Sie sind auf Seite 1von 12

Journal of Applied Geophysics 74 (2011) 2637

Contents lists available at ScienceDirect

Journal of Applied Geophysics


j o u r n a l h o m e p a g e : w w w. e l s ev i e r. c o m / l o c a t e / j a p p g e o

Reconstruction of sub-wavelength fractures and physical properties of masonry


media using full-waveform inversion of proximal penetrating radar
Claudio Patriarca a,, Sbastien Lambot b,c, M.R. Mahmoudzadeh b, Julien Minet b, Evert Slob a
a
b
c

Department of Geotechnology, Delft University of Technology, Stevinweg 1, 2628 CN Delft, The Netherlands
Earth and Life Institute, Environmental Sciences (ELI-e), Universit catholique de Louvain, Croix du Sud 2 box 2, 1348 Louvain-la-Neuve, Belgium
Agrosphere (ICG-4) Institute of Chemistry and Dynamics of the Geosphere, Forschungszentrum JlichD-52425 Jlich, Germany

a r t i c l e

i n f o

Article history:
Received 26 October 2010
Accepted 2 March 2011
Available online 9 March 2011
Keywords:
Ground penetrating radar
Electromagnetic inverse problem
Electric properties of concrete
Material characterization
Non-destructive testing

a b s t r a c t
High-frequency, ultra-wideband penetrating radar has the potential to be used as a non-invasive inspection
technique for buildings, providing high-resolution images of structures and possible fractures affecting constructions.
To test this possibility, numerical and laboratory experiments have been conducted using a proximal, steppedfrequency continuous-wave radar system operating in zero-offset mode, spanning the 38 GHz frequency range. The
reconstruction of the material electrical properties is achieved by resorting to full-waveform inverse modeling.
Numerical experiments showed that for typical electric permittivity and electrical conductivity values of concrete
and plaster, it is possible to retrieve the physical properties of the material and to detect fractures less than 1 mm
thick. Laboratory experiments were conducted on non-reinforced concrete and plaster test slabs in different
congurations. The results showed the good potential of this method: (1) to provide a thorough fracture response
model in buildings or artworks and (2) to non-invasively characterize the samples in terms of their electromagnetic
properties.
2011 Elsevier B.V. All rights reserved.

1. Introduction
Ground penetrating radar (GPR) is used in earth science and civil
engineering for a large variety of applications (Jol, 2009). The possibility
of non-destructive testing has generated growing interest in improving
electromagnetic methods, whose prime goal is to characterize materials.
This is essential for refurbishment oriented inspections, which are
meant to enhance mechanical properties that should not be jeopardized
during assessment works.
Invasive and minor invasive tests for masonry conditions assessment exist, but they can compromise the integrity of the investigated
structures. Non-destructive testing methods can quickly provide
qualitative images at reasonable cost, but thorough physical characterization has not been achieved.
Extensive research has been carried out by many agencies and
institutions on the applications concerning radar non-destructive
evaluation of the built environment. Applications include electromagnetic modeling for detecting cracks in cement-based materials
inspection (Carino, 2008; Malhotra and Carino, 2003; Nadakuduti
et al., 2006), detecting media delamination and voids in road and bridge
pavements (Belli et al., 2009), characterizing materials in terms of
electromagnetic properties (Lambot et al., 2004a; Robert, 1998; Soutsos
Corresponding author. Tel.: +31 15 2787961.
E-mail addresses: c.patriarca@tudelft.nl (C. Patriarca), sebastien.lambot@uclouvain.be
(S. Lambot), mohammad.mahmoudzadeh@uclouvain.be (M.R. Mahmoudzadeh),
julien.minet@uclouvain.be (J. Minet), e.c.slob@tudelft.nl (E. Slob).
0926-9851/$ see front matter 2011 Elsevier B.V. All rights reserved.
doi:10.1016/j.jappgeo.2011.03.001

et al., 2001), detecting damages in concrete columns (Buyukozturk and


Yu, 2009), surveying and monitoring cultural heritage constructions
(Brown et al., 2009) and assessing masonry damages (Binda et al., 2005;
Stone, 1997). Maierhofer and Leipold (2001) adopted a multipolarization approach to determine moisture content distribution and
to detect full and empty joints in masonry structures through semiquantitative analysis. Tsoias and Hoch (2006) successfully investigated
the response of GPR wave transmission through thin layers using multipolarization for non-contact characterization of fracture aperture and
uid-ll. Their experiment, however, involved transmission of GPR
signals at various angles; this implies two sided access to the masonry,
which is not always possible in real cases. Orlando and Slob (2009) used
a multicomponent pulsed GPR system to produce 2.5D vector migration
images of cracks in the oor of a historical building. With their approach,
the structures underneath an ancient building oor could be accurately
reconstructed and different orientation cracks could be detected in
qualitative images. Drobert et al. (2008) used GPR combined with a
new capacitance technique to evaluate moisture content in concrete
samples. They showed that the two procedures are highly complementary and are not able to furnish the desired speed and accuracy at the
same time. Sambuelli and Calzoni (2010) illustrate a successful
experiment carried out on a marble block containing a controlled
fracture pattern. They demonstrate the possibility of characterizing the
fracture aperture provided the fracture lling properties are known.
Many published works deal with the characterization of masonry
structures and repairing works efcacy using GPR, but none seem to
exploit all of the information contained in the amplitude and phase of

C. Patriarca et al. / Journal of Applied Geophysics 74 (2011) 2637

the frequency spectrum. The usage of this information does not limit our
method to the concept of imaging resolution as a function of
wavelength, a fundamental subject widely discussed in geophysics
(Daniels, 2004; Slob et al., 2010; Widess, 1973). The method we use is
based on a non-contact operating scheme suitable for historical
buildings subject to further refurbishment activities, for which an
accurate diagnosis of the micro-cracks aperture and of the lling-uids
is required.
For such applications, the method proposed by Lambot et al. (2004a)
appears to be promising. It is based on a non-contact radar system, with
a single antenna simultaneously playing the role of emitter and receiver
(Section 2). The radar system is set up with vector network analyzer
(VNA) technology, for which international standard calibration is
well-dened. For that particular radar conguration, an accurate fullwaveform forward electromagnetic model was proposed, which is
based on (1) an exact solution of the three-dimensional (3-D) Maxwell
equations for waves propagating in 1-D layered media and (2) linear
system theory properly accounting for the antenna and its interactions
with the target. The electrical properties of the layered medium are
retrieved from the radar data through iterative model inversion using a
global optimization method. The method has been successfully used in a
number of applications (Lambot et al., 2008, 2009; Lopera et al., 2007a;
van den Bril et al., 2007).
In this paper we investigate the applicability of the approach to
quantitative characterization of building materials with fractures; the
analysis is limited to homogeneous multilayered media. In practice, the
homogeneous medium assumption need only hold true locally. First,
numerical experiments described in Section 3 were constructed to
investigate the information content (uniqueness of the inverse solution)
in the radar data with respect to the reconstruction of different layer
properties, including thin fractures with sub-wavelength thickness. This
analysis is repeated for a number of realistic congurations that consider
the effect of thin (sub-wavelength) horizontal fractures on the received
and emitted complex signal ratio. Through these experiments we could
also optimize the inversion procedure for different numbers of variables
and for a wide set of geometrical and electromagnetic parameters.
Second, laboratory experiments are discussed in Section 4 to characterize real concrete and plaster slabs in three layered model congurations (material, fracture lled with air, material), thereby providing
valuable insights into the stability of the inverse solution with respect to
actual measurement and modeling errors. For both the numerical and
laboratory experiments an ultrawide band was used (38 GHz).
The proposed technique for the characterization and inspection of
building materials has the following advantages: (1) it maximizes
reliability for a wide range of investigation scales, i.e., from the
millimeter (micro-crack detection) to the decimeter scales (masonry
thickness characterization), which is important information for the
refurbishment support; (2) the acquisition may be rapid even on large
surfaces, even though computational times may become signicant
under certain circumstances; (3) fast acquisition combined with high
reliability and high mobility leads to a widely exploitable technique
potentially applicable to many natural and articial materials for
damage assessment and evaluation.
2. Materials and methods
2.1. Equipment
In order to characterize materials and quantitatively reconstruct
embedded fractures, an ultra-wideband proximal GPR is used. The radar
system is set up using a vectorial network analyzer (VNA, ZVT8,
Rhode&Schwarz, Munich, Germany). The antenna used is a doubleridged broadband horn (BBHA9120C, Schwarzbeck Mess-Elektronik),
connected to the VNA via a high quality N-type 50 coaxial cable using
an SMA-N adaptor. The antenna axial length is 136 mm and the aperture
area is 66 91 mm2; its nominal frequency range is 318 GHz and its

