Translating Physics into Intelligence: Machine Learning for Reliable ESP Condition Monitoring

This research focuses on combining physics-based expert rules with machine learning (ML) to improve the detection of failure-related events in Electric Submersible Pumps (ESPs). The goal is to create a generalized framework for abnormal event detection that addresses the challenge of limited labeled data and sensor data, providing a practical solution for ESP management and maintenance planning. Physics-based rules are applied to real-time historical ESP sensor data to identify potential failure events throughout the ESP lifecycle. These rules generate failure-related event scores, which serve as training data for a supervised ML model. Instead of directly labeling events as normal or failed, the model learns from these physics-derived scores, helping it detects subtle signals even during normal operations. The training framework is model-agnostic and was benchmarked against widely used semi-supervised learning approaches for comparison. Physics-based rules showed reasonable accuracy in detecting abnormal events but were often affected by overlapping or noisy signals. Training the ML model on the aggregated rule outputs improved event detection, reduced false positives, and enhanced generalization across different ESP configurations (e.g., sensor availability) and operating environments (e.g., various reservoirs and wells). This approach also increases robustness against sensor faults and adapts to varying operational conditions without extensive retraining. The final model acts as a denoiser and generalizer of the expert rules, offering reliable early warnings of ESP abnormalities. This makes it suitable for field deployment with minimal manual data labeling or intervention. This work introduces a hybrid approach where ML enhances and refines physics-based expert systems. By using ML as a denoising layer over the physics rules, the method bridges rule-based and data-driven approaches. It provides a scalable, label-efficient solution for ESP health monitoring, with potential for broader use across the energy sector.