Enterprise AI for Preventing Gas Compressor Downtime
A major hydrocarbon producer wants to minimize the environmental impact of its upstream oil production. As the oil & gas company develops new oil fields, it has committed to increase its utilization of associated petroleum gas (APG), an unwanted byproduct of upstream oil production. Processed APG, however, can serve different purposes such as fuel for generating electricity for remote oil fields or natural gas sold directly to end customers.
To transmit processed APG, operators pass it through gas compressors to inject APG at the optimal pressure into the gas transmission network. Due to the wide variability in the amount of APG available at the oil fields, the oil & gas company installed mobile gas compressors (MGC) that can easily be transported between individual fields to meet gas transmission demands.
Within a few months of installing MGCs at one of its fields, operators became overwhelmed with daily compressor failures and over 1,500 alarms per month from each compressor. While the existing control systems provided a wealth of data, they lacked the ability to accurately predict compressor failures in advance and could not support the root cause analysis required for troubleshooting. The company also struggled to distinguish false alarms from material alarms due to the volume and complexity of the data received. The unplanned shutdowns caused significant delays and increased maintenance costs. The company needed a better solution that could accurately predict failures, identify root causes, and improve the quality of alarms.
To address these issues and increase the reliability of its compressors, the oil & gas company selected BHC3™ Reliability. Within 10 weeks, the team ingested over 55 million rows of data from 4 disparate data sources, tested over 70 features, and built 50 machine learning models. BHC3 Reliability accurately predicted 50% of unplanned shutdowns with 22 hours of lead time, reduced false alarms by 99%, and reduced non-productive time by 16%.
With BHC3 Reliability, the company can achieve more than $40M in annual savings from increased productivity, increased gas throughput, and reduced maintenance costs.
About the Company
- $40B+ in 2019 annual revenue
- 50,000+ employees
- Operates in 15+ locations globally
Over 10 weeks, a team of Baker Hughes and C3 AI experts collaborated with project managers and subject matter experts from the oil & gas company to develop a production-ready application that could predict mobile compressor failure events and provide operators with early warning signals and root cause analysis to reduce unplanned downtime.
The team began by ingesting over 55 million rows of data from four disparate source systems to create a unified, federated data image. This unified data image included 9 months of telemetry data, logs for shutdowns, alarms, and events, and asset hierarchy extracted from piping and instrumentation diagrams.
The team leveraged the data image and a prebuilt anomaly detection machine learning pipeline as the foundation to determine the optimal model for predicting compressor failures. First, the team defined true anomaly events for the compressor systems and used this definition as an input to the anomaly detection pipeline. Then the team further refined the model by collaborating with the oil & gas company’s subject matter experts to generate and test over 70 machine learning features. Overall, the team conducted 5 experiments and tested more than 50 model configurations in order to arrive at the final model.
Lastly, the Baker Hughes and C3 AI developers configured BHC3 Reliability dashboards to visualize the predictive insights generated by the model. The user-friendly dashboards, configured with readily available UI components, provided a managerial view of high-level KPIs, top AI-driven alerts, and overall asset health of compressor systems and subsystems. Users could also filter alerts and utilize the solution’s interpretability framework to deep dive into specific AI alerts to view the key components or tags contributing to an AI alert.
- 10 weeks from kick-off to production-ready application
- 55M+ rows from 9 months of operational data ingested from 4 disparate data sources
- Created unified object model to represent asset hierarchy, telemetry data, and shutdown and
- 70+ machine learning features extracted from subjected matter experts and tested
- 50+ model configurations built and tested
- 3 BHC3 Reliability application user interfaces configured
With the BHC3 Reliability application, the oil & gas company can:
- Reduce non-productive time by 16%, equivalent to $40M in savings when scaled to 50 compressors
- Reduce maintenance costs by enabling predictive, versus reactive, maintenance
- Extend compressor asset lifetime by 5 years
- Reduce high-risk emergency repairs and improve safety
By using the BHC3 Reliability application, the operators will be able to:
- Increase alert quality, with a 99% reduction in false alarms
- Accurately predict ~50% of all unplanned gas compressor outages
- Identify high-risk compressor units up to 22 hours before they shut down and proactively dispatch resources to address impending outages
- Resolve issues faster via detailed root cause analysis for each AI alert
- View top operating KPIs such as overall system health and all relevant data through a single application