Proceedings of the Joint Rail Conference

JRC2021
April 19-21, 20206, St. Louis, Missouri, USA

Put paper number here

BALLAST INSPECTION TOOL USING SEISMIC SURFACE WAVES

Charles P. Oden Timothy D. Stark

Earth Science Systems, LLC U. Illinois – Urbana Champaign
Wheat Ridge, CO, USA Champaign, IL, USA

Choon Park Carl L. Ho

Park Seismic, LLC U. Massachusetts – Amherst
Shelton, CT, USA Amherst, MA, USA

ABSTRACT INTRODUCTION
A man-portable seismic instrument has been developed for A man-portable seismic instrument has been developed for
assessing the elastic properties of railroad ballast. The Seismic assessing the elastic properties of railroad ballast. The Seismic
Ballast Inspection Tool (SeiBIT) is based on the multi-channel Ballast Inspection Tool (SeiBIT) uses seismic surface waves to
analysis of surface waves (MASW) method and is designed to resolve the seismic properties such as shear modulus at depths
resolve the ballast properties as shallow as 1 foot (30 cm) and as shallow as 1 foot (30 cm) and up to 16 feet (5 m). The
to depths greater than 16 feet (5 m). The SeiBIT consists of a system is man-portable and can be operated by a single
seismic source and five seismic receivers – each packaged inspector. It uses a tablet computer that contains all data
independently. The system uses GPS time synchronization and acquisition and processing software. A survey can be
localization so that each unit can be placed individually on the completed in about 10 minutes including a complete analysis of
track ballast surface with no location constraints other than the data. The SeiBIT builds on the earlier work in ballast
lying on the survey line. The computer controlled seismic seismic properties by [1-3].
source can generate horizontal displacements in the in-line or
cross-line directions so that either Rayleigh or Love waves can BACGROUND
be preferentially generated (respectively). The receivers use Seismic surface waves can be classified as either Rayleigh
three-component accelerometers and 24 bit digitizers. The waves which have both vertical displacements and horizontal
entire system is controlled wirelessly via WiFi using a tablet displacements in the direction of propagation, or Love waves
computer. which have horizontal displacements perpendicular to the
After field data have been collected, Love and Rayleigh propagation direction (see Figure 1). Surface wave amplitudes
wave dispersion curves (i.e., wave velocity versus frequency) decay with depth from the surface and penetrate to a depth of
are generated from survey data, which are then inverted for about a wavelength (depending on the velocity structure of the
seismic properties (e.g., shear wave velocity and layer medium), and therefore low frequency waves are used to
thickness) using a one-dimensional model. The result is a investigate deep regions while high frequency waves sample
vertical section of properties that represent the ballast and shallower regions. In layered media, surface waves are
subgrade. Software has been created for the tablet computer for dispersive meaning that wave velocities change with frequency
wireless data acquisition, MASW analysis, and dispersion [4, 5]. Figure 2 shows calculated surface wave dispersion
curve inversion. Complete results can be obtained in minutes. curves for various hypothetical layered velocity structures. For
The system response has been assessed with MASW each layered model there are multiple Rayleigh and Love wave
experiments in both the laboratory and in the field. Preliminary dispersion curves corresponding to fundamental mode (slowest)
results of surveys on railroads indicate that the Rayleigh wave and to higher order modes. The higher order modes have cutoff
method produces expected results in clean and fouled ballasts. frequencies below which they cannot be excited and always
In future testing, the MASW results will be compared with have velocities faster than lower order modes.
other methods of measuring subsurface engineering properties
in railroad environments.

1 Copyright © 20xx by ASME

Figure 1. Displacements for Rayleigh waves (top-left) and
Love waves (top-right), adapted from [5]. Rayleigh wave
displacement in vertical and horizontal directions versus depth
for different values of Poisson’s ratio (bottom), adapted from
[6].

High order Rayleigh modes may be dominant when the top

layer has a much slower velocity than the deeper layers, when
there are high velocity contrasts between layers (especially at
shallow depths), or if a low velocity layer exists between two
higher velocity layers [5]. The energy partitioning between the Figure 2. Modeled dispersion for a normal velocity profile (top
modes is controlled by the layer velocities and attenuation, row), inverse velocity profile (second row), and two mixed
source type, source coupling, and the depth of the source. The profiles (bottom two rows). Note the osculation point in the
energy distribution between modes can be strongly frequency bottom row (circled).
dependent, where one mode may dominate in a specific
frequency band. At certain frequencies where the velocity of generated by a nearby source. The travel time or phase shift is
different modes is similar (i.e., osculation points, see Figure 2), measured at various frequencies using signal processing
the energy may move from one mode to another. The observed techniques and the testing is typically repeated using various
velocity is dependent on the amplitudes of all excited modes. If source offsets from the receivers. The method works well as
one mode is dominant, its velocity can be extracted using the long as only fundamental modes are excited, but the analysis is
phase-shift coherence mode discussed below, and then inverted not automated and requires a trained user to conduct a phase
if the mode number can be properly identified. Love waves, in unwrapping process to generate dispersion curves [1]. The
contrast, are less prone to higher order modes. They are always MASW method uses an array of receivers which enables the
dispersive because they require a slow layer over a faster layer detection and analysis of both fundamental and higher order
to be developed, and require a lower frequency to image to a modes, and may employ one or more source offsets. The
particular depth than Rayleigh waves. Sites with a rapid MASW method uses a coherence calculation that simplifies the
increase in velocity with depth or with a low velocity layer process of generating dispersion curves that can be automated if
overlying stiff sediments often generate higher order Rayleigh only fundamental modes are present. Both the SASW and
modes, but only fundamental mode Love waves. Love waves MASW methods start by extracting dispersion curves from
require a velocity profile that continuously increases with depth survey data and then use an inversion routine to reverse or
– a condition that may not be valid in the shallow subsurface. invert calculations made by a forward model calculator. The
Two common surface wave survey methods are the forward modeling routine calculates theoretical dispersion
spectral analysis of surface waves (SASW), and the multi- curves for hypothetical models, and the model parameters are
channel analysis of surface waves (MASW). The SASW varied until the calculated curve matches the measured curve as
method measures the travel time (or rather phase shift) of well as possible. The model with the best fitting dispersion
surface waves between two receivers due to a seismic signal curve is said to best represent the subsurface. Normally this

