American Oil and Gas Reporter - November 2023 - 23

AI, Machine Learning
Solving Mysteries
Of What Lies Beneath
" In other words, this technology can
change the rules of subsurface object
and hazard identification.
"
Recently, I had the opportunity to attend the Mary Kay O'Connor
Process Safety & Risk Conference in College Station,
Tx. The Texas A&M Engineering Experiment Station is home
to the Mary Kay O'Connor Process Safety Center in association
with the Institution of Chemical Engineers. This year, the organization
hosted its 26th annual international symposium, the
crossroads for process safety where industry, academia, government
agencies and other stakeholders come together to discuss
critical research issues in process safety. Experts from across the
world gathered for this three-day event to share the latest information
on the hottest topics aimed at making the process industry
a safer place.
One unique technology I encountered was software that
leverages artificial intelligence to locate underground objects
and hazards without having to dig or pig. The presentation,
" Scalable Machine Learning: Bridging the Divide, " was delivered
by Gary Young from the Texas Transportation Institute.
According to Young, AI and machine learning are gaining
rapid acceptance in research and application to many geophysical
data processing, imaging and interpretation workflows. In
many instances, such applications can encourage crossover applications
from one industry to another: oil and gas exploration
to near-surface and vice-versa.
Young showcases an example that uses three-dimensional
geophysical methods for imaging near-surface cultural targets.
While commonly used in mining and oil and gas exploration, the
process frequently has been deemed redundant in terms of adding
value in near-surface applications, where the key objective tends
to be mapping lateral extent or relying on key " signatures " such
as diffraction hyperbolas to identify a target's location.
Reports of significantly improved resolution in inversion results
using AI and ML in oil and gas applications have opened
the possibility of using AI-based geophysical inversions for new
applications in near-surface geophysics. The prohibitively high
computational cost involved in applying the method for largescale
3-D inverse problems has been a significant bottleneck for
widespread commercial use of AI-based 3-D inversion in all
realms of geophysics.
Fundamental changes to the form of the simulated input data
used for training and the corresponding design of the architecture
of the hidden layers enable approximately O(n) (where " n "
is the number of layers in the network) reduction in the computational
complexity of the training architecture. Combined with
multiple graphics positioning units (GPUs) distributed deep
learning algorithms optimized specifically for training largescale
ML data, it significantly improves the resolution of inversion
results relative to conventional least squares imaging. At
the same time, computational efficiency improves by an order
of magnitude compared with several commonly used opensource
ML architectures and platforms.
When deployed for inversion of dense, closely spaced, high-resolution
handheld magnetometer data collected over a buried pipe
in a Texas field, the resolved 3-D geometry and location that used
the new algorithm showed better than sixfold improvement compared
with conventional 3-D least squares inversion. When applied
to an 18-fold more extensive dataset collected by a drone-based
magnetometer over a California field, the buried complex metallic
pipelike structure was resolved using little more than two days of
compute time. A similar exercise was undertaken in the Google Collab
GPU platform using state-of-the-art Google tensor flow, which
would have taken three-six months to complete, suggesting as much
as hundredfold improvement in computational efficiency.
This method was benchmarked against the Los Alamos National
Laboratory's (LANL) open-source full waveform inversion
dataset. LANL trained 24,000 seismic datasets simulated
from various 2-D velocity models using 32 P100 Tesla GPU machines
in two hours. When inferenced on 6,000 previously unseen
test models, the root mean square error (RMSE) in the
inverted normalized velocity was 0.018. The current workflow
on the same dataset achieved a comparable RMSE of 0.012 on
6,000 unseen test models after training 24,000 models in 50 minutes
using only four GPU (V100) machines, achieving a nearly
twentyfold computational efficiency improvement.
In addition to magnetic and seismic data, the method is being
developed for applications to electromagnetic and full tensor
gravity gradiometer data. Given the significant improvements in
resolution and computational efficiency, it is expected that successful
ground truth-based field trials of AI-based geophysical
data inversion have the potential to unlock several new application
areas while dramatically improving the business impact of
such applications in existing ones.
In other words, this technology can change the rules of subsurface
object and hazard identification. With its speed and computational
efficiency, it will be only a matter of time before
operators can pull up their software and analyze buried pipelines,
abandoned wells and other subsurface objects on the fly, saving
time, money and labor.
❒
JEREMY VISCOMI is a strategic
marketing expert for the oil and
gas industry. Among his many
roles, he serves as the MidContinent
regional lead for the
Petroleum Technology Transfer
Council, providing independent
operators with access to emerging
technology aimed at reducing
risk and growing profits.
NOVEMBER 2023 23
American Oil and Gas Reporter - November 2023

Table of Contents for the Digital Edition of American Oil and Gas Reporter - November 2023
