Journal of Applied Geophysics 97 (2013) 37–44

Contents lists available at ScienceDirect

Journal of Applied Geophysics

journal homepage: www.elsevier.com/locate/jappgeo

A three dimensional approach for tracking cracks in bridges using GPR

Andrea Benedetto
Department of Engineering, University Roma Tre, Via Vito Volterra 60, 00146, Rome, Italy

a r t i c l e i n f o a b s t r a c t

Article history: Corrosion associated with reinforcing bars is the most significant contributor to bridge deficiencies. The cor-
Received 16 October 2012 rosion is usually caused by moisture and chloride ion exposure. The reinforcing bars are attacked by corrosion
Accepted 31 December 2012 and yield expansive corrosion products. These oxidation products occupy a larger volume than the original
Available online 20 January 2013
intact steel and internal expansive stresses lead to cracking and debonding. There are some conventional
inspection methods for the detection of the reinforcing bar's corrosion but they can be invasive and destruc-
Keywords:
Bridge deck inspection
tive, often laborious, and lane closure is required and it is difficult or unreliable for any quantification of cor-
Bar corrosion rosion. For these reasons, bridge engineers always prefer more to use the ground penetrating radar (GPR)
Ground penetrating radar technique. In this work a novel numerical approach for three dimensional tracking and mapping of cracks
Bridge maintenance in the bridge is proposed. The work starts from some interesting results based on the use of the 3D imaging
3D imaging technique in order to improve the potentiality of the GPR to detect voids, cracks or buried objects. The
GPR signal processing numerical approach has been tested on data acquired on a bridge by using a pulse GPR system specifically
designed for bridge deck and pavement inspection. The equipment integrates two arrays of Ultra Wide
Band ground coupled antennas, having a main working frequency of 2 GHz. The two arrays are using anten-
nas arranged with a different polarization. The cracks, associated often to moisture increase and higher values
of the dielectric constant, produce a not negligible increase of the signal amplitude. Following this, the algo-
rithm, organized in preprocessing, processing and postprocessing stages, analyzes the signal by comparing
the value of the amplitude all over the domain of the radar scan.
© 2013 Elsevier B.V. All rights reserved.

1. Introduction enhancing and facilitating the increase and diffusion of cracks in the
material. In particular we can observe a first phase of debonding
One of the most critical problems that brings about bridge deck between steel and concrete, the delamination of the concrete cover
deterioration and concrete slab damage and affects the regularity of material from the reinforcement and then a rapid phase of cracking,
surface pavement comes from the corrosion of the steel bar reinforce- starting from the steel concrete contact to the top layers of the structure
ment in bridges. (Neville and Brooks, 1987). The cracks generated at the steel cement
The corrosion of steel is generally activated by moisture and chlo- contact propagate with time to the surface facilitating the water infiltra-
ride ion exposure. More in depth the presence of excessive moisture tion (Hubbard et al., 2003).
and chloride ions in the concrete adjacent to the reinforcing bars is The basic motivation of this paper is to convince that an early
the main cause of corrosion. In these cases chemical oxidation reac- detection of the bar corrosion and a full correct tracking of hidden
tions occur and the steel bars are attacked. Particularly acidic solu- cracks are strategic in order to anticipate the main delamination. It
tions of moisture and chloride, coming from the top of the concrete results in a significant reduction of the maintenance costs and in a rele-
deck, ingress to the reinforcement steel and contribute to depassivate vant extension of the life of the bridge and the reinforced structures in
the alkaline environment in concrete. This is the start of the progres- general.
sive corrosion. The solution reacts with the steel generating oxidation For this scope the proposed approach is based on the use of the
processes and products such as FeO, Fe2O3, Fe3O4 and other oxides. In ground penetrating radar. In fact this electromagnetic tool offers many
the initial passivity stage these oxides are generated just along the advantages in the inspection of the bridge deck, the localization of the
reinforcement bars. Once the corrosion is activated the process pro- steel reinforcement bars and corrosion diagnosis. This depends on the
gresses very rapidly. The oxides yield expansion and the volume fact that there is a relevant contrast among the dielectric characteristic
increases. In general it accelerates the formation and development of original materials and oxides as well as among water, steel and con-
of cracks in all the surrounding concrete. crete dielectric properties (Halabe et al., 1993a; Holt and Eales, 1997). In
Meanwhile, because the produced oxides occupy a larger volume fact it can be observed that the bridge engineers that prefer to use the
with respect to the original intact steel, expansive stresses are generated ground penetrating radar (GPR) technique rather than traditional
inspection methods are always more (Alongi et al., 1993; Benedetto
E-mail address: [email protected]. and Pensa, 2006; Chen et al., 1994; Chung et al., 1992; Maser, 1989;

