LA-UR-17-23271 (Accepted Manuscript)

Propagation of a Finite-Amplitude Elastic Pulse in a Bar of Berea

Sandstone: A Detailed Look at the Mechanisms of Classical
Nonlinearity, Hysteresis, and Nonequilibrium Dynamics

Remillieux, Marcel
Ten Cate, James A.
Ulrich, Timothy James II

Provided by the author(s) and the Los Alamos National Laboratory (2018-03-19).

To be published in: Journal of Geophysical Research: Solid Earth

DOI to publisher's version: 10.1002/2017JB014258

Permalink to record: http://permalink.lanl.gov/object/view?what=info:lanl-repo/lareport/LA-UR-17-23271

Disclaimer:
Approved for public release. Los Alamos National Laboratory, an affirmative action/equal opportunity employer, is operated by the Los Alamos
National Security, LLC for the National Nuclear Security Administration of the U.S. Department of Energy under contract DE-AC52-06NA25396.
Los Alamos National Laboratory strongly supports academic freedom and a researcher's right to publish; as an institution, however, the
Laboratory does not endorse the viewpoint of a publication or guarantee its technical correctness.
 Confidential manuscript submitted to replace this text with name of AGU journal

Propagation of a finite-amplitude elastic pulse in a bar of Berea sandstone: A

detailed look at the mechanisms of classical nonlinearity, hysteresis, and
nonequilibrium dynamics

Marcel C. Remillieux1*, T. J. Ulrich2, Harvey E. Goodman3, and James A. Ten Cate1

1
Geophysics Group (EES-17), Los Alamos National Laboratory, Los Alamos, NM 87545,
USA.
2
Detonation Science and Technology Group (Q-6), Los Alamos National Laboratory, Los
Alamos, NM 87545, USA.
3
Chevron Energy Technology Company, Houston, TX, USA.

*
Corresponding author: Marcel C. Remillieux (mcr1@lanl.gov)

Key Points:
 The propagation of an elastic pulse is used to decouple the effects of classical
nonlinear elasticity and nonequilibrium dynamics.
 The 1D wave equation with classical nonlinearity and attenuation captures the
spectral content of the pulse up to the second harmonic.
 Nonequilibrium dynamics is quantified by tracking the arrival time of an elastic pulse
as a function of its amplitude.

This article has been accepted for publication and undergone full peer review but has not
been through the copyediting, typesetting, pagination and proofreading process which may
lead to differences between this version and the Version of Record. Please cite this article as
doi: 10.1002/2017JB014258
© 2017 American Geophysical Union. All rights reserved.
Abstract
We study the propagation of a finite-amplitude elastic pulse in a long thin bar of Berea
sandstone. In previous work, this type of experiment has been conducted to quantify classical
nonlinearity, based on the amplitude growth of the second harmonic as a function of
propagation distance. To greatly expand on that early work, a non-contact scanning 3D laser
Doppler vibrometer was used to track the evolution of the axial component of the particle
velocity over the entire surface of the bar as functions of the propagation distance and source
amplitude. With these new measurements, the combined effects of classical nonlinearity,
hysteresis, and nonequilibrium dynamics have all been measured simultaneously. We show
that the numerical resolution of the 1D wave equation with terms for classical nonlinearity
and attenuation accurately captures the spectral features of the waves up to the second
harmonic. However, for higher harmonics the spectral content is shown to be strongly
influenced by hysteresis. This work also shows data which not only quantifies classical
nonlinearity but also the nonequilibrium dynamics based on the relative change in the arrival
time of the elastic pulse as a function of strain and distance from the source. Finally, a
comparison is made to a resonant bar measurement, a reference experiment used to quantify
nonequilibrium dynamics, based on the relative shift of the resonance frequencies as a
function of the maximum dynamic strain in the sample.

1 Introduction
From a mechanical point of view, sedimentary rocks and man-made materials like concrete
are complicated solids. They may be described as a disordered network of mesoscopic-sized
“hard” elements (e.g., grains with characteristic lengths ranging from tens to hundreds of
microns) cemented together by a “soft” bond system. Such systems belong to a wider class of
materials referred to as Nonlinear Mesoscopic Elastic Materials (NMEMs) [Guyer and
Johnson, 1999]. The microscopic-sized imperfections at the interfaces between the “hard”
and “soft” subsystems are believed to be responsible for several peculiar properties related to
nonlinear and nonequilibrium dynamics, including the dependence of elastic parameters and
attenuation on strain amplitude, slow dynamics, and hysteresis with end-point memory
[Winkler et al., 1979; TenCate et al. 2000; Ostrovsky and Johnson, 2001]. Understanding and
predicting these properties, especially the cementation (soft bond system) is at the heart of
numerous applications including nuclear waste repository, oil and gas exploration, and
monitoring of infrastructure integrity.

Increasingly more elaborate techniques have emerged and evolved over the last 20 years to
observe and quantify the effects of nonlinear and nonequilibrium dynamics in these materials.
Nonlinear elasticity, in the classical sense, is usually modeled by an expansion of the elastic
energy as a power series with respect to the strain tensor [Murnaghan, 1937; Landau and
Lifschitz, 1986]. Practically, in a medium with such nonlinearity, a sinusoidal elastic wave
experiences distortions, eventually leading to shock formation if damping is sufficiently
small. In the frequency domain, harmonic generation is observed as some of the energy
contained at the source frequency is transferred to other frequencies (i.e., higher harmonics).
As part of this process, the wave oscillates around the same equilibrium state as observed in
the sample before the source was activated. In the context of elastic wave propagation,
classical nonlinearity was studied experimentally [Meegan et al., 1993; TenCate et al., 1996]
and theoretically [Thurston and Shapiro, 1967; McCall, 1994; Van Den Abeele, 1996] in the
early studies on the dynamics of NMEMs.

© 2017 American Geophysical Union. All rights reserved.

Quasi-static stress-strain measurements on rocks and other NMEMs have been made since
the early 1900s [Adams and Coker, 1906]. These measurements not only show that rocks are
nonlinear but they are also hysteretic. Repeated looping up and down in stress-strain space
results in a curved, banana-shaped path [Zoback and Byerlee, 1975; Holcomb, 1981].
Attempts to model this quasi-static hysteresis in NMEMs were going on simultaneously but
independently as the attempts to model pulse propagation in similar samples [Van Den
Abeele et al., 1997; Gusev et al., 1997; Van Den Abeele et al., 2000; Gusev, 2000; Aleshin et
al., 2004]. Practically, a sinusoidal elastic wave propagating in a medium with pure hysteretic
nonlinearity progressively evolves into a triangular wave. In the frequency domain, such
evolution is characterized by the growth of the odd harmonics, with a quadratic dependence
with respect to the amplitude at the fundamental frequency. The effects of classical and
hysteretic nonlinearity on a sinusoidal waveform are summarized in Figure 1.

Besides hysteresis effects, the classical theory of nonlinearity cannot account for the other
peculiar wave phenomena observed in NMEMs. The first repeatable observations showing
nonclassical effects were made in resonant bar experiments [Johnson et al., 1996; Ten Cate
and Shankland, 1996; Remillieux et al., 2016], ultimately leading to a quantification
technique known as nonlinear resonant ultrasound spectroscopy (NRUS). In these
experiments, a long thin bar with free boundary conditions, which is representative of a 1D
unconstrained system, is excited with a harmonic signal at one end while the elastic response
is recorded at the opposite end. The source signal sweeps a frequency range around one
resonance frequency of the bar with increasing source amplitudes and the variation of the
resonance frequency can be tracked as a function of the maximum strain in the sample. In a
linear elastic material, the resonance frequency is not strain dependent and the resonant peak
does not change with increasing drive amplitude. In NMEMs, the resonance frequency starts
decreasing as a function of strain, and very noticeably so when the strain goes above a certain
level, typically near 10-6 for sandstones at ambient conditions.

