Real Time Failure Prediction in Electrical Submersible Pumps

As oil fields age, artificial lift systems are required to maintain production levels when reservoir pressures get too low. Failure of an ESP in a well can stop production or even lead to a dangerous event. ESPs are fitted with downhole monitoring units that transmit streams of data back to the surface including: motor temperature, pump intake pressure, intake temperature motor vibration, and motor current. This paper describes a system to monitor large, geographically diverse arrays of oil wells with ESPs. Sensor measurements are transmitted, normalized, and integrated using a distributed communications network and stream processing system. Analytic models for predicting failure are created offline using historical data analysis, and executed in real time against live sensor data using the stream processing system. When failure is predicted, alerts are dispatched to both a live data operator console and a visual analytic platform. An implementation is described on a major North American oil production system using the TIBCO Fast Data Platform including TIBCO StreamBase CEP, TIBCO Spotfire Analytics, and TIBCO Live Datamart.