Ultrasonics 42 (2004) 781786

www.elsevier.com/locate/ultras

Numerical time-domain simulation of diusive ultrasound in concrete

Frank Schubert *, Bernd Koehler
Fraunhofer-Institute for Nondestructive Testing, Branch Lab, EADQ, Kruegerstrasse 22, D-01326 Dresden, Germany

Abstract
Certain aspects of diusive ultrasound elds in concrete are still unknown and thus, systematic parameter studies using numerical
time-domain simulations of the ultrasonic propagation process could lead to further insights into theoretical and experimental
questions. In the present paper, the elastodynamic nite integration technique (EFIT) is used to simulate a diusive reverberation
measurement at a concrete specimen taking aggregates, pores, and viscoelastic damping explicitly into account. The numerical
results for dissipation and diusivity are compared with theoretical models. Moreover, the inuence of air-lled pores in the cement
matrix is demonstrated.

2004 Elsevier B.V. All rights reserved.
Keywords: Diusive ultrasound; Concrete; Time-domain simulation; Elastodynamic nite integration technique

1. Introduction
For elastic wave propagation in the ultrasonic regime,
concrete represents a strongly heterogeneous medium
consisting of statistically varying aggregates and pores
embedded in a cement matrix. Additionally, structural
components like reinforcing bars and tendon ducts aect
the wave propagation process signicantly. In nondestructive testing, the interpretation of the received
ultrasonic signals may become extremely dicult due to
multiple scattering, strong attenuation, and mode conversions. Even the use of sophisticated averaging and
reconstruction techniques like synthetic aperture methods only allows the detection of defects being signicantly larger than the maximum aggregate size. Thus,
localised damage zones developing at length scales
smaller than the size of the majority of aggregates cannot be detected with typically used frequencies of less
than 100 kHz.
Increasing the frequency up to a few hundred kHz
changes the character of the wave propagation process
in concrete dramatically. While wave propagation in the
low-frequency region below 100 kHz is dominated by
the coherent eld, ultrasound becomes diusive at
higher frequencies which means dominance of the

Corresponding author. Fax: +49-351-2648219.

E-mail address: schubert@eadq.izfp.fraunhofer.de (F. Schubert).

0041-624X/$ - see front matter
2004 Elsevier B.V. All rights reserved.
doi:10.1016/j.ultras.2004.01.040

incoherent scattered eld. This is shown in Fig. 1 where

the elastic wave eld caused by a 200 kHz ultrasonic
pulse was detected at a 9 9 cm2 outer surface of a
concrete specimen using scanning Laser vibrometer [1].
Yet before the rst parts of the primary wave eld reach
the lower surface of the sample, the acoustic energy
propagation is almost diusive with coherent wave
fronts no longer visible.
In this regime ultrasonic spectral energy density
Ez; t evolves in accordance with diusion equations. In
a specimen with lateral dimensions being small compared to the longitudinal dimension, z, a one-dimensional diusion equation with dissipation can be applied
[2,3]. The solution of this equationafter taking the
natural logarithmis given by
ln Ez; t 0:5 ln t C0 

z2
 rt;
4Dt

where D is ultrasonic diusivity, r the dissipation or

absorptivity, and C0 a factor connected with the initial
energy deposition at z 0 and time t 0.
Measuring Ez; t at a certain sensor location, z 6 0,
for dierent time windows and frequency bands and
performing a least-squares t to the data allows the
determination of the three parameters in Eq. (1) which
completely characterise the diusion process. The dissipation, r, describes the exponential decay at late times
while diusivity, D, is related to the arrival time of the
maximum energy density.

782

F. Schubert, B. Koehler / Ultrasonics 42 (2004) 781786

Fig. 1. Formation of a diusive ultrasound eld in a concrete specimen caused by 200 kHz pulse excitation. The rst picture shows aggregates and
pores at an outer surface of the sample (9 9 cm2 ). The following three pictures represent time snapshots of the ultrasonic wave eld obtained by
scanning laser detection at the same surface. The ultrasonic transducer is located at the top surface of the specimen.

