Leaks from oil pipelines can cause enormous damage to the environment. Bridger Pipeline LLC wanted to ensure that it could respond rapidly and effectively to shut off leaks before they have a negative impact.
Working with IBM, Bridger Pipeline LLC built a deep-learning AI solution running on the IBM® Watson Machine Learning Accelerator platform. The solution uses historical data to aid in the recognition of leaks and to detect abnormal conditions.
Enablesdetection of potential leaks in one-fifth of the previous time using AI
Reducesthreshold at which alarms are sounded, increasing sensitivity
Cutsincidence of false alarms, reducing controller fatigue
Business challenge story
Stepping up responsiveness
Pipelines provide a safe and efficient way to transport hydrocarbons, enabling reductions in road and rail traffic that in turn cut emissions. While Bridger Pipeline LLC was already transporting close to 100 percent of oil to its destination, the company believed that new technologies could help it eliminate the remaining fractions-of-a-percent in losses.
Tad True, Vice-President, says, “We’re a family-owned company with a lot of pride in what we do, so we take any kind of impact on the environment incredibly seriously. Besides the enormous clean-up cost, a pipeline that leaks is completely unacceptable to us from an environmental standpoint. One of our core mission statements is to operate safely, and we’re constantly investing in new capabilities to eliminate leakage.”
With approximately 3,500 miles of pipeline to manage, Bridger Pipeline LLC was facing a significant challenge in detecting and resolving leaks in a timely way. The company moves 450,000 barrels through this network every day, and had built up a sophisticated monitoring system using smart meters and satellite-enabled surveillance systems, bringing huge volumes of real-time data into its control center 24 hours a day, 7 days a week.
“The experts in our control center must predict how turning on a pump or closing a valve will affect the flow of oil hundreds of miles away,” says True. “They rely on sensor data to support their experience and intuition, but our existing system was generating too many false alarms. One major risk in our industry is controller fatigue: people may start to ignore genuine alarms because they believe them to be a fault in the detection system.”
Bridger Pipeline LLC reviewed existing third-party solutions for tightening up leak detection but found them to be both costly and inflexible. As a growing company in a developing basin, Bridger Pipeline LLC is constantly changing and adding to its pipeline network, but off-the-shelf solutions proved impractical. Tad True explains: “The existing systems require a team of people to study your network, then spend up to nine months on implementation. And every time you change your pipeline network, you have to bring those people back out to update the leak-detection system.”
Pushing boundaries with AI
Aiming to build its own solution based on AI technology, Bridger Pipeline LLC spoke with several potential providers before choosing to work on a paid proof-of-concept engagement with IBM.
“What appealed to me about IBM’s approach was the number of questions the consultants asked,” recalls True. “They knew what their deep-learning platform was capable of, and they wanted to understand our challenges and all the data we could provide.”
IBM built a test environment based on its IBM Watson Machine Learning Accelerator offering – a suite of AI software featuring open-source machine-learning frameworks such as TensorFlow and Caffe, optimized for the IBM Power Systems™ platform. After a six-month project, Bridger Pipeline LLC was confident that the solution was up to the task of closely monitoring the system’s condition and detecting leaks, giving the go-ahead for a full production environment.
Bridger Pipeline LLC and IBM are now working to deploy the AI solution on an IBM Power® Systems Accelerated Compute Server (AC922) with two IBM POWER9™ processors (for a total of 44 processor cores) and four NVIDIA Tesla V100 graphic processing units (GPUs) – expandable to six GPUs and up to 2TB RAM. The Power Systems AC922 uses next-generation NVIDIA NVLink technology to provide up to 5.6 times more bandwidth between CPUs and GPUs than would be possible with PCIe Gen3 technology used in x86-architecture servers. In practical terms, this bottleneck-eliminating interconnect, combined with the fact that POWER9 supports up to two times more simultaneous threads than competing x86 processors, delivers exceptional performance to meet the demands of deep-learning and AI workloads.
The joint team is training the solution using historical data on flow rate, pressure, volume and other metrics throughout the network, and will initially run it on a 144-mile section of pipeline to benchmark its performance against the existing system. Bridger Pipeline LLC expects to add a second IBM Power Systems AC922 server to provide a failover capability and more headroom for growth.
“We’re really excited to be pioneering this type of solution and deploying it to the rest of our network in time,” comments True. “We think it’s going to improve our capabilities tremendously.”
Moving towards zero loss pipelines
Based on the successful proof-of-concept exercise for Watson Machine Learning Accelerator on the Power Systems AC922 server, Bridger Pipeline LLC anticipates a dramatic reduction in the number of false alarms and a significant increase in sensitivity.
“The threshold for alarms is of course tightly linked to the number of false alarms, because as you increase the sensitivity of any system, you bring a larger number of events into scope,” says True. “The great thing about the Watson Machine Learning Accelerator solution is that the system is intelligent enough to filter out vast numbers of false alarms, giving us greater sensitivity while reducing the number of incidents our controllers need to investigate.”
Bridger Pipeline LLC was already very confident in the ability of its controllers to detect catastrophic failures and take mitigating actions almost instantaneously to prevent damage to the environment. The new Watson Machine Learning-based solution augments the capabilities of the controllers by giving them visibility into smaller leaks that might otherwise go undetected.
“The IBM solution on Watson Machine Learning Accelerator augments our expert controllers’ abilities, enabling them to detect leaks in one-fifth of the previous time, which is vital as we work towards our ultimate goal of zero loss from the network,” says True. “What we’ve achieved in just a short time is pretty remarkable. We believe we’re already competitive with the commercial systems we initially considered, and, unlike those solutions, ours requires no costly or time-consuming re-work if something changes in the network. And thanks to the deep-learning approach, we expect that the accuracy of the solution will continue to improve as the volume of data grows.”
With the Watson Machine Learning Accelerator solution cutting down the total number of alarms, there should also be less risk of controller fatigue or overload. Each alarm can then be analyzed faster and more comprehensively, and actual leaks addressed sooner. In the coming years, Bridger Pipeline LLC plans to build up its internal skills in AI and deep learning so that it can extend the solution to address applications such as predictive infrastructure maintenance, and the avoidance of slack flow in pipelines. The company also has plans to commercialize and offer its leak-detection solution to other oil and gas companies.
“We’re already seeing the benefits of the Watson Machine Learning Accelerator solution in reducing false positives and enabling us to identify even the smallest of potentially harmful leaks,” concludes True. “We strive to be the best stewards of the environment that we can, and we’re proud to be working with IBM to push the boundaries of AI applications in our industry.”
About Bridger Pipeline LLC
Bridger Pipeline LLC and two affiliated pipeline companies operate more than 3,500 miles of oil and gas pipelines in the Western United States. The company is based in Casper, Wyoming, USA. The family-owned companies transport of more than 450,000 bbls of oil equivalent per day. Above all, the company is committed to operating safely, serving its customers, and being a good steward of the environment.
Take the next step
To learn more about IBM Power Systems solutions for AI and deep learning, please contact your IBM representative or IBM Business Partner, or visit the following website: ibm.com/marketplace/deep-learning-platform
Founded in 2008 and headquartered in Scottsdale, Arizona, TechMileage is a United States-based custom software development business with a focus on advanced analytics services and serves the Fortune 500 as well as mid-size and large enterprises. For more information, visit http://www.techmileage.com/
Sirius is a Platinum IBM Business Partner and a national integrator of technology-based business solutions that span the data center and lines of business. For more information, visit https://www.siriuscom.com/