0926-9851/$ – see front matter © 2013 Elsevier B.V. All rights reserved.
http://dx.doi.org/10.1016/j.jappgeo.2012.12.010
38 A. Benedetto / Journal of Applied Geophysics 97 (2013) 37–44

Morey, 1998; Scullion et al., 1994). They report also that it is certainly Villela and Romo, 2010); in fact, one array includes sensors with parallel
more efficient and effective (Barnes and Trottier, 2004; Halabe et al., polarization with respect to the scanning direction (VV array), the other
1993b, 1997; Rhazi et al., 2007). It is not to be neglected that the GPR has sensors in orthogonal polarization with respect to the scanning
technology helps greatly in reducing the interferences with traffic direction (HH array). Overall the system collects 16 profiles within a
while measuring and testing. In fact the work zone results are very tem- single scan (8 HH+ 8 VV). The distance between two adjacent profiles
porary and limited in the occupied space along the road lane, so that the is 10−1 m only and VV and HH traces are collected over the same line;
data can be acquired rapidly, more safely even during traffic. total scan width is 0.7 m. Fig. 1 shows the system during the survey on a
Using novel GPR systems with multiple antenna arrays the data bridge.
can be acquired with an optimal spatial resolution. More in depth
the actual tomography that can be generated from the GPR radargram 2.2. Experimental site and survey procedure
is characterized by a very fine longitudinal resolution that is usually
in the order of 10 −1 ÷ 10 −2 m and depends on the sampling step For the validation of the algorithm that will be presented in detail
adopted during the survey (typically in the order of 10 −2 m). The lat- in Section 3, in order to carry out the survey and acquire the neces-
eral resolution depends on the way of operating during the survey sary data, a bridge over the main road network in central Italy that
and in particular it depends on the distance between two parallel appears in an advanced damaged condition has been selected.
longitudinal scans. This is stated by the operator or, in the case that The first cause of damaging is the weakening of the reinforcement
multiple antenna arrays are used, it is determined also by the distance due to the corrosion that is enhanced both by the passing high loads
between two consecutive antennas and it is in general in the order of and by the cold weather in the winter season when the bridge is fre-
10 −1 m (Barnes and Trottier, 2009; Benedetto et al., 2012; Narayanan quently exposed to icing problems. The length of the bridge is 393 m.
et al., 1998). The estimated traffic volume is very low (less than 1500 vehicles per
Basing on the radargram with this level of resolution it is possible to day) but the percentage of heavy trucks is over 30%. The transversal
generate an accurate three dimensional tomography. This is a novel section of the bridge is 12.4 m wide. The bridge has one carriageway
source of information to detect and follow the cracks because it permits composed by two lanes with different flow directions. The lanes are
the correlation of data in a full three dimensional domain. In fact two di- 3.75 m wide and the shoulders are 1.0 m. Fig. 2 shows the bridge
mensional radargrams, typically B-scan and C-scan that are respectively platform.
vertical and horizontal tomographies in the x–z section and the x–y The bridge is reinforced by longitudinal and transversal bars. The
plane, are not correlated to the third dimension, respectively y and z. longitudinal bars have diameters of 14 and 80 mm, and the transversal
On the contrary the availability of a 3D tomography makes it possible ones 8 and 20 mm. The top asphalt layers over imposed to the concrete
to correlate all the data in a full 3D grid and it is consequently possible slab are 100 mm thick in total, the binder layer is 70 mm thick and the
to follow cracks out of the 2D domain (Di Donato et al., 2010). top layer is 30 mm.
The objective of the paper is to formulate, firstly validate and dis- The RIS Hi Bright system is dragged at the speed of 3.0 km/h by the