2. Numerical time-domain simulation of a diusion

process
For the numerical investigations we simulated a real
experiment similar to that performed by Anugonda
et al. [2]. They used several cylindrical concrete specimens with diameters of 5 cm and a length of 46 cm. The
maximum aggregate size was 12.7 mm. The beams were
excited at one end by a contact transducer and the signals were detected at the other end by a point-like pin
transducer. Their experiment was done in narrow-band
technique, i.e. they used several narrow-band pulses
with varying center frequencies from 100 to 900 kHz
covering the whole frequency band of interest.
For the numerical calculations we used the elastodynamic nite integration technique (EFIT) which is
particularly suitable for simulation of ultrasonic waves
in strongly heterogeneous media [46]. For reasons of
computational eort, we used a two-dimensional model
of the beam according to a plane strain propagation
process. In contrast to the experiment described above, a
broadband excitation with frequencies from zero to

about 1.5 MHz was used. To avoid problems connected

with the generation of Rayleigh waves at the source
position, we used a transducer aperture with the same
width than the lateral beam dimension. The point-like
sensor was located at the other end of the beam at a
distance of 45 cm from the source.
The concrete was modelled by using gravel and sand
aggregates with an area fraction of about 66% embedded in a homogeneous cement matrix and according to a
standardised grading curve with a maximum aggregate
size of 12 mm. The acoustic parameters of the constituents of the model are summarised in Table 1. To
incorporate dissipative loss we used the KelvinVoigt
model of viscoelasticity introducing two additional
parameters into the governing equations, i.e. volume
and shear viscosity, gV and gS . For reasons of simplicity
and interpretation of the results, we used the same values for both, aggregates and cement matrix, being aware
that in fact dissipation is likely dominated by the viscoelasticity of the cement paste alone.
Air-lled pores were modelled as voids with stressfree boundary conditions. The porosity was 0.7% with

F. Schubert, B. Koehler / Ultrasonics 42 (2004) 781786

783

Table 1
Material parameters as used for the numerical simulations
Material

cL (m/s)

cT (m/s)

q (kg/m3 )

gV (Pa s)

gS (Pa s)

Cement
Aggregates

3950
4300 215

2250
2475 124

2050
2610 130

65
65 3.25

17
17 0.85

Air-lled pores were modelled as voids with stress-free boundary conditions. The values for cL , cT , and q are typical values for concrete well known
from eld measurements. They uctuate from one grain to another by 5% (maximum) around the given mean value. The values for gV and gS are only
roughly estimated. They were chosen in such a way that the signal damping in the simulation was close to typical experimental observations as given
in [2].

pore sizes uniformly distributed in the range between 0.3

and 2 mm.
Fig. 2 shows the numerical concrete model consisting
of about one million cells (grid spacing Dx  150 lm)
and various time snapshots of the wave propagation
process after pulse excitation at the bottom end of the
beam. The pictures are taken at equidistant time intervals of 22.9 ls and represent the absolute value of the
particle velocity vector. The coherent wave front is
clearly visible. Due to scattering, the higher frequencies
are strongly attenuated during propagation and thus,
the wave contains only the lower frequencies when it
arrives at the sensor position at z 45 cm. Due to mode
conversions at the lateral stress-free boundaries, transverse head waves followed by secondary ultrasonic

waves are generated and a signicant energy transfer

from the primary wave front to the rst secondary wave
is evident. One can already see from one of the last
snapshots that the low-frequency parts of the diusive
eld following the coherent wave fronts arrive earlier at
the top end of the beam than the high-frequency parts
which are still localised in the lower beam region.
Fig. 3 shows the calculated time-domain signal at the
receiver position. The rst arrival of the primary
coherent wave front is at t  107 ls. The rst secondary
wave (with larger amplitude) arrives at t  126 ls. Although these coherent phenomena are interesting, they
are not taken into account for further consideration. In
fact, the diusive eld behind the coherent wave fronts is
used for evaluation.

Fig. 2. Two-dimensional EFIT simulation of diusive ultrasound in a concrete beam. The rst picture shows the numerical concrete model consisting
of aggregates and pores embedded in a cement matrix (porosity 0.7%). The model contains 333 2997 grid cells. The following pictures are snapshots
of the elastic wave eld calculated in equidistant time steps of Dt 22:9 ls. They represent the absolute value of the particle velocity vector.

