Ancient Stone Bridge Surveying by Ground-Penetrating

Radar and Numerical Modeling Methods

Mercedes Solla1; Belén Riveiro2; Henrique Lorenzo3; and Julia Armesto4

Abstract: Bridges are considered necessary engineering structures because they connect separated lands to improve economic and social de-
Downloaded from ascelibrary.org by University of Central Florida on 11/25/20. Copyright ASCE. For personal use only; all rights reserved.

velopment. In Spain, many of the bridges in service within the network of transport are masonry arch bridges built in ancient times. In addition to
their age, the stability of these remaining bridges is questionable because of the changing loading conditions; therefore, they require periodic
assessment of the condition state. Moreover, some of these bridges are considered a part of the cultural heritage of a region, so nondestructive
evaluation is required to preserve their historical character. In this work, a medieval stone bridge in the Galician territory of Spain was evaluated
using ground-penetrating radar, supported by a detailed geometric survey performed through a terrestrial laser scanner. The results revealed
unknown geometrical data and hidden characteristics, including the thickness of ring stones in the interior of the vault, as well as the presence
of ancient arches and restorations. To assist in the interpretation, finite-difference time-domain modeling was used, where realistic models were
built from the accurate geometry provided. The synthetic data obtained were compared with the field data, which allowed for the identification
of unknown structural details. DOI: 10.1061/(ASCE)BE.1943-5592.0000497. © 2014 American Society of Civil Engineers.
Author keywords: Nondestructive evaluation (NDE); Ground-penetrating radar (GPR); Laser scanning; Finite-difference, time-domain
(FDTD) modeling; Masonry arch bridges.

Introduction technique, while still allowing for preservation of their historical

character (Melbourne et al. 2006). There are different NDT tech-
Stone masonry construction was the dominant technique for cen- niques used in the evaluation of masonry arch bridges such as sonic/
turies during the Roman Era, and this represents the most common ultrasonic, electromagnetic, and electrical methods, as well as in-
typology built throughout history in Europe (Ural et al. 2008). There frared thermography (Clark et al. 2003; Colla et al. 2001), but each
are still a large number of these bridges in service within the trans- method provides different information regarding the physical, mech-
port network. Many of these structures have experienced unknown anical, or electrical properties of the medium (Alampalli 2010).
structural degradation because of increases in traffic load and intense Ground-penetrating radar (GPR) is a quite commonly used NDT
vibrations that are significantly higher than what was estimated for method in bridge inspection that produces an overall internal image
their original design (Brencich and De Francesco 2004). Therefore, and can provide a compromise between either high penetration
these older bridges require special attention and ongoing diagnosis depth, high resolution, or some middle ground between both in
of their changing structural integrity to help with their preservation a relatively quick time (Flint et al. 1999). Mainly, this technique
and conservation. has provided promising information about hidden geometry, bridge
Ancient stone arch bridges usually represent the traditional ar- foundations, ring stone thickness, relatively moist areas, and fill
chitectural heritage of a region (Ural et al. 2008), and they are often conditions in stone masonry bridges (Lubowiecka et al. 2011; Orbán
listed as historical monuments, valued tourist attractions, or local and Gutermann 2009; Fernandes 2006). To date, however, there
landmarks (Mabon 2002). Consequently, the use of nondestructive are few published studies, probably because of the complexity of
testing (NDT) is recommended for their diagnosis to prevent po- various geometrical structures and the difficulty in interpreting the
tential structural damage that can result from more invasive testing resultant signals. In heterogeneous structures like stone bridges, the
analysis and interpretation of the GPR data can be complicated be-
1 cause of the existence of many reflections such as multiples, corner
Assistant Professor, Dept. of Materials Engineering, Applied Mechan-
reflections, and the diffractions from the heterogeneous backfill. For
ics and Construction, School of Industrial Engineering, Univ. of Vigo,
Torrecedeira 86, CP 36208 Vigo, Spain (corresponding author). E-mail: such circumstances, numerical modeling has become a valuable
[email protected] interpretational tool to more thoroughly understand the radar wave
2
Assistant Professor, Dept. of Materials Engineering, Applied Mechan- propagation phenomena in which the finite-difference, time-domain
ics and Construction, School of Industrial Engineering, Univ. of Vigo, (FDTD) technique has evolved into one of the most advanced
Torrecedeira 86, CP 36208 Vigo, Spain. E-mail: [email protected] modeling tools (Jol 2009). Nevertheless, although it is known that
3
Associate Professor, Dept. of Natural Resources and Environmental FDTD can be an effective tool used to improve GPR signal inter-
Engineering, School of Forestry Engineering, Univ. of Vigo, A Xunqueira pretations (Millard et al. 1998), few studies have been published to
s/n., CP 36005 Pontevedra, Spain. E-mail: [email protected] date for the evaluation of masonry arch bridges, with some notable
4
Associate Professor, Dept. of Natural Resources and Environmental exceptions (Solla et al. 2012; Diamanti et al. 2008).
Engineering, School of Mining Engineering, Univ. of Vigo, Rua Maxwell,
This work presents the GPR evaluation of the medieval Monforte
Lagoas-Marcosende, CP 36310 Vigo, Spain. E-mail: [email protected]
Note. This manuscript was submitted on September 7, 2012; approved Bridge. This is a sixteenth century masonry arch bridge in the
on April 5, 2013; published online on April 8, 2013. Discussion period open southeast of Galicia. The bridge has four semicircular or segmental
until June 1, 2014; separate discussions must be submitted for individual arches between a 7.0- and 11.0-m span, triangular buttresses, and
papers. This paper is part of the Journal of Bridge Engineering, Vol. 19, No. an upward arching profile, as shown in Fig. 1(a). This bridge was
1, January 1, 2014. ©ASCE, ISSN 1084-0702/2014/1-110–119/$25.00. selected because it is an ancient structure that provides contemporary,