cuss an algorithm for the automatic tracking of bridge cracks in a full operator; six longitudinal scans cover one lane of the road (Fig. 3). In
3D domain, basing on a high resolution GPR tomography. fact, the wideness of the system being 0.8 m, six scans cover 4.8 m
that is approximately the wideness of a half carriageway: lane and
2. Experimental background and evidences shoulder.
The productivity of the system can reach about 1000–2000 m2/h
2.1. Instrument depending on the traffic flow, the operability conditions and the method
adopted (e.g., if it is a one way scan or if the GPR acquires data going in
In order to generate a full 3D tomography of the bridge structure one direction and coming back in the opposite direction).
an advanced GPR system with a novel array of antennas specifically On the top layer some longitudinal cracks are visible close to the
designed for a bridge inspection has been used. The system is called center line and at the lane borders (Fig. 4).
RIS Hi Bright. This system is efficiently and effectively used for bridge
deck and pavement inspection. In general similar systems can mea- 2.3. Crack detection
sure pavement and concrete thickness, detect moisture and concrete
damage, locate slab rebar and pipes (e.g., Benedetto, 2010; Dolgiy and The surface of the bridge has been photographed by using a high res-
Zolotarev, 2006; Shihab and Al-Nuaimy, 2005; Simi et al., 2012; Utsi olution camera in order to catalogue all the visible cracks appearing on
and Utsi, 2004). In this work, with the scope of detecting cracks in the top layer of the pavement. The average thickness of the cracks is
concrete structures, it has been used to extract accurate 3D tomogra- 5.5 mm, while the length over the surface in several cases is greater
phy of the bridge pavement, concrete slab and any other structural than 4 m. The zone of the bridge that appears more cracked is in prox-
elements. imity of the joint.
The system is specifically designed for bridge applications and, for The diagnosis of the crack by GPR results is controversial. In fact
this reason, optimizes the polarization of the array of antennas in the thickness of the cracks is generally less or comparable to the
order to increase the resolution, thanks to a very dense data collec- GPR resolution that depends on the signal wave length (in this case,
tion. At the same time the efficiency of the survey and the safety in considering that the signal frequency is 2 GHz and the dielectric per-
the work zone are increased by the system as well as the time needed mittivity of the dry pavement material is from 4 to 10, the expected
to perform the acquisition. resolution is in the order of 20 to 50 mm). So it happens that it is dif-
More in depth, the equipment integrates two arrays of Ultra Wide ficult, sometimes impossible, and generally unreliable to deconvolve
Band ground coupled antennas, having a main working frequency of the signal in order to detect any additional reflection due to the
2 GHz and featuring a very large bandwidth (more than 100% of the cracks. However the changing of the mean dielectric properties of
main frequency). The high frequency and the large bandwidth permit the materials induced by the presence of the cracks modulates the
to reach a resolution of 1.25 × 10−2 m (quarter of wavelength criteria). signal amplitude amplifying it. In effect an increase of the signal am-
Each antenna includes a transmitter, a transmitting dipole, a receiver plitude has been observed corresponding to the presence of cracks
and a receiving dipole. Basically, each antenna is therefore capable of (Barnes et al., 2010; Benedetto et al., 2012).
emitting an EM pulse and collecting the backscattered signal reflected Basing on this evidence it is possible to extract the geometry of the
by buried objects. The two arrays within the RIS Hi Bright are using cracks automatically from a three dimensional tomography checking
antennas arranged with a different polarization (Simi et al., 2012; the signal amplitude.
 A. Benedetto / Journal of Applied Geophysics 97 (2013) 37–44 39

Fig. 1. RIS Hi Bright survey on a bridge.