2 Copyright © 20xx by ASME

process is done using Rayleigh wave data, and historically,
Love wave surveys have not been conducted because the source
is more complex and horizontal receivers are less common and
thus more expensive. However Love waves are much less
susceptible to higher order modes and their inversion is more
stable. When Love wave dispersion curves are available, using
both Rayleigh and Love dispersion curves reduces non-
uniqueness in the inverted properties due to difficulties
distinguishing slow-thin layers from fast-thick layers. Both
SASW and MASW analysis use 1D wave propagation and
assume that the subsurface is laterally homogeneous in both
directions. These methods are non-invasive and provide good
shallow resolution, with decreasing resolution as depth
increases.

INSTRUMENTATION
The SeiBIT consists of a seismic source and five seismic
receivers. The system components are shown in Figures 3 and
4. The source and receivers can use cleats to couple to soil and
ballast, and in some cases skid runners might be used. For the
case of railroad ballast some surfaces can be very rough or very
well compacted, and a better approach is to use coupling plates
that are attached to the ballast with lag screws and the source
and receivers are then attached to the coupling plates. The
SeiBIT system is compatible with both impulse and stepped- Figure 3. Top row: SeiBIT seismic source and drive plate.
frequency sources. The system contains a computer-controlled Center row: receiver and bottom plate. Bottom row: coupling
source to drive horizontal displacements using a forceful plates (shown upside down) can be used for very hard and/or
magnetostrictive driver that can be programmed to generate rough surfaces. Receivers magnetically attach to the coupling
impulses or drive a fixed frequency. For a standard SeiBIT plates.
survey, the source steps through a number of discrete
frequencies and separate receiver records are obtained for each
frequency. This stepped-frequency approach provides good
signal-to-noise ratios at high frequencies (above 100 Hz),
which in turn provides the shallow resolution needed for
railroad ballast surveys. Alternatively, a hammer and a strike
plate can be used which makes the systems more portable. In
some cases it may be difficult to obtain shallow resolution
when using an impulsive source because their spectral content
emphasizes lower frequencies and there is less energy at higher
frequencies (e.g., above 100 Hz, [7]). The receivers use three-
component accelerometers to sense motion in all directions.
The entire system is wireless and controlled by a tablet
computer. The system also has an integrated GPS system that
can measure the location of the receivers to within 1 cm relative
to a GPS base station – which can be a separate unit, one of the
receivers, or the source. The tablet PC has software for
controlling the system and conducting a survey, as well as data
analysis and inversion routines for calculating layered seismic
properties. The entire system has been designed to be man-
portable and easy to use.

Figure 4. Block diagram of SeiBIT source module and receiver

(seismometer) module.