Broadly speaking, the drop in resonance frequency indicates material softening with
increasing driving strain. However, the dependence of the resonance frequency on strain is
not trivial [TenCate et al., 2004; Pasqualini et al., 2007]. At very small strains (e.g., <10-7 for
a sandstone at ambient conditions) the material exhibits an instantaneously reversible
decrease of the resonance frequency, behavior dominated by classical nonlinearity. At higher
strains (e.g., >10-6 for a sandstone at ambient conditions), the decrease of the resonance
frequency is governed mostly by a process related to the nonequilibrium dynamics of the
evolving strain fields within the rock. In summary, under a continuous harmonic excitation,
the elastic moduli begin to soften and eventually reach a metastable steady state, i.e. the
elasticity of the material reaches a new but unstable equilibrium state that is tuned to the
dynamic strain amplitude induced by the excitation. Conversely, if elastic energy is no longer
injected into the system, the material recovers back to its original elastic properties, with a
rate generally proportional to the logarithm of time [TenCate et al., 2000a, 2000b]. Notably,
conditioning times are much faster than recovery times [TenCate et al., 2000a; TenCate,
2011] and this temporal asymmetry gives rise to all sorts of interesting observations. For
instance, up and down frequency sweeps around the resonance frequency do not lead to the
same resonance curves when the sweeping rates are identical but relatively fast. On the other
hand, when the sweeping rates are slow enough, the upward and downward resonance curves
merge. In between the two strain regimes described above, classical nonlinearity and
nonequilibrium dynamics cannot be disentangled from each other in the context of a
resonance experiment where the effects from these mechanisms are smeared over many
dynamic cycles.

© 2017 American Geophysical Union. All rights reserved.

Dynamic acoustoelastic testing (DAET) is a technique developed more recently, designed to
delineate the mechanisms of nonlinear and nonequilibrium dynamics with high sensitivity,
regardless of the strain regime. This technique is based on a pump and probe scheme in
which high-frequency (HF) pulses are used to probe the various phases of a mechanical
system set by a low-frequency (LF) pump [Renaud et al., 2011; Rivière et al., 2013].
Typically, the sample is a long thin bar and the dynamic pump is obtained by exciting the
sample near a resonance mode of vibration, thus approaching the conditions used for NRUS.
The probe is obtained by tracking the relative change in the arrival times of HF pulses
traveling across the shortest dimension of the bar, with the assumption that during the short
travel of these pulses, the strain field induced by the pump is varying so slowly that it can be
assumed to be constant. Through a careful synchronization of the HF pulse emissions with
the pump signal, a stroboscopic effect can be obtained to capture the nonlinear and
nonequilibrium dynamics of the material over a full cycle of the LF pump [Lott et al., 2017].
The analysis of the relative change of the arrival times of the HF pulses as a function of the
LF strain induced by the pump at the probe location gives a complete representation of
classical nonlinearity, with high sensitivity up to at least the second-order terms of
nonlinearity, hysteresis, and nonequilibrium dynamics in the conditioning and recovery
phases [Rivière et al., 2013]. In Figure 2 of their paper (but also in Figure 3 of Renaud et al.
[2012]), most of the drop in elasticity experienced by the rock and measured locally by the
HF probes during the conditioning phase appears to follow instantaneously the envelope of
the LF pump: a process that could be referred to as “fast dynamics”. When the pump reaches
a steady state (between 0.05 and 0.1s in Rivière et al. [2013]), the material continues to soften
over ten of milliseconds: a process that could be referred to as “slow dynamics”. Likewise,
when the pump is switched off and the material recovers back to its original properties, “fast”
and “slow” dynamics can be observed clearly in the time series. These findings are consistent
with those of Ten Cate et al. [2000a]. Another result of interest is found in the study of
Rivière et al. [2015] where the link and proportionality between the material softening
observed in the conditioning phase, the size of the hysteresis, and the frequency shift
observed in the nonlinear resonance experiments was clearly established. This result was later
confirmed by an independent set of experiments where it was demonstrated that, given
appropriate corrections to account for the fact that the pump and the probe are in orthogonal
directions, the material softening obtained from DAET in the conditioning phase is
equivalent to that obtained from NRUS [Lott et al., 2016]. Last, Lott et al. [2017]
demonstrated experimentally that there are additional subtleties to consider that are related to
the type of strain field induced by the LF pump (e.g., axial motion versus torsional motion)
and the type of waves used as HF probes (e.g., P-waves and S-waves with various
polarizations).

Up to this point, the modeling of hysteresis and nonequilibrium dynamics was not addressed.
Between the first theoretical description of hysteresis in NMEMs through the Preisach-
Mayergoyz formalism by Guyer et al. [1995] and one of the most recent description
developed by Pecorari [2015], there has been a considerable number of assumptions and
models introduced with most reported and described by Guyer and Johnson [2009]. Likewise,
nonequilibrium dynamics has been described phenomenologically using many models,
including the “soft-ratchet” model proposed originally by Vakhnenko et al. [2004] and
recently improved by Berjamin et al. [2017]. It is also worth mentioning the damage
viscoelastic model used recently by Lyakhovsky et al. [2009] to simulate nonlinear resonance
experiments. Slow dynamics is not included in this model though and it is not clear if the
effects on the wave propagation that are summarized in Figure 1 can be captured. To the

© 2017 American Geophysical Union. All rights reserved.

authors’ knowledge, a unified model capable of capturing simultaneously nonlinear and
nonequilibrium effects observed in various types of experiments does not exist.

In this paper, we propose to revisit the experiment conducted by TenCate et al. [1996], this
time taking data over many more points on the surface, without contact, using a scanning 3D
laser vibrometer as a receiver. With this extensive high fidelity data, we will actually be able
to isolate and determine the relative contributions of the mechanisms of classical
nonlinearity, hysteresis, and nonequilibrium dynamics in a pulse propagation experiment.
This paper is organized as follows. Section 2 gives the necessary theoretical background on
classical nonlinear elasticity, derived from the 1D elasticity theory. Quantification of classical
nonlinearity, without and with accounting for damping in the material, is given in Section 3.
In this section, the elastic wave propagation with classical nonlinearity and attenuation is
resolved numerically and compared to experimental data to show that classical nonlinearity is
not sufficient to model the full elastic behavior of the material. Section 4 examines and
quantify nonequilibrium dynamics in the context of this elastic pulse propagation experiment.
Findings are compared to the resonant bar experiment, which is how nonequilibrium
dynamics has been traditionally quantified. Section 5 concludes.

2 Theoretical background
In the 1D theory of classical nonlinear elasticity, the stress (σ) can be expressed as a power
series with respect to strain (ε) as,

  M 0 1     2  ... (1)

Where M0 is the linear elastic modulus, β and δ are, respectively, the coefficients of first- and
second-order nonlinearity.

In the absence of energy dissipation and external forces, the propagation of an elastic wave in
an infinite 1D medium with first-order nonlinearity (β coefficient only) is governed by the
following equation [McCall, 1994],

 2u x, t  2  2u x, t  2   u  x, t  
 
2

 c   c    (2)
t 2 x 2 x  x  

where u is the displacement and c is the wave speed. Note that geometric nonlinearity is
ignored in this equation as it is assumed that material nonlinearity is dominant. In the
frequency domain, this results in the generation of harmonics. The nonlinear coefficient β can
be retrieved from a measurement of the displacement amplitude U2 at the second harmonic
generated at a distance x from a pure tone source signal as [Meegan et al., 1993],

2U 2 c 2
 (3)
U12 2 x

where ω is the angular frequency (ω = 2πf) and U1 is the displacement amplitude at the
fundamental frequency. In experiments, we often use acceleration data measured with
accelerometers or particle-velocity data measured with laser vibrometers. In the frequency
domain, the magnitude of the particle velocity V is related to the magnitude of the

© 2017 American Geophysical Union. All rights reserved.

displacement U as, U = V/ω. Eq. (3) can then be adapted to particle velocity measurements
using a factor ω,

2V2c 2
 (4)
V12x

where V1 and V2 are now the magnitudes of the FFT of the particle velocity at the
fundamental frequency and at the second harmonic, respectively.

Energy dissipation can easily be introduced in Eq. (2) by adding a term used in Burger’s
equation [Hamilton and Blackstock, 1998],

 2u x, t   2  u x, t   2  2u x, t  2   u  x, t  

 
2

b 2  c  c    (5)
t 2
x  t  x 2
x  x  

In this equation, b is a damping coefficient expressed as,

c2
b (6)
Q

where Q is the quality factor (i.e., inverse of loss factor).