An example of the detected crack is shown in Fig. 5. This crack lays corresponds to the direction of scan x, limited to N, the index j corre-
in the x–y plane and it has been observed in the real truth according sponds to the orthogonal direction y, limited to M, and k is the vertical
to the radar tracking. coordinate z, limited to L. Fig. 6 shows in an assonometric draft the po-
sition in the space of the generic element of the matrix a(i,j,k) that is the
3. Algorithm amplitude of the signal at the coordinate i,j,k, corresponding to x,y,z.
The preprocessing stage is aimed to remove the effect of attenua-
The algorithm is organized in three main steps: preprocessing, tion of the GPR signal. This attenuation is due to the energy absorp-
processing and postprocessing. tion and dissipation that increase as the inspection depth increases.
GPR data are used to feed the algorithm in the form of a three Then, before data processing, the attenuation of the signals has to
dimensional matrix of real single precision numbers A(N,M,L) that be compensated following Benedetto et al. (2012). For this purpose,
are the values of the signal amplitude in the point i,j,k. The index i an empirical inverse attenuation function f(z) has been calibrated

Fig. 2. Section of the bridge platform.

40 A. Benedetto / Journal of Applied Geophysics 97 (2013) 37–44

Fig. 3. GPR scans alignments.

and it is used to amplify the values of amplitude as z (in meter) As mentioned, the cracks, associated often to moisture variability
increases as it follows: and changing values of the dielectric constant, produce a not negligi-
ble increase of the signal amplitude. Following this, the algorithm
2
f ðzÞ ¼ 26:55505⋅z þ 1:97619⋅z þ 0:09: ð1Þ analyzes the matrix at the process stage. In detail the algorithm ana-
lyzes the signal by comparing the value of the amplitude a′(i,j,k) XA′
Using this attenuation function and sampling the z coordinate in the position of the 3D domain i,j,k all over the GPR scan with a
according to k, it is possible to calculate the amplified matrix A′ as it threshold T.
follows: The algorithm analyzes the signal amplitude all over the domain of
the radar scan for i = 1,N (longitudinal length of the scan), for j =1,M
0
a ði; j; kÞ ¼ aði; j; kÞ⋅f ðzk Þ ð2Þ (transversal length of the scan) and for k = 1,L (depth of the scan),
where N, M, L are the boundary domain limits. Starting from the generic
where zk is the value of z corresponding to the sampling index k. The position of the scan where the algorithm finds a signal amplitude great-
coefficients of the function f, that is variable in z in order to compen- er than the threshold, a′(i,j,k) > T, the 6 positions, left (j − 1), right
sate the attenuation effects, have been calibrated all over the depth of (j + 1), up (k − 1), down (k + 1), forward (i + 1) and backward (i − 1),
the scan comparing the real truth with the GPR prediction (Benedetto are checked in order to find the position where the amplitude reaches
et al., 2012). the maximum value a′max(i*,j*,k*).

Fig. 4. Picture of the top cracks and their localization.

 A. Benedetto / Journal of Applied Geophysics 97 (2013) 37–44 41

Fig. 5. Detected crack from tomography in an x–y plane.

