Case Study: Predicting Electrical Submersible Pump Failures Using Artificial Intelligence and Physics-Based Hybrid Models

Abstract

Abstract Unplanned electric submersible pump (ESP) failures cause significant revenue loss due to deferred production, and increased operational cost associated with emergency workovers. Timely prediction of failures due to critical operating conditions can help operators take proactive actions to avoid sudden ESP failures. Traditional condition-based monitoring (CBM) methods generate overwhelming alarms based on deviation from fixed thresholds. Due to the dynamic behavior of the reservoir, these alarm thresholds need to be fine-tuned frequently to generate useful insights. Moreover, these alarms often yield a short lead time to a critical event or failure, leaving little room for taking preventive actions. A major operator in Latin America was experiencing similar operational challenges. In this case study, we show how ESP Predictive Failure Analytics (PFA) helped this operator to identify critical events while the ESPs were running and to take proactive actions to extend ESP run life. Methods, Procedures, Process PFA uses artificial intelligence (AI), physics-based and expert knowledge-based methods to predict short-term damage events, such as a broken shaft, short-circuit, grounded downhole failure, as well as long-term damage events, such as pump low efficiency, sand, scale, deposition, or gassy conditions that can lead to failure. PFA combines these damage events probabilities with parametric survival analysis to predict ESP Remaining Useful Life (RUL). Results, Observations, Conclusions PFA In this study, we demonstrate PFA results for two ESPs. PFA raised a broken shaft/ missed pump stages alarm for ESP run# 4 in well 1 after a sudden motor current and production rate decline. Due to this signal, PFA estimated a significant reduction in RUL. After observing these results, the operator increased the motor speed to improve production. PFA then raised another alarm for downthrust condition, reducing the RUL further. PFA enabled the operator to quickly schedule a workover, reducing downtime. For ESP run# 7 in well 2, PFA estimated a steady decline in RUL due to motor temperature and vibration increase over time. The operator avoided the imminent failure by reducing the pump speed, which helped decrease the motor temperature and increase the ESP run life. Methods, Procedures, Process PFA is a unique approach that combines AI, physics-based, and expert knowledge-based approaches to provide interpretable predictions of ESP failure. Unlike traditional CBM approaches, PFA generates fewer alarms with longer lead times, enabling the operator to take preventive measures to avoid sudden failure. In addition, PFA ranks ESPs based on criticality, which enables effective surveillance of large number of wells.