See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/323374057

Operational modal identification in the presence of harmonic excitation

Article  in  Applied Acoustics · February 2018

DOI: 10.1016/j.apacoust.2018.02.017

CITATIONS READS

6 323

5 authors, including:

Maamar Asia Maher Abdelghani

Institut Français de Mécanique Avancée University of Sousse
7 PUBLICATIONS   14 CITATIONS    36 PUBLICATIONS   512 CITATIONS   

SEE PROFILE SEE PROFILE

Thien-Phu Le Vincent Gagnol

Université d'Évry-Val-d'Essonne SIGMA Clermont
59 PUBLICATIONS   537 CITATIONS    58 PUBLICATIONS   494 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

FGVV project View project

In-operation modal identification View project

All content following this page was uploaded by Maamar Asia on 13 March 2018.

The user has requested enhancement of the downloaded file.

 Applied Acoustics xxx (xxxx) xxx–xxx

Contents lists available at ScienceDirect

Applied Acoustics
journal homepage: www.elsevier.com/locate/apacoust

Operational modal identification in the presence of harmonic excitation

Asia Maamara,⁎, Maher Abdelghanib, Thien-Phu Lec, Vincent Gagnola, Laurent Sabourina
a
Université Clermont Auvergne, CNRS, SIGMA Clermont, Institut Pascal, 63170 Aubière cedex, France
b
Université de Sousse, Institut supérieur des sciences appliquées et de technologie de Sousse, Sousse, Tunisia
c
LMEE, Université d’Évry Val-d’Essonne, 91020 Evry cedex, France

A R T I C L E I N F O A B S T R A C T

Keywords: The dynamic behavior of structures can be studied by the identification of their modal parameters. Classical
Operational modal analysis modal analysis methods are based on the relation between the forces applied to structures (inputs) and their
Transmissibility functions vibration responses (outputs). In real operational conditions it is difficult, or even impossible, to measure the
Stochastic subspace identification excitation. For this reason, operational modal analysis approaches which consider only output data are pro-
Harmonic components
posed. However, most of these output-only techniques are proposed under the assumption of white noise ex-
citation. If additional components, like harmonics for instance, are present in the exciting force, they will not be
separated from the natural frequencies. Consequently, this assumption is no longer valid. In this context, an
operational modal identification technique is proposed in order to only identify real poles and eliminate spurious
ones. It is a method based on transmissibility functions.
The objective of the proposed paper is to identify modal parameters in operational conditions in the presence
of harmonic excitations. Identification is performed using a method based on transmissibility measurements and
then with the classical stochastic subspace identification method, which is based on white noise excitation. These
two methods are first applied to numerical examples and then to a laboratory test. Results validate the novel
ability of the method based on transmissibility measurements to eliminate harmonics, contrary to the stochastic
subspace identification approach.

1. Introduction structures at rest vary from those of structures in operational conditions

Modal analysis [1–3] is used to identify mode shapes, natural fre-
quencies and damping ratios under vibrational excitation. These
methods are efficient tool for detecting damage in structures, control-
ling them, and determining their structural stability. Modal parameters
are initially identified using experimental modal analysis (EMA) [4].
This technique exploits the Frequency Response Function (FRF) of the
structure, which represents the relation between the excitation and the
vibrational response of the structure. Resonance frequencies appear as
peaks in the measured frequency response functions. From these FRFs,
modal parameters are identified using various curve-fitting techniques.
Many excitation forms and experimental setups exist, their choice de-
pending on the structure’s complexity. Usually, the EMA is carried out
under impact hammer and/or shaker excitation. The major drawback to
experimental modal tests is that both artificially applied forces and
resulting structural vibration responses need to be measured. In prac-
tice, the measurement of the exciting force is not always possible. In
EMA, tests are performed at rest, and the dynamic properties of
[5], which can significantly influence the identified modal model.
For these reasons, modal identification techniques are developed
and operational modal analysis (OMA) is proposed [6,7], where the
modal properties are estimated from responses only. Various opera-
tional modal identification techniques are proposed, example the Sto-
chastic Subspace Identification approach (SSI) [8,9]. However, OMA
methods have limitations when applied to practical cases. One limiting
constraint of OMA is that the non-measured excitation of the system in
operation must be a stochastic realization (white noise) [6]. This im-
plies that if harmonic components are present in addition to random
excitation, standard OMA procedures cannot be applied in a straight-
forward way. Harmonic components are sometimes considered as vir-
tual modes in the identification, but when the harmonic excitation
frequencies are close to eigenfrequencies, the standard OMA ap-
proaches may break down [10].
Several indicators for the separation of structural and harmonic
modes in output-only modal identification are proposed. One of the
most widely-used methods is based on the Probability Density Function