Of course this value must be over the threshold T to be accepted as The algorithm results are very sensitive to the assumed value of
belonging to the crack. At the next step the algorithm restarts the the threshold T for filtering the signal amplitude during the process-
tracking procedure assuming a′max(i*,j*,k*) as the new generic value ing stage. This is the reason why different values of the threshold
a′(i,j,k). The procedure ends as the entire domain has been numeri- have been checked in order to track the cracks correctly.
cally scanned. If the value of the threshold is very low the algorithm identifies
Fig. 7 shows the basic cell in the domain where the algorithm is potential cracks, following them with continuity, all the sequences
applied. The central point a′(i,j,k) is in black, the signal amplitude of the contiguous points in the 3D domain where the amplitude is
check is extended to all the six points in a gray tone. greater than the threshold. It can happen that the number of false
Fig. 8 shows the flow chart explaining how the algorithm process- positive detections increases enormously. On the contrary if the
es the GPR signal, and compares the amplitude cell by cell. value of the threshold is assumed to be very high, the algorithm can
The potential cracks that are detected by the signal amplitude easily neglect the cracks as that result would be invisible to the algo-
comparison are collected in a new matrix B(N,M,L). The generic ele- rithm because the amplitude of the signal is less than the threshold. In
ment of this matrix b′(i,j,k) XB is zero if there is no crack in the this case the number of false negative detections can be relevant.
point i,j,k of the domain, otherwise the element b(i,j,k) is equal to At the moment it is impossible to calibrate on a full real scale the
an integer number that identifies the specific detected crack. threshold experimentally because there is no certain way to detect all
In the postprocessing stage the matrix B(i,j,k) containing the the cracks in a bridge except for a few very expensive and impractical
potential cracks, that have been automatically detected, is filtered in ways such as for example the X-ray technique that is generally used
order to select the cracks that have to be considered reliable and to on the samples of materials with a very small dimension. This is the
discharge possible false alarms. This filtering is made in the first reason why we have proposed here a calibration of the threshold
approximation only considering the predicted length of the cracks value on some observations that should be improved in the future
that, on the basis of experience, should be no shorter than 0.40 m. as new data could be available. The observations are limited to the
In detail the algorithm initializes a pointer (integer number) to surface of the pavement (plane x–y and z = 0), by counting and digi-
0 and sums to it the number of times that a same value of b(i,j,k) talizing the visible actual cracks.
appears all over the matrix, pertaining to the same crack. Fig. 9 shows in four plots the positioning as the automatically
A matrix C, dimensioned C(H,2) where H is the number of poten- tracked cracks are different as the value of the threshold changes. In
tial cracks as detected by the algorithm, is generated: the generic particular it can be seen that if the threshold assumes the maximum
element C(l,1) is the identification number of the crack and the
generic element C(l,2) is the length of the crack. Finally all the cracks
shorter than the minimum length are filtered.

4. Results and discussion

The algorithm has been applied for validation to the data acquired
on the bridge with the instrument and following the procedure that
has been presented previously in Section 2.

Fig. 6. The generic element a(i,j,k) of the three dimensional matrix of the GPR scan. Fig. 7. Basic cell in the grid and algorithm process.
42 A. Benedetto / Journal of Applied Geophysics 97 (2013) 37–44

value of 8.0 the algorithm neglects many possible cracks and, by a

visual comparison with the real truth, it can be concluded that it
underestimates grossly the current poor condition of the bridge.
Otherwise if the value of the threshold is assumed to be the minimum
value of 2.5, the 3D map of the cracks would appear very confusing
and reasonably the algorithm would have detected many cracks
that do not exist and thus producing a high level of false alarms.
The calibration of the threshold value is here based on a visual
comparison, that it is obviously possible only for the visible cracks
to appear on the surface, between the algorithm predictions and the
real truth. Moreover the calibration is extended to the comparison
among the characteristics of several visible cracks that have been
detected, photographed and digitalized and the characteristics of
the predicted cracks. The value of the threshold has been calibrated
to the value of 4.0.
In particular the characteristics of the cracks that have been com-
pared with the outcomes of the numerical model applied to the GPR
data are the length of the cracks and their geometrical shape. This
comparison is possible by overlapping the digitalized real cracks
and their numerical images from the model.
In more detail each predicted crack is compared to the real crack,
if it exists, by the position in the bridge (x,y,z), the length and the
shape. If the number of the predicted cracks Nc is very high and a
qualitative comparison is not possible, an Objective Function is used
in order to calibrate the value of T. The Objective Function depends
on (1) the distance between the starting points of the predicted and
real cracks that should tend to zero, (2) the difference between the
lengths of the predicted and real cracks that should tend to zero
Fig. 8. Flow chart of the algorithm. and finally (3) the difference between their tortuosity that is the

Fig. 9. Three dimensional maps (plot positioning) of cracks automatically tracked by algorithm using different values of the threshold.
 A. Benedetto / Journal of Applied Geophysics 97 (2013) 37–44 43

Fig. 10. Comparison between real cracks and numerical prediction.

rate between the measured length of the crack and the distance be- For example Fig. 10 shows the overlapping of cracks that is finally
tween its starting and ending points; also this difference should used for calibrating the threshold of the algorithm.
tend to zero. The calibrated value of T is the value that minimizes In Fig. 10 the existing cracks are digitalized in black and the most
the Objective Function. relevant have been coded by ψi. By overlapping, it is evident that the