3 Copyright © 20xx by ASME

 The SeiBIT source can preferentially generate Rayleigh or conventional GPS for time synchronization. This reduces the
Love waves depending on its orientation. To conduct a survey survey time to about 5 minutes.
the source and receivers are placed along a line along which The typical seismic source used in MASW studies is a
surface waves will propagate – from source to the receiver vertical impulse made with a hammer or weight drop device.
array. With three-component receivers, any ground motion that The goal was to build a shear source that could be oriented in-
is not in the preferred direction (i.e., vertical or in-line line to produce Rayleigh waves or cross line to produce Love
horizontal for Rayleigh waves or lateral horizontal for Love waves. Using a polarized source with polarized receivers
waves) can be omitted (at least partially) from the analysis. enables the separate Rayleigh and Love wave surveys, which
The SeiBIT system has the potential to provide three reduces uncertainty in the subsurface properties estimated by
independent measurements of depth dependent (or frequency analyzing both Rayleigh and Love survey results. The shear
dependent) seismic properties – Rayleigh wave velocities, Love source uses a magnetostrictive actuator to drive the drive plate
wave velocities, and Rayleigh wave ellipticity that together against the ground (see Figure 3). The source needs a weight
provide a very rich dataset for exploring the subsurface. on top to aid in coupling the source to the ground, and a weight
The block diagrams for the SeiBIT source and receivers that functions as an inertial reaction mass. This weight can be
(seismometers) are shown in Figure 4. The devices are provided by weight plates which reduce system portability, or
powered using 12 VDC LiFePO4 batteries, which were selected by the operator standing on the source. The weight plates are
for their ability to produce large currents and withstand more rigidly attached to the source and make a more effective
thousands of charge-discharge cycles. The batteries provide 8 reaction mass than an operator standing on the unit. But in
hours of operation and can be recharged overnight. The GPS practice we find that both methods can be used effectively
system provides precise time synchronization between the (more details in the next section).
nodes, and can optionally provide position. The transmitter The system performance has been verified using laboratory
drives a magnetostrictive actuator that can provide up to +/- and outdoor test beds. The time GPS synchronization of the
250 lbs of force over a small displacement (+/- 25 μm). The transmitter and receiver array has been verified and the
receivers use three-component accelerometers to measure maximum timing skew in the entire system is on the order of 10
seismic waves because we have determined that high quality μs, which is much less than the receiver sample interval of 200
MEMS accelerometers provide better signal-to-noise ratios μs. The total harmonic distortion (THD) of the source driver
than geophones for frequencies above about 100 Hz. has been measured at less than 6% over the range of 40 to 500
The original plan was to use UHF radios to provide time Hz. The THD of the receiver electronics is less than 1.1 %.
synchronization, but it was difficult to obtain consistent Finally, a test was performed on the receivers to insure that they
reception with the prototype system. It was also planned to had a uniform amplitude and phase response over the band of
attach the receivers to a piece of nylon webbing that would interest when subjected to external vibration. Initially, there
constrain their relative positions to the proper offset (i.e. one were some resonance issues with the prototype receiver
receiver in each successive crib). However ESS had recently housings above a few hundred Hertz, but those were resolved
integrated inexpensive RTK GPS technology into some of its by filling the enclosures damping material. Between 40 and
products and realized that they would be useful in the SeiBIT 500 Hz the difference of the amplitude response between all
system for providing precise time synchronization across all receivers is less than 25% and the phase response difference is
nodes in the system. Additionally, the RTK GPS solution can less than 3%. Since the primary goal is to measure the phase
measure relative location between the nodes to one cm differences between the receivers, this amplitude variation is
accuracy. Adding this GPS technology to the SeiBIT system acceptable. The production electronics and enclosures will be
would allow the nodes to be freely placed wherever convenient. smaller so these issues will be further minimized.
The fixed offsets provided by the webbing solution could be The acquisition software has been designed to be fully
problematic in locations with uneven or non-standard ties functional and still be easy to use. Users can acquire and
spacing. analyze the data using the Windows® tablet computer. Figure
One drawback to the RTK GPS system is that it increases 5 shows some screenshots from the software. The software
the time required to make a survey. The RTK GPS can take automatically organizes a WiFi mesh network and reports GPS
more than five minutes for the system to lock-in, and placing status and battery status for each unit. Users can configure the
the source and receivers on the track and running through the source frequencies so that the entire sequence runs
stepped-frequency sequence typically takes less than five automatically at the push of a button. After the data collection
minutes, so using the RTK GPS more than doubles the survey phase, a dispersion diagram can be generated and then inverted
time. One option is to use a separate RTK GPS base station on to estimate layer specific seismic properties [8]. All of these
the side of the track so that it can lock-in before the SeiBIT operations can be done on-site without the need to process data
source and receivers are deployed on the track, but this requires in the office.
yet another piece of equipment that hinders portability.
Because measuring the receiver offsets from the source with a SURVEY DESIGN
tape measure is a quick operation, we opted to not to use the The optimal survey layout depends on the desired depth of
RTK GPS system for positioning in early testing, but retain investigation and the range of expected seismic wavelengths.

