Available via license: CC BY 4.0
Content may be subject to copyright.
EWSHM 2024
11th European Workshop on Structural Health Monitoring
This work is licensed under CC BY 4.0
1
Media and Publishing Partner
Ground penetrating radar for moisture assessment in
railway tracks: An experimental investigation
Giacomo ARCIERI 1, Thomas RIGONI 2, Cyprien HOELZL 1, David HAENER 3, Eleni
CHATZI 1
1 ETH Zurich, Zurich, Switzerland, {giacomo.arcieri, hoelzl, chatzi}@ibk.baug.ethz.ch
2 ETH Zurich, Zurich, Switzerland, thomas.rigoni@usys.ethz.ch
3 Swiss Federal Railways SBB, Bern, Switzerland, david.haener@sbb.ch
Abstract. The assessment of the condition of railway substructure is particularly
linked to the tracking of degradation mechanisms that are tied to the phenomenon of
water infiltration, causing moisture accumulation in the underlying layers. Such
deterioration incurs substantial costs, while reducing the useful life of the track
infrastructure. In tracking such processes, a promising solution is found in the use of
Ground Penetrating Radar (GPR) technology, which employs non-invasive radar
pulses to map the subsurface, and possibly detect water infiltration. Despite previous
efforts in this direction, the technology remains relatively under-explored and lacks
comprehensive studies to establish its suitability for this task. In this work, we present
a detailed experimental analysis carried out on a controlled railway track section,
which was built by the Swiss Federal Railways for this campaign. Our findings
demonstrate that GPR systems can effectively detect moisture infiltration in railway
tracks, although several challenges are to be addressed for the development of
accurate, automated procedures. The experimental data is compared against numerical
simulations of the physical configurations of the test railway track, which were
conducted via the gprMax software.
Keywords: Ground Penetrating Radar (GPR), Structural Health Monitoring (SHM),
Railway Infrastructure, Non-Destructive Evaluation, gprMax.
Introduction
The health monitoring of railway networks offers a means to ensuring high-quality service,
avoiding safety risks, and optimally planning maintenance actions to minimize life-cycle
costs [1]. Monitoring of the substructure is particularly linked to the tracking of degradation
[2], which often stems from water infiltration, causing moisture accumulation in the ballast
and underlying layers. Such deterioration induces substantial expense and reduced useful life
of the track infrastructure.
More info about this article: https://www.ndt.net/?id=29747
e-Journal of Nondestructive Testing - ISSN 1435-4934 - www.ndt.net
https://doi.org/10.58286/29747
2
Clearly, early and accurate detection of such moisture accumulation offers
tremendous potential for optimizing asset management of railway networks. However,
common methods for measuring and tracking the deterioration of the railway substructure
are invasive, costly, and time-consuming [3], posing challenges which reduce the ability to
adopt preventive maintenance strategies. It is, thus, important to improve condition
assessment schemes; one means to doing so is via affordable and non-invasive procedures.
Ground Penetrating Radar (GPR) [4] technology employs non-invasive radar pulses
to map the subsurface for detection of water infiltration. Roberts et al. [5]–[7] first showed
the large potential of this technology in detecting railroad ballast deterioration via
comparison of GPR data and ground truth ballast condition. Fontul et al. [8] use GPR data to
characterize track sections and layers of deteriorated areas that are detected by track geometry
measurements. Benedetto et al. [9] analyze the different dielectric permittivity at varying
fouled ballast levels from GPR measurements in laboratory experiments. Ciampoli et al. [10]
study the effect of the presence of fouling and water infiltration on GPR data of the ballast
layer of railway tracks via a test-site campaign with ground truth deteriorated ballast levels.
Given this potential, mobile solutions, such as train-based GPR systems, offer a
highly promising direction toward development of preventive and automated monitoring of
railway infrastructure. Nevertheless, despite previous efforts in this direction, the technology
remains relatively under-explored and lacks comprehensive studies which verify its
suitability for this task. Ciampoli et al. [11] report some of the main challenges and the current
state-of-the-art related to GPR data interpretation for railway ballast investigation and
concludes that further research is still required. They report on no consensus currently found
regarding a standardized procedure for automatic inference of reliable indicators of railway
health from GPR observations. Moreover, previous works focus on detection of deterioration
effects in the ballast layer. This is already an outstanding pathological condition of railway
tracks, often caused by degradation of underlying layers. Further studies that investigate the
capability of inferring deterioration symptoms in subjacent layers are thus necessary.