110 / JOURNAL OF BRIDGE ENGINEERING © ASCE / JANUARY 2014

J. Bridge Eng., 2014, 19(1): 110-119

 Fig. 1. Monforte Bridge: (a) general view from the upstream side; (b) breakwater built at the left margin from upstream (images by Mercedes Solla)

strategic road access to the city. Therefore, the structure is supporting data collection direction. By using the CO mode, the technician
Downloaded from ascelibrary.org by University of Central Florida on 11/25/20. Copyright ASCE. For personal use only; all rights reserved.

traffic loads much heavier than in former times, and diagnosis is moves the transmitting and receiving antennas over the ground along
required to ensure its stability and durability. Moreover, Alvarado a specific direction with a constant distance between them. The
et al. (1989) have referenced the probable existence of two additional distance between the profiles was set to 1 m. For the 250-MHz data,
arches within the structure, indicating modifications have been made the trace-distance interval and the total time window were equal to 2
to the original structure that are no longer externally visible. It is cm and 200 ns, respectively, and were defined by 515 samples per
expected that these two nonvisible arches remain enclosed within the trace. For the 500-MHz data, the trace-distance interval and the total
breakwater built in the nineteenth century at the left margin from time window were equal to 2 cm and 100 ns, respectively, this ti-
upstream [Fig. 1(b)]. Accordingly, the main purpose was to analyze me defined by 678 samples per trace. The survey was conducted
the potential of GPR in obtaining previously unknown, inner details using the antenna mounted on a survey cart with an encoder-based
that can be valuable information for developing future conservation wheel serving as a distance measurement instrument (DMI) to
and strengthening techniques. This work also included the use of measure the profile lengths, as well as to control the 2-cm spatial
FDTD modeling for an exhaustive interpretation of the field GPR data. sampling, or trace-distance interval, along each GPR profile.
What is more, to obtain a rigorous comparison between both field and ReflexW 5.6 software (Sandmeier 2007) was used to process the
synthetic data, the key input for the computation of FDTD modeling data, applying the following sequence: time-zero correction, temporal
was a raster image containing a geometric description of the filtering (dewow), gain application (gain function with linear and
structural elements to simulate. A higher accuracy and spatial res- exponential part), spatial filtering (subtracting average), and a band-
olution (pixel size) in the image used for the radar simulation will pass (butterworth) filter. The objective was to correct the down-
provide more realistic results. In that sense, some previous works shifting of the radar section caused by the air-ground interface and to
(Solla et al. 2011) presented orthoimages as suitable and accurate amplify the received signal, as well as to remove both low- and high-
sources of information for FDTD. Terrestrial photogrammetry and frequency noise in the vertical (temporal) and horizontal (spatial)
laser scanning are the two geomatic methods most commonly used for directions. In addition, corrections were applied for topography and
the creation of metric images of built-up structures. Particularly, the tilt of the antenna to improve the accuracy in imaging subsurface
terrestrial laser scanning (TLS) has offered an important development features because the bridge has an upward-arching profile [Fig. 1(a)].
in the last decade because of the huge amount of geometric data it can The elevation variations of the GPR profiles were obtained from the
collect automatically. This technique was demonstrated to be very laser scanning data. Without performing these static corrections, the
useful as surveying methods, with millimeter accuracy, in civil perceived locations of subsurface features are incorrectly positioned
engineering applications as demonstrated by González-Jorge et al. within each GPR profile, and the resultant shapes of features are
(2011) and Lindenbergh et al. (2009). distorted. Table 1 shows the processing sequences applied and the
parameters used for the filters.
Methodology Laser Scanning Approach
This work aims toward evaluation of a masonry arch bridge structure For the geometric reconstruction of Monforte Bridge, a long-range
by conducting a NDT of its interior using GPR. The approach also terrestrial laser scanner (Riegl, Horn, Austria) was used. Model
included the use of advanced FDTD modeling by creating simu- LMS Z-390i, whose technical specifications can be seen at Riveiro
lations to make the signal more meaningful and to improve the in- et al. (2011), was used because it has a range of measurement between
terpretation of the field GPR data. By using the precise geometry 1.5 and 400 m in optimal illumination conditions, which ensured the
provided by a laser scanning survey of the structure, more realistic measurement of the bridge geometry from riverbanks. This instrument
synthetic models were built to get the best approximations compared has a nominal accuracy (1s) of 0.6 cm and average repeatability of 0.4
with the field data. cm at a range of 50 m, which can be improved if the measured points
are fitted to surfaces (Gordon and Lichti 2007). By mounting a cali-
brated camera on the top of the laser scanner, red/green/blue (RGB)
GPR Survey
information could be added to each point. When the instrument was
Data were collected with a RAMAC/GPR system (Malå Geoscience, stationed in the field, the first task was the camera mounting cali-
Järfälla, Sweden) using shielded antennas with center frequencies bration. By means of this operation, the external orientation of the
of 250 and 500 MHz, respectively. These two antenna center fre- camera (position and orientation) was obtained relative to the scanner
quencies were selected to achieve both resolution and penetration needs coordinate system.
for analyzing both the backfill and the bridge foundations. Two parallel The restricted access to the structure forced the design of an opti-
longitudinal profiles were collected, in opposite directions along the mized network of scanner stations. This network was formed by seven
pathway of the bridge (longitudinal lines along the pathway of the bridge different scanner positions (Fig. 2) to ensure that the complete surface
in Fig. 2), with each antenna. Data were acquired using the common of the bridge was recorded. Fig. 3 shows a summary of the detailed
offset (CO) mode with the antenna polarization perpendicular to the scans performed to cover the whole bridge surface. Additionally, the

