Pushing the PI System data to the cloud for greater insights
- ChallengeNeeded to use PI System data to perform advanced analytics but had difficulty moving data to the cloud.
- SolutionSet up operational data pipelines to move PI System data into the cloud to populate AI and ML models.
- BenefitsA reliable, scalable anomaly detection solution that gives users the ability to generate insights via self-service tools.
Occidental Petroleum’s onshore and offshore operations produce over 714,000 barrels of oil equivalent per day. Any equipment failures can put lives at risk, making real-time visibility into asset performance and anomaly detection critical for avoiding catastrophes. To gain that visibility, Occidental uses the PI System as the foundation for real-time asset monitoring. However, the operational data must be moved to the cloud to be used in machine learning models, so Occidental built a homegrown data transfer solution. When that solution proved to be cumbersome and costly, they turned to OSIsoft service provider, Embassy of Things, to automate data movement and processing and enable engineers to identify issues, perform root cause analysis, and take preventative action.
Anomaly detection with the PI System
Monitoring and managing onshore and offshore oilfield performance is no small feat. With numerous assets spread across large geographic locations in a high-risk environment, every second counts. That’s why Occidental created the Intelligent Production Surveillance Optimization (IPSO) program. Designed as real-time surveillance and reservoir management solution, IPSO works to optimize onshore and offshore well performance and calculate diagnostics through automated algorithms.
Under the IPSO umbrella, the Advanced Analytics and Emerging Technology team created the APA Anomaly Detection System to perform root cause analysis and take preventative action. Applied to high-value operational assets, such as pipeline pumps or compressors, the APA Anomaly Detection System leverages real-time asset data from the PI System servers located at onshore and offshore sites. Each server is equipped with Asset Framework (AF), the contextualization layer of the PI System, to create a consistent equipment and asset data model. “Once the right data hierarchy was in place, it was time to push that PI System data to the cloud to train and run predictive models,” said Eric Garcia, cloud engineering manager at Occidental.
The homegrown hurdle
Making PI System data available in cloud services would allow engineers to use that operational data within machine learning models to perform advanced analytics and anomaly detection. However, the models first needed to be trained, which required Occidental to perform an initial backfill of a large historical PI System database. Not only that, Occidental needed to stream real-time PI System data into the trained models so team members could quickly detect issues and take preventative action. During the data transfer process, it was important to retain meta-data values in Asset Framework (AF), the contextualization layer of the PI System, to ensure data mapped properly.
Initially, IT engineers developed a custom solution for data movement. The solution pulled PI System data into cloud services using a Python-based cascade of scripts and databases. Unfortunately, the homegrown transfer process was complex, had no software support, and required IT engineers to have in-depth knowledge of the PI System. The backfill process also required a manual extraction and upload. “Our initial response was to build a homegrown system,” said Garcia. “But we realized that to get the most out of our PI System data we needed to find another way.”
Opening the data pipeline
To build a long-term, scalable solution, Occidental created data pipelines to the cloud that would allow engineers to use PI System data for specific business cases. Using Twin Talk, Embassy of Things’ solution to automate data movement, Occidental’s new architecture pushes and pulls PI System data between applications. Twin Talk runs inside the process control network, while providing people outside of the network a granular view of PI System data. Asset hierarchies are mapped to operational data sets, allowing users to create operational data pipelines using PI System tags. Each pipeline addresses a specific business case, and users can access pipelines for their individual needs.
Next, the team addressed the large data backfill. Occidental curated groups of operational data sets relying on AF, and those data sets were used to automatically perform backfill operations. Initially, 7,750,000 values were transferred in just 85.45 seconds. Moving forward, Twin Talk will load the transferred data into Big Query where it can be processed or transferred to other databases and used to train AI and ML models.
Now Occidental has hundreds of data pipelines that stream PI System data into cloud services, enabling engineers to perform real-time monitoring and cross-site analytics, as well as leverage AI and ML models to detect anomalies and predict asset failure. With PI System data, Occidental is on its way to shorter maintenance cycles, increased asset reliability and uptime, and self-service data options so users can quickly and easily get the insights they need.
For more information about Occidental and the PI System, watch the full presentation here.