4 Copyright © 20xx by ASME

The receiver spread should be located far enough from the
source that near-field effects of the source are not present and
body waves have attenuated so that surface waves dominate.
Xia et al. (2004) recommend that the distance between the
source and first receiver should be at least a half wavelength of
the lowest frequency in the survey. Conversely, if the receivers
are too far from the source (i.e., more than two wavelengths)
then higher order modes are more likely and signal to noise
ratio can be poor [1]. If the receivers are too close to the
source, near field effects can cause the phase shift between
proximal receivers to be different from that of distal receivers.
Conversely, if the phase shift between distal receiver pairs is
different than that of proximal pairs, then body waves or higher
order modes may be affecting the measurement [9, 10].
Another consideration is that the receiver spread should be
configured to avoid spatial aliasing, meaning that the receiver
spacing should be half of the smallest wavelength and the
spread should span the longest wavelength. These constraints
can be relaxed when the fundamental mode is well delineated
by the dispersion diagram (see Field Tests section), because the
spatially aliased regions of the diagram can be avoided and the
velocity of long wavelengths can be measured with spreads
shorter than a wavelength.
The dispersion curves in Figure 2 show that the number of
higher order modes increases with frequency. These modes
have a deeper depth of investigation than the fundamental mode
and their amplitude increases with increasing source-receiver
offset. Long receiver arrays with several tens of receivers are
needed to be able to distinguish high order modes from lower
order modes because coherence peaks with long arrays produce
coherence images that are more focused. In an effort to keep
this system portable the number of SeiBIT receivers has been
limited to five, and therefore only the fundamental mode will
be considered in the SeiBIT data analysis. Source geometries
and receiver spreads should be selected to minimize the
generation and reception of higher order modes. The presence
of higher order modes can be indicated by sudden jumps in the
ratio of horizontal to vertical motion (H/V) of Rayleigh waves.
A series of experiments were conducted using different
source coupling techniques and receiver offsets in order to
investigate source coupling versus excitation frequency and
also the minimum required distance between the source and
first receiver. Because access to in service track is a limited
resource with time limitations, we opted to use an open field for
these initial tests as shown in Figure 6. The following
configurations were tested.
1. Shear source directly on the soil using a human
operator (160 lbs.) as a coupling and reaction mass. Figure 5. Tablet software data acquisition and processing
2. Shear source directly on the soil using fixed weights screens.
(100 lbs.) as the coupling and reaction mass.
3. Shear source sitting on ballast sized crushed rock with Surprisingly, there was no significant loss when coupling
a human operator as a coupling and reaction mass. through the crushed rock, and no significant advantage to using
4. Shear source sitting on ballast sized crushed rock weights over a human. Although the weights are cumbersome
using fixed weights as the coupling and reaction mass. to transport, they relieve the operator from needing to stand on

5 Copyright © 20xx by ASME

 contact with the ballast because at some frequencies the volume
heard by ear was louder and distorted which could indicate that
the driven displacement has significant harmonics. This
however may be tolerable because the main signal component
is the fundamental drive frequency and spectral filters are
employed at the receivers to remove out of band noise.
The receiver offsets were selected to be in intervals of nominal
timber tie spacing in North American railroads (i.e., 19 inches
or 48 cm). Using a five receiver spread with offsets of [3, 4, 5,
6, 7]∙19 inches (or [3, 4, 5, 6, 7]∙48 cm), the experimental
results shows that the phase delay at each frequency in the
range of 40 to 400 Hz was constant over the array, except for
the first receiver at frequencies at or below 60 Hz (a non-linear
phase offset may indicate that the receivers are too close to the
source). Figure 7 shows some data recorded by the SeiBIT
system for the [3, 4, 5, 6, 7] array where linear phase offsets
can be observed and the SNR levels are typically in the range
of 50 to 400 over the typical operating frequency range. Since
a good signal was obtained for all receivers over all frequencies
ranges, it was concluded that this would be the standard spread
geometry used on a track. Note that since the receivers use
accelerometers rather than geophones, low frequency signals
below 100 Hz are generally more attenuated than the high
frequency signals. The shear wave velocities of the soils as
determined by the MASW method were between 100 and 200
m/s, which is similar to the expected velocities of railroad
ballast. For ballast surveys, the selected stepped-frequency
excitation ranges from 40 Hz to 250 Hz in 10 Hz intervals.
Assuming a Rayleigh wave velocity of 100 m/s and a shallow
inversion resolution of a half a wavelength (see section Seismic
Data Processing and Inversion), then the survey can resolve
depths as shallow as 20 cm. If the Rayleigh wave velocity of
200 m/s and the deep inversion resolution is one wavelength,
Figure 6. Different source coupling techniques: shear source then the survey can measure as deep as 5 meters.
directly on the soil using a human operator as a coupling and The SeiBIT’s programmable source is capable of
reaction mass (left), and Shear source sitting on ballast sized conducting surveys with different source excitations including
crushed rock using fixed weights as the coupling and reaction impulsive, swept frequency, and stepped frequency. Park et al.
mass (right). [9, 10] successfully investigated the use of a swept frequency
source, however the stepped-frequency approach was selected
for this application so that the system could dwell at each
frequency for sufficient time to obtain a good signal-to-noise
ratio. In the stepped-frequency approach, the duration
(typically one second) of small-amplitude constant frequency
chirps can be extended to generate as much or more energy
than a large-amplitude short-duration source such as a hammer.
The SeiBIT’s impulsive source is significantly weaker than a
hammer source, so when an impulsive source is desired a
traditional hammer source is recommended. The Field Tests
section compares the stepped-frequency results with those from
a traditional hammer source.

Figure 7. Receiver data recorded by the SeiBIT system. Note FIELD TESTS
the phase offset between receivers. There are several test methods that can provide comparison
data for evaluating MASW surveys. Perhaps the easiest
the transmitter while testing. At certain excitation frequencies method is the dynamic cone penetrometer (DCP) whose
it seems that the source cleats may not maintain continuous measurements can be used to estimate both shear strength and

6 Copyright © 20xx by ASME

the California Bearing Ratio (CBR) for a range of different
ballast types and conditions [11, 12]. The lightweight
deflectometer (LWD) provides a convenient measure of
Young’s modulus (or LWD modulus [13]), but only provides
one value at its fixed depth of investigation (i.e., ~2 times the
base plate diameter). An LWD base plate with a diameter
larger than about 16 inches (40 cm) is not practical on railways
with 19 inch (48 cm) cribs, so the depth of investigation is
limited to about 32 inches (80 cm). Furthermore, the LWD
measurement does not make small-strain measurements so its
modulus readings cannot be compared directly with MASW
measurements. Some MASW practitioners have used a shear
wave logging probe in a borehole that penetrates the region of
investigation [14], but that requires a borehole which is not
practical for ballast surveys. Subsequently DCP measurements
are recommended for comparison when available.