In this work, we report on an extensive experimental analysis conducted on a
controlled railway track section, built by the Swiss Federal Railways for this campaign. The
humidity condition of the railway track was artificially altered to reach different levels of
water content. GPR measurements were gathered under the varying conditions, with ground
truth assessed through lab tests on collected samples. GPR data with complete ground truth
labels deliver a rare benchmark, which can enhance understanding of this technology. In
particular, this work focuses on GPR data interpretation for detection of different moisture
levels of the substructure, not considered in the literature thus far. Our findings show that
GPR systems can effectively detect moisture infiltration in railway tracks, although some
challenges are to be addressed for the development of accurate, automated procedures.
Finally, we reproduce the experimental results with numerical simulations of the physical
configuration of the test railway track. The simulated GPR data generated via the gprMax
[12] software, under the varying moisture conditions, is thus compared with the ground truth
experimental data.
1. The Railway Track
A typical railway track comprises multiple components, as illustrated in Fig. 1. The rails are
tightened onto sleepers, which can be comprised of three materials, namely steel, concrete,
and wood. Sleepers and rails lie on a ballast layer, i.e., (mainly) crushed limestone of roughly
3-7 cm diameter (pristine), of varying thickness, typically within 20-35 cm.
3
Fig. 1. General structure of a railway track.
Beneath the ballast layer lies the substructure of the track, which can in turn be
composed of different layers, depending on the age of construction, travel speed, loads, and
other parameters. In Fig. 1 a railway track typical of the Swiss railway network is reported,
with the substructure formed by a single formation layer termed “PSS” (from the German
Planumsschutzschicht). The latter is a highly compacted layer of gravel and sand, usually 25-
40 cm thick, of very low water permeability when in perfect condition. The substructure
plays a key role in the degradation of the railway track, as it bears loads from the
superstructure, facilitate water drainage from the ballast, acts as a filter to block fines from
the subsoil, and reduces the impact of freezing. Therefore, early detection of defects in the
track substructure is essential for optimizing asset management. Finally, beneath the
substructure generally lies the subsoil (natural ground).
Fig. 2. Degradation effects in railway tracks: accumulation of water (left), fouling (center), vegetation (right).
The main degradation effects in the railway track are fouling and water accumulation,
as displayed in Fig. 2. The former is primarily located in the ballast and is defined as the
presence of fine material infiltration, for example due to ballast attrition or ascending fines
from underlying layers. The latter pertains to standing water due to a blocked drainage. This
is typically caused by a weakened substructure and/or presence of fouling, which traps the
water in the railway track. In fact, water only accumulated in the ballast layer, with no
presence of fouling and pristine substructure is usually drained off relatively quickly.
However, persistent water accumulation leads to permanent damages of the substructure with
subsequent distortion of the track geometry. As a result, a renewal of the railway track
becomes the inevitable long-term solution. Other typical problems due to the degradation
process of the railway tracks are the presence of vegetation, roots, and gaps, which are
however generally consequences of the moisture accumulation.
2. Ground Penetrating Radar
GPR is a non-destructive geophysical method used to map the subsurface. A GPR system is
composed of an antenna, which is in turn composed of a transmitter and a receiver. The
transmitter sends an Electro-Magnetic (EM) impulse into the subsurface, with frequency that
depends on the specifics of the antenna. This is commonly in the range of 100MHz to 2GHz,
4
where low frequency waves penetrate deeper at the cost of lower resolution and vice versa
for high frequency waves. The EM signal disseminate downwards, with velocity that
depends on the dielectric permittivity of the material encountered:
=
√
where is the velocity of light in free space. The velocity of the signal thus reaches its
maximum in air (= 1) and its minimum in water (=80).
Changes in the dielectric permittivity (i.e., changes of encountered materials) result in a
portion of the signal being reflected, with its amplitude depending on the magnitude of the
change. The reflected signal is recorded by the receiver for a predetermined duration (e.g.,
50 ns), termed Two-Way-Traveltime (TWT). The single waveform recorded by the receiver
forms the so-called A-scan. By moving the antenna along an -direction, several A-scans are
recorded, which, once stacked, form the so-called B-scan.