The amplitude of a monochromatic wave is attenuated with distance as,

V x   V0 exp  x  (7)

where γ is the attenuation coefficient. This attenuation coefficient should be measured as a

function of frequency, which we did not do in the experiments reported in this paper. The
alternative is to use the damping model proposed in Eqs. (5) and (6). Using this model, the
attenuation coefficient can be expressed as,

b 2
 3 (8)
2c

Nonlinear elasticity and damping are competing effects in the context wave propagation.
While nonlinear elasticity tends to generate harmonics, damping will attenuate these
harmonics since attenuation increases with frequency. Such competition can be quantified by
the Gol’dberg number [Gol’dberg, 1957; Hamilton and Blackstock, 1998] which is a
dimensionless parameter measuring the strength of the nonlinearity relative to that of
dissipation. More specifically, it is the ratio of the absorption length (inverse of absorption
coefficient) for a small-amplitude, linear signal at the fundamental frequency to the shock
formation distance and can be expressed using the parameters defined previously as,

V1
 (9)
 c2

© 2017 American Geophysical Union. All rights reserved.

Essentially, if this number is much smaller (resp. larger) than 1, the propagation is dominated
by damping (resp. nonlinearity). When this number is close to one, there is a strong
competition between nonlinearity and damping during the propagation.

3 Quantifying and modeling classical nonlinearity

3.1 Experimental setup

The setup used in the pulse propagation experiments is shown in Figure 2. Experiments were
conducted on a sample of Berea sandstone (Cleveland Quarries, Amherst, Ohio) with a length
of 1794 mm (70.63 in) and a diameter of 39.6 mm (1.56 in), in fact the same sample used by
TenCate et al. [1996]. The sample rested on foam to approach free (unconstrained) boundary
conditions. Elastic waves were generated in the sample using a PZT-5A (Navy type II)
piezoelectric disk with a diameter of 45 mm (1.77 in) and a thickness of 10 mm (0.394 in).
The transducer and an inertial backload were epoxied onto one flat end of the sample. Figure
3 shows the time history and magnitude of the FFT of the normalized source signal used in
the experiments. The impulsive signal initially consisted of a sinusoidal wave with a
frequency of 22.4 kHz tapered at the start and end by half Gaussian windows. The output
signal from the signal generation card (internal arbitrary waveform generator in Polytec PSV-
3D-500 scanning laser vibrometer) was amplified 20 times by a voltage amplifier (TEGAM
2350) before being used as input signal to the transducer. The propagation of the elastic pulse
was recorded at 119 points along a line, on the surface of the sample, with a regular spacing
of approximately 5 mm between the points, starting at 10 mm from the source and extending
600 mm from the source. The Cartesian components of the particle velocity were recorded on
the surface of the sample with a Polytec PSV-3D-500 scanning laser vibrometer. Data were
acquired with a sampling frequency of 2.56 MHz. At each scanning point, waveforms were
averaged 500 times to increase the signal to noise ratio. Each signal was acquired for a
duration of 6.4 ms (16384 samples): response to intermittent sine bursts with approximate
duration of 0.6 ms and intervals of 5.8 ms. Note that the short duration between acquisitions
leads to conditioning effects. However, this experiment focuses on the quantification of
classical nonlinearity and conditioning effects will be ignored.

3.2 Experimental results and data analysis

Figure 4 shows three selected time histories and magnitudes of the FFT of the axial
component of the particle velocities for a source signal with a peak amplitude of 16Vpp
(before amplification) at various distances from the source. The time series have been
truncated to show only the direct pulse; the reflection from the far end, opposite to the source,
is also recorded but does not overlap with the direct pulse. The time series have been
synchronized so that they can be plotted on the same time scale and the calculation of the
FFT is not affected by the arrival time of the pulse at a measurement point. A full movie of
the pulse propagation (both axial and radial components) recorded over the 119 measurement
points is provided in the supplementary material. The first waveform and corresponding FFT
show that the vibrational response measured at 10 mm from the source is similar to the source
signal, with some minor differences caused by the frequency response of the piezoelectric
transducer – even though the transducer is operated well away from its first resonance
frequency, its frequency response is not flat and modulates the source signal. More
importantly, the vibrational response at this location does not exhibit a significant harmonic
content, which attests to the linearity of the source and quality of the bond between the source

© 2017 American Geophysical Union. All rights reserved.

and the sample. The next two sets of plots show how the sinusoidal waveform distorts with
propagation distance, exhibiting growth of the second harmonic (i.e., peaks observed at 44.8
kHz) and increasing higher frequency content (i.e., between 60 and 80 kHz). The evolution of
the waveform with propagation distance is consistent with analytical predictions on the
effects of classical nonlinearity and hysteretic nonlinearity: the β term in the classical theory
of nonlinearity in Eq. (1) induces a steepening of the waveform (eventually leading to an N-
wave in the absence of damping) whereas hysteretic nonlinearity induces a triangulation of
the waveform, as shown in Figure 1. Note that, TenCate et al. [1996] used a higher-order
term of classical nonlinearity, i.e., δ term in Eq. (1), to model the observed triangulation of
the waveform. At relatively small amplitudes, the distortion produced by this term results in
what appears to be a triangulation. However, as the amplitude increases, the original
sinusoidal waveform is bowed and not triangulated [Kadish et al., 1996]. With this term, the
amplitude at the third harmonic (3f0) depends on the cube of the amplitude at the fundamental
frequency (f0) whereas, as will be shown in Section 4, we observe a quadratic dependence, as
expected for hysteretic nonlinearity [Van Den Abeele et al, 2000]. In 1996, theories of
hysteretic nonlinearity and nonequilibrium dynamics in geomaterials (and experimental
evidence supporting these theories) were just emerging and could not be used in the analysis
so it seemed natural to use the classical description of nonlinearity to model the experimental
observations.

Classical nonlinearity was quantified by reducing the spectral data at all 119 measurement
points with the expression given in Eq. (4). Figure 5a shows the variation of 2V2c2 as a
function of V12ωx, based on a wave speed of 2010 m/s. First, note that there is an offset at
V12ωx = 0. This offset is not due to the nonlinearity of the system but to the finite duration of
the pulse (i.e., bandwidth). The term 2V2c2 is calculated from the spectrum amplitude at 44.8
kHz (second-harmonic amplitude) at various distances from the source. For the source signal
(see Figure 3), which is a linear signal, the amplitude at the second harmonic (i.e., at 44.8
kHz) is two orders of magnitude smaller than the amplitude at the fundamental frequency but
is not zero. The same applies to the signals recorded on the sample very close to the source.
For this analysis, what is of interest is not the offset at V12ωx = 0 but the growth of the second
harmonic amplitude with propagation distance, which is quantified by the slope of the data
reported in Figure 5a. The slope of the linear fit through this data set is β = 370. Experiments
conducted years earlier on the same sample by TenCate et al. [1996] led to β = 400. This
close agreement is remarkable given that the source, source signals, and receivers were
different in the two experiments. In the work of TenCate et al. [1996], the elastic pulse
generated by the source was centered at 12.4 kHz and the vibrational signals were recorded
by a small number of accelerometers along the pulse propagation. This agreement also
indicates that the material properties of the sample have not changed significantly over the
long time separating the two experiments. In both studies, however, the effects of damping
are ignored in the quantification of classical nonlinearity.

It is possible to correct the measured spectra by the amount of damping in the system using
the model given in Eqs. (5)-(8). Figure 5b shows the variation of 2V2c2 as a function of V12ωx
for the 119 scanning points, where the amplitudes at the fundamental frequency and second
harmonic have been corrected for damping based on a quality factor Q = 60. The quality
factor was estimated from the width (full width at half maximum) of the resonance curve of a
purely longitudinal mode near 20 kHz in the current experiment. The slope of the linear fit of
this corrected data is β = 660.

© 2017 American Geophysical Union. All rights reserved.

Last, it is worth mentioning the high variability of the data around the linear fit for V12ωx > 5
in Fig. 5(a) and V12ωx > 10 in Fig. 5(b). In this experiment, the Gol’dberg number is Γ = 0.6
(using c = 2010 m/s, V1 = 15 mm/s near the source, f0 = 22.4 kHz, Q = 60, and β = 660). This
value is close to 1 and suggest that the competition between nonlinearity and damping is
strong, which results in data variability after a certain distance. Such variability would likely
be reduced if the experiment had been conducted in a sample with much smaller damping
(e.g., metallic component with distributed damage).

