Journal of Sound and Vibration 428 (2018) 13e25

Contents lists available at ScienceDirect

Journal of Sound and Vibration

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

An identification method for frequency dependent material

properties of viscoelastic sandwich structures
K.S. Ledi*, M. Hamdaoui, G. Robin, E.M. Daya
Universit

e de Lorraine, CNRS, Arts et M

etiers ParisTech, LEM3, F-57000 Metz, France

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

Article history: In this paper, an inverse identification method of the viscoelastic material properties
Received 24 October 2017 (shear modulus G0 and loss factor hC ) of a symmetric three layered viscoelastic sandwich
Received in revised form 28 March 2018 beam is proposed. Experimental vibration tests are performed to determine resonant
Accepted 24 April 2018
frequencies and loss factors for different bending modes. The inverse approach consists
Handling Editor: I. Trendafilova
into an iterative procedure that determines the mode shapes given the material param-
eters and then computes the viscoelastic properties from the modes using a Rayleigh
quotient until convergence on the material properties is met. As a result, the frequency
Keywords:
Inverse problem
dependent viscoelastic material properties of sandwich beams are determined in an
Rayleigh quotient automated fashion. The method is successfully compared to the Ross-Kerwin-Ungar for-
Finite element model mulas and to a standard optimization approach. A fit of the viscoelastic material properties
Vibrations is performed providing analytical expressions for these quantities over a wide range of
Viscoelasticity frequencies.
Sandwich structures © 2018 Elsevier Ltd. All rights reserved.

1. Introduction

Constrained viscoelastic sandwich structures are largely used to reduce and control noise and vibrations in aerospace and
automotive [1e3]. Thanks to their high stiffness and strength to weight ratios, good fatigue properties, good thermal and
acoustical insulation and ease of mass production, they are also popular in engineering applications as load-carrying
structural members [4,5]. Viscoelastic sandwich structures are usually three layered structures composed by two thin
elastic face sheets that sandwich a viscoelastic core. The face sheets material can be fiber-reinforced plastics [6e8], heat-
resistant steel [9] or aluminum alloys [10,11] which are bonded to the core with an adhesive. The core keeps the faces
separated, stabilizes them and contributes to the flexural stiffness, out-of-plane shear and compressive strength and in-
troduces a structural damping. As the performance of viscoelastic sandwich structures depends on the material properties of
the faces and of the core, it is of importance to characterize material core parameters [12]. In the case of homogeneous and
isotropic material, only two elastic values are necessary. It has been established in Refs. [13,14] that the pairs (G,K) or (E,K) are
the most suitable to measure. Viscoelastic material characterization is generally done through measurements of the uniaxial
shear modulus (G) and the compressibility modulus (K).
In the frequency domain, the resonant method and the non-resonance methods like Dynamical Mechanical Analysis
(DMA) are the most well known. The Oberst method [15e17] is used to measure the viscoelastic properties of relatively soft
and highly damping materials. This method, described in Ref. [18], consists in resonance trials on beams damped by a

* Corresponding author.
E-mail address: koffi[email protected] (K.S. Ledi).

https://doi.org/10.1016/j.jsv.2018.04.031
0022-460X/© 2018 Elsevier Ltd. All rights reserved.
14 K.S. Ledi et al. / Journal of Sound and Vibration 428 (2018) 13e25

viscoelastic material layer (free or stressed). Once the modal parameters (resonant frequency and loss factor) are determined,
the storage modulus and the viscoelastic loss factor are calculated using analytical relationships [19,20]. Barbosa and Farage
[21] used a finite element model formulation based on GollaeHughes method (GHM) to describe the viscoelastic material.
The parameters in the GHM rheological model are determined thanks to experimental tests based on Ross-Kerwin-Ungar
relationships. Barkanov et al. [22] worked on an inverse technique to characterize the nonlinear mechanical properties of
viscoelastic core layers in sandwich panels. Based on vibration tests, this technique consists in a numerical model and an
identification procedure that makes use of a response surface method in order to decrease the computational efforts. Elkhaldi
et al. [23] presented a viscoelastic parameters identification method for the generalized Maxwell model. They used automatic
differentiation with a gradient algorithm to minimize a least-square error between numerical and experimental values.
Elejabarrieta [24,25] identified the parameters of a fractional derivative model for viscoelastic sandwich beams by minimizing
an error between predicted frequency response curves (FRF) and measured ones at specified control frequencies. Wassereau
et al. [26] identified experimentally flexural and shear complex moduli of sandwich beams using the force analysis technique
along with Timoshenko beam's theory and homogenization. Sun et al. [27] proposed an identification method based on
response surface methodology, simplified FRF representation and an optimization approach to identify the frequency
dependent mechanical parameter of unconstrained viscoelastic plates. In the aforementioned works, usually the viscoelastic
layers are characterized independently; which means that the mechanical properties are determined layer by layer. These
characterization methods do not take into account the interfaces between the layers or the possible imperfections of these
interfaces.
In this paper, we propose a new inverse identification method based on a Rayleigh quotient for viscoelastic core layer
material properties in sandwich structures. In our study, the viscoelastic layer's properties (shear modulus and loss factor) are
determined from a numerical model of the three layered sandwich beam based on vibration tests performed on the whole
structure. Our identification method is a non-destructive method that takes into account the structure in its globality.
In section 2, the experimental setup is described. In section 3, the new inverse identification method is detailed and
validated by comparison to a standard optimization approach. In section 4, experimental vibration tests for different beam
lengths are performed and the shear modulus and loss factor are determined by our method and fitted versus frequency.