As water and air are at the two extremes of the dielectric permittivity spectrum, GPR
forms a promising technology to detect moisture accumulation in railway tracks. Likewise,
GPR can potentially be used to detect gaps or presence of fouling (i.e., spaces between ballast
stones filled with fine material, resulting in significantly higher dielectric constants), which
however lies outside the scope of this work.
3. Experiment Set-up
An experimental test is designed to investigate the suitability of the GPR technology in
detecting moisture accumulation in the substructure. For this experiment, a 1GHz GSSI
antenna is used, in order to achieve the desired performance of depth and high resolution.
Fig. 3. Design of the experiment.
The experiment is carried out on a railway track built by the SBB for this test-
campaign, displayed in Fig. 3 (first figure from the left). The bottom layer is constituted by
a PSS layer roughly 20 cm thick (measured in the center), with a transversal inclination of 3-
5%, which generally serves facilitation of water drainage. Below the PSS layer lies the
asphalted road surface. On top of the PSS lies the constructed ballast layer, formed by ballast
stones (3-7 cm diameter) in perfect and dry condition. The rails lie on the ballast layer and
include 12 sleepers of different materials (4 × steel, 4 × wood, 4 × concrete). The distance
between two sleepers is approximately 0.6m, while this distance is doubled between the
different sleeper configurations. The overall dimensions of the track setup are approximately
10m × 4m × 0.6m.
Samples of the initial dry condition of ballast stones and PSS are collected (second figure
in Fig. 3) under each of the three sections, with their exact weight and depth annotated. GPR
measurements are gathered over the entire track length (third figure in Fig. 3). Subsequently,
the first watering phase initiates and the water content of the steel section is increased by
means of an irrigation device (fourth figure in Fig. 3) and sensors placed on the PSS surface
to track when the desired water content is achieved. At the end of the watering phase, new
5
GPR measurements over the entire track are gathered, where the section over the steel
sleepers presents a significantly higher water content, while for the other sections this is
unchanged. New samples of ballast and PSS (both at the surface and at a greater depth) are
collected. The entire procedure is repeated for the wooden section and, next, for the concrete
section. All collected samples are analyzed in the laboratory to assess their ground truth water
content (fifth figure in Fig. 3). This is achieved by means of a furnace to dry out the samples
until their weight does no longer change and it is computed as percentage difference between
the initial and the final weight.
Table 1 reports the ground truth water content across all collected samples. As expected,
the water content of the ballast samples does not significantly change, as water does not
permeate the stones and the moisture on the stones’ surface is minimal. On the other hand,
the irrigation phases successfully increased the water content of the PSS, both at the surface
and at deeper locations.
Table 1. Ground truth water content of the collected samples.
Samples
Water content dry
After Watering (PSS surface)
After Watering (PSS deep)
Ballast Steel
0.2%
0.8%
-
Ballast Wood
0%
0.7%
-
Ballast Concrete
0%
0.5%
-
PSS Steel
3.8%
11.7%
9.9%
PSS Wood
3.4%
10.4%
8.8%
PSS Concrete
3.4%
9%
8%
4. Results
This section reports on the results of the GPR measurements of the previously described
experiment. Fig. 4 displays the B-scans gathered data of the initial dry condition (top left),
after watering over the steel section (top right), after watering over the wood section (bottom
left), and after watering over the concrete section (bottom right).
Firstly, it is evident that the type of sleeper, highlighted in the figures, affects the
underlying portion of the B-scan in a different manner. The proper accounting and
elimination of the effects of sleepers (and rails) in GPR measurements forms ongoing
research [11]. In the figure of the initial dry condition, the sleepers and the PSS surface are
marked with dashed black and red lines, respectively. After watering over the steel sleepers,
the water content of the PSS layer increases from 3.8% to 11.7% on the surface and to 9.9%
at a deeper location. The higher water content leads to a higher dielectric permittivity of the
layer. As a result, the signal takes longer time to travel in the PSS. In the top right figure, it
is reported the PSS surface prior to watering (higher line) and after watering (lower line).
