Geotechnical Site Characterization via Deep Neural Networks: Recovering the Shear Wave Velocity Profile of Layered Soils
- Creators
- Ayoubi, Peyman
- Seylabi, Elnaz
- Asimaki, Domniki
Abstract
The mechanical property of soils is a vital part of seismic hazard analysis of a site. Such properties are obtained by either in-situ (destructive) experiments such as crosshole or downhole tests, or by non-destructive tests using surface wave inversion methods. While the latter is more favorable due to the cost-efficiency, there are challenges mostly due to computational need, non-uniqueness of inversion results, and fine-tuning parameters. In this article, we use a deep learning framework to circumvent the above-mentioned limitations to output soil mechanical properties, requiring dispersion data as input. Our trained model performs with high accuracy on the test dataset and shows satisfactory performance compared to the ensemble Kalman inversion technique. We finally propose a framework to extend the method to higher dimensions by numerically solving the wave equation in a two-dimensional medium.
Additional Information
License CC0 1.0 Universal. Submitted: October 02, 2020; Last edited: October 05, 2020. Author asserted no Conflict of Interest.Attached Files
Submitted - 1DInverNet_Neurips_2020_Preprint_v2.pdf
Files
Name | Size | Download all |
---|---|---|
md5:669a26d61bc7dc24b13cd4b0db996301
|
1.5 MB | Preview Download |
Additional details
- Eprint ID
- 105764
- Resolver ID
- CaltechAUTHORS:20201002-151457704
- Created
-
2020-10-05Created from EPrint's datestamp field
- Updated
-
2021-11-16Created from EPrint's last_modified field