784

F. Schubert, B. Koehler / Ultrasonics 42 (2004) 781786

Fig. 3. Calculated time-domain signal after broadband pulse excitation

using the model from Fig. 2. The signal gives the normal particle
velocity, vz , at the receiver position at z 45 cm. The arrival of the
primary coherent wave front is at t  107 ls. The rst secondary echo
arrives at t  126 ls.

3. Results for dissipation and diusivity

The calculated data was analysed using standard
timefrequency techniques [2,3]. We used 40 time windows of length Dt 37:5 ls and 10 frequency windows
of width Df 100 kHz. The latter were centred at 100,
200, 300, . . ., 1000 kHz. The results are shown in Fig. 4
together with the corresponding least-squares curve ts
to Eq. (1), i.e. the solution of the 1-D diusion equation.
It can be seen that the late time curve ts as well as the
early time ts for the high frequencies are quite good
while the early time ts for the lower frequencies are
rather bad. This is most likely caused by the fact that in
the low-frequency regime wave propagation is still
dominated by the coherent eld and thus, the diusion
equation is not well suited to describe energy propagation. This assumption is conrmed by the fact that for
the low-frequency data in Fig. 4 the positions of the

Fig. 4. Results of the numerical experiment (as described in Figs. 2 and

3) for the logarithm of energy density as a function of time and frequency (from 100 to 1000 kHz in steps of 100 kHz each). The solid
lines are least-squares curve ts to Eq. (1), i.e. the solution of the 1-D
diusion equation.

maxima of ln E coincide with the arrival times of the

primary and secondary coherent wave fronts, respectively.
The diusivity, D, and the dissipation, r, can directly
be determined from the curve ts shown in Fig. 4. The
results for various numerical experiments with dierent
parameters are given in Figs. 5 and 6. According to the
comments above the diusivity results for the low frequencies are expected to have larger errors than the
high-frequency results.
The dissipation is signicantly increasing with
increasing frequency (Fig. 5). This rise is stronger than
linear which is in accordance with the KelvinVoigt
model of viscoelasticity (solid line). Following this
model, the theoretical dissipation for the longitudinal
wave is given by r 2p2 f 2 gV 4=3gS =qc3L , where
we used the mean values from Table 1. It is important to
point out that for concrete as a heterogeneous medium,
this curve should be interpreted as the ensemble average
over numerous dierent scatterer arrangements. In

Fig. 5. Results of the numerical experiment for dissipation as a function of frequency (obtained at a single material conguration in each
case). The theoretical mean value according to the KelvinVoigt model
is given by the solid line. The porosity in this case was 0.7%.

Fig. 6. Results of the numerical experiment for diusivity as a function

of frequency and porosity (obtained at a single material conguration
in each case).

