Machine learning at UND pumps up North Dakota’s oil production

From drill bit to pipeline, the Big Data work of Assistant Professor Minou Rabiei matters for N.D.’s oil and gas industry

Minou Rabiei’s research in the College of Engineering and Mines at UND is fueled by the economics of enhanced oil extraction, as well as the goals of increased safety and industry compliance. Rabiei wants to use Big Data to create a more effective and efficient industry. Photo by Mike Hess/UND Today.

As she works with algorithms on her computer at UND, Minou Rabiei exerts a few ounces of pressure with every keystroke. That’s where the miracle of leveraging begins.

For Rabiei is an assistant professor of petroleum engineering. And because she uses Big Data to figure out how to squeeze more oil out of North Dakota’s Bakken formation, Rabiei’s keyboard is one of the world’s strongest amplifiers, turning her keystrokes into the practical equivalent of kilotons of additional force.

Big Data matters in oil and gas, because the industry is comprised of so many disciplines and complex operations. Those disciplines span the entire process of resource recovery and refinement, from the geological exploration of oil and gas reserves to delivering fuel to consumers.

Each of these areas yields millions of data points every day, said Rabiei, and that number is only growing as technology advances. In the UND College of Engineering and Mines, Rabiei is one of many professors trying to rein in and harness that data to increase extraction.

Rabiei’s own specialty in processing all of that data is machine learning and artificial intelligence. She’s bringing those skills to bear on a complex industry where a 1 percent increase in recovery could result in potentially billions of dollars in new revenue.

Rabiei’s idea is that if well-trained algorithms – complex computation sets – can enhance the resource-extraction process in real time, they can also bolster safety and regulatory compliance, all of which could create better earnings for the industry. And that, in turn, would mean increased revenue for North Dakota’s economy.

“Every industry is faced with a huge amount of data, and learning and gaining the capacity to use this data provides a huge competitive advantage,” Rabiei said. “So, the more advanced and optimized the operations are in the oil and gas industry, the more advantage we can provide to the economy as a whole.”

One area of Rabiei’s machine learning research is in the highly specialized drilling process, where misinterpretations of underground conditions can cause costly delays in oil and gas production. Above, Rabiei is pictured in one of the College’s drilling simulator labs. Photo by Mike Hess/UND Today.

Decisions informed by data

To illustrate machine learning’s effectiveness, Rabiei described the responsibility of a drilling engineer to pick the best-suited drill bit for a new operation. In many cases, a producing company brings in another service provider to start drilling a well – a highly specialized task.

“Traditionally, picking the optimal drill bit is the task of an experienced drilling engineer who, based on previous knowledge from different formations they’ve worked with, knows the type of rock and the issues they might encounter,” Rabiei said.

But sometimes such a level of knowledge or experience isn’t available. What Rabiei researches is the ability of an algorithm to perform the same task, using massive amounts of previously collected data from drill sites.

“If a new engineer, for example, wants to design an operation in a new formation, he or she can use previous knowledge accumulated in a knowledgebase and integrate different parameters,” she continued. “There are so many factors that need to be considered when finding the best type of drill bit.”

Using such a knowledgebase, an algorithm brings all of the available information – current and previous – together to suggest the best type of bit to use. Rabiei said such an algorithm can use that data to help identify, and even predict, potential defects and drilling issues that might be encountered in the operation. (When it comes to the 24/7 process of mining, any stoppages or delays contribute to losses on the entire effort.)

What’s great about today’s technology is the fact that data from the drilling – such as the pressures, temperatures and fluid characteristics of the earth being bypassed – can be gathered in real time, she said. That data can then be fed into an algorithm, also in real time.

“If drill operators can be proactive in terms of predicting problem areas in the formation, before reaching that location, they could be able to change the speed of drilling, or they could change the characteristic of drilling mud,” Rabiei said. “That saves a lot of money for both the producing company and the drilling service provider.”

In this Jan. 2020 photo, Vamegh Rasouli, Continental Resources Distinguished Professor of Petroleum Engineering, stands in front of components for the new oil drilling and completion lab. Soon, the lab will be operational and capable of recreating a full-scale rig site. UND archival image.