2. Experimental measurements

2.1. Viscoelastic sandwich beams

The viscoelastic sandwich is composed of three layers: two elastic face layers and one viscoelastic core layer (Fig. 1). The
three layers are glued to become an assembly. The viscoelastic material is studied in situ, i.e. when it is sandwiched between
two elastic faces.
In this paper two types of sandwich beams (beam 1 and beam 2), distinguished by their respective core thicknesses, are
considered. Geometrical and material properties are reported in Table 1.

2.2. Experimental setup

The experimental setup to measure sandwich beam frequency response curves is shown in Fig. 2. It is composed of a
shaker, a laser vibrometer and a sandwich beam sample.
The shaker generates vibrations that are controlled and measured by a controller device (UCON system). The range of
frequency excitation is between 4 and 1500 Hz. The vibration amplitude of the shaker is controlled thanks to an acceler-
ometer (PCB Piezotronics - 352C33). The amplitude of the shaker is chosen to minimize nonlinear vibrations typically
0,02 mm displacement between 5 and 157, 5 Hz and a 1g acceleration between 157,5 and 1500 Hz. The amplitude of the
beam's vibration is measured by a laser vibrometer, pointed on the free extremity of the beam.
The boundary condition of the beam is clamped-free. The beams are attached to the shaker thanks to an apparatus, which
adjusts the length of the clamping. Four lengths are tested (318, 354, 400 and 475 mm).

Fig. 1. Sandwich beam's geometry.

 K.S. Ledi et al. / Journal of Sound and Vibration 428 (2018) 13e25 15

Table 1
Material properties and dimensions for the beams.

Elastic layers
Material Aluminum
Young's modulus Ef ¼ 6:9*1010 Pa
Poisson's ratio nf ¼ 0:3
Density rf ¼ 2766 kg=m3
Thickness hf1 ¼ 1 mm and hf2 ¼ 0:5 mm
Viscoelastic layer
Material Polyurethane dielectric resin: RE 12461 Polyol e RE 1010 Isocyanate
Poisson's ratio nc ¼ 0:3
Density rc ¼ 1550 kg=m3
Thickness hc1 ¼ 1 mm and hc2 ¼ 2 mm
Whole beam
Length L ¼ 500 mm
Width b ¼ 30 mm
Thickness h ¼ 3 mm

2.3. Experimental results

Fig. 3 shows the spectrum of vibration displacement amplitude at the free end of beam 2. For each mode, the resonant
frequency and the loss factor are determined. The damping is calculated by the half power bandwidth method. This method
remains accurate for low damping values ðh < 0:4Þ [28]. In this example, six resonant frequencies (10.43 Hz, 66.71 Hz,
186.48 Hz, 359.99 Hz, 591.60 Hz, and 873.27 Hz) and the corresponding loss factors (0.075, 0.064, 0.074, 0.066, 0.084, and
0.100) are measured.

3. New identification approach

3.1. Finite element model

3.1.1. Kinematic model

In Fig. 1, x, y and z symbols denote respectively the coordinates along the length, the width and the thickness directions.
The symbol zi represents the ith middle plane coordinate, (z2 ¼ 0 for the viscoelastic core layer). The subscript i ¼ 2 refers to
the core layer while the subscripts i ¼ 1 and i ¼ 3 refer respectively to the top and bottom elastic faces.
The following assumptions, common to several authors have been used:

 Plane transverse sections stay plane during bending,

 Shear effect is taken into account in the visco-elastic layer and neglected in the elastic layers,
 Transversal displacement w is the same for the three layers,
 The three layers are perfectly glued thus ensuring the continuity of the displacement field.

The kinematic fields are based on Rao's zig-zag model [29] which consists on a classical EulereBernoulli's beam theory for
the elastic faces and Timoshenko's beam theory for the viscoelastic layer. The faces layers have the same thickness, mass
density and Young's modulus. In this study, the face material is assumed to be isotropic, homogeneous and elastic. The core
material is also isotropic, homogeneous and linearly viscoelastic with a real Poisson's ratio and a complex frequency
dependent shear modulus Ec* which is related to the visco-elastic relaxation function of the material R by:

Zþ∞
Ec* ðuÞ ¼ iu RðtÞeiut dt (1)
0

Note that, this model is sufficient when the thickness ratio hc =hf  10 and the Young's modulus ratio Ec =Ef  0:01 [30].
The schematic representation of the zig-zag model is shown on Fig. 1.
Based on these assumptions, the displacement and strain fields are presented for the elastic layers i ¼ 1; 3 as follows

Ui ðx; z; tÞ ¼ ui ðx; tÞ  ðz  zi Þw;x (2.a)

Wi ðx; z; tÞ ¼ wðx; tÞ (2.b)

16 K.S. Ledi et al. / Journal of Sound and Vibration 428 (2018) 13e25

Fig. 2. Experimental setup to measure resonant frequencies and loss factors of sandwich beams.