27

isotropic gain ranges from 10 to 16 dBi. The Open-Short-Match


calibration kit Rohde&Schwarz ZV-Z21 was used to calibrate the VNA
at the connection between the cable and the antenna feed point.
The procedure followed for the antenna calibration can be found in
Jadoon et al. (2011). Measurements were recorded as the frequency
dependent complex ratio S11() between the received and emitted
signal, where is the angular frequency ( = 2f), f being the natural
frequency. The signal was sampled at 2501 frequencies over the 38 GHz
range, using 2 MHz frequency step, 0 dBm transmission power and an
averaging factor of 50; the average antenna aperture distance from the
slabs required for the EM model used in this study is 15 cm. The high
directivity of the antenna (40 3-dB beam width in the E-plane and 34 in
the H-plane at 5 GHz) makes it suitable for measurements on reduced
size slabs.
2.2. Forward and inverse modeling
The main advantage of the described radar set up is that it allows an
accurate modeling of the radarantennamasonry system. The antenna
is modeled as a point device with complex frequency dependent global
reection and transmission coefcients, which can be obtained through
calibration measurements. This calibration procedure allows an exact
modeling through the solution of the 3-D Maxwell's equations for wave
propagation in 1-D multilayered media. Masonry materials are treated
here as dispersive media, whose electromagnetic properties vary with
frequency.
In both the numerical and laboratory experiments a boundary
condition was given by a perfect electric conductor (PEC) placed on
one side of the investigated media (air-multilayer masonry-PEC), thus
reducing the complexity of the model conguration.
2.2.1. The radar model
Several assumptions on the material properties and simplications
of the physics are made to allow accurate modeling; the details can be
found in Lambot et al. (2004a), Lopera et al. (2007b). Here we simply
list them: (1) the investigated target is situated in the far eld (the
Fraunhofer region) of the antenna emitting point, therefore the antenna
can be modeled as a point source and receiver that interacts with its
surroundings; (2) the high directivity of the TEM horn antenna makes it
reasonable to approximate the investigated material as horizontally
innite, and to neglect the border effects of the slabs, which in our model
are assumed to be homogeneous, isotropic and non magnetic; (3) the
received signal mainly propagates in the axial direction of the antenna,
whose actual radiation pattern is not needed. The fractures are modeled
as layers with sub-wavelength thickness.
Under these assumptions, the radar model expressed in the
frequency domain can be written as (Lambot et al., 2004a):
S11 =

b
H Gxx
= Hi +
a
1Hf Gxx

where S11() is the measured (raw radar data) frequency dependent


ratio between the backscattered b() and incident a() electric elds at
the calibration plane of the VNA. Hi(), H() and Hf () are the
characteristic antenna transfer functions accounting for the antenna
propagation effects and antennamaterial interactions, while Gxx() is
the impulse reection response of the airmaterial system modeled as a
multilayered medium. The physical antenna is reduced to a point source
with a scattering matrix, represented by the transfer functions. These
functions act as reection and transmission coefcients and account for
all variations of impedance from the calibration plane of the VNA
(antenna connector) to the air and vice versa. The characteristic antenna
transfer functions can be determined by solving the system of Eq. (1) to
these three unknown functions, by performing S11() measurements at
different heights above a copper sheet for which the Green's functions
Gxx() can be computed (Lambot et al., 2004a). Once the antenna

28

C. Patriarca et al. / Journal of Applied Geophysics 74 (2011) 2637

transfer functions are known, the presence of the antenna can be


removed from the raw radar data and the impulse reection response of
the air-masonry is obtained using Eq. (1). In Fig. 1 (ac) the antenna
transfer functions are shown for the double-ridged broadband horn
antenna discussed before.
2.2.2. Physical properties of concrete and plaster
The constitutive parameters governing electromagnetic wave
propagation in the materials are the electric permittivity (Fm 1),
electrical conductivity (Sm 1), and magnetic permeability (Hm 1).

a
0.8

|Hi|

0.6
0.4
0.2
0

f (GHz)

H (rad)

0
2
3

5
f (GHz)

x 10

|H|

2
1
0

5
f (GHz)

H (rad)

2
0
2
3

5
f (GHz)

x 10

The relative electric permittivity, to which we will refer hereafter, is


dened as r = /0, where 0 8.85 10 12 Fm 1 is the electric
permittivity of the free space. In this paper, we assume equal to the
permeability of free space 0 = 4 10 7 Hm 1.
An effective electric permittivity is usually dened, because both the
macroscopic electric permittivity ( = j) and electrical conductivity ( = + j) are complex quantities that always occur together in
Maxwell's equations. The wave velocity inside lossless materials is
controlled by the real component of the electric permittivity (Tsui and
Matthews, 1997); however, in dissipative materials both real and
imaginary parts of electric permittivity determine propagation velocity
and attenuation. These losses are partly due to the migration collision of
free charges, and are represented by the real part of the electric
conductivity ; the imaginary component of the electrical conductivity
acts as a real part of permittivity inuencing the wave propagation
velocity and amplitude attenuation.
The presence of water is one of the main factors controlling the
value of the electromagnetic properties in soil materials (Topp et al.,
1980) as well as in masonry materials (Lai et al., 2009). The value of
the electric permittivity has been observed to vary during the cement
hydration process (Lai et al., 2009) when the oxides build complex
compounds, the cement paste undergoes hardening and the free
water decreases, becoming progressively bounded to the crystalline
phases. The curing temperature indirectly affects the electric
permittivity of the cement paste by controlling its rate of hydration
(Morsy, 1999); moreover, an increase of the hydration temperature
results in an increase of the electrical conductivity (Morsy, 1999).
Once hardening is completed, a small conductivity increase can be
seen with increasing temperature, but no signicant change in
permittivity (Soutsos et al., 2001).
A continuous variation of the electromagnetic properties in the
investigated prole can be emulated by discretizing the multilayered
media as a function of the minimal wavelength used. A multilayer
structure can be built by assigning a different value of these parameters
to discrete thicknesses. At GPR frequencies it is known that materials
exhibit frequency dependence of their electromagnetic properties (West
et al., 2003). In practice, it is necessary to take these effects into account
for reliable electromagnetic modeling. Similar to results obtained by
Lambot et al. (2004b) for a sandy soil, we observed that the frequency
dependence of the materials electric permittivity is negligible. However,
the electrical conductivity, as clearly illustrated in Section 4, shows a
frequency dependence that can be locally approximated by a linear
equation (Lambot et al., 2005) in the 38 GHz frequency range:


9
f = 3GHz + a f 3 10

|H |

Hf (rad)

f (GHz)

2
0
2
3

f (GHz)
Fig. 1. Antenna transfer functions. (a) Return loss Hi(); (b) transmitting and receiving
response function H(); (c) feedback loss Hf(). These transfer functions were
obtained solving an over-determined system of seven equations to three unknowns
according to Eq. (1) over the entire frequency range.

where 3GHz is the reference apparent electric conductivity at the


minimum frequency of 3 GHz, and a represents the constant slope of
(f). The assumption of a frequency independent complex permittivity and a frequency dependent complex conductivity has a direct
consequence on the electromagnetic model (Bradford, 2007). In fact,
the effective conductivity, a function of the real component of the
electric conductivity and the imaginary component of the electric
permittivity, leads to a frequency dependent wave attenuation model.
2.2.3. Model inversion and objective functions
When inadequate prior information on the model parameters is
available, those are treated as deterministic uncertain quantities. Then,
the inverse problem is solved through a maximum likelihood approach
that requires the minimization of an objective function; this reduces to
the classical least squares problem, since we assume that in our data
the observation errors are normally distributed, independent, and
homoscedastic (Carrera and Neuman, 1986; Lambot et al., 2004a). In
this study the inversion procedure is formulated as a classical, non-

C. Patriarca et al. / Journal of Applied Geophysics 74 (2011) 2637

linear least-squares problem that is solved iteratively using global


optimization.
For each scenario, the correct values of the parameter vector
b = [r, n, n, hn], for n = 1, , N are determined so that the objective
function, dened as (b):
1
 12
max 
 T
2
B Gxx Gxx  C
B fmin
C
C
b = B
B fmax 
C

@
 T 2 A
Gxx 

29

TEM

0f

h0

air

fmin

() and G = Gxx ; b are the vectors


is minimized, where GTxx = Gxx
xx
containing the observed and simulated air-target Green's functions, and
b is the parameter vector dened before. The topography of the objective
function is inherently characterized by multiple local minima due to the
non-linearity of the problem. To overcome this problem, we use the
global multilevel coordinate search (GMCS, (Huyer and Neumaier,
1999)) in sequence with the NelderMead Simplex algorithm (Lagarias
et al., 1998; Lambot et al., 2002) to rene the solution by means of a local
search.
The task of the GMCS is to nd the global minimum of the objective
function. The search space is divided into smaller boxes, until a userdened maximum level (smax) has been reached. In the next paragraph,
the tradeoff between the solution accuracy and the required computational time is discussed.

h1

h0

The objective of the numerical experiments is twofold: (1) to


analyze the uniqueness of the inverse solution for different model
congurations, and (2) to determine an optimal parametrization of
the GMCS optimization algorithm as a tradeoff between the ability to
nd the inverse solution and computation time. The considered
models are shown in Fig. 2 as single and double-slab models with
different electrical properties and layer thicknesses. In particular, the
retrieval of fractures thin compared to the wavelength is investigated
with the double-slab model, with a thin air layer separating the two
slabs. Similar model congurations are set up for the laboratory
experiments (Section 4). For all these model congurations, numerical Green's functions Gxx() are generated and subsequently inverted
for the parameters of interest. The benet of this numerical approach
is that the inverted parameters can be compared to the true
parameters, which are known exactly. The errors in the estimated
parameters are expressed in terms of the standard deviation sx:

sx =

!1
2

where N is the total number of scenarios, n refers to a particular


scenario, whereas x and x are the retrieved and exact parameters,
respectively. The inversion results are presented in Table 1, which also
reports the arithmetic mean x and the standard deviation.
3.1. A: Single slab
A three-layer model is adopted here as sketched in Fig. 2a (airmaterial-PEC). The relevant parameters to be retrieved, both electric and
geometrical, are the relative permittivity of the slab r, 1, which can take
values of 2.5, 3.5 and 5; the reference electrical conductivity of the slabs
log10 3GHz, 1, as dened in Eq. (2) (2.5 and 1.2 Sm 1, more precisely
the units refer to rather than to its logarithm, however, we will use
these units in the text); the constant slope of the conductivity log10a
(12 and 10 Fm 1, similarly as discussed above); the thickness of the
slabs h1 (0.04 and 0.068 m) and the source height h0 (0.15 and 0.20 m).

slab 1
PEC

3. Numerical experiments

1 N
2
xT xn
N n=1 n

r,1 , 1( f )

hf

TEM

air

h1

r,1 , 1 ( f )

slab1

h2

r,2 , 2 ( f )

slab2

air
PEC

Fig. 2. Scenarios simulated in numerical experiments: (a) single slab lying on a PEC and
(b) double slab on PEC. TEM represents the transverse-EM horn antenna, h0 is the
antenna height, h1 and h2 are the rst and second slab thickness, hf stands for fracture
thickness. Similarly r and subscripts refer to the single or coupled slabs conguration.
Subscript number indicates the progression of the encountered media. For PEC r = 1
and pec = . S indicates the source position.

Every possible parameter ordering generates a new scenario, resulting in


48 different combinations. Inversions were performed in a large
parameter space (2 r, 1 10; 4 log 3GHz, 1 1 Sm 1; 15
log a 8 Fm 1; 3 10 2 h1 13 10 2 m; 0.15 h0 0.25 m).
Fig. 3 represents 2-D slices of the 5-D objective function topography
for several parameter pairs. Stars represent the true parameter values.
For each parameter a discretization of 100 values was adopted, resulting
in 10,000 objective function computations per plot. First, we can observe
a signicant oscillating behavior of the objective function for the
presence of local minima. This justies the use of an advanced global
optimization algorithm for minimizing the function. For all parameter
pairs involving , a minimum valley can be observed in the direction of
, denoting a relatively low sensitivity for this value of the conductivity
compared to the other parameters. In general, the minimum with
respect to the electric permittivity r is well dened, except for the
parameter pair r h1 (Fig. 3(c)), where the two parameters are
negatively correlated. This can be explained by the opposite effect of the
electric permittivity and layer thickness on the wave propagation time.
Yet, a global minimum can be identied because the signal amplitude is
used; in addition, both the global transverse electric and magnetic
reection coefcients are function of n and hn + 1 but not of hn (Slob and
Fokkema, 2002). As expected, the antenna height (h0) is always well
dened because we know the physical parameters of air.