Figure 8. Dispersion diagrams and inversion results for open

field experiment. The blue dots are the velocities picked from
the dispersion diagrams and the red dots are the modeled
velocities from the inversion procedure. The right panel shows
the inverted shear wave velocity profile.

The first field test with the SeiBIT was in an open field as
shown in Figure 6. This site was used to vet the
instrumentation, acquisition software, and analysis routines
before any work was conducted on a rail line. A stepped-
frequency survey was performed and the inversion results using
Rayleigh wave data. These results are reasonable for dry
unconsolidated sediments. When calculating the dispersion
diagrams, the receiver pair closest to the source was omitted
from the stack for frequencies less than 80 Hz to reduce near
field effects, and the farthest pair was omitted for frequencies
above 180 Hz to reduce higher order mode effects. Note also
that there is a kink in the Rayleigh wave dispersion curve at
about 160 Hz that could be due to the presence of a higher
order mode. Including Love wave data above 80 Hz increased
the misfit between the modeled and measured dispersion
curves. Love waves require a continuously increasing velocity Figure 9. Pictures of SeiBIT survey at TTCI and near
profile and it is likely that this assumption is violated in many Hickman, NE. Top: using human operator with source, center:
shallow subsurface environments. If higher order mode curves using fixed weights with source, and right: using a hammer
are to be included in the inversion, their mode number must be source at the clean ballast site.
properly identified. With such a small receiver spread (i.e. 5
receivers) the ability to resolve higher order modes is limited. A second field test was conducted on the FAST track at the
Similar results were obtained when using either cleats or lag- Transportation Technology Center near Pueblo, CO. In an
screwed mounting plates, and when using a human reaction effort to obtain ground, truth deflection data were taken at three
mass versus weight plates. locations using mechanical displacement device that records the

7 Copyright © 20xx by ASME

maximum deflection seen between the rail and the ballast (see
Figure 9). Unfortunately there was no significant difference in
the deflection readings from site to site that could be used to
compare to different MASW results. Additionally, as the
survey was being conducted it was apparent that there were
coupling problems at some sites. At one site the fouling in the
ballast was compacted so thoroughly that the cleats on the
source and receivers could not be inserted into the ballast and
coupling was poor. At another location, new ballast had been
recently installed with a rather large aggregate size (AREMA
#3, up to 2.5 inches (6.3 cm) along the minor axis and up to 5
inches (12.5 cm) along the major axis) that the cleats could not
penetrate into the ballast for good coupling.
The data quality from the TTCI surveys was very poor and
there was little similarity between the surveys using the man-
coupled source and the weight coupled source, indicating that
the measurement is not repeatable. Furthermore it is difficult to
pick continuously coherent velocities from the dispersion
diagrams. These problems could stem from several possible
sources such as 1) multiple reflections along the line caused by
the ties, 2) energy coupling to the superstructure above the
ballast, 3) the presence of higher order modes, or 4) poor
coupling of the sensors to the ballast medium. In subsequent
testing it was learned that the issue was indeed due to poor
coupling and that gravity coupling on ballast is insufficient. As Figure 10. Dispersion diagrams and inversion results for a field
a result of these issues, we designed coupling plates that can be experiment at a track location near Hickman, NE with heavily
attached to the ballast using lag screws, and then the source and fouled ballast. The blue dots are the velocities picked from the
receivers can be attached to the plates. The receivers attach to dispersion diagrams and the red dots are the modeled velocities
the plates using magnets to facilitate fast setup. from the inversion procedure. The right panels show the
A final set of field tests were made near Hickman, Nebraska on inverted shear wave velocity profile. A stepped-frequency
in-service BNSF track that has been instrumented with strain source with a human reaction mass was used in the top plots, a
gauges and accelerometers [15]. Measurements were taken at stepped-frequency source with weigh plates was used in the
two locations: one with heavily fouled ballast and another with middle plots, and the lower plots used an impulsive hammer
clean ballast (see Figure 9). The receiver coupling plates were source.
large enough to make it slightly inconvenient to find a sequence
of cribs that were large enough for mounting the plates while poorer quality than the stepped-frequency data. In Figures 10
avoiding contact with the tie, and the receiver spread had to and 11 spatial aliasing is observed for velocities less than about
skip a crib in some instances. The future recommendation is to 100 m/s where the high amplitude streaks emanate from the
use both smaller receivers and coupling plates. The dispersion origin (approximately), and this effect is more pronounced with
diagrams are shown in Figures 10 and 11 for the fouled and the impulsive source. Different source types and coupling
clean sites respectively. A second survey was conducted at the mechanisms affect the generation of higher order modes, and
clean ballast site a few meters along the track from the first this may be what is causing the poor dispersion diagrams when
clean ballast survey site, and the results were very similar. using an impulsive source. The phase velocities and inverted
There was no visible difference in the ballast condition at the model velocities were lower when using the hammer source.
two clean ballast survey locations. As with previous tests, the Love wave dispersion diagrams are
At each site, a stepped-frequency dataset was taken using a not consistent with the Rayleigh wave diagrams and are not
human reaction mass and another was taken using weight plates used in the inversion. When calculating the dispersion
for the reaction mass. The final dataset at each site was taken diagrams, the receiver pair closest to the source was omitted
using an impulsive hammer source (a 5 lb. sledge hammer) from the stack for frequencies less than 80 Hz to reduce near
where the blows were made in the vertical direction to excite field effects, and the farthest pair was omitted for frequencies
Rayleigh waves and the horizontal direction to excite Love above 180 Hz to reduce higher order mode effects. The
waves. The data show good Rayleigh wave repeatability for similarity of the inverted models for the different source types
both types of stepped-frequency surveys, and it is concluded is encouraging.
that the weight plates are not needed. Although the received
signal level of the hammer source data is very good, the
dispersion diagrams from the impulsive source data have

