Engineering Structures 22 (2000) 1159–1166

www.elsevier.com/locate/engstruct

Comparison of techniques for modal analysis of concrete structures

a,*
J.M. Ndambi , B. Peeters b, J. Maeck b, J. De Visscher c, M.A. Wahab b,
J. Vantomme a, G. De Roeck b, W.P. De Wilde c
a
Department of Civil Engineering, Royal Military Academy, 30, renaissance avenue, B-1000 Brussels, Belgium
b
Department of Civil Engineering, K.U.Leuven, W. De Croylaan 2, B-3001 Heverlee, Belgium
c
Department of Mechanics of Materials and Constructions, Vrije Universiteit Brussel, Pleinlaan 2, B-1050 Brussels, Belgium

Received 7 January 1999; received in revised form 7 April 1999; accepted 11 June 1999

Abstract

This paper describes different experimental techniques for obtaining modal parameters of structures. Attention is focused on
those techniques that may be applicable to in situ concrete structures (e.g. bridges). In a first stage, experiments are made on
reinforced concrete beams of 6 meters length. The beams are excited using three types of excitation methods: impact hammer
excitation and two different electromagnetic shaker signals: pseudo-random and swept-sine signals. The modal parameters are
determined either by performing curve-fitting procedures on series of measured frequency response functions or by applying the
stochastic subspace identification technique to the time response signals of the structure. The influence of the non-linear behaviour
of the concrete beams is investigated by performing measurements at different excitation amplitudes.
It appears that modal parameter estimates are affected by excitation techniques, data acquisition parameters and processing
methods. The main cause of this is the non-linear behaviour which is observed even at very low vibration amplitudes. However,
the influence on resonant frequencies and mode shapes is negligible. This is not the case for the modal damping ratios, so that the
estimation of these parameters is unreliable.  2000 Elsevier Science Ltd. All rights reserved.

Keywords: Modal analysis; Modal parameter; Reinforced concrete; Dynamic measurement; Dynamic characteristics

1. Introduction investigated changes in dynamic characteristics with

simulated damage, using both model and real structures.
The comparison of techniques used for the dynamic A literature review can be found in [1]. One of the main
analysis of concrete structures is a part of a large experi- problems encountered is that modal parameter estimates
mental program. The aim of this program is to investi- are affected by different variables which are sometimes
gate how dynamic characteristics can be used to detect difficult to evaluate and control, such as environmental
and evaluate damage in concrete structures, such as conditions, excitation techniques, data acquisition para-
bridges. meters, data processing methods and human factors. If
In concrete structures, damage is associated with the variation of the modal parameters due to these
structural modifications, which can be observed through experimental parameters is of the same order than the
changes in modal parameters: resonant frequencies, variations due to damage, wrong conclusions could be
modal damping ratios, mode shapes and their deriva- drawn concerning the state of health of the structure.
tives. Modal testing techniques are used to determine The importance of the dynamic loading effect due to
these modal parameters, by performing curve fitting pro- excitation devices is discussed in [2]. It is shown that
cedures either on series of measured frequency response the resonant frequencies shift with different excitation
functions (FRFs), or on time histories of the structures devices and the user has to be careful in the interpret-
dynamic response. Several researchers have already ation of the results obtained.
In this paper, the results obtained from different exci-
tation techniques, data acquisition parameters and data
* Corresponding author. Tel.: +32-2737-6422; fax: +32-2737-6412. processing techniques are compared and discussed.
E-mail address: [email protected] (J.M. Ndambi).

0141-0296/00/$ - see front matter  2000 Elsevier Science Ltd. All rights reserved.
PII: S 0 1 4 1 - 0 2 9 6 ( 9 9 ) 0 0 0 5 4 - 1
1160 J.M. Ndambi et al. / Engineering Structures 22 (2000) 1159–1166

2. Modal parameter estimation methods: matrix; {f(t)} is the excitation force vector; {x(t)} is the
theoretical aspects response vector.
Eq. (1) can be reformulated in a state-space descrip-
In this research, two methods are used to extract the tion of the system [7]:
modal parameters from the dynamic measurements: the {Ẋ(t)}⫽[A]{X(t)}⫹[B]{u(t)} (2)
Frequency Domain direct Parameter Identification
(FDPI) method implemented in the CADA-X system [3] where:

冋 册
and the Stochastic Subspace Identification method (SSI).
The first method is a frequency domain technique, where −[M]−1[C] − [M]−1[K]
[A]⫽ : state matrix;
the modal parameters are derived from series of meas- [I] [0]

冋 册
ured frequency response functions. The second method
is a time domain technique, and the dynamic character- [M]−1
[B]⫽ : input matrix;
istics are derived from the measured time response sig- [0]

再 冎
nals.
A study on the free vibration characteristics of three- {ẋ(t)}
X(t)⫽ : state space vector;
dimensional elastic solid has been proposed in [9] and {x(t)}
[10]. Liew, Hung and Lim [9,10] propose an analytic
energy functional based method for the calculation of {u(t)}⫽{f(t)} : input vector.
eigenfrequencies and mode shapes of elastic solids. This The transfer functions for acceleration over force are
method is based on the minimisation of energy func- easily derived by taking the Laplace transform of Eq.
tional according to Ritz procedure leading to the govern- (1):
ing eigenvalue equation. The reinforced concrete beams
tested could be considered as stress-free elastic solids. p2[I]⫹p[M]−1[C]⫹[M]−1[K][H(p)]⫽p2[M]−1 (3)
This method could thus be applied for theoretical trans- The equations hold for all values of p, and thus for all
versal bending and torsion eigenfrequencies and also values of jw. If sufficient measurements are available,
mode shapes calculations. this equation can form the base for an overdetermined
The next two paragraphs give a short theoretical set of equations in the unknown matrices:
description of the FDPI and the SSI methods. [M]−1[C], [M]−1[K], [M]−1. A least square solution will
It should be noted that these two methods are based estimate these matrices.
on the same theory. The main difference resides in the
fact that for the FDPI method, the dynamic character-
istics are derived from the measured frequency response 2.1.2. System poles and participation factors
functions, whereas for the SSI method, the dynamic calculation
characteristics are obtained by applying curve fitting pro- The mass modified damping matrix and the mass
cedures on the time response signals. The second method modified stiffness matrix form the kernel of the state
does not need the input signal, but assumes that the input matrix [A]. Its eigenvalues are the system poles (in com-
is a white Gaussian noise. plex conjugate pairs) and its eigenvectors yield the
modal vectors (also in complex conjugate pairs).
This will yield the diagonal matrix [⌳] of poles and
2.1. Frequency domain direct parameter identification a matrix [⌿] of eigenvectors. The participation factors
method are estimated using the two matrices [⌳] and [⌿].

This method allows the estimation of the natural fre- 2.1.3. Expansion of the reduced data
quencies, damping values and mode shapes of several Before estimating the system matrices, all available
modes simultaneously [3]. It works in three steps: the data are condensed via a projection on their principal
first one is the estimation of the system matrices, the components (reduction phase). For all response stations,
second step derives the system poles and participation a maximum of N principal components are first calcu-
factors and the third step is the expansion of the reduced lated and then analyzed. The obtained eigenvectors
data in order to obtain the complete model. matrix [⌿] thus represents the modal matrix for this set
of fictitious response stations. The complete eigenvectors
2.1.1. Estimation of system matrices of the structure are obtained by expanding the reduced
The starting point is the second order differential equ- mode shapes [7].
ation for dynamically loaded structures: The data reduction procedure offers two advantages:
[M]{ẍ(t)}⫹[C]{ẋ(t)}⫹[K]{x(t)}⫽{f(t)} (1) the calculation time is drastically decreased for the esti-
mation of model parameters and the number of contribu-
where: [M], [C], [K] is the mass, damping and stiffness ting modes is more easily determined.
 J.M. Ndambi et al. / Engineering Structures 22 (2000) 1159–1166 1161