⁎
Corresponding author.
E-mail addresses: essia.maamer@sigma-clermont.fr (A. Maamar), maher.abdelghani@gmail.com (M. Abdelghani), thienphu.le@univ-evry.fr (T.-P. Le),
vincent.gagnol@sigma-clermont.fr (V. Gagnol), laurent.sabourin@sigma-clermont.fr (L. Sabourin).

https://doi.org/10.1016/j.apacoust.2018.02.017
Received 30 June 2017; Received in revised form 18 January 2018; Accepted 15 February 2018
0003-682X/ © 2018 Elsevier Ltd. All rights reserved.

Please cite this article as: Maamar, A., Applied Acoustics (2018), https://doi.org/10.1016/j.apacoust.2018.02.017
A. Maamar et al. Applied Acoustics xxx (xxxx) xxx–xxx

(PDF) of harmonic and structural responses. The PDF of a structural two transmissibility functions measured at the same output points i and
response is a distribution with only one peak, and the PDF of a har- j but with two different input excitation locations k and l .
monic response is a distribution with two peaks. This difference was
ΔTijkl (s ) = Tijk (s )−Tijl (s ) (3)
originally illustrated by Lago [11]. Kurtosis criteria have also been used
to identify harmonic components and structural modes, [12–16]. Kur- And consequently the poles of its inverse:
tosis is defined as the fourth central moment of the PDF, normalized
1
with respect to the standard deviation. In addition to the above-men- Δ−1Tijkl (s ) =
tioned methods, knowledge of the damping ratios is an a priori in- Tijk (s )−Tijl (s ) (4)
dicator to distinguish between harmonics and structural poles. Gen-
The PolyMAX method [23] is then investigated in order to calculate
erally, the damping ratios of real poles vary between 0.1% and 2%. This the system poles from Δ−1Tijkl (s ) . Generally, additional poles can be
information enables modes with negative and high damping to be
present in the Δ−1Tijkl (s ) functions. Structural poles λr can easily be de-
eliminated [7]. The Modal Assurance Criterion (MAC) [17] is also an
termined by performing a singular value decomposition of the trans-
effective tool to distinguish a structural mode from a harmonic one. The
missibility matrix T [19]. If we consider, for example, four different
MAC value between a structural mode shape and a mode shape corre-
loading conditions k ,l,m and n , the transmissibility matrix is the fol-
sponding to a harmonic component will show a low correlation. Spe-
lowing:
cific numerical filters have also been developed [18] in order to elim-
k
inate harmonic components from the measured response. However, in ⎡T1r (s ) T1lr (s ) T1mr (s ) T1nr (s ) ⎤
practice filters are not perfect, and if the harmonic frequency is close to ⎢T k (s ) T2lr (s ) T2mr (s ) T2nr (s ) ⎥
resonant frequencies, the filtering will disturb the response so that the ⎢ 2r ⎥
T = ⎢T k (s ) T3lr (s ) T3mr (s ) T3nr (s ) ⎥
3r
identified modal parameters are perturbed. ⎢ k ⎥
In order to overcome the white noise excitation assumption and ⎢T4r (s ) T4l r (s ) T4mr (s ) T4nr (s ) ⎥
⎢
⎣ 1 1 1 1 ⎥ ⎦ (5)
consequently to identify modal parameters in the presence of harmonic
excitation, an operational modal identification method is proposed by In fact, in system poles λr the rank of matrix T is one; consequently,
Devriendt et al. [10,19–21]. This method is based on transmissibility σ1 > σ2 > σ3 > σ4 ⩾ 0 and 1/ σ2 tends to ∞.
measurements. The significant advantage of this approach is its in-
dependence from the nature of the excitation. Consequently, the pre-
2.2. Stochastic subspace identification method (SSI)
sence of harmonics will not disturb the identified modal model.
The main objective of this paper is to identify the modal parameters
The dynamic behavior of a discrete mechanical system consisting of
of structures in the presence of harmonic components using the
n masses connected through springs and dampers is described by the
Transmissibility Function-Based method (TFB). The decision con-
following matrix differential equation:
cerning whether a particular mode is structural or not is based on a
singular value decomposition of the system’s transmissibility matrix. Mq¨ (t ) + C2 q ̇ (t ) + Kq (t ) = f (t ) (6)
The identified eigen-parameters are then compared to those obtained
M,C2 and K ∈ n × n
are the mass, damping and stiffness matrices. q (t )
using the classical SSI method, which is based on the white noise ex-
∈ n is the displacement vector at continuous time t. Vector f (t ) ∈ n is
citation assumption. The idea behind this comparison is to demonstrate
the excitation force.
that the TFB method is a particular OMA which can eliminate harmo-
SSI is a method that converts a 2nd order problem into two 1st order
nics and identify only the real poles of the structure. The paper is or-
problems. Eq. (6) can be converted into the following state equation:
ganized as follows. In Section 2, a brief description of the TFB and SSI
techniques is provided. In Section 3, the OMA technique based on x ̇ (t ) = Ac x (t ) + Bc f (t )
transmissibility measurements is applied to a numerical model and to a y (t ) = Cx (t ) + Df (t ) (7)
cantilever beam test. Results are then compared with those obtained via
the SSI technique. Finally, Section 4 concludes the paper. The state matrix Ac in continuous time, the load matrix Bc and the
output matrix C are given by:

2. Operational modal identification techniques 0 11 ⎞ q

Ac = ⎛ −1K − M−1C , Bc =
⎛ 0−1⎞, C = (11 0), x = ⎛ ⎞
⎝ − M 2⎠ ⎝ M ⎠ ⎝ q ⎠̇
2.1. Modal identification method based on transmissibility functions (TFB)
D is the feedback matrix (zero in the case of mechanical systems). The
first equation in (7) is called the state equation and models the dynamic
OMA approaches are generally based on the assumption of white
behavior of the system. The second equation is called the observation or
noise processes for operational excitations. However, this assumption is
output equation. Eq. (7) can be converted to following discrete-time
hard to respect in real situations [10,22]. The method based on trans-
stochastic state-space model [8]:
missibility functions (TFB) is independent from the nature of the ex-
citation, and solves the problem of the presence of harmonic compo- xk + 1 = Axk + wk
nents. The use of transmissibility functions was proposed by Devriendt yk = Cxk + vk (8)
et al. [19] as a new approach in operational modal analysis. The ex-
pression of the frequency response at a point i under an excitation at a where yk = y (k Δt ) is the sampled output vector, xk = x (k Δt ) is the
point k is written as: discrete state vector, wk is the process noise due to the unknown ex-
citation of the structure, vk is the measurement noise and k is the time
Xik (s ) = Hik (s ) Fk (s ) (1) instant, A = exp (Ac Δt ) is the discrete state matrix. In order to obtain
the modal parameters, an eigenvalue decomposition (EVD) of the ma-
A transmissibility function is defined as the ratio between the mo-
trix A is performed:
tion response Xik (s ) and the reference motion response X jk (s ) under a
single force located at k . A = ΨΛ d Ψ−1 (9)

Xik (s ) Hik (s ) Fk (s ) H (s ) ∈  n×n

is the eigenvector matrix and Λ d = diag (λi ) ∈ is the di-  n×n
Tijk (s ) = = = ik
X jk (s ) Hjk (s ) Fk (s ) Hjk (s ) agonal matrix containing the discrete time eigenvalues μi . The con-
(2)
tinuous time state Eq. (7) is equivalent to the second order matrix
System poles are zero points resulting from the subtraction between equation of motion (6). Consequently, they have the same eigenvalues