30

C. Patriarca et al. / Journal of Applied Geophysics 74 (2011) 2637

Table 1
Results of numerical experiments for single slab model conguration. For each parameter the correct values, the mean x, the standard deviation std and the objective function value
(b) are given as a measure of the accuracy.
Parameter

r, 1

Correct value
x
std
(b)

2.50
2.49
5.32e5
1.47e5

3.5
3.49
2.07e5
7.00e6

5.0
5.0
3.54e5
5.74e6

log103GHz, 1(Sm 1)

log10a(Fm 1)

2.5
2.5
3.61e5
2.81e6

12.0
12.0
6.14e5
3.45e6

1.2
1.2
1.95e5
1.48e5

Because in numerical experiments the problem solution is known,


the global multilevel search algorithm is terminated when the objective
function value has reached a tolerance value (Huyer and Neumaier,

Fig. 3. 38 GHz response surfaces of the objective functions log


correspond to the true parameter values.

10()

h0(m)
10.0
10.0
1.02e5
1.37e5

0.15
0.15
5.95e8
1.07e5

h1(m)
0.20
0.20
3.02e8
8.03e6

0.04
0.04
3.75e7
9.95e6

0.068
0.068
2.28e7
8.39e6

1999) of the global minimum equal to 10 3. This means that the GMCS
is stopped when a relative error b 0.1% in the optimal objective function
value is reached. The optimal value refers to the global search algorithm;

for numerical data in the (a) r , (b) r h0, (c) r h1, (d) h0 , (e) h1 , (f) h0 h1 planes. Stars

C. Patriarca et al. / Journal of Applied Geophysics 74 (2011) 2637

local search, in fact, systematically improves the results. Global search


stop is also imposed when a user-dened maximum number of function
calls is reached. The number of function evaluations that guarantees
numerical convergence is determined by setting a maximum threshold
value between the exact parameter value and the inverted one. This
choice gives insight into the minimum number of iterations that have to
be performed to obtain a successful convergence as a function of the
number of unknowns. The maximum number of function evaluations
was set to 50 thousand, while the maximum splitting levels smax was set
to 50. The number of function evaluations actually used and the split
level reached were recorded. They can vary depending on the parameter
of interest, as the less demanding inversions demonstrated. Inversion
results are presented in Table 1. For all the simulated scenarios the
mean values x match the solutions, and the standard deviation for all the
parameters show that errors achieved are physically negligible.
3.2. Conguration B: Overlapped slabs
We used a ve-layer model to investigate the response functions of a
multilayered medium, as shown in Fig. 2b (air-materialair-material
PEC). To limit the number of unknowns, inversions were run assuming
the linear increase of the electrical conductivity, the antenna height and
the second layer thickness as known parameters. The thickness of the
underlying layer is a parameter occurring in the global TE and TM modes
reection coefcients. However, In this model conguration we intend
to retrieve geometrical and electric parameters, i.e., the electric
permittivity of the rst and second layers (r, 1 = 2.5 and 5; r, 2 = 3.5
and 6), the reference electrical conductivity of the rst and second layer
log10 3GHz, 1 = 2.5 and 1.2; log10 3GHz, 2 = 2.5 and 1 Sm 1),
the rst layer thickness (h1 = 0.04 and 0.068 m), and the air layer
thickness hf (1 10 3 and 5 10 3 m), resulting in 6 unknowns.
The number of simulated scenarios is thus 64; the parameter space used
for the fracture retrieval is 5 10 4 hf 3 10 2 m, while for other
parameters it is the same as discussed in Subsection 3.1.
The effectiveness of the inverse method in retrieving the fracture
thickness is tested for apertures below the commonly considered
resolution limit of one quarter of the wavelength. The possibility for
the sub-bandwidth resolution is given by the effective removal of the
antenna effects, and by the subsequent exploitation of the signal
amplitude; this information is then requested to t a layered earth
model through the global and local reection coefcients for a zerooffset conguration, which interprets the amplitude oscillations as the
sum of the contribution given by each layer in the model.
Fig. 4 shows the response surfaces for the ve layer model. The
relation between all the parameters with respect to the fracture
aperture, and the effect of the second layer permittivity on the rst
layer thickness are analyzed. Again local minima occur as a result of the
objective function oscillation. Fig. 4 (ab) illustrates the lack of local
minima and a relatively simple topography, where the global minimum
is found easily. The topography is more complex when the electrical
conductivity is shown. A good sensitivity that ensures a unique solution
is depicted with respect to 1 in Fig. 4 (c); the second layer electrical
conductivity 2 has no inuence on the air layer thickness retrieval as
can be seen in Fig. 4 (d). As expected, an inverse correlation exists within
the parameter pair h1 hf as shown in Fig. 4(e): an increase in one layer
thickness is coupled by a decrease in the other. The slope of this
correlation is explained by the difference in the two permittivity values:
a small increase of the rst layer thickness h1 corresponds to a large
decrease of the air layer to account for the propagation time inside the
two materials. Fig. 4(f) conrms that the bottom layer parameters do not
inuence the fracture thickness retrieval as the rst layer does. A
correlation exists between the fracture thickness and the rst layer
properties. The relatively simple objective function topography of Fig. 4
(f) allows us to treat the second layer thickness as a known parameter.
Alternatively, the second layer conductivity, because of the presence of
an unambiguous global minimum (Fig. 4(d)), could be assumed to be

31

known without simplifying the inverse problem but reducing its


dimensionality. Both those parameters, h2 and 2, affect the TE and TM
reection coefcients computation; therefore, the bottom layer effect
with respect to the fracture thickness retrieval can only be partially
investigated because we treat its thickness as a known parameter.
Results of numerical simulations are given in Table 2. In an effort to
reduce the computation time, we set again the global search stopping
criterion for objective function values less than 10 3.
4. Laboratory experiments
The numerical experiments described in Section 3 helped to
understand the theoretical potential of the radar system, before
performing laboratory measurements in a real physical and noisy
environment. Several physical models are analyzed, rst performing
measurements on single slabs for characterization purposes, then piling
up the concrete slabs with increasing void spacing between them. The
experiments are repeated with concrete overlapped by a plaster slab.
These laboratory investigation mimic real structure/object studies
for damage diagnosis. Concrete and plaster slabs are used to test the
capability of identifying reections from multilayers and eventually
from discontinuities occurring inside the same material, in slightly
different media, or at the interface between different media. These
discontinuities are represented by sub-wavelength air layers running
parallel at different depths with respect to the rst reection interface.
With this technology it may be possible to detect multiple dielectric
leaves in multilayered walls, such as internal hidden openings in
concrete, detachments in the plaster/concrete interface and adherence
loss between layers. The determination of electromagnetic parameters
makes this technology suitable for geometric locations of subsurface
features.
Two 17 dm3 slabs made from concrete (0.5 0.5 0.068 m3) and one
10 dm3 plaster slab (0.5 0.5 0.04 m3) are used for laboratory tests.
The two Portland concrete based specimens (referred to as C1 and C2)
differ slightly in mixing proportions; an exact characterization as a
function of the mix is beyond the scope of this study, which focuses on
the ability of material and fracture characterization without a priori
information. The properties of the samples are thus expected to be
different, because of the cement hydration effect and different initial
water-to-binder ratios. The only ne aggregate used in the mix is quartz
river sand. The sandcement volume ratios used are 4:1 to which water
was added until a homogeneous mix was obtained. The slab specimens
were cast in a timber mould lying horizontally on the oor;
subsequently, the slabs were manually compacted inferring vibrations
and using a smoothing rod in order to obtain at interfaces. The
specimens were cured for 106 days and left to air dry at room
temperature before measurements were performed. The drying time
recommended by the manufacturer for such concrete is 28 days to reach
a value of 32.5 n 52.5 MPa, where n indicates the material uniaxial
compressive strength; however, complete concrete drying may require
up to 6 months. The plaster samples were cast using commercially
available product (calcium sulphate hemihydrate). The powder-towater weight ratio recommended by the manufacturer is 1:0.7, but a
ratio equal to 1:2 was used instead, to avoid air bubbles formation inside
the mix and, at the same time, prevent crack formation.
Two main scenarios are considered: measurements are taken for
each slab separately to determine their electromagnetic properties, as in
this case the inverse problem can accurately be solved (Fig. 2 (a),
Table 3); subsequent measurements were taken using overlapped slabs
(Fig. 2 (b), Table 4) with increasing void spacing (from approximately 1
to 22 mm) with the intent of simulating varying opening interfaces with
a rough planar surface inside the masonry. To minimize the computation of the full-waveform inversion, 6 MHz time steps were taken, with a
total of 834 frequencies to be evaluated; this selection was made in
order to preserve most of the signal information; no gain functions were
applied to enhance the signal-to-noise ratio.