Up to this point, dispersion and its effects on the wave propagation was not discussed. The
movie of the wave propagation provided in the supplementary material shows that double
peaks appear and disappear in the waveform along the propagation path. It seems that not all
frequency components are traveling at the same speed. This is mostly likely caused by
dispersion, which affects the wave propagation in the frequency range of the third harmonic
and above. This is also why the amplitude spectra do not show a clear peak around the third
harmonic. To mitigate the dispersion effects, the experiment described above was repeated
but this time using a source signal with a lower frequency, namely 15.7 kHz. The time history
and corresponding magnitude of the FFT of the axial component of the particle velocity
measured at 590 mm from the source, for this source frequency, are shown in Figure 6. The
amplitude spectrum now exhibits a clear peak at the third harmonic. Also of interest is the
fact that the amplitude at the third harmonic grows faster than the amplitude at the second
harmonic. The piezoelectric transducer is less efficient at 15.7 kHz than at 22.4 kHz and
consequently the amplitude of the elastic wave generated in the sample using the same input
voltage is about twice as small. Because of the smaller frequency and smaller wave amplitude
in this experiment, the Gol’berg number decreased to Γ = 0.2 (using c = 2010 m/s, V1 = 7.5
mm/s near the source, f0 = 15.7 kHz, Q = 60, and β = 660) and damping tends to dominate the
propagation. The elastic wave does not have a sufficient amplitude for efficient second-
harmonic generation, which would indicate that classical nonlinearity has been activated, but
it is sufficient to activate hysteretic nonlinearity. We could then study the effects of amplitude
and frequency on the activation of the various mechanisms but this is beyond the scope of
this paper. As a more modest achievement, we repeated the experiment at 17.1 kHz. The
axial component of the particle velocity collected at 590 mm from the source at this
frequency is shown in Figure 7. The peak at the third harmonic is not as clear as previously
because dispersion starts to play a role in the wave propagation but the amplitude at the
second harmonic is larger than previously thanks to the larger source amplitude and higher
frequency.

3.3 Modeling the 1D wave propagation with classical nonlinearity and damping
The simple 1D elastic wave equation with damping given in Eq. (5) is now used to model the
experiments to see how well it fits the data. We use the following parameters in the model: a
wave speed c = 2010 m/s, a damping coefficient b = 0.525 (based on a quality factor Q = 60
and a center frequency of the impulsive source signal f = 22.4 kHz), and a nonlinear
parameter β = 660. The wave equation is projected onto a finite-element space using the
“Mathematics, General Form PDE” module of the software package COMSOL Multiphysics
5.2a. The geometry consists of a line with a length of 1794 mm. One boundary is imposed a
flux condition (source) while the opposite boundary is free and so imposed a Neumann
condition, i.e., ux, t  x  0 . The impulsive source signal is the measured signal depicted in
Figure 4 for a receiver taken at 10 mm from the source. This signal is normalized to have a
peak amplitude of 40.5. This approach allows us to take into account the response of the
transducer in the 1D model. The computational domain is discretized into Lagrange quadratic

© 2017 American Geophysical Union. All rights reserved.

elements with a uniform length of 3 mm, which ensures at least 6 elements per wavelength up
to 110 kHz. The solution is integrated in time with a time step of 100 ns using the
Generalized-α method [Chung and Hulbert, 1993], which is a fully implicit, unconditionally
stable, and second-order accurate method with control over the dissipation of spurious high-
frequency modes. The time step is small enough to satisfy the Courant–Friedrichs–Lewy
stability condition.

Figure 8 (left) shows 6 typical measured and simulated time histories of the axial component
of the particle velocity at various positions along the sample. The agreement between the
simulations and the experiments is excellent at all points. The model captures the arrival
times, amplitudes, and overall shapes of the waveforms observed in the experiments, for both
the direct and reflected pulses. Some small discrepancies are observed and expected in the
later portion of the reflected pulses at some positions because the model ignores the presence
of the source and backload, and resulting interaction with the reflected pulse. The time series
depicted in Figure 8 (left) are now truncated and synchronized to retain only the direct, first
arrival pulses. The measured and simulated magnitudes of the FFT of the direct pulses are
shown in Figure 8 (right). The FFTs again show just how well the simple model works; the
growth of the second harmonic is very well captured. In fact, after correcting the amplitudes
at the fundamental frequency and second harmonic for damping, the variation of 2V2c2 as a
function of V12ωx for the 6 simulated signals are aligned perfectly with the linear fit shown in
Figure 5b, as expected. However, it is quite clear that very soon (~ 65 mm), the model fails to
capture the 3rd harmonic and higher frequency behavior between 60 and 80 kHz. This is
expected this the model does not include terms related to hysteretic nonlinearity and does not
include the effects of dispersion.

Practically, at 22.4 kHz, the dynamic behavior of the rock is dominated by classical
nonlinearity and attenuation while hysteresis and other nonequilibrium effects are far less
important. As shown in the previous section, this is not necessarily true at lower frequencies
where the source amplitude is smaller. With the quality of the data taken in these
experiments, it is instructive to see at what point the effects of nonequilibrium dynamics and
hysteresis do start to become apparent.

4 Quantifying nonequilibrium dynamics

4.1 Experimental setup

An alternative to the experiment reported in the previous section is to monitor the evolution
of the waveform at a fixed position on the sample for various source amplitudes. In essence,
this is a similar experiment to that reported by Cabaret et al. [2015] where the propagation of
torsional pulses is studied in a 1D chain of 70 beads. Disordered bead packs and granular
lattices serve as simplified paradigms for understanding some of the mechanisms responsible
for nonlinear and nonequilibrium dynamics. In their study, the physical mechanism at stake is
the friction between the beads. Consequently, the wave propagation is mostly affected by
hysteretic nonlinearity and is very well captured by a model based on the Preisach-
Mayergoyz formalism. Here, we focus on the propagation of a longitudinal elastic wave in a
disordered consolidated granular material. The nonlinear and nonequilibrium effects involved
in the wave propagation in such a medium are not only produced by friction between the
constituents but also possibly by clapping and other microscopic interactions.

© 2017 American Geophysical Union. All rights reserved.

The same experimental setup and source signal (see Figures 2 and 3) as in the previous
section are used for this experiment. Waveforms were also averaged 500 times but the
duration between acquisitions was set to 320 ms to allow the sample to start recovering back
to its original elastic properties between each average. Figure 9 shows the time histories of
the axial component of the particle velocity measured at 500 mm from the source for 20
source amplitudes ranging from 1 to 20 Vpp in steps of 1Vpp (before amplification). The
original sinusoidal waveform observed at the lowest amplitudes progressively evolves into a
triangular wave, as a result of hysteresis, with additional distortion caused by classical
nonlinearity and dispersion, as discussed in the previous section. Most notably, the
waveforms experience a significant delay in arrival time as the source amplitude increases.
This delay is observed not only at the extrema (peaks and troughs) but also at the zero-
crossings. Classical nonlinearity would not induce any delay at the zero-crossings (where the
strain is equal to zero) because it produces an instantaneous variation of the modulus without
time constants involved. This delay is a direct consequence of nonequilibrium dynamics,
specifically the strain-induced material softening experienced during the conditioning phase.
Note that, because of the material softening and the corresponding decrease of the wave
speed from the source to the measurement point, the pulse is stretched. This stretch is
quantified by monitoring the fundamental frequency f0, which decreases by 0.53% from the
smallest to the largest source amplitude.

The data shown in Figure 9 can be further reduced to quantify nonequilibrium dynamics, by
estimating the time delay as a function of strain amplitude at the zero-crossings of the
waveform, as in the work of Cabaret et al. [2015]. The waveform measured at the lowest
source amplitude is used as a reference. By definition, at the zero-crossings, the strain
amplitude is equal to 0. For the analysis, the strain amplitude experienced by the sample at
the strain extremum (peak or trough) preceding (1/4 of cycle before) the zero-crossing of
interest is considered. A total of 28 sets of zero-crossing points are considered. In each set, 20
source amplitudes are used for the analysis, with the lowest source amplitude used as a
reference and the higher amplitudes compared to this reference.

The relative time delay between the signals also provides the relative change in the
longitudinal wave speed, Δci/0/c0, where the superscript “i” denotes the ith source amplitude
and “0” the lowest source amplitude for which the sample is assumed to behave linearly.
Finally, at the perturbation level, the relative change in the Young’s modulus E (the modulus
involved in the propagation of a longitudinal wave in a long thin bar) is related to the relative
change in the speed of the longitudinal wave as,

E i 0 E0  2 ci 0 c0 (10)

The relative change in the elastic modulus over the propagation path of the waveform can be
tracked as a function of the maximum strain amplitude at the measurement point. Along the
axial direction (x-direction), the strain component of interest can be expressed as,

vx
 xx  (11)
c

© 2017 American Geophysical Union. All rights reserved.

