Prediction failures in electric submersible pumps (ESP) for improved production performance

Electric Submersible Pump (ESP) installations are widely used in oil extraction worldwide. Failures of ESP in the field require costly repair interventions and lead to downtime in oil wells. However, it is important to note that ESP's technical limits exceed the average failure time, necessitating predictive models to optimize logistics for repair crew scheduling and pre-failure equipment stockpiling. This not only reduces well downtime but also enhances extraction efficiency. This paper presents an approach to predict ESP failures using machine learning (ML) techniques. The authors focus on developing a pipeline for feature engineering and model evaluation, specifically predicting if a failure will occur within the next three months. The novelty of this work lies in employing time-series features with a focus on sequences. Analysis of the computational results supports the adaptability of the proposed approach in predicting ESP failures in oil wells. Importantly, this study demonstrates model performance that is comparable to the industry's state-of-the-art for similar tasks, highlighting the importance of leveraging telemetry data in these contexts. The ML model was evaluated and compared with a baseline in terms of accuracy in predicting the date of equipment failure. The impact of different features on the probability of equipment failure was also analyzed.