SBIR-STTR Award

Developing and Applying Deep Learning Time-Frequency Denoising Tools to Das Data to Enhance Passive Seismic Signals for Seismic Hazard Analysis and Reservoir Characterization
Award last edited on: 1/14/2023

Sponsored Program
SBIR
Awarding Agency
DOE
Total Award Amount
$249,935
Award Phase
1
Solicitation Topic Code
C54-23c
Principal Investigator
Whitney Trainor-Guittion

Company Information

Zanskar Geothermal & Minerals Inc

292 East 4075 North Street
Provo, UT 84604
   (712) 309-5306
   N/A
   www.zanskar.us
Location: Single
Congr. District: 03
County: Utah

Phase I

Contract Number: DE-SC0022733
Start Date: 6/27/2022    Completed: 4/26/2023
Phase I year
2022
Phase I Amount
$249,935
Better seismic characterization tools are needed to assess the seismic hazards of those sites before major investments are made to prevent large economic losses (and possible human life losses) that would occur if moderate to large earthquake events are induced. Fiber optic sensing for seismic monitoring (distributed acoustic sensing; DAS) is revolutionizing seismology today but despite undeniable advantages, distributed acoustic sensing remains an underutilized tool, due to hurdles associated with data management and processing. The proposer plans to adapt and optimize novel deep learning time-frequency denoising tools to DAS array data to enhance the signal-to-noise (SNR) ratios of DAS waveforms. Such denoising work is intended to improve the detectability of certain signals of interest, including local, regional and teleseismic earthquakes, microseisms, converted phases from local structures, and long-duration and emergent signals (e.g., tremor-like signals). The Phase I effort will adapt and optimize a recently developed, convolutional neural network (CNN) applied to time-frequency domain data. During the Phase I effort, the proposer will assemble a large database of a diverse range of signals of interest (e.g., local, regional and teleseismic earthquakes, microseisms, converted phases from local structures, and tremor-like signals) recorded by publicly available DAS deployments that will be used to create training, validation, and test datasets. These datasets will be used to modify, optimize and train Convoluted Neural Networks (CNN) to produce masks for each signal of interest that can be applied to raw DAS data to improve signal-to-noise ratios (SNR) without distorting waveforms. We will assess CNN performance against industry standard spectral filters and a pre-trained version of a CNN by utilizing the test DAS dataset and comparing filtered waveform properties. The Phase I effort will lead to a Phase II DAS field campaign, to be conducted at a proposed deep saline formation carbon sequestration site in the Appalachian Basin of SW Pennsylvania. Commercial and Other

Benefits:
Industry and market users in the carbon storage sector that will benefit from improved characterization include those looking to develop deep saline, depleted hydrocarbon, deep coalbeds, or young mafic rock reservoirs (which can overlap geothermal reservoirs, see CarbFix) for geologic CCS. Potential customers or beneficiaries of this development include oil and gas majors (e.g., ExxonMobil, Chevron, and Shell), several of whom are currently partnered on developing one of the largest geologic CCS sites in the world (the Gorgon Carbon Dioxide Injection Project in Australia, which is modeled to capture 3-4 million tons of CO2 per year). Significantly, the de-risking of CCS projects will contribute to the administration’s effort to achieve net-zero emissions by 2050.

Phase II

Contract Number: ----------
Start Date: 00/00/00    Completed: 00/00/00
Phase II year
----
Phase II Amount
----