JOURNAL OF BRIDGE ENGINEERING © ASCE / JANUARY 2014 / 111

J. Bridge Eng., 2014, 19(1): 110-119

Downloaded from ascelibrary.org by University of Central Florida on 11/25/20. Copyright ASCE. For personal use only; all rights reserved.

Fig. 2. Plan view of the test site showing the seven different scanner positions used to cover the whole structure; in addition, the two parallel
longitudinal GPR profiles acquired, in opposite directions, along the pathway of the bridge (images by Belén Riveiro)

Table 1. Data Processing (ReflexW) Applied to GPR Data Acquired with scene, control points could be detected for the registration of
the 250- and 500-MHz Antennas different scans and also for the mounting and calibration of the
Step 250-MHz details 500-MHz details camera (if needed).
• High-resolution scan of control points: This high-resolution scan of
1 Time-zero correction Time-zero correction control points is obtained using the previous, lower-resolution scan
2 Dewow filtering Dewow filtering where the position of control points was detected. These control
3 Gain function (linear 1.44 Gain function (linear 1.73 points were manufactured of reflective material that scatters the
and exponential 1.44) and exponential 1.44) laser radiation clearer than the surrounding points. These targets
4 Subtracting average Subtracting average were easily identified in the point cloud because of the amplitude of
(100 average traces) (200 average traces) the laser echo (intensity). Once identified, a forced scan of each
5 Band-pass Butterworth Band-pass Butterworth point was created using the maximum resolution of the instrument,
(low cutoff: 120 MHz; (low cutoff: 260 MHz; so the center of gravity of the target was accurately identified by
upper cutoff: 350 MHz) upper cutoff: 600 MHz) digital image processing for the subsequent matching.
6 Static corrections Static corrections • Color acquisition: Because RGB was needed for the construc-
(velocity 5 0:11 m=ns) (velocity 5 0:11 m=ns tion of orthoimages, a calibration of camera position had to be
performed. This was done through a spatial resection (Luhmann
2006) that practically consists of identifying at least four control
geometric basis for the subsequent FDTD simulation would be the points in both scan [three-dimensional (3D) space] and photog-
planar orthoimages of bridge walls, so it was carefully checked that raphy [two-dimensional (2D) space] for the computation of the
hidden areas or shadows did not appear for the final model in these coordinate transformation. After calibration, images were taken
masonry walls. of the bridge, adjusting illumination conditions until a homoge-
The data acquisition from each station consisted of the following neous panorama of the bridge structure for the subsequent pro-
steps: duction of orthoimages was obtained.
• Overview scan: One that collects points of the whole field of view • Detailed scan: Finally, a point cloud of the bridge was collected
of the scanner with the minimum resolution. In this first scanned with the resolutions obtained as shown in Fig. 3.

112 / JOURNAL OF BRIDGE ENGINEERING © ASCE / JANUARY 2014

J. Bridge Eng., 2014, 19(1): 110-119

Downloaded from ascelibrary.org by University of Central Florida on 11/25/20. Copyright ASCE. For personal use only; all rights reserved.

Fig. 3. Detailed scans to complete survey of Monforte Bridge with size of point cloud, time of scanning, and angular resolution (images by Belén
Riveiro)

The postprocessing of data started with the alignment of different

scanner positions in a common coordinate system. From each sta-
tion, the instrument collects data in a coordinate system centered in
the scanner so each point cloud belongs to a different coordinate
system. Because of the control points precisely measured from each
station (minimum of four common targets), it is possible to compute
a 3D conformal transformation (Luhmann 2006) between each point
cloud and the global point cloud.
Once the global point cloud is mounted, noisy points were de-
leted from the model. First, a manual cleaning was performed to
delete all those points not belonging to the structure of interest.
Then, automatic filters had to be applied to homogenize the density
of the point cloud. Because the geometric input data for the FDTD
modeling are the orthoimages, the point cloud model had to be
converted into a model of surfaces by means of a 2:5D triangulation
(Rebay 1993). Because of the errors in the alignment of scans,
a double surface effect in the overlapped areas of walls may occur.
To avoid the influence of this error in the triangulation, a 2:5D raster
filter was applied that generates a unique 2:5D rasterized point cloud
of the given source data. Fig. 4. Three-dimensional realistic model of the (a) Monforte Bridge;
After triangulation, the model was textured with the RGB in- (b) orthoimages produced for the FDTD modeling (images by Belén
formation recorded by the digital camera so the 3D realistic model of Riveiro)
the bridge could be created [Fig. 4(a)]. To transfer this geometric