2.2. Stochastic subspace identification method where: k苸N; {uk} is the discrete input vector; [A2]=e[A]⌬t
is the discrete state matrix; {Xk}={X(k⌬t)} is the discrete
This method allows the determination of the modal state vector; ⌬t is the sampling time;
parameters of a structure using only the response [B3]=[A2⫺I][A]⫺1[B] is the discrete input matrix.
measurements. The detailed knowledge of the excitation Knowing that data measured are always influenced by
is replaced by the assumption that the system is excited noise, the stochastic components (noise) are included in
by white Gaussian noise. these equations and the deterministic-stochastic discrete
The starting point of this method remains also the time state space model is obtained:
second order differential Eq. (1). The force vector {f(t)}
{Xk+1}⫽[A2]{Xk}⫹[B3]{uk}⫹{wk}
is factorized into a matrix [B2], describing the locations
of the inputs and a vector {u(t)} that gives the inputs as {Yk}⫽[C2][Xk]⫹[D]{uk}⫹{vk} (7)
a function of time.
Using the following definitions: where: {wk} is the process noise vector due to disturb-

冋 册
ances and modelling inaccuracies; {vk} is the measure-
[M]−1[B2] ment noise vector due to sensor inaccuracy.
{f(t)}⫽[B]{u(t)}; [B]⫽ , These two noise vectors wk and vk are not measurable,
[0]
but are assumed to be white and to have zero mean. The
and the matrices A and B described in the previous para- white noise means that the noise is totally unpredictable.
graph, Eq. (1) can be also transformed into the time state In the case of ambient vibration testing, the input uk
Eq. (2). remains unmeasured.
The number of elements of the state space vector is Assuming that the input is white Gaussian noise
the number of independent variables needed to describe allows one to model the input implicitly by the noise
the state of the system. In practice, not all the degrees terms wk and vk. The stochastic state space model is then
of freedom (DOFs) are monitored. If it is assumed that defined by:
the measurements are evaluated at only l sensor
locations, and these sensors can be accelerometers, velo- {Xk+1}⫽[A2]{Xk}⫹{wk}
city or displacement transducers, the observation equ- {Yk}⫽[C2]{Xk}⫹{vk} (8)
ation is:
It is shown in [9] how this stochastic state space model
{Y(t)}⫽[Ca]{ẍ(t)}⫹[Cv]{ẋ(t)}⫹[Cd]{x(t)} (4) can be identified using the measured time response sig-
where: {Y(t)} is the response data vector; [Cd], [Cv], nals. As a result of this identification, we do not obtain
[Ca]: output matrices for displacement, velocity and the continuous matrix [A] but the discrete matrix [A2].
acceleration. As for the continuous matrix [A], the discrete matrix [A2]
With the following definitions: can also be decomposed by an eigenvalue decompo-
sition. It is proved in [6] that A2 and A have the same
[C2]⫽[[Cd]⫺[Ca][M]−1[K]; [Cv]⫺[Ca][M]−1[C]]; [D] eigenvectors and their eigenvalues are related as:
⫽[Ca][M]−1[B2] lnl2i
li ⫽ (9)
Eq. (4) can be transformed into: ⌬t
Y(t)⫽[C2]{x(t)}⫹[D]{u(t)} (5) where: λi is the eigenvalue of the continuous state matrix
where: [C2] is the output matrix, [D] is the direct trans- [A]; λ2i is the eigenvalue of the discrete state matrix [A2].
mission matrix. The mode shapes at the sensor location (defined as
Eq. (5) is the second time state space equation. columns of ⌽) are the observed parts of the system eig-
It is proved in [6] that the continuous time state Eq. envectors (⌿) and thus are obtained using the obser-
(2) is equivalent with the second-order matrix equation vation Eq. (4):
of motion and that they have the same eigenvalues and [⌽]⫽[C2][y] (10)
eigenvectors. These last parameters can be obtained by
an eigenvalue decomposition of the continuous time where: [C2] is the output matrix, [⌿] contains eigenvec-
state matrix [A]. tors of the state matrix [A] as columns.
In practice, measurements are available at discrete
time instants. Thus, the continuous time state Eqs. (2)
and (5) have to be sampled. After sampling, the follow- 3. Experimental work
ing equations are obtained:
A series of dynamic measurements are performed on
{Xk+1}⫽[A2]{Xk}⫹[B3]{uk}
reinforced concrete beams using different excitation
{Yk}⫽[C2]{Xk}⫹[D]{uk} (6) techniques. The two previously described data pro-
1162 J.M. Ndambi et al. / Engineering Structures 22 (2000) 1159–1166

cessing methods are applied on the measured data to

determine the dynamic characteristics of the beams.
These measurements permit to evaluate the influence of
excitation techniques and data processing methods on
the estimated dynamic characteristics of the beams.