8 Copyright © 20xx by ASME

 visible reflections, and higher frequencies and/or slower
velocities will increase the effect. Results to date indicate that
it may be difficult to obtain reliable readings at depths less than
about 30 cm (1 foot), but more study is needed through field
experiments and possibly numerical modeling.
As in the previous field tests, the quality of Love wave data
at the rail road sites was poor. This could be because 1) the
shear velocity does not increase steadily with depth, 2) there is
interference from ties, 3) the lateral geometry that is not infinite
as assumed and waves might reflect off of the shoulders, or 4)
there could be interference from Rayleigh waves and higher
order modes. The rough ballast surface causes some receivers
to tilt as much as 20-30 degrees, which enables interference to
Love waves from Rayleigh waves. In future work it is
recommended to either use self leveling mounts or account for
tilt in the data analysis. Although Love wave surveys can
theoretically provide more information to better constrain
inversions, they can also provide data that is inconsistent from
Rayleigh wave dispersion curves if poor quality data are used
or the survey assumptions are violated. Due to these
difficulties it is suggested that Love waves not be used because
consistent results cannot be obtained in a wide variety of
environments. This has the benefit of reducing survey time.

Figure 11. Dispersion diagrams and inversion results for a field

experiment at a track location near Hickman, NE with clean
ballast. The blue dots are the velocities picked from the
dispersion diagrams and the red dots are the modeled velocities
from the inversion procedure. The right panels show the
inverted shear wave velocity profile. A stepped-frequency
source with a human reaction mass was used in the top plots, a
stepped-frequency source with weigh plates was used in the
middle plots, and the lower plots used an impulsive hammer
source.

Generally, the high frequency regions of the dispersion

diagrams above 200 Hz do not show correlated energy at the
railroad test locations, but the open field site without the ties
and rail does show correlated energy to about 350 Hz (see
Figure 8). From first principals, seismic waves will reflect off
interfaces with dissimilar impedances (i.e., density times
velocity), and periodic structures such as repeating ties tend to
cause complex wave fields with energy traveling in both
directions through or along the structure. Figure 1 shows the
displacement versus depth for Rayleigh waves in a
homogeneous half-space, where most of the wave energy is
contained in depths down to about half of a wavelength. For a
Figure 12. Effective seismic velocities for a sandy medium
50 cm (20 inch) wavelength (e.g., 200 Hz wave traveling versus porosity (top-left) and saturation (top-right). Effective
through a medium with a velocity of 100 m/s), ties with a seismic velocities for a clay medium versus porosity (bottom-
thickness of 18-22 cm (7-9 inches) will interfere with about 40-
left) and saturation (bottom-right).
80% of the particle motion (visually estimated from Figure 1).
In terms of reflections along the direction of propagation,
Table 2 shows the measured shear velocity at the two
objects in a medium that are larger than about 1/8 of a
Hickman sites along with the shear modulus calculated using
wavelength can produce coherent reflections [16]. With a 50 equation (2). Effective medium models for shallow
cm (20 inch) wavelength, objects with thicknesses as small as unconsolidated sediments are shown in Figure 12 [17, 18], and
6.3 cm (2.5 inches) in the direction of propagation can cause
the shear wave velocities measured at the Hickman sites with

