Fault diagnosis system for prognosis and health monitoring of offshore process equipment

초록

Theoffshoreplant'sprocessequipmentisdesignedtooperateinaminimizedfailurestateduringitslifecycle.Itis designed to ensure the safe operation of equipment through reliability information such as MTBF(mean time betweenfailure)andHAZID/HAZOPanalysis.Inordertopreventcriticalfailureoftheprocessandtoevaluate the possibility of failure during operation of the equipment, anomaly monitoring and fault diagnosis are also essential. The inspection has been driven based on the reliability-based maintenance (RBM) approach. The RBM involvetheregularinspectionsonthebasisofthereliabilitydataobtainedfromriskanalysesofaparticularsystem andrelatedequipment.Also,inmostcases,thepossibilityoffailureoranomalyisdeterminedwhenthemeasured fault indication signal exceeds a predetermined threshold value. The reliability-based method often fails to detect the failure or abnormal operation. Therefore, it is necessary to extract a significant pattern from the measured signal and diagnosis the failure. In this study, we propose a diagnosis system for fault diagnosis by using the vibration data obtained from the actual facility and optimize the operation of rotatory equipment from the viewpointofPHM(prognostichealthmanagement).PHMisimplementedbythesequentialprocessoftheanomaly detectionandthefaultdiagnosis.Anomalydetectionandthefaultdetectionofrotatoryequipmenthasbeencarried out using the pattern recognition and machine learning technique. We extract the significant patterns from the measured signals using machine learning that can expect the pattern of normal operation. By analyzing the statistical property of pattern, abnormal events are detected. Bayesian network is useful for the isolation of fault pattern, and identification of faults and normal operation. This paper presents the procedure of classification schemesforfaultdiagnosiswithBayesiannetwork.Therefore,theBayesiannetworkmethodisusedtofigureout thenormalandabnormalpatterns.Thediagnosticsystemclassifiesthecharacteristicsofthedatathroughstat