32

C. Patriarca et al. / Journal of Applied Geophysics 74 (2011) 2637

Fig. 4. 38 GHz response surfaces of the objective functions log 10() for numerical data in the (a) r, 1 hf, (b) r, 2 hf, (c) hf 1, (d) hf 2, (e) h1 hf, (f) hf h2 planes. Stars
correspond to the true parameter values.

4.1. Conguration A: Single slabs results and discussion


Frequency dependence of the electromagnetic parameters is
considered in this study. Dedicated inversion for the electric permittivity

and for the electric conductivity, with the aim of getting insight into the
behavior of the tested materials over the whole frequency bandwidth,
are performed for numerical Green's functions and for the three slabs
C1, C2 and P1. The numerical responses are generated by assigning a

Table 2
Results of numerical experiments for double slab model conguration. For each parameter the correct values, the mean x, the standard deviation std and the objective function value
(b) are given as ameasure of the accuracy.
Parameter

r, 1

Correct value
x
std
(b)

2.5
2.5
8.14e6
1.17e5

r, 2
5.0
4.99
1.1e05
1.54e6

3.5
3.49
1.84e5
3.01e6

6.0
5.99
1.08e3
1.20e5

log103GHz, 1(Sm 1)

log103GHz, 2(Sm 1)

h1 (m)

2.5
2.5
6.26e5
8.69e7

2.5
2.49
1.58e4
1.50e6

0.04
0.04
2.62e5
2.24e5

1.2
1.2
8.97e5
1.21e5

1.0
1.0
4.81e5
9.82e6

hf (m)
0.068
0.068
1.85e7
1.96e6

0.005
0.005
3.46e7
2.70e6

0.001
9.94e4
1.61e5
1.79e5

C. Patriarca et al. / Journal of Applied Geophysics 74 (2011) 2637


Table 3
Results from the inversion of laboratory measurements on single slabs. 3GHz and a are
the reference electrical conductivity and the linear gradient as dened in Eq. (2).
Conguration

log 3GHz
(Sm 1)

log a
(Fm 1)

h0
(m)

h1
(m)

h1meas
(m)

(b)

Slab C1
Slab C2
Slab P1

4.19
4.69
2.52

1.38
1.17
2.54

10.56
10.54
11.99

0.184
0.185
0.185

7.34e2
6.97e2
3.86e2

6.68e2
6.65e2
3.95e2

1.67e1
1.07e1
5.39e2

5.5
5
4.5

4
3.5

Table 4
Results from the inversion of laboratory measurements on overlapped concrete and
plaster slabs.

Slab C1 on C2

Real value

Slab P1 on C2

Real value

h0 (m)

h1 (m)

h2 (m)

hf (m)

hfmeas (m)

(b)

0.188
0.187
0.185
0.198
0.195
0.189
0.184

0.185
0.183
0.181
0.186
0.183
0.187
0.184

7.17e2
7.18e2
7.19e2
7.23e2
7.18e2
7.20e2
7.26e2
6.68e2
3.79e2
3.78e2
3.78e2
3.67e2
3.71e2
3.76e2
3.75e2
3.95e2

6.97e2
6.95e2
6.92e2
6.92e2
6.88e2
6.89e2
6.87e2
6.65e2
6.97e2
6.99e2
7.00e2
6.88e2
6.90e2
6.98e2
6.98e2
6.65e2

4.74e3
6.08e3
8.00e3
1.06e2
1.51e2
1.92e2
2.35e2

2.70e3
4.80e3
6.69e3
1.28e2
1.42e2
1.94e2
2.23e2

2.39e3
3.53e3
6.41e3
8.96e3
1.50e2
2.20e2

3.37e3
5.21e3
7.66e3
1.01e2
1.71e2
2.20e2

0.166
0.163
0.159
0.142
0.137
0.146
0.139
0.137
0.130
0.125
0.122
0.126
0.107
0.115

Fracture aperture is given by the irregularities on the slabs surfaces; the parameter
hfmeas was not measured for these congurations.

Concrete 1
Concrete 2
Plaster 1
Synthetic

3
2.5
2

f (GHz)

b
0.5
1

log10

constant value to the electric permittivity, and a linear behavior to the


conductivity. Results are shown in Fig. 5(a) and (b). Inversions were run
on twelve 400 MHz wide sub-bands to cover the entire frequency range,
and a constant model is assumed to represent both the electric
permittivity and the electrical conductivity in these sub-bands.
In both the numerical and laboratory data, the permittivity is well
dened in the sub-bands, and it is in agreement with the real value or
with the full-bandwidth inversions. In the case of real data, the
electric permittivity does not show frequency dispersion over the 3
8 GHz bandwith (Fig. 5(a)); inversions performed for the electric
conductivity conrm the low sensitivity with respect to this
parameter. Despite the discrepancy between the real values or the
full-bandwidth and the sub-bandwidth inversion, a signicant
frequency dependence is visible on the considered frequency range
(Fig. 5(b)). Concrete specimens and plaster show different r
signatures, although the order of magnitude is the same (Fig. 5(a))
for the same age samples. The values of the electrical conductivity
increases with frequency for all the samples considered. Results show
a linear behavior in agreement with that expressed in Eq. (2). The
mismatch between the curves and the independent marks may be
attributable to modeling errors on the retrieved values when only a
limited frequency band is considered in the inversions; moreover,
neglecting the frequency dependence in frequency sub-intervals may
be responsible for the lack of sensitivity. The sub-interval inversion
results are very consistent with the numerical model, which excludes
errors due to lack of convergence.
The electric permittivity and the electric conductivity values seem to
match fairly well with literature data on concrete properties (Lai et al.,
2009; Laurens et al., 2005; Robert, 1998; Soutsos et al., 2001; Tsui and
Matthews, 1997). Geometrical parameters accuracy could be evaluated
because the model geometry was accurately measured, i.e., the antenna
height and the slabs thicknesses. Inversions were run in a relatively large
parameter space (2 r 8; 3 log3GHz 1 Sm 1; 13
loga 8 Fm 1; 0.17 h0 0.22 m; 0.01 h1 0.10 m), where the
expected solutions were believed to occur. The average distance from
the antenna aperture to the target is 15 cm. The distance from the target