3.1. Description of the test beams

Three identical reinforced concrete beams are tested.

They were designed in order to reach the following
objectives:

앫 The first eigenfrequencies have to be as low as poss-

Fig. 2. Cross section.
ible in accordance to the rather low eigenfrequencies
generally encountered in civil engineering structures.
앫 The cross section has to be rectangular to avoid the
coupling effect between horizontal and vertical bend- geometry of the beams. The cross section is represented
ing modes. in Fig. 2.
앫 The reinforcement ratio is chosen to have a suf-
ficiently wide interval between the first crack load and 3.2. Measurement system
the failure load of the beams.
앫 Sufficient transverse reinforcement ensures that fail- 3.2.1. Experimental set-up and test procedure
ure occurs in bending, not in shear. The tests are prepared and conducted in the laboratory
of the Civil Engineering Department of the K.U.Leuven.
These four objectives result in reinforced concrete beams The beams are suspended by means of elastic springs
with 6 meters length (L), 25 cm width (w) and 20 cm to simulate free boundary conditions during the dynamic
height (h). measurements. These springs are attached at the theoreti-
The beams are reinforced with longitudinal steel cal nodal points of the fundamental (1st) bending mode
(S500) bars with 16 mm diameter, equally distributed of the beams. Springs also reduce the influence of the
over tension and compression side, corresponding to a environment and the suspension system on the vibration
reinforcement ratio of 1.4%. Transverse reinforcement behaviour of the beams. The measurements are made in
consists of stirrups with 8 mm diameter placed at every 62 points (2 rows of 31 points on each side of the top
200 mm. The eigenfrequencies are proportional to h/L2: surface) in order to capture both the vertical bending and
a height of 0.2 m and a length of 6 m result in a first torsion modes. Fig. 3 shows the dynamic measurement
free- free eigenfrequency of about 20 Hz. The measured configuration.
total mass is 750 kg corresponding to a density of For the FDPI method, the dynamic measurements are
reinforced concrete of r=2500 kg/m3. Fig. 1 shows the performed in two steps: in the first step, the force and
acceleration signals are sent through an analogue filter
amplifier, before they are stored on a digital audio tape

Fig. 1. Geometry of the beams. Fig. 3. Dynamic test configuration.

 J.M. Ndambi et al. / Engineering Structures 22 (2000) 1159–1166 1163

(Fig. 4). The second step consists of replaying the digi- with respect to the beam axis, in order to excite
tally stored signals, producing analogue signals that are all the bending and torsion modes simultaneously.
acquired by the signal analysis system for modal analy- The impacts and shaker forces are applied in the
sis. This two steps procedure is necessary, since the vertical direction (perpendicular to the longitudinal
CADA-X system is installed in the laboratory and is not axes) of the beams in order to excite vertical bend-
transportable for field measurements. ing and torsion modes only.
The sampling frequency of the digital recorder is fixed 앫 Dynamic response: twelve accelerometers with sensi-
at 6000 Hz, which is high enough to avoid the problem tivity ±100 mV/g are used to register the dynamic
of aliasing for the analysed frequency range (0–1000 response of the beams. Ten of the twelve acceler-
Hz). This sampling frequency remains the same when ometers are used in six moves in order to cover the
transferring signals from the digital recorder to the 62 measurement points.
analysis system. Ten signals are averaged when using 앫 Analogue low pass programmable filter/amplifier.
shaker excitation technique and four in the case of 앫 Digital audio tape data recorder.
impact hammer method.
For the SSI method, based on curve fitting of time
domain data [4], the output signals of the analogue filter 3.2.3. Excitation signals
are immediately stored on a portable personal computer. Two different excitation signals are adopted for the
Both methods use the same sampled data, in order to electrodynamic shaker to study their influence on the
allow comparison. estimated modal parameters of the beams. The para-
meters for each excitation signal are presented in
3.2.2. Instrumentation Table 1.
The measurements consist of capturing in the 62 The driving signals of the shaker first pass through a
measurement points the dynamic response of the struc- power amplifier before they are sent to the shaker.
ture using an applied excitation in one point (single
input/multiple output) [5]. This section describes the
devices used to perform the dynamic measurements. 4. Results and discussions