2
A. Maamar et al. Applied Acoustics xxx (xxxx) xxx–xxx

1 2
and eigenvectors. These can be obtained by an eigenvalue decomposi- In Fig. 2a, the two transmissibility functions T12 (s ) and T12 (s ) cross
tion of the continuous time state matrix: each other at two different points, which represent the system’s poles.
The subtraction between two transmissibility functions under two dif-
Ac = Ψc Λ c Ψ −c 1 (10) ferent input excitations on each of the two masses is:
where Λ c = diag (λ c ) is a diagonal matrix containing the continuous 1 2
ΔT (s ) = T12 (s )−T12 (s ) (16)
time complex eigenvalues and Ψc contains the eigenvectors.
The continuous time complex eigenvalues λ c are found from the Systems poles are poles of Δ−1T (s ) . In Fig. 2b, two peaks of the
discrete time eigenvalues λi by: system’s poles are observed at Δ−1T (s ) . The frequency domain esti-
mator, PolyMAX, is then investigated for the function Δ−1T (s ) , in order
ln(λi )
λc = to identify the damping ratio and the natural frequency of the system’s
Δt (11)
poles. Modal parameters are reported in Table 1 with their exact values.
where Δt is the sampling time. The modal parameters identified by the transmissibility function-
The eigenvalues of Ac occur in complex conjugated pairs and are based method are quite close to exact ones, thus confirming the nu-
written as: merical implementation of the TFB method. The damping ratio is more
influenced by the nature of the excitation than the natural frequency.
λ c ,λc∗ = −ξc ωc ± jωc 1−ξc2 (12) This explains the good correlation between TFB and exact values as
Natural frequencies fc and damping ratios ξc are identified from: regards the determination of natural frequencies.

ωc |λ | |R (λ c )|
fc = = c , ξc = 3.1.2. Combination of white noise excitation and fixed harmonic excitation
2π 2π |λ c | (13) In order to evaluate the influence of the presence of harmonic
The modal matrix Φ can be obtained from the eigenvector matrix using components on the exciting force, and to check the efficiency of the TFB
the relation: method, a combination of harmonic excitation and Gaussian white
(14) noise excitation with zero mean is applied to each mass separately. The
Φ = CΨ
frequency of the applied harmonic excitation is equal to 15 Hz. The Fast
Fourier Transform (FFT) of the simulated temporal response is shown in
3. Applications Fig. 3. Three peaks combining structural and harmonic modes are
present in the FFT curve. The harmonic component frequency is 15 Hz,
1 2
3.1. Numerical model and the two other modes are structural. In Fig. 4, the T12 (s ) and T12 (s )
curves cross each other on the system’s poles. In [21], it was demon-
In order to validate the numerical implementation of the transmis- strated that the ratio between two frequency responses under the same
sibility function-based method, a numerical model was considered and input force eliminates the harmonic component. Thus these harmonic
modal identification was performed in the case of impact hammer ex- components will not be present in Δ−1T (s ) . However, Δ−1T (s ) may
citation and then in the case of a combination between harmonic and contain additional poles that are not related to the system’s dynamics,
random excitations for comparison purposes. as can be seen on the curve of Δ−1T (s ) in Fig. 5. It is important to verify
whether these peaks on the curve of Δ−1T (s ) are real structural modes or
3.1.1. Impact hammer excitation not.
A two-degree-of-freedom (DoF) system is considered for numerical Step 1: Singular value decomposition of the transmissibility matrix
validation and is illustrated in Fig. 1 with its mechanical properties. T
Two impact hammer excitations were applied to both masses of the At the system poles, the rank of the proposed matrix T is one.
system separately. For each loading condition, responses in displace- Consequently, σ2 tends to zero and the peaks of 1/ σ2 indicate the sys-
ment are obtained by the integrating motion equations with the Range- tem’s poles.
Kutta algorithm. The mass displacement responses, which are sampled Step 2: Stabilization diagram
at N = 4000 points with a period Δt = 0.01 s , are presented in Fig. 1. Stabilization diagrams showing the stability of the poles as a func-
In this case, two transmissibility functions are calculated with re- tion of Δ−1T (s ) and as a function of increasing model order were used to
spect to the frequency response of each mass of the studied system distinguish the spurious modes from the physical poles. Fig. 6 shows
under the two different applied forces. such a stabilization diagram. The green point indicates that a pole is
stable with respect to damping ratio and natural frequency, and the red
X11 (s ) X (s )
1
T12 (s ) = 2
, T12 (s ) = 12 point indicates that a pole is unstable with respect to damping ratio
X21 (s ) X22 (s ) (15)
and/or natural frequency. Here it is clear that the two poles at

Fig. 1. Two-degree-of-freedom system and simulated responses under impact hammer excitation on mass 1.