Reduced data from Figure 9 are shown in Figure 10. Three distinct regimes can be identified
in the waveform: i) the earlier portion of the waveform, where the dynamic strain at the
measurement point transitions from zero to steady state; ii) the center portion of the
waveform, where the absolute value of the peak amplitude of the dynamic strain is quasi-
constant (i.e., steady state); iii) the later portion of the waveform where the dynamic strain at
the measurement point transitions from steady state back to zero, i.e., a path similar to the
earlier portion of the waveform but in reverse. In the earlier portion of the waveform, the
elastic modulus decreases progressively from its undisturbed equilibrium value to a value
4.1% smaller. The decrease of the elastic modulus is an instantaneous effect produced by the
hysteresis and referred to as “fast dynamics” in the introduction. It was already evidenced in
DAET experiments, e.g. [Rivière et al., 2013], in the propagation of a torsional elastic pulse
in a 1D granular chain [Cabaret et al., 2015], and in the interaction between two elastic waves
in a rock sample [TenCate et al., 2016]. In the center portion of the waveform, the evolution
of the material softening with the strain amplitude is essentially independent of the wave
cycle selected for the analysis. In other words, the pulse is not long enough to observe the
effects of “slow dynamics” in the conditioning phase. In the conditioning phase, the elasticity
of the material reaches a new (but unstable) equilibrium state where the modulus is 4.9%
smaller than in the undisturbed (stable) equilibrium. Note that, in this steady-state regime, the
dependence of material softening seems to intensify once the strain reaches a value of 4 to 5
microstrains. In the later portion of the waveform, the evolution of the elastic modulus
follows a different path from that observed in the earlier portion: recovery of the undisturbed
equilibrium elastic modulus occurs at a much slower rate than material softening. This result
from a pulse propagation experiment is in excellent qualitative agreement with the results
obtained in resonant bar experiments by TenCate et al. (2000a); the two experiments will be
compared quantitatively in the next section. The experiment was repeated at other positions
on the sample. Raw and reduced data at 650 mm from the source are shown in Figures 11 and
12. These data lead to the same conclusions, both qualitatively and quantitatively, as those
reached at 500 mm from the source.

The same experiment was repeated at 15.7 kHz, to measure the delays at the zero crossings as
a function of the strain amplitude but also to quantify the growth of the odd-harmonic
amplitudes with respect to the amplitude at the fundamental frequency. The time histories of
the axial component of the particle velocity for 20 source amplitudes at 500 mm from the
source are shown in Figure 13. The corresponding reduced data is shown in Figure 14. Unlike
previously, in this experiment where the strain amplitude is much smaller throughout the
pulse duration, there is no clear distinction between the early, steady-state, and late portions
of the waveform. In other words, the effects of slow dynamics are not visible. The peak strain
amplitude remains below 4 microstrains and the material softens by up to 1.8%. It is likely
that strain needs to reach a threshold amplitude for the material to exhibit unambiguously the
effects of slow dynamics. The same data set is now analyzed in the frequency domain, with
amplitude spectra shown in Figure 15. The spectra show that there is almost no contribution
from the even (second and fourth) harmonics but a significant growth of the third and fifth
harmonic. The dependence of the odd harmonics with the amplitude at the fundamental
frequency is a quadratic one, which indicates that the behavior we observe in this particular
experiment is mostly due to hysteretic nonlinearity (see Figure 1).

© 2017 American Geophysical Union. All rights reserved.

 4.2 Quantitative comparison with a resonant bar experiment
To see how well the pulse propagation measurements can quantify nonequilibrium dynamics,
a standard resonant bar experiment was performed for comparison. The arrangement of the
resonant bar experiment is depicted in Figure 16. The piezoelectric transducer described
previously was digitally swept up with a sequence of harmonic voltage signals. Each
harmonic signal was applied for 55 ms and the transient vibrational response was recorded
during the last 40 ms of the source signal, to ensure that steady-state conditions had been
reached. Vibrational spectra were constructed from the harmonic responses, using heterodyne
processing. The experiment was conducted for frequencies ranging between 0.3 and 7 kHz in
steps of 2.5 Hz and at 22 excitation amplitudes ranging between 0.25 to 10 Vpp (before
amplification). The source signals were generated by a function generator (National
Instrument PXI-5406) and amplified 20 times by a voltage amplifier (TEGAM 2350). The
axial component of the acceleration was measured on the flat end opposite to the source by an
accelerometer (PCB Piezotronics 352A60) connected to a signal conditioner (PCB
Piezotronics 482C Series). The signal measured by the accelerometer was digitized with a
sampling rate of 10 MHz (National Instrument PXI-5122).

Any resonance mode can be selected to quantify nonequilibrium dynamics as long as the
mode type (either purely longitudinal or purely torsional) is unchanged, as recently
demonstrated by Remillieux et al. [2016]. Here, we focus on the longitudinal modes to be
consistent with the pulse propagation experiments. The vibrational spectra for this experiment
are shown in Figure 17. Material softening is observed when the drive amplitude of the
source becomes sufficiently large. For a given drive amplitude, such softening can be
quantified by plotting the relative frequency shift as a function of the maximum strain in the
sample, the slope of which is the nonlinear parameter α. The maximum strain in the sample
may be inferred analytically from the measured vibrational response. For the longitudinal
modes, the strain component of interest is εxx, where x is the direction given by the axis of
symmetry of the bar. For a system with a 1D-like geometry and unconstrained boundaries,
the expression given in Eq. (11) can also be used in the context of a resonance experiment to
relate the maximum amplitude of the axial component of the particle velocity at the free end
of the sample (where data is acquired) to the maximum amplitude of the axial component of
the strain in the sample.

Figure 18 shows the shift in resonance frequency at the jth source amplitude with respect to
the smallest source amplitude as a function of the amplitude of the axial strain εxx. This
evolution is very similar to that observed in Figures 10 and 12 in the context of a pulse
propagation experiment. The relative frequency shift varies almost linearly with the
maximum strain beyond 4 microstrain. This linear dependence is more obvious when higher
strain amplitudes can be reached in the experiment (typically on the order of 15 to 20
microstrains). When a mode has enough points beyond this strain value of 4 microstrain, the
points are fitted linearly and the slope of this fit is calculated. It appears that material
softening converges to a single value (all curves superimpose) of α = 5260 ± 160 for the
modes L6 through L12. Below the 6th resonance mode, the elastic response does not have a
sufficiently large amplitude to reach the threshold strain value of 4 microstrain.

The slope of the relative change of the resonance frequency is approximately twice as small
as the relative change of the Young’s modulus observed in the pulse propagation experiment,

© 2017 American Geophysical Union. All rights reserved.

which is consistent with the analytical relationship between Young’s modulus and resonance
frequency of a longitudinal mode at the perturbation level,

E E0  2 f f0 (12)

5 Conclusions
We demonstrated that monitoring the propagation of a finite-amplitude elastic pulse in a rock
can be used to isolate the mechanisms of classical nonlinearity, hysteresis, and
nonequilibrium dynamics at time scales less than 1 ms. Such experiment contrasts with the
complexity of the pump and probe scheme in a dynamic acousto-elastic experiment and with
the incomplete description of nonlinearity in a resonant bar experiment. Note that there are
still some great benefits to conducting these experiments, for instance to study slow dynamics
at much longer time scales (say, minute to hours). From a modeling point of view, the
significance of this experiment lies in its simplicity: a 1D-like geometry with only one source
frequency (or time scale) involved. It is our hope that the experimental results reported in this
paper will be useful to the development and validation of a truly universal model of the
NMEM dynamics.

Acknowledgments
We wish to thank Martin Lott, Pierre-Yves Le Bas, and Paul Johnson for valuable discussions
and suggestions. The current policies of Los Alamos National Laboratory do not allow the
public release of experimental data presented in this study. For more information, please
contact the authors.

References
Aleshin, V., V. Gusev, and V. Yu. Zaitsev (2004), Propagation of acoustics waves of
nonsimplex form in a material with hysteretic quadratic nonlinearity: Analysis and numerical
simulations, J. Comp. Acoust., 12(3), 319–354, doi:10.1142/S0218396X04002328.

