A common issue with oil wells is that they can run dry even when sound-based measurements indicate there’s still oil present. A team from Penn State University utilized PSC’s flagship Bridges-2 supercomputer to introduce a time dimension to these seismic measurements and to study how oil dampens the sound traveling through it. Their initial analysis suggests that hidden rock formations in oil reserves can prevent all the oil from being extracted. They are now expanding their research to examine larger oil fields.
Why It’s Important
As oil exploration moves to more remote and deeper locations, it’s essential to drill smarter. Waste is costly, and today, being clean and efficient in oil and gas extraction is more crucial than ever.
Experts often rely on sound movement through the Earth to identify where oil deposits are located. These measurements also help estimate the size of a reserve. However, it’s not uncommon for a well to run dry after extracting only a small portion of the expected oil. Penn State’s Tieyuan Zhu, along with his students and postdoctoral fellows, aimed to understand why this occurs and to refine measurements of how much oil can realistically be produced from a well.
“We actually tested data from the North Sea. They began drilling in 2008 and estimated they could produce oil for 20 to 30 years. Unfortunately, after just two years, the well went dry. They were puzzled: where did the oil go? The main issue is the geological complexity of the reservoir.” — Tieyuan Zhu, Penn State
The team’s innovative approach, which analyzes more aspects of sound data than before, requires significant computing power and memory to handle the computations effectively. Fortunately, PSC’s NSF-funded Bridges-2 supercomputer provided the necessary resources, thanks to an allocation from the ACCESS network of NSF computing sites.
How PSC Helped
Oil doesn’t exist in pools underground; it permeates porous rock. Solid rock transmits sound better than oil-saturated rock. Experts can detect oil reserves by observing how sound is slowed down as it passes through these formations. Using seismic methods, they create 3D maps showing the locations of oil-saturated rock.
Despite these sophisticated maps, wells often produce less oil than expected. Zhu’s team suspected that some aspects of the geological picture were missing from the 3D imaging. They thought capturing images over different time periods—creating a sort of 4D animation—could provide a clearer understanding.
Additionally, the team incorporated more features from seismic data in their analysis. Instead of just focusing on how long sound waves took to cross the area, they also examined the amplitude of the signal, or how the presence of oil reduced its loudness.
This approach posed significant computational challenges. The supercomputer needed to process large amounts of data quickly while also storing parts of the problem in memory to avoid slowdowns. Bridges-2, with its more than a thousand powerful central processing units (CPUs) and ample regular memory nodes, was suited for this task. Each CPU node offers between 256 and 512 gigabytes of RAM—substantially more than a high-end gaming laptop.
“We have two postdocs and a graduate student using Bridges-2… The first phase was to parallelize our research code and make it more effective. The second phase involves applying the code to real field data. PSC provided a hundred thousand computing hours and the memory to store my field data—something we couldn’t do with our local resources.” — Tieyuan Zhu, Penn State
The team’s repeated measurements and broader analysis proved fruitful. They discovered that time-only images missed certain structures within the oil reserve. Features like a solid rock layer might not have significantly slowed down sound to be detected, yet they could block access to oil. Sometimes, the solution was simple: drill deeper to reach the oil. The scientists shared their findings in the journal Geophysics in September 2024, with more extensive results expected in April 2025.
This current report serves as a proof of concept for their method in a limited geological area of about 9 square miles. The team is now expanding their computations to cover broader areas, potentially dozens of square miles. They might also consider using Bridges-2’s extreme memory nodes, which feature 4,000 gigabytes of RAM each.
Summary: Penn State University’s research team utilized the Bridges-2 supercomputer to enhance oil extraction techniques by introducing a time dimension to seismic measurements. Their findings indicate that hidden geological features can cause wells to run dry prematurely, even when sound measurements suggest oil is available. This groundbreaking approach could lead to more efficient drilling methods, optimizing oil recovery in larger fields.