3
A. Maamar et al. Applied Acoustics xxx (xxxx) xxx–xxx

a b
4 4
T112 Δ
í1
T

Magnitude (dB)

Magnitude (dB)
2 2
T2
12

0 0

í2
í2

í4
5 10 15 20 0 5 10 15 20
Frequency (Hz) Frequency (Hz)
1 2
Fig. 2. (a) Transmissibility functions T12 and T12 , (b) Δ−1T under impact hammer excitation.

Table 1
60 1/σ
Identified modal parameters from the TFB method and exact values. 2
50
Mode Natural frequency (Hz) Damping ratio (%) 40
30
Exact TFB Exact TFB
20
1 3.55 3.51 1.12 1.36 10
2 8.71 8.78 2.73 2.47
0 5 10 15 20
Frequency (Hz)
0 Fig. 5. Selection of system poles by means of singular value decomposition.
Magnitude (dB)

í5
30

í10
Model order

20
í15
10
í20
0 10 20 30 40 50
0
Frequency (Hz) 0 2 4 6 8 10 12 14 16 18 20
Frequency (Hz)
Fig. 3. Fast Fourier Transform of the simulated response.
Fig. 6. Selection of system poles through a stabilization diagram by the TFB method.

frequencies 3.54 Hz and 8.72 Hz are real structural poles, and that the
other peaks are spurious ones. In order to demonstrate the novel ability Table 2
of the TFB technique to eliminate harmonics, modal parameters are also Identified modal parameters using SSI and TFB methods.

identified using the SSI method. Results obtained from the two different
Mode Natural frequency (Hz) Damping ratio (%)
approaches are presented in Table 2 and compared. The times con-
sumption of the proposed TFB approach and the classical SSI method Exact SSI TFB Exact SSI TFB
are respectively 3.544 s and 2.834 s. In the stabilization diagram ob-
1st 3.55 3.58 3.54 1.12 1.21 1.24
tained with the SSI method, shown in Fig. 7, three poles are estimated.
2nd 8.71 8.717 8.72 2.73 2.02 2.68
The SSI is a robust method, but we cannot really distinguish be- 3rd 15 15 – – 0.05 –
tween structural and non-structural poles, and this generally represents
a difficulty in the operational modal identification field. With the
transmissibility-based method, the harmonic mode is automatically
eliminated, which is not the case for SSI, that estimates it with a low
damping value. However, the modal identification of the real structural

4
T1 Δ í1 T
5 12 2
Magnitude (dB)
Magnitude (dB)

T2
12 0

0 í2

í4
í5
í
0 5 10 15 20 0 5 10 15 20
Frequency (Hz) Frequency (Hz)
1 2
Fig. 4. (a) Transmissibility functions T12 and T12 , (b) Δ−1T under a combination of harmonic and white noise excitation.

4
A. Maamar et al. Applied Acoustics xxx (xxxx) xxx–xxx

20 Table 4
Modal parameters identified using EMA.
15
Model order

Mode Natural frequency (Hz) Damping ratio (%)

10
PolyMAX TFB PolyMAX TFB
5
1st 10.569 10.57 1.05 1.059
0 2nd 67.751 67.51 0.74 0.841
0 5 10 15 20
3rd 185.959 188.99 1.59 1.547
Frequency (Hz)
4th 387.542 388.92 0.78 0.763
Fig. 7. Stabilization diagram obtained from the SSI method. 5th 626,917 626,51 0,57 0,578

M$&(TBFí3olyM$;)

Fig. 8. Description of the cantilever beam and position of the accelerometers.