JOURNAL OF BRIDGE ENGINEERING © ASCE / JANUARY 2014 / 113

J. Bridge Eng., 2014, 19(1): 110-119

 model to the subsequent FDTD modeling, orthoimages of main For simulations, the electromagnetic properties were determined
planes of the structure were produced. In particular, because of the to characterize the propagation media from the acquired field GPR
configuration of GPR surveying of the bridge, orthoimages of both data. The metric information provided by laser scanning allowed for
upstream and downstream walls were created as shown in Fig. 4(b). calculation of radar wave velocities for granitic masonry in different
zones of the structure. Knowledge of the most appropriate radar
Synthetic Models and Simulations wave velocity allows for a more complete interpretation of the field
data. Velocity was determined by adapting a diffraction hyperbola
The precise geometry [2D orthoimage; Fig. 4(b)] provided by laser (Sandmeier 2007) to the hyperbolic reflection produced by the arch-
scanning was the input to elaborate on the FDTD modeling. The air interface (Fig. 5). The arch span geometry provided by laser
purpose was to design realistic, large-scale, synthetic GPR models to scanning to define the hyperbolic shape to fit was used. Nevertheless,
assist in the interpretation of the processed field data. However, some of the reflections generated from the arch-air interface present
simulating large-scale and realistic models requires high-performance an asymmetric shape on each side of the keystone, so the average
computing to obtain results in a reasonable time. The need to discretize velocity was determined by adapting a hyperbolic shape to each half
Downloaded from ascelibrary.org by University of Central Florida on 11/25/20. Copyright ASCE. For personal use only; all rights reserved.

the volume of the problem space and the staircase approximation of of the reflection, resulting in two different signal velocities as shown
curved interfaces to the real boundary result in excessive computer in Fig. 5.
memory requirements and large execution times (Giannopoulos The obtained radar wave velocity allowed both the dielectric
2005). Thus, the synthetic models were created using a parallelized constant (ɛr ) and the dielectric conductivity (s) in the media to be
version, based on a hybrid message passing interface (MPI) and determined. By considering a nonconducting material, the dielectric
open multiprocessing (OpenMP) parallel programming, of GprMax, constant was calculated from
which is an electromagnetic wave simulator for GPR using the  2
c
FDTD method (Giannopoulos 2006). Using a mixed model of ɛr ¼ v (1)
parallelization, where different traces are computed at different
nodes of a cluster, a significant reduction in computational time is
achieved. The FDTD algorithm was previously developed using the where v 5 propagation velocity; and c 5 free-space velocity
MATLAB programming language. This approach encompasses the (30 cm=ns).
geometry of the structures in fine detail. To calculate the conductivity, it was assumed that the dielectric
The synthetic models were built with a small spatial step (grid loss tangent of materials (tan d) was unity, where s is the dielectric
cell size in the x- and y-directions), and the excitation was a Gaussian conductivity of the medium, vð2pf Þ is the angular frequency, and
pulse of the same center frequencies used in field acquisition to get ɛðɛ 0 ɛ r Þ is the effective dielectric constant
the best approximation compared with the field data. The trace- s
tan d ¼ vɛ ¼ 1 (2)
distance interval and the total time were also defined in accordance
with the real data. After simulation, the synthetic data were exported
and provided by GprMax to ReflexW, and then filtered using a similar However, the conductivity obtained holds that tan d 5 1, and granite
processing sequence to that used for the field data (Table 1). is a nonconductive medium, for which the low-loss condition

Fig. 5. Example of velocity determination by adapting a diffraction hyperbola to each half part (left and right) of the first arch reflection from the left
hand downstream; the distance in the x-direction for each radargram is in accordance with the position of the arch in the entire GPR profile (starting from
the left margin upstream) (images by Belén Riveiro)

114 / JOURNAL OF BRIDGE ENGINEERING © ASCE / JANUARY 2014

J. Bridge Eng., 2014, 19(1): 110-119

 supports that tan d  1. Reynolds (1997) has reported that the higher than the upper medium. In this sense, the dielectric constant
maximum dielectric conductivity in medium can be obtained when for river water is 81, whereas the dielectric constant for air is 1. In
multiplying the conductivity value estimated from the assumption addition, the reflection produced by an existing ground elevation
tan d 5 1 by 0.05. The maximum value thus determined allows under the fourth arch (Fig. 6, R7) from the right margin upstream
considering the approximation tan d  1 with no appreciable error. [Fig. 1(a)] was observed. This interpretation agrees with the reality,
and the reflection caused by the air-ground interface was a few
nanoseconds before the reflection from the air-water interface oc-
Results and Discussion cupying one-half of the space. The data also showed the corner
reflections (Fig. 6, R8) generated from the perpendicular interfaces
between the top of the vault and the water level (Martinaud et al.
Field GPR Data
2004).
Meaningful results were obtained that provided information from From the interpretation of the processed radargrams, it was possible
civil-structural, historical, and archaeological points of view. By to detect and map the remains of performed restorations of the bridges,
Downloaded from ascelibrary.org by University of Central Florida on 11/25/20. Copyright ASCE. For personal use only; all rights reserved.

