Condition monitoring of reciprocating compressor valves using analytical and data-driven methodologies

Condition-based health monitoring systems are a very important addition to machinery to monitor the system and assure it is running at the peak efficiency, to schedule maintenance, and prevent catastrophic failure. Many times these systems are combined with different sensors to predict when service is required for different wear parts and this keeps the machine running optimally. An accurate prediction of health is accomplished by measuring and analyzing different critical parameters and detecting when these parameters deviate from the nominal values. Recently, these systems have started to become more common on industrial compression technology. Typically, reciprocating compressor health monitoring systems only use indirect measurements, P-V diagrams, to monitor the health of the system. This research focuses on improving these monitoring systems. Specifically this research will focus on three different valve failure modes that are common in reciprocating compressors. They are liquid slugging, valve spring fatigue, and valve seat wear. These faults are investigated first through a system level model to better understand how different subsystem dynamics are related through the compressor. Also an instrument investigation is conducted to determine what types of sensors are the most effective at detecting these faults. The Bayesian classification method is used in conjunction with seeded fault training data to create a classifier that can determine the state of health of the machine. The classification approach can be integrated into health monitoring software to be used in different reciprocating compressors