0.5
Table 3
Exact natural frequency values of the cantilever beam.

0
Mode Natural frequency (Hz)
1
1st 10.8 2
2nd 67.8 3 5
4
3rd 189.85 4 3
4th 371.5 5 2
3olyM$; mode shapes 1
5th 614.8 TBF mode shapes

Fig. 10. Mode shape comparison-MAC values.

pole parameters shows a good correlation, especially on the natural

frequencies.

3.2. Laboratory experimental test

3.2.1. Experimental modal analysis

In order to check the efficiency of the OMA method based on
transmissibility measurements, an experimental test, shown in Fig. 8,
was carried out for a cantilever beam. A cantilever beam 0.757 m in
length, 0.103 m in width and 0.008 m in height is considered to identify
its modal parameters. Fig. 11. Laboratory investigation: experimental beam test.
First, an analytical computation of the first five natural frequencies
of the considered beam was performed, using the following equation: Height: h = 0.08 m
1 E h Length: L = 0.757 m
fi = αi2 Young’s modulus: E = 70,000 MPa
2π 12 ρ L2 (17)
Density: ρ = 2698 kg/m3
with:

Fig. 9. Stabilization diagram from an EMA.

5
A. Maamar et al. Applied Acoustics xxx (xxxx) xxx–xxx

Fig. 12. Sum of the measured cross-power spectra and stabilization diagram using the PolyMAX method.

The modal assurance criterion (MAC) shows a very good correlation

between the five structural mode shapes obtained by these two different
techniques (Coefficient of correlation greater than 0.9) (see Fig. 10).

3.2.2. Operational modal analysis

Fig. 13. Selection of the system’s poles by means of singular value decomposition.
α 0 = 1.8751, α1 = 4.695, α2 = 7.85, αi + 1 = (2i + 1)/2, for i > 2
Exact natural frequency values of the cantilever beam are shown in
Table 3. In order to obtain a reference modal model, an EMA was
performed using an impact hammer at well-defined locations on the
beam. Time responses were recorded using six PCB Piezoelectric ac-
celerometers, equally distributed over the full length of the beam, as
shown in Fig. 8. LMS TEST.Lab software was used to acquire the re-
sponses from the accelerometers. Modal parameters were identified
using the PolyMAX modal identification method implemented in the
LMS TEST.Lab software. From the stabilization diagram shown in
Fig. 9, modal parameters of the studied cantilever beam were identified.
The method based on transmissibility measurements is then applied to
the measured data response. The results obtained by the two different
methods are shown in Table 4. Results show a good correlation when
comparing the values of the natural frequencies and the damping ratios.
In order to compare the mode shapes identified by the two different
methods, the modal assurance criterion is calculated.
In addition to the random impact hammer shocks, the beam is now
excited by harmonic excitations provided by a motor turning at
851 rpm. The motor is fixed to the second point of the beam, as can be
seen in Fig. 11. Experimental tests were carried out under random
impact hammer shocks in the presence of harmonic excitation. Data
responses were measured using six accelerometers, equally distributed
over the full length of the beam, as described in the EMA section. Op-
erational modal analysis was performed in the frequency band [0
200 Hz]. Harmonic components are multiples of 14.18 Hz. Six cross-
power spectra are measured with reference to the second accelerometer
for each experiment, their sum leading to a total of 7 power spectra. The
sum of six cross power spectra, shown in Fig. 12, was chosen in order to
identify the modal parameters of the structure using PolyMAX. Addi-
tional peaks, corresponding to harmonic components, are present. The
frequencies of the 2nd, the 3rd and the 4th modes are approximately
two, three and ten times the frequency of the first mode (over-harmo-
nics), which indicates that these are harmonics.
The stabilization diagram is obtained using the PolyMAX method
implemented in the LMS TEST.Lab software. Here we find that the
distinction between harmonics and structural poles is difficult. It is
therefore not possible to decide if a pole is real or spurious. The same
difficulty was found when identifying the modal parameters of the
excited beam with the SSI method, which identifies both structural and
harmonic poles.
As explained in Section 1, in order to select the correct poles a
singular value decomposition of the transmissibility matrix is per-
formed, and the second singular value of T(s ) should be close to 0 for
this system’s poles.