Adams, F. D., and E. G. Coker (1906), An investigation into the elastic constants of rocks,
more especially with reference to cubic compressibility, Publ. 46, Carnegie Inst. of
Washington, Washington, D. C.

Berjamin, H., N. Favrie, B. Lombard, G. Chiavassa (2017), Nonlinear waves in solids with
slow dynamics: An internal-variable model, Proc. R. Soc. A, 473(2201), 20170024,
doi:10.1098/rspa.2017.0024.

Cabaret, J., P. Béquin, G. Theocharis, V. Andreev, V. E. Gusev, and V. Tournat, Nonlinear

Hysteretic Torsional Waves, Phys. Rev. Lett., 115(5), 054301,
doi:10.1103/PhysRevLett.115.054301.

Chung, J., and G. M. Hulbert (1993), A time integration algorithm for structural dynamics
with improved numerical dissipation: the generalized-a method, J. Appl. Mech., 60(2), 371–
375, doi:10.1115/1.2900803.

© 2017 American Geophysical Union. All rights reserved.

Gusev, V., C. Glorieux, W. Lauriks, and J. Thoen (1997), Nonlinear bulk and surface shear
acoustic waves in materials with hysteresis and end-point memory, Phys. Lett. A, 232(1–2),
77–86, doi:10.1016/S0375-9601(97)00357-5.

Gusev, V., (2000), Propagation of acoustic pulses in material with hysteretic nonlinearity, J.
Acoust. Soc. Am., 107(6), 3047–3058, doi:10.1121/1.429333.

Holcomb, D. J. (1981), Memory, relaxation, and microfracturing in dilatant rock, J. Geophys.

Res., 86(B7), 6235–6248, doi:10.1029/JB086iB07p06235.

Gol'dberg, Z. A. (1957), On the propagation of plane waves of finite amplitude, Sov. Phys.
Acoust., 3, 340-347.

Guyer, R. A., and P. A. Johnson (1999), Nonlinear mesoscopic elasticity: Evidence for a new
class of materials, Phys. Today, 52(4), 30–36, doi:10.1063/1.882648.

Guyer, R., K. McCall, and G. Boitnott (1995), Hysteresis, discrete memory and nonlinear
wave propagation in rock: A new paradigm, Phys. Rev. Lett., 74(17), 3491–3494,
doi:10.1103/PhysRevLett.74.3491.

Guyer, R. A., and P. A. Johnson (2009), Nonlinear Mesoscopic Elasticity: The Complex
Behavior of Granular Media Including Rocks and Soil, 410 pp., Wiley-VCH Verlag,
Weinheim.

Hamilton, M. F., and D. T. Blackstock (Eds.) (1998), Nonlinear Acoustics: Theory and
Applications, 455 pp., Academic Press, San Diego, Calif.

Johnson, P., B. Zinszner, and P. Rasolofosaon (1996), Resonance and elastic nonlinear
phenomena in rock, J. Geophys. Res., 101(B5), 11,553–11,564, doi:10.1029/96JB00647.

Kadish, A., J. A. TenCate, and P. A. Johnson (1996), Frequency spectra of nonlinear elastic
pulse-mode waves, J. Acoust. Soc. Am., 100(3), 1375–1382, doi:10.1121/1.415984.

Landau, L. D., and E. M. Lifshitz (1986). Theory of Elasticity, vol. 7. Course of Theoretical
Physics, 109 pp., Pergamon Press, New York.

Lott, M., C. Payan, V. Garnier, Q. A. Vu., J. N. Eiras, M. C. Remillieux, P.-Y. Le Bas, and T.
J. Ulrich (2016), Three-dimensional treatment of nonequilibrium dynamics and higher order
elasticity, Appl. Phys. Lett., 108(14), 141907, doi:10.1063/1.4945680.

Lott, M., M. C. Remillieux, P.-Y. Le Bas, T. J. Ulrich, V. Garnier, and C. Payan (2016),
From local to global measurements of nonclassical nonlinear elastic effects in geomaterials,
J. Acoust. Soc. Am., 140(3), EL231–EL235, doi:10.1121/1.4962373.

Lott, M., M. C. Remillieux, V. Garnier, P.-Y. Le Bas, T. J. Ulrich, C. Payan (2017),

Nonlinear elasticity in rocks: A comprehensive three-dimensional description, Phys. Rev.
Materials, 1(2), 023603, doi:10.1103/PhysRevMaterials.1.023603.

© 2017 American Geophysical Union. All rights reserved.

Lyakhovsky, V., Y. Hamiel, J.-P. Ampuero, and Y. Ben-Zion (2009), Non-linear damage
rheology and wave resonance in rocks. Geophys. J. Int., 178(2), 910–920,
doi:10.1111/j.1365-246X.2009.04205.x

McCall, K. R. (1994), Theoretical study of nonlinear elastic wave propagation, J. Geophys.

Res., 99(B2), 2591–2600, doi:10.1029/93JB02974.

Meegan Jr., G. D., P. A. Johnson, R. A. Guyer, and K. R. McCall (1993), Observations of

nonlinear elastic wave behavior in sandstone, J. Acoust. Soc. Am., 94(6), 3387–3391,
doi:10.1121/1.407191.

Murnaghan, F. D. (1937), Finite deformation of an elastic solid, Am. J. Math., 59(2), 235–
260, doi:10.2307/2371405.

Ostrovsky, L.A., and P. A. Johnson (2001), Dynamic nonlinear elasticity in geomaterials,

Riv. Nuovo Cimento Soc. Ital. Fis., 24S4(7), 1–46.

Pasqualini, D., K. Heitmann, J. A. TenCate, S. Habib, D. Higdon, and P. A. Johnson (2007),

Nonequilibrium and nonlinear dynamics in Berea and Fontainebleau sandstones: Low-strain
regime, J. Geophys. Res., 112, B01204, doi:10.1029/2006JB004264.

Pecorari, C. (2015), A constitutive relationship for mechanical hysteresis of sandstone

materials, Proc. R. Soc. A, 471(2184), 20150369, doi:10.1098/rspa.2015.0369.

Remillieux, M. C., R. A. Guyer, C. Payan, and T. J. Ulrich (2016), Decoupling nonclassical

nonlinear behavior of elastic wave types, Phys. Rev. Lett, 116(11), 115501,
doi:10.1103/PhysRevLett.116.115501.

Renaud, G., P.-Y. Le Bas, and P. A. Johnson (2012), Revealing highly complex elastic
nonlinear (anelastic) behavior of Earth materials applying a new probe: Dynamic
acoustoelastic testing, J. Geophys. Res., 117(B6), B06202, doi:10.1029/2011JB009127.

Renaud, G., M. Talmant, S. Callé, M. Defontaine, and P. Laugier (2011), Nonlinear

elastodynamics in micro-inhomogeneous solids observed by head-wave based dynamic
acoustoelastic testing, J. Acoust. Soc. Am., 130(6), 3583–3589, doi:10.1121/1.3652871.

Rivière, J., G. Renaud, R. A. Guyer, and P. A. Johnson (2013), Pump and probe waves in
dynamic acousto-elasticity: Comprehensive description and comparison with nonlinear
elastic theories, J. Appl. Phys., 114(5), 054905, doi:10.1063/1.4816395.

Rivière, J., P. Shokouhi, R. A. Guyer, and P. A. Johnson (2015), A set of measures for the
systematic classification of the nonlinear elastic behavior of disparate rocks, J. Geophys. Res.
Solid Earth, 120(3), 1587–1604, doi:10.1002/2014JB011718.

TenCate, J. A., K. E. A. Van Den Abeele, T. J. Shankland, and P. A. Johnson (1996),

Laboratory study of linear and nonlinear elastic pulse propagation in sandstone, J. Acoust.
Soc. Am., 100(3), 1383–1391, doi:10.1121/1.415985.

Ten Cate, J. A., and T. J. Shankland (1996), Slow dynamics in the nonlinear elastic response
of Berea Sandstone, Geophys. Res. Lett., 23(21), 3019–3022, doi:10.1029/96GL02884.

© 2017 American Geophysical Union. All rights reserved.

TenCate, J. A., E. Smith, L. W. Byers, and T. J. Shankland (2000a), Slow dynamics
experiments in solids with nonlinear mesoscopic elasticity, AIP Conf. Proc., 524(1), 303–
306, doi:10.1063/1.1309228.

