Page 76 - Machine Learning for Subsurface Characterization

62 Machine learning for subsurface characterization

implemented various unsupervised clustering methods to process ultrasonic
shear waveforms dimensionally reduced using short-time Fourier transform
followed by principal component analysis. Based on displacement
discontinuity theory, each cluster label can be associated with a certain
degree of geomechanical alteration (change of stiffness) in geomaterial due
to hydraulic fracturing. In this way the data-driven solutions are infused
with physics-based factors for improved applications. For an isotropic rock,
like Tennessee sandstone, the postfracture shear waveforms (without
requiring prefracture waveforms) are shown to be effective in generating
reliable noninvasive visualization of geomechanical alterations. Use of
short-time Fourier transform followed by robust scaling and principal
component analysis ensures that various clustering methods generate
relatively similar clustering results. K-means and agglomerative clustering-
based geomechanical alteration indices are spatially well correlated to the
acoustic-emission events recorded during the hydraulic fracturing;
however, K-means clustering generates more consistent visualization.
Density-based spectral clustering of applications with noise (DBSCAN) is
found be the most ineffective clustering method for purposes of
noninvasive visualization of geomechanical alterations.

Acknowledgments
Various workflows and visualizations used in this chapter are based upon work supported by
the U.S. Department of Energy, Office of Science, Office of Basic Energy Sciences, Chemical
Sciences Geosciences, and Biosciences Division, under Award Number DE-SC-00019266.
3
We thank the Integrated Core Characterization Center (IC ) and the Unconventional Shale
Gas (USG) Consortium at the University of Oklahoma for providing us the shear
waveforms and acoustic-emission data.

Declarations
The authors declare that they have no competing interests. AC and SM developed and tested
the clustering and feature extraction methods used in the study. AC and SM developed the
workflow for the noninvasive visualization/mapping of geomechanical alterations in
geomaterials. AC generated all the figures and tables in this study. SM arranged funding
to support the development of the workflow for the noninvasive mapping. AC and SM
prepared the first complete draft of the chapter. SM wrote various conceptual topics
related to data analysis and clustering. PB developed the laboratory setup and performed
3
3
the shear-waveform measurements at the IC Lab. CR is co-PI of the IC Lab and USG
Consortium.
References
[1] O’Connell RJ, Budiansky B. Seismic velocities in dry and saturated cracked solids. J Geophys
Res 1974;79:5412–26.