observing the 250-, 250-, and 500-MHz data obtained [Figs. 6(a–c), such as the reconstruction of the pathway [shallowest continuous line
respectively], it was possible to detect the hyperbolic reflections in Fig. 6(b)]. More than a superficial restoration, it was observed that
produced by the arch-air interfaces for all four arches (Fig. 6, R1– this bridge could be emptied and later refilled using different backfill
R4). Furthermore, two more hyperbolic reflections were identified than the original one. The white line in Fig. 6(b) indicates the reflection
(Fig. 6, R5) at the beginning of the radargram (left), which may generated at the interface between both original and new backfills,
correspond to the two nonvisible arches hidden by the construction which is a consequence of the high dielectric contrast between the
of a breakwater at the left margin (from upstream side) of the original different materials used.
structure, as reported in the literature (Alvarado et al. 1989). Reflec- The 500-MHz data showed a relatively homogeneous area es-
tions produced at the air-water interfaces were also identified (Fig. 6, pecially over the second and fourth arches of the bridge (Fig. 6, R2
R6) by the shift of the signal polarity. The change or inversion in and R4, respectively). This homogeneous area might be indicating
polarity occurs when the dielectric constant of the lower medium is the arch ring thickness. This is interesting information because the

Fig. 6. Processed (a) 250-, (b) 250-, and (c) 500-MHz field GPR data showing the interpretation of the main reflectors identified (summarizing, R1–R5
are arches, R6–R7 correspond to the air-water and air-ground interfaces, respectively, and R8 represents corner reflections); radargrams are shown from
the upstream side

JOURNAL OF BRIDGE ENGINEERING © ASCE / JANUARY 2014 / 115

J. Bridge Eng., 2014, 19(1): 110-119

 thickness of the ring stones is a geometric parameter that is crucial was detected in the form of hyperbolic reflections [Figs. 6(c), R10,
for providing the stability of these structures (Brencich and De and 7(b)].
Francesco 2004). Fig. 7(a) illustrates the 500-MHz data for the The two nonvisible arches detected are not actually part of the bridge
second arch in detail, where the fill-stone interface and its corre- structure, but they are integrated in the access road to the bridge sup-
sponding shift of the signal polarity may not be clearly appreciated porting higher traffic loads and intense vibrations. These may act
because of the confluence of many unwanted reflections. The ex- much like approach ramps abutting modern bridge decks, which more
istence of nonhomogenous backfill over the arch and some irreg- effectively and gradually transfer load from mobile roadway traffic onto
ularities in the fill-stone interface, such as the usual internal irregular the bridge itself. Therefore, it is interesting to know if these hidden
geometry of ring stones and a probable ring separation (presence of arches are emptied or filled to provide information to engineers and
voids), produce these complex reflections (Solla et al. 2012). Despite others on aspects of their design, functionality, safety, and stability. The
the usual external uniform surface of ring stones, the ring stones used acquired field data were observed in detail (Fig. 8), which allowed
to build the arch ring are all of different sizes and dimensions within analysis of the polarity of the signal to be registered at the arch-air
the structure as they provide a better structural reinforcement. Fig. 6 interfaces. Observing the traces collected over the keystones of the
Downloaded from ascelibrary.org by University of Central Florida on 11/25/20. Copyright ASCE. For personal use only; all rights reserved.

(R9) shows the reflection from the springer stone, which is produced hidden arches [Fig. 8(a)], the polarity of the signal did not show in-
by its usual larger size in relation to the other ring stones composing version because the reflection produced at the interfaces showed similar
the arch ring. In addition, a probable separation between ring stones polarity as the one produced by the direct signal at the air-ground

Fig. 7. Processed 500-MHz field GPR data in detail from the radargram in Fig. 6(c) showing (a) the homogenous area (arch ring thickness) over the
second arch from the right hand [Fig. 6(c), R2]; (b) the hyperbolic reflections by separation of the ring stones observed in the arch-air interface of the forth
arch [Fig. 6(c), R4]; the distance in the x-direction for each radargram is in accordance with the position of the arches in the entire radargram (Fig. 6)

Fig. 8. Field data (250 MHz) in detail from the radargram in Fig. 6(a): (a) hidden arches [Fig. 6(a), R5]; (b) first and second arches of the visible structure
from the left margin upstream [Fig. 6(a), R4 and R3, respectively]; reflections identified are the arch-air and air-water interfaces (R1 and R2, re-
spectively) and the corner reflections (R3); the distance in the x-direction for each radargram is in accordance with the position of the arches in the entire
radargram (Fig. 6); the traces at the right of the radargrams show the polarity of the signal for both direct signal (up) and arch-air (low) reflections