TenCate, J. A., E. Smith, and R. A. Guyer (2000b), Universal slow dynamics in granular
solids, Phys. Rev. Lett., 85(5), 1020–1023, doi:10.1103/PhysRevLett.85.1020.

TenCate, J. A., D. Pasqualini, S. Habib, K. Heitmann, D. Higdon, and P. A. Johnson (2004),

Nonlinear and nonequilibrium dynamics in geomaterials, Phys. Rev. Lett., 93(6), 065501,
doi:10.1103/PhysRevLett.93.065501.

TenCate, J. A. (2011), Slow dynamics of earth materials: An experimental overview, Pure

Appl. Geophys., 168(12), 2211–2219, doi:10.1007/s00024-011-0268-4.

TenCate, J. A., A. E. Malcolm, X. Feng, and M. C. Fehler (2016), The effect of crack
orientation on the nonlinear interaction of a P wave with an S wave, Geophys. Res. Lett.,
43(12), 6146–6152, doi:10.1002/2016GL069219.

Thurston, R. N., and M. J. Shapiro (1967), Interpretation of ultrasonic experiments on finite-

amplitude waves, J. Acoust. Soc. Am., 41(4B), 1112–1125, doi:10.1121/1.1910443.

Vakhnenko, O. O., V. O. Vakhnenko, T. J. Shankland, and J. A. Ten Cate, Strain-induced

kinetics of intergrain defects as the mechanism of slow dynamics in the nonlinear resonant
response of humid sandstone bars, Phys. Rev. E, 70(1), 015602,
doi:0.1103/PhysRevE.70.015602.

Van Den Abeele, K. E.-A. (1996), Elastic pulsed wave propagation in media with second- or
higher-order nonlinearity. Part I. Theoretical framework, J. Acoust. Soc. Am., 99(6), 3334–
3345, doi:10.1121/1.414890.

Van Den Abeele, K. E.-A., P. A. Johnson, R. A. Guyer, and K. R. McCall (1997), On the
quasi-analytic treatment of hysteretic nonlinear response in elastic wave propagation, J.
Acoust. Soc. Am., 101(4), 1885–1898, doi:10.1121/1.418198.

Van Den Abeele, K. E.-A., P. A. Johnson, and A. Sutin, (2000), Nonlinear elastic wave
spectroscopy (NEWS) techniques to discern material damage, Part I: Nonlinear wave
modulation spectroscopy (NWMS), Res. Nondestr. Eval., 12(1), 17–30,
doi:10.1007/s001640000002.

Winkler, K., A. Nur, and M. Gladwin (1979), Friction and seismic attenuation in rocks,
Nature 277(5697), 528–531, doi:10.1038/277528a0.

Zoback, M. D., and J. D. Byerlee (1975), The effect of cyclic differential stress on dilatancy
in westerly granite under uniaxial and triaxial conditions, J. Geophys. Res., 80(11), 1526–
1530, doi:10.1029/JB080i011p01526.

© 2017 American Geophysical Union. All rights reserved.

 Nonlinear Classic
Linear Nonlinear Hysteretic (α)
(1st order perturbation β)

  M 0   M 0 1      M 0 1     f   ,   
  

Stress-Strain

  

  

Strain-Time t t t

A A A
Strain
Amplitude
Spectrum

1  1 2 3 4 5  1 3 5 

No Harmonics 2nd Harmonic: Slope 2 No Even Harmonics

Hamonic
3rd Harmonic: Slope 3 3rd Harmonic: Slope 2
Amplitude
4th Harmonic: Slope 4 5th Harmonic: Slope 2
Dependence
etc. etc.

Figure 1. Effects of pure classical nonlinearity and pure hysteretic nonlinearity on the
evolution of a sinusoidal waveform in the time and frequency domains. Reproduction of a
figure originally produced by Van Den Abeele et al. [2000].

© 2017 American Geophysical Union. All rights reserved.

 Scanning heads of
3D laser Doppler
vibrometer

PZT transducer + backload

Voltage amplifier

+ Controller Sample of Berea sandstone:

+ Signal generator • Diameter: 39.6 mm (1.56 inches)
• Length: 1.794 m (70.63 inches)
+ Data management system

Figure 2. Experimental setup for monitoring the propagation of an elastic pulse in the sample
of Berea sandstone.

© 2017 American Geophysical Union. All rights reserved.

 (a) (b)
1.25 0
10
1
-1
0.75 10
0.5
voltage [V]

voltage [V]
-2
0.25 10
0
-3
-0.25 10
-0.5
-4
-0.75 10
-1
-5
-1.25 10
0 0.2 0.4 0.6 0.8 1 0 10 20 30 40 50 60 70 80
time [ms] frequency [kHz]
1
Figure 3. Normalized electrical source signal used in the pulse propagation experiments: (a)
time history and (b) magnitude of the FFT. The signal consists of a sine wave with frequency
of 22.4 kHz, tapered at the start and end by half-Gaussian windows.
20 2
20 10
15 15
f0
particle velocity [mm/s]

1
particle velocity [mm/s]

10
particle velocity [mm/s]

10
10 5
0
5 -5 0
10
-10
0 -15 2f0
-1
-5
-20
0.32 0.34 0.36 0.38 10 3f0
time [ms]

-10 -2
10
-15
-3
-20 10
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 10 20 30 40 50 60 70 80
time [ms] frequency [kHz]
20 2
20 10
15 15
particle velocity [mm/s]

1
particle velocity [mm/s]

10
particle velocity [mm/s]

10 5
10
0
5 -5 0
10
-10
0 -15
-20 -1
-5 0.32 0.34 0.36 0.38 10
time [ms]

-10 -2
10
-15
-3
-20 10
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 10 20 30 40 50 60 70 80
time [ms] frequency [kHz]
20 2
20 10
15 15
particle velocity [mm/s]

1
particle velocity [mm/s]

10
particle velocity [mm/s]

10 5
10
0
5 -5 0
10
-10
0 -15
-20 -1
-5 0.32 0.34 0.36 0.38 10
time [ms]

-10 -2
10
-15
-3
-20 10
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 10 20 30 40 50 60 70 80
time [ms] frequency [kHz]

© 2017 American Geophysical Union. All rights reserved.

Figure 4. Axial component of the particle velocity measured at representative locations along
the propagation path of the impulsive wave for a source signal at 22.4 kHz. Left column: time
histories; right column: magnitudes of the FFT; top row: 10 mm from the source; Center row:
335 mm from the source; bottom row: at 590 mm from the source.
(a) 4 (b)
x 10
8000 2
raw data raw data
linear fit linear fit
6000 1.5

2V2c2
2V2c2

4000 1

2000 0.5
Slope of Linear Fit: Slope of Linear Fit:
β = 370 β = 660
0 0
0 2 4 6 8 10 0 5 10 15 20
V21x V21x

Figure 5. The variation of 2V2c2 as a function of V12ωx, the slope of which is β. This is a
post-processing of the data set corresponding to the spectral data reported in Figure 4. (a)
without any correction for damping; (b) where the amplitudes at the fundamental frequency
and second harmonic have been corrected for damping, based on a quality factor Q = 60.

© 2017 American Geophysical Union. All rights reserved.

 1
10 10
8
8
f0
6

particle velocity [mm/s]

0
particle velocity [mm/s]

particle velocity [mm/s]

6 10
2
4 0 3f0
-1
2 -2
10 2f0
-4
0 -6
-2
-2 -8
0.3 0.32 0.34 0.36 0.38 10
time [ms]
-4
-3
-6 10
-8
-4
-10 10
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10 20 30 40 50 60
time [ms] frequency [kHz]
Figure 6. Axial component of the particle velocity measured at 590 mm from the source for a
source signal at 15.7 kHz. Left: time histories; right: magnitudes of the FFT.

© 2017 American Geophysical Union. All rights reserved.

 10 1
10 10
8 8
f0

particle velocity [mm/s]

6
0
particle velocity [mm/s]

6 4 10
2
4

velocity [mm/s]
0
-2 -1 2f0 3f0
2 -4 10
-6
0 -8
-2
-2 -10
0.3 0.32 0.34 0.36 0.38 10
time [ms]
-4
-3
-6 10
-8
-4
-10 10
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10 20 30 40 50 60
time [ms] frequency [kHz]
Figure 7. Axial component of the particle velocity measured at 590 mm from the source for a
source signal at 17.1 kHz. Left: time histories; right: magnitudes of the FFT.

© 2017 American Geophysical Union. All rights reserved.