앫 Excitation: the excitation is applied by impacting the Only the results obtained for one of the three beams
beam with an instrumented hammer or by using an are presented here and commented. Those obtained for
electrodynamic shaker. the other two beams confirmed the observations.
앫 A hammer with a sensitivity of ±0.7 mV/lb is used.
The hardness of the tip is chosen in order to gener- 4.1. Eigenfrequencies and modal damping ratios
ate a reasonable spectrum in the frequency range
of interest (0–1000 Hz). The frequency resolution Table 2 shows the modal parameters obtained with the
of 1.17 Hz is adopted. FDPI method, using the three different excitation tech-
앫 An electrodynamic shaker equipped with a force niques. Because of the expected non-linear behaviour of
sensor having a sensitivity of ±52 mV/lb is used. the structure, differences may arise in dynamic charac-
The shaker is driven by either a pseudo-random or teristics due to the differences in response accelerations
a swept-sine signal. levels attained with the various excitation techniques.
The hammer impacts are applied on the upper side Therefore, the response acceleration amplitudes are also
of the beams and the driving point of the shaker shown in the table. They correspond to the peaks ampli-
is situated on the lower side. The excitation point tude values (for each mode) observed on the frequency
is located at the end of the beams and eccentrically spectra of the time acceleration signals.
From Table 2 it can be observed that:

앫 the difference between the resonant frequencies for

the swept-sine and the pseudo-random excitation
methods is negligible (max. 0.5%);
앫 the resonant frequencies obtained with the impact
testing method are slightly lower (2%). This differ-
ence may be due to the influence of the shaker on the
dynamic behaviour; the shaker makes the structure
behave in a more rigid way, resulting in apparently
higher resonant frequencies [2]. The second expla-
nation may be the high response acceleration ampli-
Fig. 4. Acquisition of test data. tudes observed with the impact testing method. The
1164 J.M. Ndambi et al. / Engineering Structures 22 (2000) 1159–1166

Table 1
Excitation parameters: electrodynamic shaker

Signal generator Signal type Frequency range Spectral lines Frequency resolution Max excitation
amplitude

FFT Analyzer ONO- Pseudo-Random 0–500 Hz 800 0.625 Hz ±13 N

SOKKI
DIFA Swept-Sine 0–1000 Hz 16 384 0.196 Hz ±23 N

Table 2
Eigenfrequencies and modal damping ratios (FDPI method)a

Modes Excitation: Pseudo-random Excitation: Swept-sine Excitation: Impact

Freq [Hz] Damp [%] Accel [m/s2] Freq [Hz] Damp [%] Accel [m/s2] Freq [Hz] Damp [%] Accel [m/s2]

1F 20.388 0.3336 0.0681 20.442 0.3618 0.0327 19.962 0.943 0.3621

2F 57.124 0.3432 0.0915 56.600 0.3112 0.0545 56.093 0.899 0.3373
3F 112.089 0.3390 0.1010 112.525 0.4312 0.0612 110.287 0.709 0.3116
1T 169.268 0.4435 0.0762 169.670 0.2648 0.0342 168.498 0.619 0.1817
4F 183.841 0.2984 0.1151 184.133 0.2690 0.0539 181.283 0.788 0.2429
5F 273.992 0.3653 0.1631 274.718 0.3018 0.0603 270.716 0.722 0.2358
2T 363.623 0.2775 0.1139 364.099 0.3112 0.0470 362.740 0.423 0.1569
6F 373.674 0.3152 0.1451 374.585 0.3394 0.0594 369.751 0.634 0.1775
7F 480.456 0.3825 0.0646 480.563 0.4482 0.0225 475.289 0.649 0.0788

a
F: bending mode; T: torsion mode; Freq: resonant frequency, Damp: Modal damping ratio; Accel: response acceleration amplitude.

