is to present a time-frequency-based approach to vibration condition monitoring of gas compressor valves. Due to the cyclostationary nature of reciprocating compressors frequency analysis alone is typically insufficient to assess valve condition. Experimental data is collected on an instrumented Dresser-Rand ESH-1 dual-acting reciprocating compressor by seeding known leakage fault severity levels in the suction and discharge manifold valve assemblies. Fault severity is measured by the leakage hole area and the number of damaged valve elements. By applying a short-time Fourier transform signal processing method to externally mounted vibration measurements, it is possible to identify key regions of interest in the compression cycle. Statistical features are then extracted from these regions and used to train a Bayesian classifier. The presented machine learning approach achieves greater than 90% success in assessing the valve leakage health of the compressor.