 20 2
10
Experiment
15 x = 15 mm Simulation 1
f0

particle velocity [mm/s]

particle velocity [mm/s]

10
10
5 0

0
10 2f0
-5 10
-1 3f0
-10 -2
10
-15
-3
-20 10
0 0.5 1 1.5 2 2.5 3 10 20 30 40 50 60 70 80
time [ms] frequency [kHz]
20 2
10
15 x = 115 mm 1
particle velocity [mm/s]

particle velocity [mm/s]

10
10
5 10
0

0
-1
-5 10

-10 -2
10
-15
-3
-20 10
0 0.5 1 1.5 2 2.5 3 10 20 30 40 50 60 70 80
time [ms] frequency [kHz]
20 2
10
15 x = 215 mm 1
particle velocity [mm/s]

particle velocity [mm/s]

10
10
5 0
10
0
-1
-5 10

-10 -2
10
-15
-3
-20 10
0 0.5 1 1.5 2 2.5 3 10 20 30 40 50 60 70 80
time [ms] frequency [kHz]
20 2
10
15 x = 315 mm 1
particle velocity [mm/s]

particle velocity [mm/s]

10
10
5 0
10
0
-1
-5 10

-10 -2
10
-15
-3
-20 10
0 0.5 1 1.5 2 2.5 3 10 20 30 40 50 60 70 80
time [ms] frequency [kHz]
20 2
10
15 x = 415 mm 1
particle velocity [mm/s]

particle velocity [mm/s]

10
10
5 0
10
0
-1
-5 10

-10 -2
10
-15
-3
-20 10
0 0.5 1 1.5 2 2.5 3 10 20 30 40 50 60 70 80
time [ms] frequency [kHz]
20 2
10
15 x = 515 mm 1
particle velocity [mm/s]

particle velocity [mm/s]

10
10
5 0
10
0
-1
-5 10

-10 -2
10
-15
-3
-20 10
0 0.5 1 1.5 2 2.5 3 10 20 30 40 50 60 70 80
time [ms] frequency [kHz]

Figure 8. Measured and simulated axial components of the particle velocity at various
positions along the sample: (left) time histories and (right) magnitudes of the FFT. In the
waveforms, the time window is long enough to show the direct and reflected pulses. Only the
direct pulses are used to compute the FFTs.

© 2017 American Geophysical Union. All rights reserved.

 20
20
15 15
particle velocity [mm/s]

particle velocity [mm/s]

10 10
5
5
0
0
-5
-5 -10

-10 -15
-20
-15 0.32 0.34 0.36
time [ms]
-20
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7
time [ms]
Figure 9. Time histories of the axial component of the particle velocity measured at 500 mm
from the source for 20 amplitudes of a pulse centered at 22.4 kHz.

© 2017 American Geophysical Union. All rights reserved.

 1
0
-1 Slope of = 11,100 700
-2
E/E0 [%]

-3
-4
-5
-6
-7
0 1 2 3 4 5 6 7 8 9
 xx [  ]
Figure 10. Relative change in elasticity as a function of strain estimated from the propagation
of a pulse centered at 22.4 kHz and measured at 500 mm from the source, for 20 source
amplitudes. The red star, blue circle, and green cross symbols correspond to the first five, 9 th
to 21st, and last five zero-crossings of the waveforms, respectively, as indicated by the inset.
A total of 23 curves are plotted in the figure from the reduced data. Each curve consists of 19
points (20 amplitudes with one amplitude, the lowest, used as a reference for the calculation
of the time delays).

© 2017 American Geophysical Union. All rights reserved.

 20
20
15 15
particle velocity [mm/s]

particle velocity [mm/s]

10 10
5
5
0
0
-5
-5 -10

-10 -15
-20
-15 0.32 0.34 0.36
time [ms]
-20
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7
time [ms]
Figure 11. Time histories of the axial component of the particle velocity measured at 650
mm from the source for 20 amplitudes of a pulse centered at 22.4 kHz.

© 2017 American Geophysical Union. All rights reserved.

 1
0
-1 Slope of = 10,500 350
-2
E/E0 [%]

-3
-4
-5
-6
-7
0 1 2 3 4 5 6 7 8 9
 xx [ ]
Figure 12. Relative change in elasticity as a function of strain estimated from the propagation
of a pulse centered at 22.4 kHz and measured at 650 mm from the source, for 20 source
amplitudes. The red star, blue circle, and green cross symbols correspond to the first five, 9 th
to 21st, and last five zero-crossings of the waveforms, respectively, as indicated by the inset.
A total of 23 curves are plotted in the figure from the reduced data. Each curve consists of 19
points (20 amplitudes with one amplitude, the lowest, used as a reference for the calculation
of the time delays).

© 2017 American Geophysical Union. All rights reserved.

 8
8
6 6
particle velocity [mm/s]

particle velocity [mm/s]

4 4
2
2
0
0
-2
-2 -4

-4 -6
-8
-6 0.34 0.36 0.38 0.4
time [ms]
-8
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
time [ms]
Figure 13. Time histories of the axial component of the particle velocity measured at 500
mm from the source for 20 amplitudes of a pulse centered at 15.7 kHz.

© 2017 American Geophysical Union. All rights reserved.

 1
0
-1
-2
E/E0 [%]

-3
-4
-5
-6
-7
0 0.5 1 1.5 2 2.5 3 3.5 4
 xx [  ]
Figure 14. Relative change in elasticity as a function of strain estimated from the propagation
of a pulse centered at 15.7 kHz and measured at 500 mm from the source, for 20 source
amplitudes. The red star, blue circle, and green cross symbols correspond to the first five, 9 th
to 21st, and last five zero-crossings of the waveforms, respectively, as indicated by the inset.
A total of 23 curves are plotted in the figure from the reduced data. Each curve consists of 19
points (20 amplitudes with one amplitude, the lowest, used as a reference for the calculation
of the time delays).

© 2017 American Geophysical Union. All rights reserved.

 1
10 f0 0.14
raw data 0.1
0.12 quadratic fit raw data
0.1 0.08 quadratic fit
particle velocity [mm/s]

[mm/s]
0 0.08

[mm/s]
10 0.06
0.06

3 f0
A

5 f0
0.04 0.04

A
0.02
3f0 0.02
-1 0 5f0
10 1 2 3 4 5
f0
6 7
0
A [mm/s] 1 2 3 4 5 6 7
f0
A [mm/s]

-2
10

-3
10
0 10 20 30 40 50 60 70 80 90
frequency [kHz]
Figure 15. Magnitudes of the FFT of the axial component of the particle velocity measured at
500 mm from the source for 20 amplitudes of a pulse centered at 15.7 kHz.

© 2017 American Geophysical Union. All rights reserved.

 Sample of Berea sandstone:
Signal conditioner • Diameter: 39.6 mm (1.56 inches)
• Length: 1.794 m (70.63 inches)

Voltage amplifier

+ Data acquisition
+ Signal generator
+ Data management system
Accelerometer

Vibration isolation table

Figure 16. Experimental setup for conducting nonlinear resonant ultrasound spectroscopy
(NRUS) on the sample of Berea sandstone.

© 2017 American Geophysical Union. All rights reserved.

 ALL MODES ONE MODE
14 14
L12
L10 L11 12 L10
12
L9
L7 L8
particle velocity [mm/s]

particle elocity [mm/s]

10 L6 10
L5
L4
8 8
L3
6 6
L2
4 4
L1
2 2

0 0
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 5.5 6 6.5 7 5.45 5.5 5.55 5.6 5.65 5.7 5.75 5.8
frequency [kHz] frequency [kHz]
Figure 17. Vibrational spectra measured on the sample of Berea sandstone at 22 source
amplitudes. The first 12 modes of longitudinal vibration L1 through L12 are indicated. The red
star symbols denote the location of the resonances for all source amplitudes. The resonance
frequencies are not computed for the first three modes because of the poor signal-to-noise
ratio.

© 2017 American Geophysical Union. All rights reserved.

 0.5
L4
L5
[f(Aj)-f(A0)]/f(A0) [%]

0
L6
L7
-0.5
α = 5260 160 L8
-1 L9
L10
-1.5 L11
L12
-2
0 1 2 3 4 5 6 7
 xx(Aj) [ ]

Figure 18. Relative frequency shift as a function of the strain component εxx for the
longitudinal modes L4 through L12.

© 2017 American Geophysical Union. All rights reserved.