Abstract This paper presents a machine learning–driven workflow to enhance reliability in Electrical Submersible Pump (ESP) operations through early risk identification. Deployed across 250 ESPs in a South American asset, the solution integrates real-time sensor data, historical failures, and operational context to detect anomalous behavior and predict failure probability up to 90 days in advance. The system is composed of two models: a Healthy Model for anomaly detection and a Survival Model to estimate time-to-failure. These models feed into a dashboard that delivers daily alerts and drives timely field actions such as chemical cleanups, parameter adjustments, and surface recalibrations. The deployment resulted in a 72% recall rate and enabled preventive interventions in 75% of the high-risk wells during Q1 2024. On average, run life was extended by over 145 days in successfully mitigated cases. The paper summarizes key technical learnings and outlines how predictive analytics can support more efficient ESP management.
Batallas et al. (Mon,) studied this question.