Table 5
Modal parameters of the cantilever beam, identified through experimental and operational modal analysis.

Component Identified frequency (Hz) Damping ratio (%) Nature

PolyMAX SSI TFB PolyMAX SSI TFB

1st 10.14 10.22 10.12 1.12 7.85 1.128 Structural

2nd 14.10 14.46 – 0.10 0.04 – Harmonic
3rd – 25.93 – – 7.9 – Harmonic
4th 43.69 43.52 – 0.66 6.12 – Harmonic
5th 63.70 63.67 67.751 2.66 9.05 2.20 Structural
6th 184.10 183.1 185.957 0.94 0.72 0.39 Structural

6
A. Maamar et al. Applied Acoustics xxx (xxxx) xxx–xxx

M$&(TFBí3olyM$;) M$&(66,í3olyM$;)

1 1

0.5 0.5

0
0

1
1
2 3 2 3
3 2
1 3 2
3olyM$; modes shapes
3olyM$; modes shapes 1
66, modes shapes

Fig. 14. Modal assurance criterion for mode shapes identified using the SSI, TFB and PolyMAX methods.

1
⎡T12 (s )
2
T12 3
(s ) T12 4
(s ) T12 (s ) ⎤ advantage of the TFB method is its independence from the nature of the
⎢T1 (s ) T32 (s ) T32 (s ) T32 (s ) ⎥
2 3 4 excitation and its novel ability to eliminate harmonic components.
⎢ 32 ⎥ Consequently, it provides an accurate prediction of modal parameters
⎢T1 (s ) 2
T42 3
(s ) T42 4
(s ) T42 (s ) ⎥
T = ⎢ 42 ⎥ in the presence of harmonic excitations. Thus, it opens possibilities for
1 2 3 4
⎢T52 (s ) T52 (s ) T52 (s ) T52 (s ) ⎥ application to more complicated structures in the presence of harmonic
⎢T (s )
1
T62 (s ) T62 (s ) T62 (s ) ⎥
2 3 4
excitations under operational conditions.
⎢ 62 ⎥
⎣ 1 1 1 1 ⎦ (18)
Acknowledgments
Through the EMA of the studied cantilever beam, three structural
poles are identified in the frequency band [0 200 Hz], at frequencies of
This work was sponsored by the French government research pro-
10.57 Hz, 67.51 Hz and 188.99 Hz, respectively. Furthermore, in
gram through the IMobS3 Excellence Laboratory, by the European
Fig. 13, it is shown that in the considered frequency band, the three
Union through the Regional competitiveness program and by the
dominant peaks coincide with the first three natural frequencies.
Auvergne region.
The PolyMAX technique is then investigated to identify modal
parameters from the transmissibility Δ−1T (s ) functions. Results obtained
References
from the two different TFB and SSI approaches are presented in Table 5
and Fig. 14. The times consumption of the proposed TFB approach and
[1] Guillaume P. Modal analysis. Pleinlaan: Department of Mechanical Engineering,
the classical SSI method are respectively 7.519 s and 2.492 s. Vrije Universiteit Brussel; 2006. vol. 2, p. 4–6.
Similarly to the first studied case, only structural modes are selected [2] Heylen W, Sas P. Modal analysis theory and testing. Katholieke Universteit Leuven,
and estimated by the method based on transmissibility measurements. Departement Werktuigkunde; 2005.
[3] Peeters B, Ventura C. Comparative study of modal analysis techniques for bridge
Modal parameter identification appears to be reasonably robust in re- dynamic characteristics. Mech Syst Signal Process 2003;17:965–88.
lation to the determination of the natural frequencies and vibration [4] Mejri S, Gagnol V, Le T-P, Sabourin L, Ray P, Paultre P. Dynamic characterization of
mode shapes. However, the damping ratio identified by both SSI and machining robot and stability analysis. Int J Adv Manuf Technol 2016;82:1–4.
[5] Mejri S, Gagnol V, Le TP, Sabourin L, Ray P, Paultre P. Experimental protocol for the
PolyMAX methods is strongly influenced by the presence of harmonics. dynamic modeling of machining robots. 21ème Congrès Français de Mécanique,
The modal assurance criterion (MAC) shows a very good correlation Bordeaux, France. 2013.
between the structural mode shapes obtained using these three different [6] Masjedian M, Keshmiri M. A review on operational modal analysis researches:
classification of methods and applications. Proc of the 3rd IOMAC. 2009.
techniques (Coefficient of correlation greater than 0.8).