Fig. 11. Crack severity maps related to the cracks detected using different values of the threshold.
44 A. Benedetto / Journal of Applied Geophysics 97 (2013) 37–44

model well fits all the main cracks, in particular eight over eleven. Asphalt-Covered Decks, Rep. SHRP-S-325. National Research Council, Washington,
D.C.
Only in the three cases that the prediction is ambiguous, in fact Barnes, C.L., Trottier, J.F., 2004. Effectiveness of ground penetrating radar in predicting
while the cracks ψ6, ψ7 and ψ11 have been photographed and digita- deck repair quantities. Journal of Infrastructure Systems 10, 69–76.
lized, the model prediction is not here continuous and in the tomog- Barnes, C.L., Trottier, J.F., 2009. Phenomena and conditions in bridge decks that
confound ground-penetrating radar. Data Analysis Transport Research Record
raphy it is possible only to identify some spots, corresponding to an 1795, 57–61.
unexpected and anomalous increase of the signal amplitude. Other- Barnes, C.L., Trottier, J.F., Forgeron, D., 2010. Improved concrete bridge deck evaluation
wise these spots are not correlated in space and any prediction of using GPR by accounting for signal depth-amplitude effects. NDT&E International
41, 427–433.
cracking based only on these could be arbitrary. Benedetto, A., 2010. Water content evaluation in unsaturated soil using GPR signal
The analysis of the cracks is not limited to their shape and tracking, analysis in the frequency domain. Journal of Applied Geophysics 71 (1), 26–35.
but using this algorithm it is also possible to evaluate the severity of Benedetto, A., Manacorda, G., Simi, A., Tosti, F., 2012. Perspectives in bridges inspection
using GPR. Nondestructive Testing and Evaluation 27 (3), 239–251.
them. In particular the severity of the cracks has been related to the
Benedetto, A., Pensa, S., 2006. Indirect diagnosis of pavement structural damages using
value of the GPR signal amplitude. Following this approach it is possible surface GPR reflection techniques. Journal of Applied Geophysics 62 (2), 107–123.
to generate the maps of severity of the cracks. In Fig. 11 the severity Chen, H.L.R., Halabe, U.B., Sami, Z., Bhandarkar, V., 1994. Impulse radar reflection waveforms
maps related to the cracks traced and shown in Fig. 9 are presented. of simulated reinforced concrete bridge decks. Materials Evaluation 52, 1388–1392.
Chung, T., Carter, C.R., Masliwec, T., Manning, D.G., 1992. Impulse radar evaluation of
In particular on the z axis is represented the severity of the cracks, in a asphalt-covered bridge decks. IEEE Transactions on Aerospace and Electronic
parametric scale, while x and y are the spatial coordinates of the cracks, Systems 28, 125–136.
all projected on the horizontal plane. Di Donato, L., Catapano, I., Soldovieri, F., Crocco, L., 2010. Imaging of 3D magnetic
targets from multiview multistatic GPR data. Proceedings of the 13th Int. Conference
It is clear from Fig. 11 that the most severe cracks are located at on Ground Penetrating Radar, Lecce, Italy.
the joint but somewhere several cracks are also present along the Dolgiy, A., Zolotarev, V., 2006. Optimal radius estimation for subsurface pipes detected
pavement of the bridge. by ground penetrating radar. Proceedings of the 11th Int. Conference on Ground
Penetrating Radar, Columbus, Ohio, USA.
Halabe, U.B., Maser, K.R., Kausel, E.A., 1993a. Condition assessment of reinforced
5. Conclusion concrete structures using electromagnetic waves. ACI Materials Journal 92, 511–523.
Halabe, U.B., Sotoodehnia, A., Maser, K.R., Kausel, E.A., 1993b. Modeling of the electro-
magnetic properties of concrete. ACI Materials Journal 90, 552–563.
In this paper a new algorithm has been presented and finalized in Halabe, U.B., Chenn, H., Bhandarkar, V., Sami, Z., 1997. Detection of sub-surface anomalies
the automatic tracking of cracks in the bridge decks by processing the in concrete bridge decks using ground penetrating radar. ACI Materials Journal 94,
