UNIVERSITY PARK, Pa. — It is difficult for geoscientists to map sedimentary rocks' compositional and mechanical properties at high resolution, according to Yashar Mehmani, assistant professor in the John and Willie Leone Family Department of Energy and Mineral Engineering. He recently received a seed grant from the Institute for Computational and Data Sciences (ICDS) to investigate using artificial intelligence (AI) to develop a new method to model the Earth’s subsurface.
The ICDS seed grant program is designed to help Penn State scientists use the latest computational technology and cutting-edge data science techniques to deepen understanding and develop innovation across fields and disciplines. Mehmani received the grant for his proposal, "Using AI to Map Infrared Spectra to Geomechanical Properties from the Micron to Meter Scale."
"I am super excited," said Mehmani, who also is a co-funded faculty member of the Institutes of Energy and the Environment. "This seed grant is significant because the underlying idea is experimental to the point that there is a finite probability of failure. But if successful, the rewards are really high because they could potentially change how geoscientists model subsurface formations.
"What is also exciting is the promise of machine learning in this specific problem, which I have not so far formally applied in my research. The potential lies in extrapolating data from small to large and translating 'cheap but less useful' information to 'expensive but more useful' information. The speed with which this could be done opens up extraordinary possibilities," said Mehmani.
According to Mehmani, it is difficult to map sedimentary rocks' compositional and mechanical properties at high resolution because the instruments available either lack resolution or are too expensive to use on new, previously unobserved sections of a subsurface formation.
Determining the formation's mechanical properties requires drilling 100-meter-long cores of rock and then extracting smaller sample for testing. While indispensable, the approach is time-consuming, leaves gaps between measurements and must be repeated whenever a new section needs analyzing ─ even from the same formation. Mehmani proposes a new approach that would expose sedimentary rocks to infrared light and record its reflections. His team will then analyze the reflections at multiple wavelengths to understand the compositional makeup of minerals and organics within the rock. The compositional information would then be related to mechanical properties measured on lab samples using AI.
According to Mehmani, the proposed approach only needs to occur once to build the initial database for the formation. The entire process of producing the infrared spectra and mapping them to a high-resolution mechanical property could take only a few hours. This reduction of time and cost could dramatically change how subsurface formations are analyzed.
"When deployed, the AI would instantaneously translate data from a few lab samples into meter-scale information," said Mehmani. "AI is that bridge. You train it on a few small samples and when you deploy it, you get something that no instrument can measure on its own."
The use of infrared imaging builds on Mehmani's previous research, which successfully used near-infrared spectra to develop models of organic-rich shales from the Green River Formation.