hammer impacts induce some non-linearities in the Table 3 shows the different results obtained using the
dynamic behaviour of the beam resulting in different SSI method.
values of eigenfrequencies obtained. The problem of From Table 3, the same observations follow as for the
response acceleration levels is investigated in the FDPI method:
next paragraph.
앫 in contrast to eigenfrequencies, the results for the 앫 the resonant frequencies are nearly the same for the
modal damping ratios are very different for the three pseudo-random and swept-sine excitation techniques
excitation methods. This difference may be due to the and slightly higher compared to those obtained with
frequency resolution of the FRFs measured (0.625 Hz the impact technique;
for the pseudo-random excitation, 0.196 Hz for the 앫 the modal damping ratios are different for the three
swept-sine excitation and 1.17 Hz for the impact test- excitation methods, and the values obtained with the
ing method) or the response acceleration amplitudes impact testing are higher than those obtained with the
which were different for the three excitation tech- first two methods.
niques (Table 2). Experimental evidence of the depen-
dence of modal damping on frequency resolution is
found in [8], while the influence of excitation ampli- Table 3
Eigenfrequencies and modal damping ratios (SSI method)a
tude is dealt with in the next paragraph.
앫 modal damping ratios obtained from the impact test- Modes Excitation: Excitation: Swept- Excitation: Impact
ing method are systematically higher than for the two Pseudo-random sine
other methods. At first sight, this is surprising,
because the addition of the shaker to the structure Freq Damping Freq Damping Freq Damping
[Hz] [%] [Hz] [%] [Hz] [%]
would normally lead to a higher damping value. Yet,
the phenomenon, which is observed for the three con- 1F 20.48 1.07 20.42 0.93 19.99 0.77
crete test beams, can be explained by the response 2F 57.07 0.51 56.98 0.52 56.19 0.82
acceleration amplitudes obtained from this method 3F 112.10 0.37 111.91 0.40 110.8 0.69
which are higher than those obtained with the two 1T 169.2 0.37 169.0 0.42 168.5 0.47
4F 183.8 0.32 183.4 0.36 181.3 0.73
other methods (Table 2). This expresses also the non-
lenear behaviour induced by the hammer impacts in a
Freq: resonant frequency, Damping: Modal damping ratio; F:
the dynamic behaviour of the beams. bending mode; T: torsion mode.
 J.M. Ndambi et al. / Engineering Structures 22 (2000) 1159–1166 1165

Fig. 5. Relative eigenfrequencies in function of excitation amplitude.

The resonant frequencies obtained with the FDPI and These figures show that:
SSI methods are nearly the same for the three excitation
techniques used. On the other hand, the modal damping
ratios are different. This last observation shows the 앫 the resonant frequencies decrease with increasing
influence of the processing techniques on these para- excitation amplitude;
meters. 앫 the modal damping ratios increase with the excitation
amplitude, except for some modes between the first
4.2. Eigenfrequencies and modal damping ratios in and the second amplitude.
function of excitation amplitude

Both identification techniques, the FDPI and SSI The non-linear behaviour is thus observed, even at the
methods assume the structure to behave in a linear way low vibration amplitudes at which the modal analyses
[7]. In general, this assumption is accepted for steel are performed. However, for the response acceleration
structures, but for concrete with its non-linear material amplitudes varying from 0.005 to 0.065 m/s2, the fre-
behaviour, its application seems to be less appropriate. quency variations are very small and negligible (0.01%–
To investigate whether this assumption is acceptable for 0.2%). But, the increase in modal damping ratios varies
the concrete test beam, complete modal analyses were from 1 to 30%.
performed at different amplitudes of excitation using the These observations confirm the results obtained in the
swept-sine excitation. The figures presented below show previous paragraph (4.1).
the evolutions of relative eigenfrequencies (Fig. 5) and The techniques used are acceptable for the identifi-
modal damping ratios (Fig. 6) in function of the increas- cation of resonant frequencies, but not for the modal
ing response acceleration amplitude of the beam. damping ratios.

Fig. 6. Relative modal damping ratios in function of excitation amplitude.