The line of the PSS surface thus appears significantly lowered, as an effect of the increased
content of water, under the steel section, while this is unchanged in the other two sections,
which have indeed not undergone any water increase. Likewise, after irrigating over the wood
section (water content from 3.4% to 10.4%) and over the concrete section (water content
from 3.4% to 9%), the line of the PSS layer shows visible decreases. Interestingly, we notice
a decrease of the PSS surface in the GPR images that strongly correlates to the increase of
water content (e.g., the PSS under the concrete section endures the smallest increase of water
content, which indeed results in a significantly smaller decrease of the PSS line, when
compared to the other sections).
6
Fig. 4. B-Scans of the GPR measurements of the experiment: initial dry condition (top left), after watering
over the steel section (top right), after watering over the wood section (bottom left), after watering over the
concrete section (bottom right).
5. gprMax Simulation
Fig. 5. Simulation of the physical model of the test railway track.
The experimental results from the GPR measurements are finally compared in this section
with numerical simulations of the physical model of the test railway track, conducted via the
gprMax software. Fig. 5 displays the model of the test railway track, where all physical
characteristics have been faithfully reproduced. The rails lie on the concrete, wooden, and
steel sleepers. Beneath the sleepers, a ballast layer has been reproduced with spheres of
varying diameter (3-7cm), which have been compacted by means of PyChrono [13]. The
ballast in turn lies on a layer of PSS of slightly varying thickness, respecting the real-world
measurements, an asphalted layer, and finally gravel.
B-scans of the GPR measurements are simulated via gprMax by assuming a 1GhZ
dipole source-receiver (which differs from a more complex 1GHz GSSI antenna model,
which however is not yet implemented in gprMax). Fig. 6 reports the simulated B-scans of
the GPR data over the track with initial dry condition (top left), and then with higher water
content of the PSS under the sleepers of steel (top right), wood (bottom left), and concrete
(bottom right). All conditions reflect the ground truth values from the samples analyzed in
the laboratory.
7
Fig. 6. B-scans of the simulations of the GPR data via gprMax on the track model with dry condition (top left)
and increased water content under the steel (top right), wooden (bottom left), and concrete (bottom right)
sleepers.
A first important difference that can be noticed when comparing the simulated GPR
data with the real-world instances is that the signal appears to reach the rails significantly
faster for the former, around 5ns, while the latter requires 10-12ns. We assume this difference
is due to the real-world antenna and its offset, whose effect is not present in the simulations.
As a result, the rest of the image appears to be shifted, with the PSS line displayed around
17.5ns. Nevertheless, the simulated GPR images exhibit a similar decrease along the PSS
line, compared to the experimental data, when the water content is increased (the PSS line
appears more defined in the simulation results, as a result of the modeling assumptions, which
consider predefined shapes and average water contents, while in the real world the layer is
more fractal-shaped and water content increase may not be homogenous over the considered
section). While some limitations are present, this shows the tremendous potential of GPR
simulations to favor the understanding of its technology and to generate useful datasets, for
which real-world tests might be too costly.
6. Discussion and Conclusion
This work reports on an experimental investigation on the use of the GPR technology to
assess the moisture content in the railway substructure. GPR data is gathered on a test railway
track at the initial dry condition and, subsequently, at increasing water contents of the
substructure. The latter is assessed via ground truth tests in the laboratory on collected
samples.
The experimental results demonstrate the potential of adoption of GPR technology for
moisture assessment in the track substructure. It should be clear that assessing the water
content of the track substructure from a single GPR measurement remains a challenging task,
given the limited information. While, for example, the PSS layer might exhibit a stronger
reflection in presence of water accumulation, it does not seem obvious how to assess the
water content from a single image (e.g., a lowered PSS line due to the moisture effect might
8
be mistaken for the ground truth PSS depth). However, the goal of using the GPR aboard a
track measurement vehicle is to collect more GPR observations at a given location over time,
which will allow to track the evolution of the measurements along the depth of the layers that
form the track substructure, thus allowing to detect increments due to moisture accumulation.
In particular, the differences in the TWT of the signal seem to strongly correlate with the
differences of water content, enabling detection of the increase of moisture content over time.
Early detection of moisture accumulation in the substructure can thus form a potent tool
in support of preventive maintenance actions. The partial and uncertain information
corresponding to such GPR observations naturally frames the problem as a Partially
Observable Markov Decision Process (POMDP) [14]. Future work will focus on the creation
of a simulated environment of the railway track maintenance problem based on GPR
observations. This will open up large space for the implementation of innovative solutions
for the optimization of proactive maintenance policies, such as schemes that are based on
reinforcement learning [15].
