CLIENT
The client is a joint venture company operating one of the world's premier super-giant oilfields in the Middle East with a strategy driven to optimize its potential and position the field as a key player in the global oil industry.
CHALLENGE
The client faced challenges in proactively identifying ESP (Electric Submersible Pump) failures and minimizing unplanned downtime, with a focus on electrical failures.
OBJECTIVE
The main goal was to reduce unplanned downtime and production losses in an ESP operated field:
- Predict failure probability with a daily frequency.
- Estimate impact of failure predictions on a well-by-well basis.
SOLUTION
To predict ESP failures a combination of Machine Learning (ML) models was employed:
- ESP Failures Detection using Neural Networks (PyTorch) and ML.NET algorithms.
- Fuzzy logic and industry-known ESP problems for problem detection.
- Unsupervised algorithms to identify patterns of failures.
- Binary classification for predicting failure or non-failure on time-series data.
- Holistic approach combining regression, classification algorithms, and neural networks.
- Exclusive selection of highly accurate electrical failure prediction models.
A Proactive Advisory System was developed encompassing:
- Workflow automation for daily data ingestion and verification.
- ML models generate predictions every day for all wells.
- Dashboard visualization of daily failure probability for each well.
Visualization highlighted wells approaching failure and provided a report enabling potential ESP failures to be addressed in advance.
RESULT
The implemented ML models demonstrated the solution's broad applicability to various types of ESP failures. Specifically, the pilot phase yielded noteworthy results:
- AI-driven advisory allowed scheduling workovers up to 32 days before potential pump failure.
- Pilot delivered an accuracy of predictions up to 83% with an average notification window of 32 days before failure.
- The solution predicted a $5 million saving in deferred production per well, reflecting potential losses during unexpected failures and workovers.
- The success of the pilot phase paved the way for scaling the solution to all ESP wells.
