Predict Oilfield Equipment Failure and Improve Maintenance
One of the largest oil and gas producers in the United States launched the first phase of deployment of BHC3 Predictive Maintenance™ for its oilfield equipment in just 12 weeks, demonstrating the capability of software analytics to accurately forecast equipment failure and improve condition-based maintenance of their beam pumps.
BHC3 Predictive Maintenance employs machine learning based algorithms to enhance failure prediction and diagnostic capabilities. The application augments traditional systems by continuously monitoring all instrument signals, tracking complex failure modes, and detecting operating anomalies associated with impending equipment failures for a large range of oilfield equipment. In this deployment, the team integrated daily sensor readings from oilfield equipment and unstructured data (e.g., field notes and operator comments) from maintenance work orders. This comprehensive data integration and analysis gives oilfield service teams a complete weeks-ahead view of emerging equipment maintenance requirements, with detailed supporting data and diagnostic tools to support maintenance decision making.
About the Oil Producer
- Upstream, midstream, marketing, and chemical divisions
- $20 billion annual revenue
- 20,000+ production wells globally
- 500,000+ BOEPD production
- Global upstream operations across 10 countries
"We were able to predict 46% of the failures in advance and 77% of the predictions were correct.”
Lead Data Scientist
Building and deploying an AI-based application
Employing the latest developments in analytical algorithms, machine learning, data integration, and cloud-scale infrastructure, the joint team implemented three machine learning classifiers, one for each beam pump failure mode (rod, pump, and tubing) and applied 650 analytic features to predict failures. These analytics enable machine learning algorithms to unlock insight from the oil producer’s structured time-series data and unstructured maintenance records, notes, and work orders (i.e., free text). The BHC3 Predictive Asset Maintenance machine learning application employs advanced clustering, classification, and prediction analytics to identify conditions associated with failure and project failure likelihood. It also continuously incorporates user feedback to increase prediction accuracy and reduce instances of false-positive cases. As a closed-loop system that learns based on field feedback, BakerHughesC3.ai solutions allow operators to apply predictive maintenance algorithms at scale while generating increasingly accurate and relevant results for their specific equipment environments.
- 1,031 beam pump wells analyzed
- Data aggregated from 3 source systems
- 6 years of data with 300+ unique data fields
- Structured and unstructured data (e.g., field maintenance notes) processed
- 650 analytics created for failure prediction algorithms
- Separate prediction scores calculated for rod, pump, and tubing failure