Case Study: Automated Recommendations to Increase Run Life of the Electrical Submersible Pumps (ESPs) Using Machine Learning

Abstract

Abstract Unplanned ESP shutdowns and failures result in deferred production, which can lead to significant revenue losses. Most operators rely on engineers to optimize hundreds of ESPs using traditional surveillance technologies. Due to inherent limitations of these technologies, identifying critical conditions and taking remedial actions is extremely time consuming. This case study demonstrates how automated field production solutions (AFPS) has helped a major operator in North America to increase ESP run life in two of its producing assets. AFPS is an ensemble of machine learning (ML) and physics-based models that predicts critical conditions, estimates remaining useful life (RUL) and provides remedial recommendations to increase run-life of the ESP. ML models for predicting critical conditions were trained using historical timeseries sensor data, hand labeled by experts for various clinical conditions. Physics based models for detecting critical conditions were calibrated using well completions, fluid property, inflow performance, power, and correlations data. Recommendations models were trained using 5 years of event action logs data for 500+ ESPs installations in North America. In this study, we demonstrate how AFPS increased run life of an ESP with gas interference and locking conditions. Since the first few months of run life, ESP1 experienced sudden fluctuation in motor current, and increase in intake pressure and motor temperature. Its production declined to outside of the minimum recommended flow range. AFPS was able to accurately identify these critical conditions and estimated reduced remaining life. Between May to Sept. 2023, ESP1 experienced severe gas interference/gas locking symptoms. AFPS automated recommendation was enabled and implemented with expert supervision. In Aug. 2023, experts carried out three recommendations provided by AFPS, which significantly improved the gas interference condition and reduced excessive cycling due to motor underload and high motor temperature. AFPS is an innovative approach that combines ML and physics-based methods to automatically provide effective recommendations to increase ESP run-life. Unlike traditional surveillance approaches, AFPS leverages ML models to learn patterns in historical ESP operational data to reliably predict critical conditions, optimal remedial actions and remaining useful life. Thus, AFPS is scalable and robust to different operating conditions and requires minimal human interventions to avoid unplanned shutdowns and failures, resulting in deferred production.