Data-Driven Framework for Fault Diagnosis of Process Loop Components

Abstract

All industrial operations for oil, gas, power plants, manufacturing plants, and allied industries are automated. To run operations efficiently and safely, all field instruments must be working properly. Many events like mechanical failure, electrical failure, aging of equipment, operator errors, and lack of maintenance of components are leading causes of unscheduled downtime in the plant. Preventive maintenance techniques are not effective as a variety of instruments are working under different operating conditions. Condition Based Monitoring (CBM) techniques require vibration and temperature data to diagnose the performance of the instruments. CBM techniques are not real-time and are costly as an expert is required to diagnose the faults. Moreover the automated maintenance schedule required for field instruments has not been tried and tested till present. A data-driven framework is proposed to monitor the online performance of process loop components such as transmitters, current-to-pressure converter and control valve. If the type of fault is correctly diagnosed, corrective action can be taken. Real-time data for a control valve and a current-to-pressure converter has been used to develop regression and classification models. Regression models detect deviation in the calibration of a current-to-pressure converter and control valve. SVM Classification models have been developed to identify faults like air leakage, stiction, and disturbed orientation. To diagnose the performance of process loop components, data-driven models are useful. The SVM model shows 97.73% accuracy for fault diagnosis of the control valve. Process loop component maintenance is more effective using a data-driven model than preventive maintenance and CBM techniques.