Case Study: a major oil and gas company reduces ESP downtime with SparkCognition predictive maintenance

Predicting ESP Failures With Variable Data Quality

A major oil and gas company faced the challenge of predicting costly electric submersible pump (ESP) failures. They were impeded by poor data quality and quantity due to broken underground sensors, remote power outages, and insufficient data storage. They partnered with the AI solutions provider SparkCognition to overcome these barriers and implement a predictive maintenance solution.

SparkCognition implemented two custom machine learning approaches to address the data issues. Using normal behavior modeling and clustering with classification, they built models that could predict ESP failures in advance. These solutions successfully identified historical failures, with the models estimated to have saved 45 and 27 days of lost production time respectively, preventing millions in downtime costs for the oil and gas company.