116 / JOURNAL OF BRIDGE ENGINEERING © ASCE / JANUARY 2014

J. Bridge Eng., 2014, 19(1): 110-119

 interface (first and second arrows, respectively, in Traces 1 and 2). As moist zones in the stonework (because water significantly slows
seen in Solla et al. (2010), this lack of polarity inversion indicates that GPR signal propagation velocity). Pérez-Gracia (2001) also reported
these arches are emptied or filled by using a filling with lower dielectric lower values of 7:5e11:5 cm=ns for wet granitic masonry. The
constant than the material used to fill the main structure. Moreover, lowest value (7:35 cm=ns) was obtained for the right side of the first
to ratify this hypothesis, the signals registered over the hidden arches arch [Fig. 1(a)]. The bridge structure ends immediately after this
were compared [Fig. 8(a)] with those obtained over the visible arches arch, and the velocity determined can indicate the existence of more
[Fig. 8(b)]. The behavior of the signals was similar, and their polarities compact backfill or modern materials such as concrete in this part of
did not indicate inversion [Fig. 8(b)]. Therefore, it was concluded that the structure. The velocity determined by Pérez-Gracia (2001) for
the two nonvisible arches identified by GPR remain most probably concrete ranged from 5:5 to 12:0 cm=ns. On the other hand, Table 2
emptied into the access road to the main structure of the bridge. shows that the dielectric constant and dielectric conductivity were
However, this might be a difficult conclusion to support without using estimated for each corresponding calculated velocity. For this work,
some other means to determine whether the arches are empty or void the dielectric conductivity was determined assuming the central
spaces. It could then be interesting to ratify interpretation by other frequency of 500 MHz.
Downloaded from ascelibrary.org by University of Central Florida on 11/25/20. Copyright ASCE. For personal use only; all rights reserved.

nondestructive evaluation (NDE) methods or other minor destructive Furthermore, these synthetic models were constructed from the
techniques such as localized coring. contour of the arch ring defined by the orthoimages. For that purpose,
the images had to be reprocessed before they were introduced into the
algorithm for FDTD modeling. This operation consisted of classifying
Integration of Laser Scanning and Simulations the materials presented in the image according to the different hy-
Advanced FDTD modeling was used to analyze the potential of the potheses of simulation. As a result, an image was obtained that was
GPR to identify the arch ring thickness, which is one parameter formed by different objects (defined by pure colors). The model was
influencing the stability of the arch. created with a very small spatial step equal to 2 cm, and the excitation
The synthetic models were built by considering the electro- was a Gaussian pulse of 500-MHz center frequency. The trace step
magnetic properties for granitic masonry calculated from the field and the total time window were 5 cm and 120 ns, respectively, and the
data. The purpose was to get the best approximation to reality for this entire model was defined by 451 traces. Following this approach, the
dielectric estimate. This was done through the laser scanning data by model defined the arch ring geometry in fine detail. The electro-
comparing the known distances in both the point cloud and the radar magnetic properties assumed for granite stone and fill media were
signal. Through this operation, the radar wave velocity could be those obtained from the field data (Table 2).
calibrated. Because of the radar data being gained in 2D space, the
3D model was simplified into the 2D representations corresponding Synthetic GPR Data and Comparison with Field
to the main planes of the bridge, that is, the orthoimages. Fig. 4(b) GPR Data
shows the orthoimages of both the upstream and downstream walls
of the bridge, where the main geometric parameters that define Observing the synthetic data obtained [Fig. 9(a)], the fill-stone in-
bridge geometry can easily be measured. Through these images, the terface was not clearly identified [Fig. 9(a), R1] because of the amount
geometry of the arch rings could be defined, and subsequently, the of reflections generated at this interface by multiple sources such as
radar wave velocity was estimated by the procedure of the diffraction a probable ring separation, the usual irregular geometry of the ring
hyperbola, explained previously in the section methodology (Fig. 5). stones in the interior of the structure, and the existence of non-
The radar wave velocity was estimated in different zones of the homogeneous fill material over the arch. A regular interface [Fig. 9(b)]
bridge, which ranged from 7:35 to13:0 cm=ns (Table 2). The typical was simulated to ratify this hypothesis, and the synthetic data easily
velocity reported in the literature (Pérez-Gracia 2001) for dry gra- showed the fill-stone interface and its consequent shift in the polarity
nitic masonry is approximately 12:0e13:5 cm=ns. However, the of the signal compared with the direct signal. This inversion in polarity
velocities estimated for this bridge were lower, which is likely was not clearly observed when the interface was simulated like ir-
a consequence of heterogeneities in the masonry (such as differences regular geometry ring stones [Fig. 9(a)], where the confluence of many
in composition and density) in addition to the probable presence of diffractions [see the trace in Fig. 9(a)] made the inversion difficult to
appreciate. As observed in the field data [Fig. 7(b)], the hyperbolic
reflections produced by the separation between ring stones at the
Table 2. Radar Wave Velocity (n), Dielectric Constant (ɛ r ), and Dielectric intrados surface [inner curve of the arch; Fig. 9(a), R4] was also
Conductivity (s) Values Calculated from the GPR Field Data identified. The springing of the arch on each side is nonvisible behind
the cutwaters [Fig. 1(a)], but, observing the field data, a reflection most
Medium n ðcm=nsÞ ɛr s ðmS=mÞ probably caused by the largest size of the springer stone in relation
Arch 1 with the other ring stones composing the arch was identified [Fig. 6(c),
Right 7.35 16.70 23.20 R9]. To verify this, a larger springer stone [Fig. 9(a)] was simulated,
Left 9.90 9.20 12.80 and the reflection obtained [Fig. 9(a), R3] was similar to that observed
Arch 2 in the field data. This information is useful for engineers because the
Right 9.50 10.00 13.90 use of a larger springer stone provides additional reinforcement to the
Left 13.00 5.30 7.40 arch.
Arch 3 The most appropriate recognition of the complex pattern of
Right 11.00 7.40 10.30 reflections generated at the arch-stone interface caused by the ir-
Left 10.70 7.90 11.00 regular shape of the ring stones allows for the detection of inner
Arch 4 structural damage such as delaminations and arch ring separation.
Right 11.25 7.10 9.90 As seen in Fig. 10, additional simulations were built to analyze how
Left 10.55 8.10 11.30 the GPR can identify an arch ring separation in both regular and
Note: For calculations, both right and left parts of the reflection at the arch- irregular interfaces. Observing synthetic radargrams obtained in Fig.
air interfaces were considered. Arches are numbered from the right margin at 10, the hyperbolic reflection produced by the existence of the
the upstream side in Fig. 1(a). simulated cavity were identified in both cases [Figs. 10(a and b)].

