A Feature Extraction Method for Prognostic Health Assessment of Gas Compressor Valves

Abstract

Abstract This paper presents features derived from the pressure-volume (PV) diagram that is useful in estimating different valve faults in reciprocating compressors with a strong potential of remaining useful life prediction. The PV diagram is expected to deviate depending on valve wear conditions. The valve degradation types explored in this work are leakage, seat wear, and spring fatigue in the suction and discharge assemblies of a Dresser-Rand ESH-1 compressor commonly used in the petrochemical industry. The proposed method estimates the polytropic exponent on the compression and expansion phase as well as the discharge and suction valve loss power and uses them as features for a quadratic discriminant analysis. The features are created through in-cylinder pressure, suction pressure, discharge pressure, and crank angle measurements collected on a single-stage, dual-acting compressor operating on air with wear precisely machined and seeded into the poppets of the inlet and outlet valves. A very high classification accuracy is achieved in distinguishing the wear types, severity, and location with strong prognostic trends.