Enterprise AI for Detecting Control Valve Anomalies at Offshore Upstream Production
A major oil and gas operator approached Baker Hughes and C3 AI (BHC3) to apply AI-enabled technology to identify control valve failures with advance warning to reduce unplanned outages on offshore production facilities.
To manage valve performance, the oil & gas company used a rules-based alarm system. However, the existing control systems relied on preset thresholds and due to varying operating conditions, generated a high volume of false alarms per year. Operators faced nearly 30 alarms per valve each month across 49 valves. Furthermore, the existing system was difficult to use, and operators did not know how to configure or apply analytics to improve alarm quality. The upstream processing facility had limited redundancy of critical control valves. As a result, control valve failures caused several unplanned maintenance events in 2020, resulting in compressor turbine shutdown, significant production deferments, and increased maintenance costs from urgent workovers.
The company needed a better solution to help operators accurately predict control valve anomalous state and failures in advance, reduce alarms, and better leverage the high volume of data from control systems.
- Create a unified, federated data model integrating disparate data sources (e.g., asset data, sensor tags, outage events)
- Deliver a user-friendly application that applies machine learning to identify high-risk control valves and predict unplanned outages in advance
- Reduce the number of alarms created by rules-based systems
- Demonstrate scalability of application to onboard and manage a large number of control valves and machine learning models
- Reduce non-productive time caused by unplanned downtime to increase throughput
About the International Oil & Gas Company
- $40B+ in 2020 annual revenue
- 50,000+ employees
- Operates in 15+ locations globally
A team of Baker Hughes and C3 AI experts collaborated with project managers and subject matter experts to onboard 49 control valves, build, train, and test over 350 machine learning models, and deploy the Shell Predictive Maintenance Control Valves Module that could detect anomalous valve condition in advance and provide operators with early warning signals to reduce unplanned downtime. The team ingested over 100 million rows of data from 4 disparate source systems to create a unified and federated data image. They then trained more than 350 models in just 7 days to arrive at the final model for each valve. Baker Hughes and C3 AI developers configured Shell Predictive Maintenance Control Valves dashboards and screens to visualize the predictive insights generated by the models. The user-friendly dashboards, configured with readily available UI components, provided a managerial view of high-level KPIs, top AI-driven alerts, individual model predictions, tag analysis, bulk model training, and other ML model management capabilities. With the proven success of this initial phase, the oil & gas company plans to implement the production-ready Shell Predictive Control Valves for 600 control valves in the same production unit.
- 6 weeks from kick-off to production-ready application
- 100M+ rows from 3 years of operational data ingested from 5 disparate data sources
- 350+ machine learning models built and tested
- Up to 9 trained machine learning models per control valve with one champion model
- 6 screens configured in application user interfaces