[7] Gagnol V, Le T-P, Ray P. Modal identification of spindle-tool unit in high-speed
machining. Mech Syst Signal Process 2011;25:2388–98.
[8] Van Overschee P, De Moor B. Subspace algorithms for the stochastic identification
4. Conclusions
problem. Automatica 1993;29:649–60.
[9] Peeters B, De Roeck G. Reference-based stochastic subspace identification for
In this paper, an operational modal identification method is pro- output-only modal analysis. Mech Syst Signal Process 1999;13:855–78.
posed for ambient vibration testing in the presence of harmonics. This [10] Mohanty P, Rixen DJ. Modified SSTD method to account for harmonic excitations
during operational modal analysis. Mech Mach Theory 2004;39:1247–55.
operational modal identification method, based on transmissibility [11] Lago T. The difference between harmonics and stochastic narrow band responses.
functions (TFB), is first applied to a two-degree-of-freedom system in Oral presentation at the SVIBS symposium, Stockholm. 1997.
order to validate its numerical implementation, and then investigated [12] Antoni J. Blind separation of vibration components: principles and demonstrations.
Mech Syst Signal Process 2005;19:1166–80.
via an experimental beam test. Identified modal parameters are com- [13] Antoni J. The spectral kurtosis: a useful tool for characterising non-stationary sig-
pared to those found by a classical OMA approach based on the white nals. Mech Syst Signal Process 2006;20:282–307.
noise excitation assumption, namely the SSI method. In the two dif- [14] Brincker R, Andersen P, Moller N. An indicator for separation of structural and
harmonic modes in output-only modal testing. European conference on system
ferent cases, the time consumption of the TFB approach is higher than identification and structural health monitoring. 2000. p. 265–72.
that of the SSI method. This can be explained by the additional stage of [15] Jacobsen N-J, Andersen P, Brincker R. Eliminating the influence of harmonic
the signals preprocessing. It should be noted that the efficiency of the components in operational modal analysis. IMAC conference. 2007.
[16] Jacobsen N-J, Andersen P, Brincker R. Using efdd as a robust technique for de-
proposed TFB method also depends on the modal identification capa-
terministic excitation in operational modal analysis. International operational
city from transmissibility functions. Even these functions are in- modal analysis conference. 2007. p. 193–200.
dependent of excitation’s nature, the TFB could be thus less efficient [17] Kim BH, Stubbs N, Park T. A new method to extract modal parameters using output-
only responses. J Sound Vib 2005;282:215–30.
when modal density is high as in mid-frequency range. The main

7
A. Maamar et al.
[18] Randall R, Peeters B, Antoni J, Manzato S. New cepstral methods of signal pre-
processing for operational modal analysis. International conference on noise and
vibration engineering. 2012.
[19] Devriendt C, Guillaume P. The use of transmissibility measurements in output-only
modal analysis. Mech Syst Signal Process 2007;21:2689–96.
[20] Devriendt C, Guillaume P. Identification of modal parameters from transmissibility
measurements. J Sound Vib 2008;314:343–56.
[21] Devriendt C, De Sitter G, Vanlanduit S, Guillaume P. Operational modal analysis in
View publication stats
Applied Acoustics xxx (xxxx) xxx–xxx
the presence of harmonic excitations by the use of transmissibility measurements.
Mech Syst Signal Process 2009;23:621–35.
[22] Hermans L, Van der Auweraer H. Modal testing and analysis of structures under
operational conditions: industrial applications. Mech Syst Signal Process
1999;13:193–216.
[23] Peeters B, Van der Auweraer H. Polymax: a revolution in operational modal ana-
lysis. 1st International operational modal analysis conference, Copenhagen,
Denmark. 2005. p. 26–7.