F. Schubert, B. Koehler / Ultrasonics 42 (2004) 781786

contrast to that, the numerical experiment was carried

out by using one single sensor position and one certain
arrangement of aggregates and pores. Thus, a quantitative dierence between the theoretical curve and the
simulation data as seen in Fig. 5 can be expected. Apart
from that, the quadratic frequency dependence of r is
reproduced well by the numerical data.
In addition to the viscous model, the calculation was
repeated by setting both viscoelastic parameters to zero.
The resulting absorptivities are not vanishing totally but
have only very small values (Fig. 5, lower curve). This
also demonstrates the consistency of the evaluation
method.
To investigate the dependency of the diusion coefcient on frequency, we used three dierent concrete
models with varying porosity (0%, 0.7%, and 1.4%). As
one can see from Fig. 6, diusivity is signicantly lowered by increasing porosity. This is consistent with diffusion theory since a larger number of pores means
stronger scattering and thus, a further delayed energy
propagation. It is striking that the dierences between
the nonporous model and the model with 0.7% porosity
are signicantly larger than the dierences between 0.7%
and 1.4% porosity.
The general fact that small changes in porosity have a
dramatic eect on the diusivity seems to be surprising
at rst view. However, one has to keep in mind that the
largest pores (2 mm diameter) are not negligible compared to the smallest wavelengths of 48 mm (for frequencies between 500 and 1000 kHz at a wave speed of
cL  4000 m/s). Moreover, it is well known from preceding investigations that although size and concentration of air-lled pores are clearly smaller than that of the
aggregates, porosity plays a signicantif not dominantrole for scattering of ultrasonic waves in concrete
[7]. This is caused by the large impedance mismatch
between air and matrix. Apart from that it should also
be taken into account that scattering at a 2-D plane
strain concrete model is quantitatively dierent from
scattering at a 3-D model, as discussed in [7]. As a
consequence, the numerical results obtained so far
should only be compared qualitatively with 3-D experiments.
In order to predict diusivity in concrete, Turner and
Anugonda proposed a statistical scattering theory using
a two-phase approach of cement matrix and aggregates
with perfect bonding between the two phases [2,8]. They
found a good agreement with experimental data. So far,
we did not nd a satisfactory agreement between our
numerical 2-D results and the 3-D AnugondaTurner
model. The diusion coecients of the upper curve in
Fig. 6 should be comparable to the AnugondaTurner
model due to the lacking porosity but the values are
signicantly larger then predicted by the model. This is
probably caused by the fact that scattering at aggregates
and pores in the numerical 2-D model is larger than in

785

3-D if using the same volume/area fractions for aggregates and pores [7].
Moreover, the strong rise of diusivity at low frequencies as predicted by the model [2], cannot be found
in our numerical results. This is maybe caused by the
domination of the coherent eld in this regime but the
exact reason for this discrepancy should be investigated
further. A 2-D version of the AnugondaTurner model
or alternatively a 3-D simulation would help to answer
these questions. Independent from that it is worth noting that preceding experiments and simulations clearly
revealed that porosity in concrete cannot be neglected
compared to the aggregates [7]. This is also conrmed by
the results shown in Fig. 6. As a consequence, appropriate enhancements of the AnugondaTurner model
need to be taken into consideration.

4. Conclusions
The results of the numerical experiments are plausible and consistent with the assumptions of a diusive
energy transport. While absorptivities obtained at the
discrete model are in a good agreement with the
KelvinVoigt model, the diusion coecients do not t
well to the AnugondaTurner model so far. However,
the numerical results reveal a signicant impact of
porosity on the diusion coecient. Due to the fact
that microstructural damage (e.g. by micro cracks) is
expected to have similar eects on wave scattering than
porosity, the use of high-frequency diusive ultrasound
could lead to new nondestructive methods for characterising material properties and damage in concrete
structures.
In the future, an ensemble averaging of the numerical
data by using various models with dierent scatterer
arrangements should lead to even more stable and reliable results. Moreover, a direct calculation of the energy
density from the elastic eld components will possibly
lead to further insights into still unsolved aspects of
diusivity in concrete.
Acknowledgements
The authors would like to thank an anonymous reviewer for useful comments to improve the presentation
of the manuscript.

References
[1] B. Koehler, M. Kehlenbach, R. Bilgram, in: Acoustical Imaging,
vol. 27, Kluwer Academic/Plenum Publishers, Dordrecht & New
York, 2004, p. 315, in print.
[2] P. Anugonda, J.S. Wiehn, J.A. Turner, Ultrasonics 39 (2001) 429.
[3] R.L. Weaver, Ultrasonics 36 (1998) 435.

786

F. Schubert, B. Koehler / Ultrasonics 42 (2004) 781786

[4] F. Schubert, in: Proceedings of Ultrasonics International 2003,

June 30July 3, 2003, Granada, Spain, Ref UI288.
[5] F. Schubert, H. Wiggenhauser, R. Lausch, in: Proceedings of
Ultrasonics International 2003, June 30July 3, 2003, Granada,
Spain, Ref UI361.

[6] F. Schubert, R. Marklein, in: Proceedings of 2002 IEEE Ultrasonics Symposium, Munich, Germany, 2003, p. 778 (article 5G-5, on
CD).
[7] F. Schubert, B. Koehler, J. Comput. Acoust. 9 (4) (2001) 1543.
[8] J.A. Turner, P. Anugonda, J. Acoust. Soc. Am. 109 (2001) 1787.