Using Deep Learning Models to Characterize Subsurface Physical Parameters at Modeled Underground Chemical Explosion Sources
Description:
The characterization of underground explosion sources is important for constraining seismic source features, such as yield and seismic moment. There is little understanding, however, of how subsurface characteristics at the source, such as emplacement or ground material (e.g., basalt, tuff, salt), affect the observed far-field seismic waveforms (FFSWs) we record and analyze or the methods through which we extract source features, such as source time function inversion. Additionally, it is not well studied how nonlinear phenomena near the source are affected by varying subsurface characteristics, and in turn, how these nonlinearities influence our often-linear source models. We present work that explores how subsurface characteristics at the source affect FFSWs by leveraging a nonlinear shock physics code coupled to a linear wave propagation code. Our methodology allows us to parametrically explore subsurface characteristics and generate many synthetic seismic data sets at the local scale (distances < ~700 m). We vary source depth, chemical explosion size, and receiver locations in addition to the physical properties at the source and some nonlinear modeling parameters.
We use our synthetic labeled data to train multiple deep neural networks (DNNs) to predict different subsurface characteristics that have been found to have the greatest effect on the FFSWs, including source emplacement, ground material, yield strength, and fracture pressure, and we explore different DNN architectures to optimize results. We then leverage our trained DNNs to perform a sensitivity analysis that quantifies how the subsurface characteristics affect the FFSWs. Once developed further, these DNN models can be used to better characterize explosion yield or moment by considering predicted near-source characteristics.
SNL is managed and operated by NTESS under DOE NNSA contract DE-NA0003525.
Session: Advancements in Forensic Seismology and Explosion Monitoring [Poster Session]
Type: Poster
Date: 5/2/2024
Presentation Time: 08:00 AM (local time)
Presenting Author: Jennifer
Student Presenter: No
Invited Presentation:
Authors
Jennifer Harding Presenting Author Corresponding Author jlhardi@sandia.gov Sandia National Laboratories |
Leiph Preston lpresto@sandia.gov Sandia National Laboratories |
Mehdi Eliassi meliass@sandia.gov Sandia National Laboratories |
Scott Gauvain sjgauva@sandia.gov Sandia National Laboratories |
|
|
|
|
|
Using Deep Learning Models to Characterize Subsurface Physical Parameters at Modeled Underground Chemical Explosion Sources
Category
Advancements in Forensic Seismology and Explosion Monitoring