Page 217 - Machine Learning for Subsurface Characterization

Deep neural network architectures Chapter 7 187

subsurface NMR log is limited due to the financial expense and operational
challenges in its subsurface deployment. In the absence of a downhole NMR
logging tool, we propose four deep neural networks (DNNs) architectures
that process the conventional “easy-to-acquire” subsurface logs to generate
the NMR T2 distributions, which approximate the fluid-filled pore size
distributions. This chapter implements the variational autoencoder (VAE)
assisted neural network (VAE-NN), generative adversarial network (GAN)
assisted neural network (GAN-NN), long short-term memory (LSTM)
network, and variational autoencoder with a convolutional layer (VAEc)
assisted neural network (VAEc-NN) to synthesize the NMR T2 distribution
response of fluid-filled porous subsurface formations around the wellbore.
NMR log contains valuable pore- and fluid-related information that cannot be
directly estimated from other conventional logs, such as porosity, resistivity,
mineralogy, and saturation logs. Acquisition of NMR log is more expensive
and requires better well conditions as compared with other conventional logs.
During data acquisition, NMR logging tool is run at a relatively slower speed
of 2000 ft/h. Moreover, the NMR logging tool is usually run in bigger-
diameter uniform boreholes around 6.5-in. in diameter. NMR signals due to
the inherent physics have a poor signal-to-noise ratio [11]. Furthermore, the
acquisition of high-quality NMR log in subsurface borehole environment
requires highly trained wireline field engineers. Consequently, the oil and gas
companies do not deploy NMR logging tool in each well in a reservoir due to
the above mentioned financial and operational challenges involved in running
the NMR tool. In the absence of NMR logging tool, neural networks can
synthesize the entire NMR T2 distribution spanning 0.3–3000 ms along the
length of a well by processing the “easy-to-acquire” conventional logs. An
accurate synthesis of NMR T2 distribution will assist the geoscientists and
engineers to quantify the pore size distribution, permeability, and bound fluid
saturation in water-bearing and hydrocarbon-bearing reservoirs; thereby,
improving project economics and subsurface characterization capabilities.

2 Introduction to nuclear magnetic resonance (NMR)
measurements

2.1 NMR relaxation measurements
In the absence of an external magnetic field, the directions of nuclear magnetic
moment associated with the spin of hydrogen nuclei in the pore-filling fluid are
randomly oriented. In an external magnetic field, the nuclear magnetic moments
associated with spins will get aligned in specific orientations. Such ordered
nuclei, when subjected to EM radiation of the proper frequency, will absorb
energy and “spin-flip” to align themselves against the field at a higher
energy state. The energy transfer takes place at a wavelength that
corresponds to radio frequencies, and when the spin returns to its base level,