1166 J.M. Ndambi et al. / Engineering Structures 22 (2000) 1159–1166

measurement of the excitation makes it a promising

technique for large concrete structures, since ambient
excitation (wind, traffic) can be used instead of for-
ced excitation;
앫 the impact testing technique remains the simplest
method for modal testing. However, whatever the
used data processing technique may be, the non-linear
behaviour of the analysed structures could make this
technique less efficient when the modal damping
ratios have to be measured with accuracy.
Fig. 7. Experimental mode shapes of the third reinforced concrete
beam. These conclusions are limited to the case of reinforced
concrete beams dynamically tested in free boundary con-
ditions.
5. Modes shapes

Both the previously modal parameter estimation Acknowledgements

methods discussed permit to clearly identify mode
shapes and nodal line patterns. Fig. 7 represents the 6 This work is a part of a research project G-0243.96
first eigenmodes of one of the three reinforced con- sponsored by the Flanders Fund for Scientific Research.
crete beams. Its financial support is gratefully acknowledged.
The smoothness of the mode shapes is a result of the
high quality of the measurements.
References
6. Conclusions [1] Doebling SW, Farrar CF, Prime MB, Shevitz DW. Damage
identification and health monitoring of structural and mechanical
Three reinforced concrete beams were subjected to systems from changes in their vibration characteristics, a litera-
vibration testing in free boundary conditions. Complete ture review. Research report LA-13070-MS, New Mexico: Los
experimental modal analyses were performed using dif- Alamos National Laboratory, 1996.
[2] Olbrechts T, Vandepite D, Sas P, Heylen W. Influence of exci-
ferent excitation methods. The modal parameters were tation systems on the dynamic behaviour of test structures. In:
determined using the FDPI and SSI methods. Proc. of 21th Int. seminar on Modal Analysis (ISMA 21),
It appears from these experiments that: Belgium, 1996:881–91.
[3] CADA-X. Modal Analysis: user manual, Rev. 3.4. Belgium:
앫 the modal parameter estimates are affected by data LMS International, 1996.
[4] Peeters B, Wahab MA, De Roeck G, De Visscher J, De Wilde
acquisition and processing techniques, more for WP, Ndambi JM, Vantomme J. Evaluation of structural damage
modal damping ratios and less for resonant fre- by dynamic system identification. In: 21th Int. seminar on Modal
quencies. The modal damping ratios are very sensitive Analysis (ISMA 21), Belgium: 1996.
and difficult to measure; [5] Bencat J. Statical and Dynamical Testing of Structures. In: Proc.
앫 non-linear behaviour is observed, even at very low of 2nd RILEM Int. Conference on Diagnosis of concrete struc-
tures, Slovak Republic, 1996:259–66.
vibration amplitude. However, for the excitation [6] Peeters B, De Roeck G. Stochastic Subspace System Identifi-
amplitudes used in these experiments, it is observed cation of a Steel Transmitter Mast. In Proc. of the 16th Int. Modal
that the variations due to the non-linear behaviour are Analysis Conference, Santa Barbara, 1998:130–36.
small and negligible for the resonant frequencies, in [7] Heylen W, Lammens S, Sas P. Modal analysis theory and testing.
contrast to the modal damping ratios; K.U. Leuven (Belgium): PMA, Mechanical Engineering Dept,
1995.
앫 the resonant frequencies obtained are nearly the same [8] Garreau G. Etude de paramètres influençant les essais dyna-
for the two processing methods used (FDPI and SSI). miques sur une poutre en béton. Engineering thesis, Royal Mili-
On the other hand, the modal damping ratios are very tary Academy, Institut Superieur Industiel de Bruxelles, 1998.
different. The fact that these parameters are influ- [9] Liew KM, Hung KC, Lim MK. Free vibration studies on stress-
enced by several experimental parameters do not let free three-dimensional elastic solids. ASME Journal of Applied
Mechanics 1995;62:159–65.
us to say which of these processing methods is the [10] Liew KM, Hung KC, Lim MK. Vibration stress-free hollow cyl-
best for their calculation; inders of arbitrary cross section. ASME J of Appl Mech
앫 the fact that the SSI method does not require the 1995;62:718–24.