Well-equipped for research

At UND’s College of Engineering and Mines, 10 existing labs in the Department of Petroleum Engineering include two drilling simulators, a multiphase flow and pipeline simulation lab and a virtual reality lab containing myriad scenarios.

Soon, UND will also operate the world’s largest full-scale oil drilling and completion lab. The lab’s impressive capabilities to replicate a real rig site will enable Rabiei to refine her machine learning processes in previously impractical ways.

“What we will soon be able to do is repeat different experiments in terms of drilling a type of rock from a certain formation,” Rabiei said. By testing samples from an area they’re examining (such as from the Williston Basin), researchers can acquire all of the properties and characteristics of the rock. The lab can also recreate conditions deep within the earth that a drill would encounter, which makes the data all the more valuable.

The repeatability of the “simulated” drilling processes is key to building inputs for Rabiei’s machine learning algorithms.

It’s similar to seeing a doctor, Rabiei said. You have symptoms, and you explain those symptoms to the doctor. Then the doctor, based on his or her experience and previously seen cases, can tell you what type of illness or ailment you’re likely experiencing.

Replace the doctor with an AI-capable computer, and you’ll see how the process works in petroleum engineering. And for Rabiei, training algorithms to maintain high levels of accuracy in their “diagnoses” is a crucial aspect of her work.

“That’s basically the same thing we are doing, in terms of predicting possible issues during drilling operations,” said Rabiei of her medical analogy.

As North Dakota sits atop some of the largest reserves of oil-bearing shale in the world, any amount of increased extraction from the unconventional rock can have major economic implications. Image courtesy of Tim Evanson/Creative Commons.

Work of heightened importance

As it stands, North Dakota is the second-largest oil producer in the Union and also home to one of the world’s largest reserves of oil-bearing shale. Extraction from such a rock formation is known in the industry as “unconventional,” which essentially means it’s hard to economically recover the vast amount of oil and gas that’s present.

Considering all of the North Dakota shale oil that is known, only around 8 percent reaches that “economical” threshold. More could be recovered, but right now, the cost of doing so would exceed the oil’s value.

Vamegh Rasouli

“Any technology or any type of analysis that tells us how to gain even a fraction of a percentage from that remaining 92 percent is huge, in terms of volume and profit,” said Vamegh Rasouli, Continental Resources Distinguished Professor and chair of UND’s Department of Petroleum Engineering.

That’s why the industry leaders are turning to UND for its energy research capabilities, both in the College of Engineering and Mines as well as the Energy & Environmental Research Center just east of UND’s main campus.

Through the support of the North Dakota Industrial Commission (NDIC) and North Dakota’s Oil & Gas Research Council, department faculty have the ability to fund Ph.D. students and additional research projects, many of which also involve machine learning and artificial intelligence. That research effort, in turn, directly impacts the state’s energy industry.

In March, the state committed to an additional three years of funding. That financial support also supported the establishment of the new drilling and completion lab.

Brent Brannan, director of the NDIC Oil and Gas Research Program, said UND continues to evolve as a leading research institution through the Department of Petroleum Engineering’s work, as well as its contributions to the state’s oil and gas industry.

“Partnerships between public and private entities provide a direct benefit towards the advancement of energy development in North Dakota,” Brannan told UND Today. “The innovative teaching and support for undergraduate students results in excellent job placements, while the establishment of the Ph.D. and Master’s programs have successfully contributed to numerous publications, presentations, and applied research activities.”

Said Rasouli, “this support has been a great assistance to us in terms of increasing our number of Ph.D. students, as well as our research profile. The NDIC’s heavy interest in data-driven solutions indicates the importance of this research area, in terms of the needs of North Dakota.”

Thus, Rabiei’s research area carries great importance, especially considering that UND has a Grand Challenges goal committed to the harnessing of Big Data, said Rasouli.

“Dr. Rabiei’s research is key to the department, to the College, as well as the University, because it contributes to a number of UND efforts,” he continued. “The statewide interest in her expertise only heightens that importance.”