References
[1] C. Hoelzl, V. Dertimanis, M. Landgraf, L. Ancu, M. Zurkirchen, and E. Chatzi, On-board monitoring
for smart assessment of railway infrastructure: A systematic review. Elsevier Inc., 2022.
[2] C. Hoelzl et al., “Fusing Expert Knowledge with Monitoring Data for Condition Assessment of Railway
Welds,” Sensors, vol. 23, no. 5, pp. 1–31, 2023, doi: 10.3390/s23052672.
[3] W. Shao, A. Bouzerdoum, S. L. Phung, L. Su, B. Indraratna, and C. Rujikiatkamjorn, “Automatic
classification of ground-penetrating-radar signals for railway-ballast assessment,” IEEE Trans. Geosci.
Remote Sens., vol. 49, no. 10 PART 2, 2011, doi: 10.1109/TGRS.2011.2128328.
[4] M. Robinson, C. Bristow, J. Mckinley, and A. Ruffell, “Ground Penetrating Radar,” Br. Soc.
Geomorphol., vol. 5, pp. 1–26, 2013.
[5] R. Roberts, I. Al-Qadi, E. Tutumluer, and A. Kathage, “Ballast Fouling Assessment Using 2 Ghz Horn
Antennas–Gpr and Ground Truth Comparison From 238 Km of Track,” 2006.
[6] R. Roberts, I. Al-qadi, and E. Tutumluer, “Railroad Structure Characterization using GPR in Alaska :
A Case History,” in 12th International Conference on Ground Penetrating Radar, 2008, pp. 4–12.
[7] R. Roberts, J. Rudy, I. Al-Qadi, E. Tutumluer, and J. Boyle, “Railroad Ballast Fouling Detection Using
Ground Penetrating Radar – A New Approach Based on Scattering from Voids,” in Ninth European
Conference on NDT, 2006, p. 8p.
[8] S. Fontul, E. Fortunato, F. De Chiara, R. Burrinha, and M. Baldeiras, “Railways Track Characterization
Using Ground Penetrating Radar,” in Procedia Engineering, 2016, vol. 143, doi:
10.1016/j.proeng.2016.06.120.
[9] A. Benedetto, F. Tosti, L. Bianchini Ciampoli, A. Calvi, M. G. Brancadoro, and A. M. Alani, “Railway
ballast condition assessment using ground-penetrating radar – An experimental, numerical simulation
and modelling development,” Constr. Build. Mater., vol. 140, pp. 508–520, 2017, doi:
10.1016/j.conbuildmat.2017.02.110.
[10] L. B. Ciampoli, A. Calvi, and F. D’Amico, “Railway ballast monitoring by GPR: A test-site
investigation,” Remote Sens., vol. 11, no. 20, pp. 1–14, 2019, doi: 10.3390/rs11202381.
[11] L. B. Ciampoli, A. Calvi, F. D’Amico, and F. Tosti, “GPR Data Collection and Processing Strategies
for Railway Ballast Evaluation,” 2020, doi: 10.1109/TSP49548.2020.9163588.
[12] C. Warren, A. Giannopoulos, and I. Giannakis, “gprMax: Open source software to simulate
electromagnetic wave propagation for Ground Penetrating Radar,” Comput. Phys. Commun., vol. 209,
pp. 163–170, 2016, doi: 10.1016/j.cpc.2016.08.020.
[13] S. Benatti et al., “PyChrono and gym-chrono: A Deep Reinforcement Learning Framework Leveraging
Multibody Dynamics to Control Autonomous Vehicles and Robots,” 2022.
[14] G. Arcieri, C. Hoelzl, O. Schwery, D. Straub, K. G. Papakonstantinou, and E. Chatzi, “Bridging
POMDPs and Bayesian decision making for robust maintenance planning under model uncertainty: An
application to railway systems,” Reliab. Eng. Syst. Saf., vol. 239, no. November 2023, p. 109496, 2023,
doi: 10.1016/j.ress.2023.109496.
[15] G. Arcieri, C. Hoelzl, O. Schwery, D. Straub, K. G. Papakonstantinou, and E. Chatzi, “POMDP
inference and robust solution via deep reinforcement learning: An application to railway optimal
maintenance,” 2023. [Online]. Available: http://arxiv.org/abs/2307.08082.