Conguration

33

1.5

Concrete 1
Concrete 2
Plaster 1
Synthetic

2
2.5
3
3

f (GHz)
Fig. 5. Frequency dependence of the relative electric permittivity r (a) and the electrical
conductivity (b) for the three specimens. The symbols represent the parameters
obtained from the inversion on a 400 MHz frequency sub-range considering no losses
occurring; lines represent the true parameters in case of numerical data, and the
parameter retrieved in the full-bandwidth inversions in case of real data.

to the antenna phase center is also provided in data inversions. The


equivalent point source of the antenna is situated at approximately
3.5 cm inward from the antenna aperture (Lambot et al., 2004a). The
measured and modeled Green's function are shown in the frequency
(Figure 6(a)) and time domains (Fig. 6(b)) for a single concrete slab
(C2). The frequency and time domain responses for a single plaster slab
(P1) are shown in Fig. 6(c) and (d), respectively.
In the frequency domain the phase is accurately described for both
samples. The Green's function amplitude is also well reproduced, and
especially for the concrete specimen (Fig. 6(a)) the effect of dielectric
losses is clear: the amplitude is strongly attenuated at high frequencies.
The larger period of the oscillation observed in the |Gxx| of the plaster
sample (Fig. 6(c)) reveals that the two-way travel time in the analyzed
slab is smaller than in the concrete slab. A decrease in the amplitude is
less easy to observe because of the smaller linear gradient in the
electrical conductivity frequency dependence (Table 2).
The rst arrival is registered at t =1.19 ns and t= 1.20 ns in Fig. 6(b)
and (d), respectively. The rst interface (air-sample) gives a slightly
higher reection for the concrete specimen because of the higher
dielectric contrast with air, but the time domain reections are comparable. The PEC surface reection is less evident in the concrete sample,
where a bigger attenuation leads to a smaller reection peak at 2.20 ns.
Smaller r, reduced thickness (about 4 cm) and less dissipation in the
plaster result in a much stronger reection from the PEC at 1.60 ns.
While the inverse Fourier transform gxx is perfectly modeled for the
plaster sample (Fig. 6(d)), a mist appears in the concrete sample before
and after the PEC reection peak. This may be due to the fact that a

34

C. Patriarca et al. / Journal of Applied Geophysics 74 (2011) 2637

6000

400

Measured
Modeled

4000

200

2000
0

100
4

f (GHz)
4

Gxx(rad)

Measured
Modeled

300

gxx

|Gxx |

0
100

200

0
300

2
4
4

400

d
Measured
Modeled

800

Measured
Modeled

200
3

f (GHz)
4

Gxx(rad)

400

5000
0

600

gxx

|Gxx |

10000

t (ns)

f (GHz)

0
200

400

0
600

2
4
3

f (GHz)

800

t (ns)

Fig. 6. Measured and modeled Green's functions presented in the frequency (amplitude |Gxx| and phase Gxx) and in the time gxx domains for the antenna illuminating one concrete
slab C2 (a and b), and one plaster P1 slab (c and d).

homogeneous medium is assumed for concrete, while inlayer heterogeneities may not be negligible. The prole could be depth-dependent
because the concrete was cast on a horizontal mold; gravity caused
zonation may be responsible for the inhomogeneity observed in the
measured data. Laboratory experiments results are shown in Table 3.
The objective function value increases in laboratories experiments, as
can be observed by comparing the b values of Tables 3 and 4 with the
values reported in Tables 1 and 2. The numerically generated signal to be
inverted for, exactly satises the model requirements; conversely, in
laboratory data, additional information or noise can be present which
are not described by the model.
4.2. Conguration B: Overlapped slabs results and discussion
To limit the number of unknowns, the electromagnetic properties
retrieved in single slabs congurations are used as known parameters
in the inverse problem; the geometrical properties of the conguration are optimized here. This approach was adopted to test the ability
of thin fracture retrieval given some a priori information like the
electromagnetic parameters. The measured and modeled Green's
functions for only four congurations are presented in Figs. 7 and 8, in
the frequency and in the time domain.
In Fig. 7(a) the Green's functions for the concrete sample C2 overlapping the concrete sample C1 are shown. The effect of the frequency
dependence of the electrical conductivity is still clearly visible as a
decrease in amplitude of the Green's function with increasing frequency.
Compared to the single slab conguration, a higher order oscillation is
visible in all the frequency domain |Gxx| plots. This oscillation tends to
damp at high frequencies, even though its presence is still observable
both in the measured and modeled signals. In general, the curve is very

well reproduced by the model, but the signal presents high local
oscillations that are not accounted for. At high frequency, where inlayer
heterogeneities become detectable as thin layers with a thickness less
than 1/10 of the minimal wavelength (less than 3.7 10 3 m), the
mismatch is larger. High order oscillations are present in all the
measured signals where concrete slabs are overlapped, and to a much
lesser extent when plaster overlaps concrete (Figs. 7 and 8). In the time
domain, except for the rst reection, the concrete over concrete
modeling of the late arrivals are not very well captured. Conversely, for
plaster over concrete slabs, the peak amplitudes of the rst reections
are not well described, while late arrivals are properly accounted for.
Impedance mismatches within the setup elements can also cause
uncertainties that appear as ripples superimposed on the measured
data, as shown in Figs. 7 (a) and 8 (a). The phase shape in Fig. 7 (a) is well
reproduced, except locally at 7.5 GHz, where a minor mismatch (also
evident in the amplitude plot) is not properly accounted for. In the time
domain (Fig. 7 (bd)) the three reections are well distinguishable. A
comparison between the plots of Figs. 6 and 7 (b) reveals exactly the
same time arrivals of the gxx rst two peaks occurring at 1.22 ns and
2.25 ns. The difference in the peak magnitude at 2.25 ns is due to weaker
reection generated by the concrete/air interface compared to the
concrete/PEC reection; a third reection from the copper sheet is
observed to occur at 3.23 ns in Fig. 7 (b). This reection peak is much
more attenuated because of the damping on the two-way travel path of
the wave through the samples with relative high values of the electric
permittivity (r, 1 = 4.19 ; r, 2 = 4.69). The mismatch observed in the
time domain of Fig. 6 (b) at 2 ns is also present here, and slightly affects
modeling of the signal in the second slab, even though it can be
considered negligible with respect to the main reection events. Higher
contrast in the electric permittivity of concrete and air should lead to

C. Patriarca et al. / Journal of Applied Geophysics 74 (2011) 2637

6000
4000

b
Measured
Modeled

100
4

f (GHz)
4

Gxx(rad)

Measured
Modeled

200

2000
0
3

400
300

gxx

|Gxx |

35

0
100

200

0
300

2
4
3

400
0

6000
4000

d
Measured
Modeled

400

Measured
Modeled

100
4

f (GHz)
4

Gxx(rad)

200

2000
0
3

300

gxx

|Gxx |

t (ns)

f (GHz)

0
100

200

0
300

2
4
3

f (GHz)

400
0

t (ns)

Fig. 7. Measured and modeled Green's functions presented in the frequency and in the time domains for the antenna illuminating two overlapped slabs: concrete slab C2 on C1 (a and
b), and plaster slab P1 on C2 (c and d); the air layer between the two slabs is given by the irregularities on the slab surfaces and is in the order of a few millimeters (~ 1 mm). The
dashed vertical lines individuate the main events discussed in the text.

