Page 68 - Machine Learning for Subsurface Characterization

54 Machine learning for subsurface characterization

FIG. 2.6 Comparison of the effect of feature engineering. Noninvasive visualization of
geomechanical alterations in the postfracture Tennessee sandstone sample in the axial plane
(above) and frontal plane (below) obtained by K-means clustering of the shear-waveform dataset
transformed using STFT followed by PCA (left) and stationary statistical methods followed by
PCA (right). In Figs. 2.6–2.10 the underlying black dots represent the location of acoustic
emissions produced during the hydraulic fracturing of the sample under uniaxial stress. In
Figs. 2.6–2.10 the thin, dotted lines represent the scanning positions of the transducer
assemblies, and white dotted line represents the drilled borehole in the sample. In Figs. 2.6–2.10
the gray regions represent sections not scanned by the transducer assemblies. Hotter colors and
larger geomechanical indices indicate larger geomechanical alteration.

(dark gray in print version)] in the entire lower half of the sample. STFT-based
visualization exhibits a small region of high alteration at the bottom section of
the sample due to issues with sample preparation and sensor placement. In the
axial plane the STFT-derived GA index indicates maximum alteration [yellow
(light gray in print version) and red (dark gray in print version)] in the middle of
the sample—the same region contains highest density of acoustic-emission
hypocenters. Statistical feature-derived clustering shows diffused regions of
low alterations [pink (light gray in print version) and blue (dark gray in