9 Copyright © 20xx by ASME

the SeiBIT system are consistent with these models. These significant energy at other velocities on the dispersion
models indicate that the shear velocity increases mainly due to diagram because this indicates that there is likely
decreasing porosity and to a lesser extent due to decreasing energy at higher order modes. This does not apply to
moisture content. The heavily fouled site had very high spatial aliasing effects near the bottom of the plot.
moisture content and very high clay content. Since the porosity 2. Check for lateral inhomogeneity by conducting two
of clay is quite high, the effective shear velocity will decrease surveys – each with the source on the opposite side of
with increasing clay content. The ratio of the track deflection at the receiver spread.
the two sites is 4.33, and the shear modulus ratio is 1.67. The 3. Inspect HVSR curves and look for discontinuities
difference is most likely due to different modulus values for when higher order modes are suspected.
large and small strain regimes. Track modulus values from the 4. Insure that a good SNR has been obtained for all
emplaced strain gauges and accelerometers were not available recorded waveforms for all frequencies. The
at the time of this writing. We had planned to use a DCP at this acquisition program displays this information.
site to obtain ground truth, but unfortunately it was reallocated 5. Omit Love wave dispersion curves from the inversion
at the last minute and not available at the field site. if they do not exhibit continually decreasing velocity
Nevertheless, the measured low strain shear modulus values are with increasing frequency, and/or if they have a shape
similar to the SASW results reported by [19] where a similar that is significantly different than the Rayleigh wave
receiver spread was used, but differ from the shallow results curves.
from the Ballast Seismic Property Analyzer (BSPA) by a factor 6. Insure that the receivers are well coupled to the ground
of 4-7 which uses a much smaller receiver spread. Tamrakar et and that they are not tilted significantly.
al. [1] used the SASW method and reported Rayleigh wave 7. Using a denser and longer receiver spread will reduce
phase velocities near 200 m/s in clean ballast without ties. On a spatial aliasing affects and increase the resolution of
moderately fouled track they found Rayleigh phase velocities the dispersion diagrams. This in turn will provide
near 150 m/s when using a receiver spread parallel to the rails more information to the inverse modeling step
and about 200 m/s when the spread was perpendicular to the subsequently increase subsurface resolution. A denser
rails. Anbazhagan et al. [7] conducted MASW surveys on and longer receiver spread may also increase system
railway lines and used an impulsive sledgehammer source to portability by enabling the use of a hammer source.
excite frequencies from 25 to 100 Hz. They constructed a test
track with a range of different ballast conditions including clean CONCLUSIONS
ballast, clay fouled ballast and coal fouled ballast. They found A wireless MASW instrument has been developed that
that the measured shear velocities ranged from 100 to 150 m/s. eliminates cumbersome wires in the field and is easy to setup.
The SeiBIT results are consistent with all of these other SASW The use of three-component receivers allows filtering of ground
and MASW measurements with the exception those made with motion for either Rayleigh or Love waves. The receivers are
the BSPA device which has a much smaller volume and depth compatible with a stepped-frequency source or an impact
of investigation. hammer source. A wireless computer-controlled source has
been built that can produce shear excitations in the in-line
Table 2. Measured ballast properties at BNSF sites near direction to generate Rayleigh waves or cross-line direction for
Hickman, NE. Love waves. Experimental results indicate that the use of a
Site name Description Shear Density Low Track stepped-frequency excitation provides better results than a
velocity (kg/m3) strain deflection hammer blow. Software has been written for a laptop computer
(m/s) shear that controls wireless networking, data acquisition, and position
modulus measurement of the sensors. The software also provides data
Hickman Standing water 120 m/s 2030 29.2 MPa 1.65 cm processing and inversion for Rayleigh and Love waves. The
fouled in some spots, kg/m3 4.24 ksi 0.65 in software has been tested at several field sites.
very muddy Field testing has been performed at two different railroad
high clay sites: the FAST track at the TTCI and a BNSF track near
content fouling Hickman, Nebraska. As a result of testing at TTCI, the ground
up to surface coupling mechanism was updated which lead to successful
Hickman Recently 155 m/s 2030 48.8 MPa 0.38 cm testing at the Hickman site. These tests vetted the electronics
clean reconstructed kg/m3 7.08 ksi 0.15 in hardware, mechanical systems, data acquisition software, and
ballast data processing routines. The system was designed to be very
flexible and have the capability to collect different types of
Based on the field test results and theory of the MASW surface wave data with different survey configurations. The
method, a few guidelines can to help insure high quality best performance was obtained by collecting Rayleigh wave
surveys. data using a stepped-frequency source and coupling plates
1. Omit dispersion curve points at frequencies where anchored to the ballast with lag screws.
there is a discontinuity in the curve or where there is

10 Copyright © 20xx by ASME

 In future work further systems optimizations are planned to Track Ballast Subsurface Using Seismic Surface
enhance portability and ease of use. The relationships between Survey Method: Model and Field Studies”. Journal of
shear modulus and other engineering properties such are shear Testing and Evaluation, v. 39, n. 5, Paper ID
strength and track modules will be studied. JTE103472.