higher reection responses, but the backscattered signal has to travel


inside concrete once again before being registered.
In Fig. 7(c) the amplitude and phase of the complex Green's
function for plaster slab P1 overlapping concrete slab C2 are presented.
The phase (Fig. 7(c)) is perfectly reproduced over the entire frequency
range, and the amplitude t is only affected by noise at high
frequencies. The results are very consistent with the two single slab
plots shown before: a high period oscillation (due to the thin plaster
sample) hosts a lower period oscillation due to the presence of concrete
that thus affects the signal. The inverse Fourier transform is represented
in Fig. 7(d). As expected, the reection from the fracture is not damped
as much as in the previous case: it occurs at 1.63 ns, whereas the rst
reection is at 1.20 ns. This reection is not masked by the main
reection, and is of the same order of magnitude. The PEC amplitude
reection (2.62 ns) is also much less attenuated because the wave
travels in a medium (plaster) with low electric permittivity
(r, 1 = 2.52) and conductivity, and the plaster slab is thinner compared
to concrete. The change in signal amplitude at different frequencies is
due to dispersion effects.
We further consider a high aperture fracture (22 mm) effect
occurring in the same conguration (Concrete C1 on concrete C2).
Results of the modeled and measured complex Green's functions are
plotted in Fig. 8 in both frequency and time domains. The modeled and
measured phases again show an excellent agreement. The low period
oscillation superimposed on the higher period oscillation (Fig. 8(a)) is
much less marked for wide fractures in concrete. Its presence is still
observable at lower frequencies, but it becomes less evident after 6 GHz,
where signicative attenuation effects make the signal shape atter,

masking up the fracture effect from the |Gxx| signal. In time domain, four
main reections are expected from the four main interfaces. The rst
reection peak is well visible in Fig. 8 (b) at 1.19 ns, the second reection
occurs at 2.10 ns, while the reection from the third interface occurs at
2.25 ns, giving a t3 2 = 0.15 ns, in agreement with the theoretical t
for a wave traveling in a 22 mm air interval that should be equal to
t = 0.15 ns. The last reection from the PEC occurs at 3.33 ns, as shown
in Fig. 8 (b).
The match between the measured and modeled amplitude of the
Green's functions for the plaster overlying concrete (Fig. 8(c)) is good
in all the frequency range, presenting only local oscillations. In the
time domain for such congurations it is more difcult to distinguish
between the second and third reections, while the last reection
from the copper sheet is clearly visible at 2.73 ns. The reection from
the plasterair interface (the second, occurring at 1.72 ns) has the
same magnitude as the reection from the airplaster interface (the
rst in Fig. 8 (d) at 1.19 ns), but it overlaps the airconcrete reection
(the third) expected at 1.89 ns.
Inversions results are reported in Table 4. The mean error in the
slab C1 thickness retrieval is 7.8%, amounting to about 4 mm overestimation of the actual thickness. The C2 thickness is retrieved with a 5%
error. The fracture thickness is generally overestimated by a few
millimeters, with larger errors arising in concrete than in plaster
congurations (Table 4). The plaster slab thickness is retrieved with a
3.5% mean error with respect to the real size measured with millimeter
accuracy. The objective function values increase with decreasing
fracture aperture, showing a loss of accuracy when the resolution limit
is approached.

36

C. Patriarca et al. / Journal of Applied Geophysics 74 (2011) 2637

6000

Measured
Modeled

4000

100
3

f (GHz)
Gxx(rad)

200

0
300

2
3

f (GHz)

400

Measured
Modeled

t (ns)

400

4000

Measured
Modeled

300
200

2000

100
3

f (GHz)
4

gxx

|Gxx |

6000

Gxx(rad)

0
100

Measured
Modeled

200

2000
0

400
300

gxx

|Gxx |

0
100

200

0
300

2
4
3

400

f (GHz)

t (ns)

Fig. 8. Measured and modeled Green's functions presented in the frequency and in the time domains for the antenna illuminating two overlapped slabs: concrete slab C2 on C1 (a and
b), and plaster slab P1 on C2 (c and d); the air layer between the two slabs is 22 mm. The dashed vertical lines individuate the main events discussed in the text.

5. Conclusion
The problem of localizing and characterizing hidden fractures and
detachments suffers from limitations, and has been assessed mainly
qualitatively in non-destructive testing research. We have proposed a
quantitative analysis based on full-waveform inversion of the GPR
signals. An exact model capable of reproducing the true crack
response has been adopted; this exploits the magnitude and phase
of the reection coefcient to estimate the crack depth and aperture,
provided the electric properties of the lling are known. The results of
this study limit to common calcium sulphate based plaster and
Portland cement based concrete without iron rebars.
Numerical and laboratory experiments showed that the proposed
system and large bandwidth are appropriate to detect 1 mm planar
fractures in concrete or between plaster and concrete. The real electric
properties and the geometry of the problem are retrieved by true
amplitude inversion of the ltered radar data on single slabs.
Frequency dependent analysis of reection data can provide
relevant information without a priori knowledge of the material.
The interfaces of a multilayered structure are clearly visible in the
time domain; detachments generate strong reections.
In the multi-layered congurations the thickness of the layers are
well retrieved given their electric properties. Fracture thickness is
retrieved with an error less than 10% for thickness around 1/3 of the
minimum wavelength used. On the other hand, it appears impossible
to determine quantitatively the thickness of the air layer directly from
the time and frequency domain representations. A rst analysis of
frequency and time domain plots is useful, because the Green's
functions indicate single or multi-layered conguration. This indication can be obtained from the reection peaks in gxx, and from the

number of oscillation periods in |Gxx|. Once the problem is laid out, a


great advantage is given by the full-waveform inversion algorithm
that is capable of retrieving sub-wavelength fractures in different
permittivity materials with a very good approximation.

Acknowledgments
The research leading to these results has received funding from the
European Community's Seventh Framework Programme [FP7/20072013] under grant agreement no. 213651. The research was further
supported by the DIGISOIL project, nanced by the EC under the 7th
Framework Programme for Research and Technology Development,
Area Environment, Activity 6.3 Environmental Technologies and
FNRS (Belgium). We also thank the editor and two unknown
reviewers for their helpful comments.

References
Belli, K., Zhan, H., Wadia-Fascetti, S., Rappaport, C., 2009. Comparison of the Accuracy of
2D vs. 3D FDTD Air-Coupled GPR Modeling of Bridge Deck Deterioration. Research
in Nondestructive Evaluation 20, 94115.
Binda, L., Zanzi, L., Lualdi, M., Condoleo, P., 2005. The use of georadar to assess damage
to a masonry Bell Tower in Cremona, Italy. NDT and E International 38, 171179.
Bradford, J.H., 2007. Frequency-dependent attenuation analysis of ground-penetrating
radar data. Geophysics 72, J7J16.
Brown, J., Nichols, J., Steinbronn, L., Bradford, J., 2009. Improved GPR interpretation
through resolution of lateral velocity heterogeneity: example from an archaeological site investigation. Journal of Applied Geophysics 68, 38.
Buyukozturk, O., Yu, T.Y., 2009. Far-eld radar NDT technique for detecting GFRP
debonding from concrete. Construction and Building Materials 23, 16781689.
Carino, N., 2008. Nondestructive test methods. Concrete Construction Engineering
Handbook. CRC Press, p. 584.