GPR data. The proposed approach and algorithm use real GPR data 396–408.
Holt, F.B., Eales, J.W., 1997. Nondestructive evaluation of pavements. Concrete International
and make it possible to track in a three dimensional domain the
9, 41–45.
cracks and to evaluate their severity by comparing the amplitude of Hubbard, S.S., Zhang, J., Monteiro, P.J.M., Peterson, J.E., Rubin, Y., 2003. Experimental
the GPR reflected signals. It is a relevant contribution in the improve- detection of reinforcing bar corrosion using nondestructive geophysical techniques.
ment of the actual methods and equipment that are applied for bridge ACI Materials Journal 100, 107–123.
Maser, K.R., 1989. New Technology for Bridge Deck Assessment, Phase I Report, Rep. No.
investigation. In fact the measurement is non destructive, causing no FHWA-NETC-89-01. Massachusetts Institute of Technology, Center for Transportation
damage in the structure, it is relatively rapid, reducing the interfer- Studies, Federal Highways Administration, McLean, VA.
ences with the traffic flow, it is cost effective and it is very resolute, Morey, R.M., 1998. Ground Penetrating Radar for Evaluating Subsurface Conditions for
Transportation Facilities, Synth. of Highway Practice 255, NCHRP. National Academy
giving information at the scale of 10 −2 m. However the most impor- Press, Washington, D.C.
tant point of strength and innovation of the model is the capacity of Narayanan, R.M., Hudson, S.G., Kumke, C.J., 1998. Detection of rebar corrosion in bridge
following the exact geometry of the crack in a three dimensional decks using statistical variance of radar reflected pulses. Proceedings of the 7th Int.
Conference on Ground Penetrating Radar, Lawrence, Kansas, USA.
space, in fact the existing bridge deck inspection procedures are Neville, A.M., Brooks, J.J., 1987. Concrete Technology. Longman Scientific & Technical,
able to map critical areas unless they identify the single crack. Using England.
the proposed method is possible to improve the effectiveness and Rhazi, J., Dous, O., Laurens, S., 2007. A new application of the GPR technique to
reinforced concrete bridge decks. Proceedings of the 4th Middle East NDT Conference
efficiency of any rehabilitation and restoration. The most critical and Exhibition, Kingdom of Bahrain.
point in the algorithm application is fixing the proper value of the Scullion, T., Lau, C., Chen, Y., 1994. Implementation of the Texas Ground Penetrating
threshold of detection. It has been fixed here by comparing the Radar System, Res. Rep. 1233-1. Texas Transportation Institute, Texas, USA.
Shihab, S., Al-Nuaimy, W., 2005. Radius estimation for cylindrical objects detected by
existing visible cracks on the bridge pavement surface and the out-
ground penetrating radar. Subsurface Sensing Technologies and Applications 6, 1–16.
comes of the algorithm. In this case only in the 27% of the cases that Simi, A., Manacorda, G., Benedetto, A., 2012. Bridge deck survey with high resolution
there is an ambiguous detection, mainly a false negative detection. ground penetrating radar, ground penetrating radar (GPR). Proc. 14th International
The availability of additional data in the future could make it possible Conference on GPR, pp. 489–495.
Utsi, V., Utsi, E., 2004. Measurement of reinforcement bar depths and diameters in
to improve the capability of the algorithm in the automatic 3D track- concrete. Proceedings of the 10th Int. Conference on Ground Penetrating Radar,
ing of bridge deck cracks. The actual results appear however very Delft, The Netherlands.
promising. Villela, A., Romo, J.M., 2010. Experimenting with different polarization arrays in a test site.
Proceedings of the 13th Int. Conference on Ground Penetrating Radar, Lecce, Italy.

References
Alongi, A.J., Clemena, G.G., Cady, P.D., 1993. Condition evaluation of concrete bridges
relative to reinforcement corrosion. Vol. 3: Method of Evaluating the Condition of