Page 297 - Machine Learning for Subsurface Characterization

Classification of sonic wave Chapter 9 259

materials with and without discontinuities indicate the velocity anisotropy of
the material containing discontinuities.

4 Methodology for developing data-driven model for
the noninvasive characterization of static mechanical
discontinuities in material

We perform three major tasks in chronological order: Step 1 create thousands of
numerical models (realizations) of material containing various network of static
discontinuities (Figs. 9.12 and 9.13A); Step 2 perform fast-marching simula-
tions of compressional wavefront that starts from one source and propagates
through the material containing discontinuities to create a dataset of compres-
sional wavefront travel times measured at multiple receivers (Fig. 9.13B),
which is combined with user-assigned label that categorically represents the
overall spatial characteristics of the embedded network of static discontinuities;
and Step 3 train several data-driven classification methods (Fig. 9.13C) on the
dataset of compressional wavefront travel times with associated user-assigned
labels to learn to noninvasively characterize the material containing discontinu-
ities using only the compressional wavefront travel times measured at various
receivers/sensors.
In this study, we focus on noninvasive characterization of materials containing
embedded network of static discontinuities using limited sonic measurements.

FIG. 9.12 A realization of material containing discontinuities generated using DFN model and the
locations of 1 source and 28 sensors along the boundary of the material. FMM is used to simulate the
compressional wavefront propagation from the source to sensors through the material.