ACKNOWLEDGMENTS [8] Wathelet, M., Chatelain, J.-L., Cornou, C., Di Giulio,

The authors would like to acknowledge Hugh Thompson G., Guillier, B., Ohrnberger, M. and Savvaidis, A.,
of the U.S. Department of Transportation’s Federal Railroad 2020. “Geopsy: A User-Friendly Open-Source Tool
Administration and Ted Sussmann of U.S. Department of Set for Ambient Vibration Processing”. Seismological
Transportation’s Volpe National Transportation System Center Research Letters, v. 91, n. 3, pp. 1878–1889.
for their support, advice and assistance during this activity. The
authors would also like to acknowledge the Transportation [9] Park, C. B., Miller, R. D. and Xia, J., 1999.
Technology Center and the BNSF Railway for providing access “Multichannel Analysis of Surface Waves”.
to revenue service track. This work was funded by the Federal Geophysics, v. 64, n. 3, pp. 800-808.
Railroad Administration’s SBIR program.
[10] Park, C. B., Miller, R. D. and Xia, J., 1998. “Imaging
REFERENCES Dispersion Curves of Surface Waves on Multichannel
[1] Tamrakar, P., Azari, H., Yuan, D., and Nazarian, S., Record”. 68th Annual International Meeting Society of
2017. “Implementation of Spectral Analysis of Surface Exploration Geophysicists, Expanded Abstracts, pp.
Waves Approach for Characterization of Railway 1377-1380.
Track Substructures”. Transportation Geotechnics, v.
12, pp. 101–111. [11] Ayers, M., Thompson, M., and Uzarski, D., 1989.
“Rapid Shear Strength Evaluation of In Situ Granular
[2] Stark, T.D., Nazarian, S., Ho, C.L., and Tutumluer, E., Materials”. Transportation Research Board 68th
2013. “Seismic Testing for Track Substructure (Ballast Annual Meeting, Jan., 1989, Washington D.C., Paper
and Subgrade) Assessment for Passenger/Freight no. 880387.
Corridors”. Proceedings of the ASME 2013 Joint Rail
Conference (JRC2013), Knoxville, TN, April. [12] Salgado, R., and Yoon, S., 2003. “Dynamic Cone
Penetration Test (DCPT) for Subgrade Assessment”.
[3] Stark, T.D., Dehlin, T.J., Nazarian, S., Azari, H., Yuan, USDOT, Federal Highway Administration, Report
D., and Ho, C.L., 2014. “Seismic Wave Modeling for FHWA/IN/JTRP-2002/30, SPR-2362, 90 p.
Track Substructure Assessment”. Proceedings of
ASME 2014 Joint Rail Conference (JRC2014), Paper [13] Bendaña, L., Yang, W., McAuliffe, D., and Lu., J.,
No. JRC2014-3776, Colorado Springs, CO, April. 1994. “Interpresting Data from the Falling-Weigh
Deflectometer”. USDOT, Federal Highway
[4] Shearer, P., 2009. “Introduction to Seismology, 2nd Administration, Research Report 160, Report
Edition”. Cambridge University Press, New York, FHWA/NY/RR-94/160, 54 p.
USA, 412 p.
[14] Pei, D., 2007. “Modeling and Inversion of Dispersion
[5] Foti, S., Hollender, F., Garofalo, F., Albarello, D., Curves of Surface Waves in Shallow Site
Asten, M., Bard, P., Comina, C., Cornou, C., Cox, B., Investigations”. U. Nevada – Reno, PhD Diss., 165 p.
Di Giulio, G., Forbriger, T., Hayashi, K., Lunedei, E.,
Martin, A., Mercerat., D., Ohrnberger, M., Poggi V., [15] Stark, T. D., Thompson, H. B., Sussman, T. R.,
Renalier, F., Sicilia, D., and Socco, V., 2018. Banister, R., Ren, Y., and Boucias, Z., 2018.
“Guidelines for the Good Practice of Surface Wave “Monitoring Fouled Ballast Sites”, Transportation
Analysis: a Product of the InterPACIFIC Project”, Bull Research Board 99th Annual Meeting, Washington,
Earthquake Eng, n. 16, p. 2367–2420. DC., TRB 19-04943, 12 p.

[6] Athanasopoulos, G., Pelekis, P., and Anagnostopoulos, [16] Widess, M., 1973. “How Thin is a Thin Bed”.
G., 2000. “Effect of Soil Stiffness in the Attenuation of Geophysics, v. 38, n. 6, p. 1176-1180.
Rayleigh-Wave Motions from Field Measurements”.
Soil Dynamics and Earthquake Engineering, n. 19, p. [17] Oden, C. P., Stark, T. D., Park, C., and Ho, C. L.,
277–288. 2020. “Portable Stiffness/Elastic Modulus
Measurement System – Phase I”. Federal Railroad
[7] Anbazhagan, P., Indraratna, B., and Amarajeevi, G., Administration, SBIR Final Report, June 2020, 46 p.
2011. “Characterization of Clean and Fouled Rail

11 Copyright © 20xx by ASME

[18] Shen, J., Crane, J., Lorenzo, J., and White, C., 2016. [19] Sussmann, T. R., Thompson, H. B., Stark, T. D., Wilk,
“Seismic Velocity Prediction in Shallow (<30 m) S. T., and Ho, C. L., 2017. “Use of Seismic Surface
Partially Saturated, Unconsolidated Sediments Using Wave Testing to Assess Track Substructure
Effective Medium Theory”. J. Environmental and Condition”. Construction and Building Materials, v.
Engineering Geophysics, v. 21, n. 2, pp. 67-78. 155, pp 1250-1255.

12 Copyright © 20xx by ASME