Real-time anomaly detection methodology in the electric submersible pump systems

Trips and failures are common occurrences in the Electric Submersible Pump (ESP) systems. Considering the high maintenance cost of the intervention and workovers post failures, ESPs connected to real-time monitoring and surveillance systems are used to fulfill proactive responses and early detection of incidents. Large amounts of dynamic and historical production data acquired from the surface and downhole sensors can help perform diagnostics and prognostics on ESPs. This study applies the Hotelling T-square statistic (T2) and Squared Prediction Error (SPE) equations combined with Principal Component Analysis (PCA) method to construct a predictive model based on the collected ESP data. This predictive model provides a solution that enables to detect the developing ESP trips and failures in real time. A new input matrix corresponding to any period in the whole ESP life cycle is fed to the predictive model to obtain prediction results. The model validation analysis demonstrates that the predictive model can identify the stable operating regions of ESP and detect the potential ESP trips and failures before the actual failure time. This study concludes that this proposed diagnostic model can serve as a real-time platform to identify dynamic changes and therefore monitoring potential ESP trips and failures in real-time.