JOURNAL OF BRIDGE ENGINEERING © ASCE / JANUARY 2014 / 117

J. Bridge Eng., 2014, 19(1): 110-119

Downloaded from ascelibrary.org by University of Central Florida on 11/25/20. Copyright ASCE. For personal use only; all rights reserved.

Fig. 9. FDTD modeling to analyze the detection of the arch ring thickness, where (a) irregular and (b) regular interfaces were simulated; the usual
largest size of the springer stone was also simulated

Fig. 10. FDTD modeling to analyze how the GPR can detect ring stone separation in both (a) irregular and (b) regular interfaces; circles indicate the
hyperbolic reflections produced for each case

The reflection is even distinguished for the case of an irregular shape 500-MHz antenna allowed the identification of the potential thick-
at the extrados of the arch ring [Fig. 9(a)]. ness of the arch ring in the inner part of the structure. The velocity of
propagation of the radar wave was determined in different zones of
the bridge to range from 7.35 to 13.0 cm/ns. The lowest values in
Conclusions velocity obtained here potentially delineate the presence of moist
areas in the stonework. The diffraction hyperbola procedure was
In this work, the medieval bridge of Monforte, in the northwest of used to determine velocities in which, for a more appropriate esti-
Spain, is evaluated by using NDT to protect the historical character mation, the hyperbolic shapes to fit were the arch span geometries
of the structure. For surveying, GPR was used with central fre- provided by accurate laser scanning data.
quencies of 500 and 250 MHz. The processed GPR data have The laser scanning survey provided complete information about
revealed valuable information that was previously unknown, such as the geometric characteristics of the masonry structure. After post-
hidden structural details and evidence of restoration tasks. The ex- processing of the laser data, not only was accurate geometry obtained,
istence of two hidden arches was identified in the left margin (from but textural information improved the quality of the geometric model.
upstream) under the access road to the bridge. The analysis of the In fact, based on the textured model, orthoimages were produced for its
radar wave revealed that these nonvisible arches remain emptied subsequent integration in a process of radar response simulation
inside the structure. Moreover, the higher resolution provided by the (FDTD). Accurate geometry also provides better static and angular

118 / JOURNAL OF BRIDGE ENGINEERING © ASCE / JANUARY 2014

J. Bridge Eng., 2014, 19(1): 110-119

 corrections for field GPR data processing, analysis, and interpretation. Giannopoulos, A. (2006). “GprMax2D/3D: Electromagnetic simulator for
As the orthoimages become more accurate, more realistic simulations ground probing radar.” Æhttp://www.gprmax.orgæ (Apr. 25, 2012).
can be performed and, consequently, the better the interpretation of González-Jorge, H., Riveiro, B., Armesto, J., and Arias, P. (2011).
GPR data. Each sequential improvement in one process helps provide “Standard artifact for the geometric verification of terrestrial laser
scanning systems.” Opt. Laser Technol., 43(7), 1249–1256.
better data for the next. Additionally, the accurate geometric model
Gordon, S. J., and Lichti, D. D. (2007). “Modeling terrestrial laser scanner
results proved to be very useful for the calibration of the radar wave data for precise structural deformation measurement.” J. Surv. Eng.,
behavior through the constitutive materials of the bridge under study. 133(2), 72–80.
Simulating the structure in fine detail results in excessive com- Jol, H. M. (2009). Ground penetrating radar: Theory and applications,
putational requirements, but the mixed model of parallelization Elsevier Science, Amsterdam, Netherlands.
used here has proven to provide valuable additional information in a Lindenbergh, R., Uchanski, L., Bucksch, A., and van Gosliga, R. (2009).
reasonable amount of time. Some simulations were made to evaluate “Structural monitoring of tunnels using terrestrial laser scanning.” Rep.
the potential of the GPR to define the arch ring thickness. The Geodesy, 2(87), 231–239.
knowledge of this dimensional parameter is essential for analyzing the Lubowiecka, I., Arias, P., Riveiro, B., and Solla, M. (2011). “Multidisci-
Downloaded from ascelibrary.org by University of Central Florida on 11/25/20. Copyright ASCE. For personal use only; all rights reserved.