C. Patriarca et al. / Journal of Applied Geophysics 74 (2011) 2637


Carrera, J., Neuman, S.P., 1986. Estimation of aquifer parameters under transient and
steady state conditions: 1. Maximum likelihood method incorporating prior
information. Water Resources Research 22, 199210.
Daniels, D.J., 2004. Ground Penetrating Radar, 2nd Edition. The Inst. Electrical Eng, London.
Drobert, X., Iaquinta, J., Klysz, G., Balayssac, J.P., 2008. Use of capacitive and GPR techniques
for the non-destructive evaluation of cover concrete. NDT and E International 41, 4452.
Huyer, W., Neumaier, A., 1999. Global optimization by multilevel coordinate search.
Journal of Global Optimization 14, 331355.
Jadoon, K.Z., Lambot, S., Slob, E.C., Vereecken, H., 2011. Analysis of horn antenna transfer
functions and phase-center position for modeling off-ground GPR. Geoscience and
Remote Sensing, IEEE Transactions 114.
Jol, H.M., 2009. Ground Penetrating Radar Theory and Applications. Elsevier,
Amsterdam, p. 524.
Lagarias, J.C., Reeds, J.A., Wright, M.H., Wright, P.E., 1998. Convergence properties of the
NelderMead Simplex method in low dimensions. SIAM Journal on Optimization 9,
112147.
Lai, W.L., Kou, S.C., Tsang, W.F., Poon, C.S., 2009. Characterization of concrete properties
from dielectric properties using ground penetrating radar. Cement and Concrete
Research 39, 687695.
Lambot, S., Javaux, M., Hupet, F., Vanclooster, M., 2002. A global multilevel coordinate
search procedure for estimating the unsaturated soil hydraulic properties. Water
Resources Research 38 (11), 1224. doi:10.1029/2001WR001224.
Lambot, S., Slob, E.C., van den Bosch, I., Stockbroeckx, B., Vanclooster, M., 2004a. Modeling
of ground-penetrating radar for accurate characterization of subsurface electric
properties. IEEE Transactions on Geoscience and Remote Sensing 42, 25552568.
Lambot, S., Slob, E.C., van den Bosch, I., Stockbroeckx, B., Scheers, B., Vanclooster, M.,
2004b. Estimating soil electric properties from monostatic ground-penetrating
radar signal inversion in the frequency domain. Water Resources Research 40,
W04205. doi:10.1029/2003WR002095.
Lambot, S., van den Bosch, I., Stockbroeckx, B., Druyts, P., Vanclooster, M., Slob, E.C.,
2005. Frequency dependence of the soil electromagnetic properties derived from
ground-penetrating radar signal inversion. Subsurface Sensing Technologies and
Applications 6, 7387.
Lambot, S., Slob, E., Vanclooster, M., Vereecken, H., 2008. Hydrogeophysical techniques
for soil pollution characterization and monitoring: ground penetrating radar. In:
Simeonov, L., Sargsyan, V. (Eds.), Soil Chemical Pollution, Risk Assessment, Remediation and Security. NATO Security Through Science Series. Springer, Amsterdam,
pp. 183202.
Lambot, S., Slob, E., Rhebergen, J., Lopera, O., Jadoon, K.Z., Vereecken, H., 2009. Remote
estimation of the hydraulic properties of a sand using full-waveform integrated
hydrogeophysical inversion of time-lapse, off-ground GPR data. Vadose Zone Journal
8, 743754.
Laurens, S., Balayssac, J.P., Rhazi, J., Klysz, G., Arliguie, G., 2005. Non-Destructive
Evaluation of Concrete Moisture by GPR: Experimental Study and Direct Modeling.
R I L E M Publications, pp. 827832.
Lopera, O., Milisavljevic, N., Lambot, S., 2007a. Clutter reduction in GPR measurements
for detecting shallow buried landmines: a Colombian case study. Near Surface
Geophysics 5, 5764.

37

Lopera, O., Slob, E.C., Milisavljevic, N., Lambot, S., 2007b. Filtering soil surface and
antenna effects from GPR data to enhance landmine detection. IEEE Transactions on
Geoscience and Remote Sensing 45, 707717.
Maierhofer, C., Leipold, S., 2001. Radar investigation of masonry structures. NDT and E
International 34, 139147.
Malhotra, V.M., Carino, N.J., 2003. Handbook on Nondestructive Testing of Concrete
Second Edition. CRC Press.
Morsy, M.S., 1999. Effect of temperature on electrical conductivity of blended cement
pastes. Cement and Concrete Research 29, 603606.
Nadakuduti, J., Genda, C., Zoughi, R., 2006. Semiempirical electromagnetic modeling of
crack detection and sizing in cement-based materials using near-eld microwave
methods. Instrumentation and Measurement, IEEE Transactions 55, 588597.
Orlando, L., Slob, E., 2009. Using multicomponent GPR to monitor cracks in a historical
building. Journal of Applied Geophysics 67, 327334.
Robert, A., 1998. Dielectric permittivity of concrete between 50 Mhz and 1 Ghz and GPR
measurements for building materials evaluation. Journal of Applied Geophysics 40,
8994.
Sambuelli, L., Calzoni, C., 2010. Estimation of thin fracture aperture in a marble block by
GPR sounding. Bollettino Di Geosica Teorica Ed Applicata 51, 239252.
Slob, E.C., Fokkema, J., 2002. Coupling effects of two electric dipoles on an interface.
Radio Science 37, 1073. doi:10.1029/2001RS2529.
Slob, E., Sato, M., Olhoeft, G., 2010. Surface and borehole ground-penetrating-radar
developments. Geophysics 75, A103A120.
Soutsos, M.N., Bungey, J.H., Millard, S.G., Shaw, M.R., Patterson, A., 2001. Dielectric
properties of concrete and their inuence on radar testing. NDT and E International
34, 419425.
Stone, W.C., 1997. Electromagnetic Signal Attenuation in Construction Materials,
NISTIR; 6055. U.S. Dept. of Commerce, Technology Administration, National
Institute of Standards and Technology, Gaithersburg, Md.
Topp, G., Davis, J.L., Annan, A.P., 1980. Electromagnetic determination of soil water
content: measurements in coaxial transmission lines. Water Resources Research
16, 574582.
Tsoias, G.P., Hoch, A., 2006. Investigating multi-polarization GPR wave transmission
through thin layers: implications for vertical fracture characterization. Geophysical
Research Letters 33 5 pp.5 pp.
Tsui, F., Matthews, S.L., 1997. Analytical modelling of the dielectric properties of
concrete for subsurface radar applications. Construction and Building Materials 11,
149161.
van den Bril, K., Grgoire, C., Swennen, R., Lambot, S., 2007. Ground-penetrating radar
as a tool to detect rock heterogeneities (channels, cemented layers and fractures) in
the Luxembourg Sandstone Formation (Grand-Duchy of Luxembourg). Sedimentology 54, 949967. doi:10.1111/j.1365-3091.2007.00868.x.
West, L.J., Handley, K., Huang, Y., Pokar, M., 2003. Radar frequency dielectric dispersion
in sandstone: implications for determination of moisture and clay content. Water
Resources Research 39 (2), 1026. doi:10.1029/2001WR000923.
Widess, M.B., 1973. How thin is a thin bed? Geophysics 38, 11761180.

Das könnte Ihnen auch gefallen