stability of the arch. However, a complex pattern of reflections usually plinary approach to the assessment of historic structures based on the
case of a masonry bridge in Galicia (Spain).” Comp. Struct., 89(17–18),
occurs in the interface between the filling and the extrados surface
1615–1627.
(outer curve of the arch), mainly because of the irregular shape of the Luhmann, T. (2006). Close range photogrammetry: Principles, methods
ring stones. Simulations demonstrated that regular interfaces and and applications, Wiley, Dunbeath, U.K.
subsequent arch ring thickness are better identified than irregularly Mabon, L. (2002). “Assessment, strengthening and preservation of masonry
shaped interfaces. For simulation, the model was elaborated from the structures for continued use in today’s infrastructure.” Rep. 86, In-
contour of the arch ring delineated from the laser scanning data. The ternational Association for Bridge and Structural Engineering, Lisbon,
most realistic scenario for modeling provided a better understanding Portugal, 448–459.
of the scattering or diffraction events produced at the interface, which Martinaud, M., Frappa, M., and Chapoulie, R. (2004). “GPR signal for the
allowed for the detection of other potential structural damage such as understanding of the shape and filling of man-made underground ma-
arch ring separation (or presence of voids). sonry.” Proc., 10th Int. Conf. on Ground Penetrating Radar, Delft,
Netherlands, 439–442.
MATLAB [Computer software]. MathWorks, Natick, MA.
Acknowledgments Melbourne, C., McKibbins, L. D., Sawar, N., and Sicilia-Gaillard, C. (2006).
Masonry arch bridges: Condition appraisal and remedial treatment,
Construction Industry Research and Information Association, London.
This work was financially supported by the Spanish Ministry of Sci-
Millard, S. G., Shaw, M. R., Giannopoulos, A., and Soutsos, M. N. (1998).
ence and Innovation (Grant No. BIA2009–08012) and the Spanish “Modeling of subsurface pulsed radar for nondestructive testing of
Center for Technological and Industrial Development (Grant No. structures.” J. Mater. Civ. Eng., 10(3), 188–196.
IDI–20101770). Orbán, Z., and Gutermann, M. (2009). “Assessment of masonry arch
railway bridges using nondestructive in-situ testing methods.” Eng.
Struct., 31(10), 2287–2298.
References Pérez-Gracia, V. (2001). “Radar de Subsuelo. Evaluación para aplicaciones
en arqueología y en patrimonio artistico.” Ph.D. thesis, Polytechnic
Alampalli, S. (2010). “Special issue on bridge inspection and evaluation.” Univ. of Catalonia, Barcelona, Spain (in Spanish).
J. Bridge Eng., 15(4), 349–351. Rebay, S. (1993). “Efficient unstructured mesh generation by means of
Alvarado, S., Durán, M., and Nárdiz, C. (1989). Puentes históricos de Delaunay triangulation and Bowyer-Watson algorithm.” J. Comput.
Galicia, Colegio Oficial de Ingenieros de Caminos, Canales y Puertos, Phys., 106(1), 125–138.
Xunta de Galicia, Santiago de Compostela, Spain (in Spanish). Reynolds, J. M. (1997). An introduction to applied and environmental
Brencich, A., and De Francesco, U. (2004). “Assessment of multispan ma- geophysics, Wiley, London.
sonry arch bridges. I: Simplified approach.” J. Bridge Eng., 9(6), 582–590. Riveiro, B., Morer, P., Arias, P., and de Arteaga, I. (2011). “Terrestrial
Clark, M. R., McCann, D. M., and Forde, M. C. (2003). “Application of laser scanning and limit analysis of masonry arch bridges.” Construct.
infrared thermography to non-destructive testing of concrete and ma- Build. Mater., 25(4), 1726–1735.
sonry bridges.” NDT Int., 36(4), 265–275. Sandmeier, K. J. (2007). ReflexW manual v.6.1, Sandmeier Scientific
Colla, C., McCann, D. M., and Forde, M. (2001). “Radar testing of a ma- Software, Æhttp://www.sandmeier-geo.de/download.htmlæ (Jul. 6, 2012).
sonry composite structure with sand and water backfill.” J. Bridge Eng., Solla, M., Lorenzo, H., Novo, A., and Caamaño, J. C. (2012). “Structural
6(4), 262–270. analysis of the Roman Bibei bridge (Spain) based on GPR data and
Diamanti, N., Giannopoulos, A., and Forde, M. C. (2008). “Numerical numerical modelling.” Autom. Construct., 22, 334–339.
modelling and experimental verification of GPR to investigate ring Solla, M., Lorenzo, H., Novo, A., and Rial, F. I. (2010). “Ground-penetrating
separation in brick masonry arch bridges.” NDT Int., 41(5), 354–363. radar assessment of the medieval arch bridge of San Anton, Galicia,
Fernandes, F. (2006). “Evaluation of two novel NDT techniques: Micro- Spain.” Archaeol. Prospect., 17(4), 223–232.
drilling of clay bricks and ground penetrating radar in masonry.” Ph.D. Solla, M., Lorenzo, H., Riveiro, B., and Rial, F. I. (2011). “Non-destructive
thesis, Minho Univ., Guimarães, Portugal. methodologies in the assessment of the masonry arch bridge of Traba,
Flint, R. C., Jackson, P. D., and McCann, D. M. (1999). “Geophysical Spain.” Eng. Fail. Anal., 18(3), 828–835.
imaging inside masonry structures.” NDT Int., 32(8), 469–479. Ural, A., Oruç, Ş., Dogangün, A., and Ískender Tuluk, Ö. (2008). “Turkish
Giannopoulos, A. (2005). “Modelling ground penetrating radar by GprMax.” historical arch bridges and their deteriorations and failures.” Eng. Fail.
Construct. Build. Mater., 19(10), 755–762. Anal., 15(1–2), 43–53.

JOURNAL OF BRIDGE ENGINEERING © ASCE / JANUARY 2014 / 119

J. Bridge Eng., 2014, 19(1): 110-119