εi ðx; z; tÞ ¼ ui ;x  ðz  zi Þw;xx (2.c)

h þh
where z1 ¼ z3 ¼ f 2 c , Ui is the axial displacement of the ith layer, ui and w represent respectively the axial displacement of
the middle surface of the ith layer and common transverse displacement in the three layers. The quantity εi is the normal
strain in the ith layer.
The relatives quantities for the central layer i ¼ 2 are given by:

U2 ðx; z; tÞ ¼ uðx; tÞ þ zbðx; tÞ (3.a)

W2 ðx; z; tÞ ¼ wðx; tÞ (3.b)

ε2 ðx; z; tÞ ¼ u;x þ zb;x (3.c)

g2 ðx; z; tÞ ¼ b þ w;x (3.d)

where u is the axial displacement of the middle surface of the central layer, b the rotation of the normal to the middle surface
of the central layer and g2 is the shear strain. The assumption of continuity of displacement field at interfaces gives us
 
hc h
u1 ¼ u þ b  f w;x (4.a)
2 2
 
hc h
u3 ¼ u  b  f w;x (4.b)
2 2

The generalized Hooke's stressestrain law gives us the expression of the axial force Ni and the bending moment Mi of
layers i ¼ 1; 3

Ni ðx; tÞ ¼ Ef Sf ui ;x (5.a)

Mi ðx; tÞ ¼ Ef If w;xx (5.b)

where Sf and If are the area and the quadratic moment of the elastic faces cross-section, respectively. For the viscoelastic
central layer, we have:

_x
N2 ðx; tÞ ¼ Sc RðtÞ*u; (6.a)

M2 ðx; tÞ ¼ Ic RðtÞ*b_ ;x (6.b)

where Sc and Ic are the area and the quadratic moment of the viscoelastic layer cross-section, RðtÞ is the visco-elastic
relaxation function. The shear stress T induced by the shear strain g2 in central layer is:

Sc  
Tðx; tÞ ¼ _ x þ b_
RðtÞ* w; (7)
2ð1 þ nc Þ

3.1.2. Finite element discretization

The principle of virtual work gives us the equation of motion of a sandwich beam for free vibrations analysis:
 K.S. Ledi et al. / Journal of Sound and Vibration 428 (2018) 13e25 17

Fig. 3. Spectrum of vibration displacement amplitude at the free extremity for beam 2.

ZL h   i
Ndu;x þ Mb db;x þ Mw dw;xx þ Tðdw;x þ dbÞ þ 2rf Sf þ rc Sc w€ dw dx ¼ 0: (8)
0

Equation (8) shows that bending and longitudinal motion can be decoupled which gives the following equation for the
bending motion:

ZL h   i
Mb db;x þ Mw dw;xx þ Tðdw;x þ dbÞ þ 2rf Sf þ rc Sc w€ dw dx ¼ 0: (9)
0

where

hc Ef Sf hc  
Mb ¼ M2 þ ðN1  N3 Þ ¼ Ic R*b_ ;x þ hc b;x  hf w;xx (10)
2 2
!
hf Ef Sf h2f Ef Sf hf hc
Mw ¼ M1 þ M3 þ ðN3  N1 Þ ¼ 2Ef If þ w;xx  b;x (11)
2 2 2

For free vibration analysis, the following harmonic form [24] is considered:

wðx; tÞ ¼ WðxÞeiut
(12)
bðx; tÞ ¼ BðxÞeiut

Thus, Equation (9) becomes:

ZL   ZL " ! #
Ef Sf hc   Ef Sf h2f Ef Sf hf hc
Ic Ec* ðuÞB;x dB;x þ hc B;x  hf W;xx dB;x dx þ 2Ef If þ W;xx dW;xx  B;x dW;xx dx
2 2 2
0 0
ZL "  
#
Sc Ec* ðuÞ
þ ðW;x þ BÞðdW;xx þ dBÞ  u 2rf Sf þ rc Sc W dW dx
2
2ð1 þ nc Þ
0
¼ 0: (13)

An unidimensional finite element approach for the discretization of Equation (13) is used. Each element has two nodes and
each node has three degrees of freedom: the transverse displacement ðWÞ, the slope (W;x ) and the rotation ðBÞ.
The elemental column vector of nodal displacements is:
n o
T
U ðeÞ ¼ W1 W1;x B1 W2 W2;x B2 : (14)
18 K.S. Ledi et al. / Journal of Sound and Vibration 428 (2018) 13e25

The transversal displacement of the sandwich beam and the rotation are expressed in function of the nodal displacements
by standard shape functions [31]:
n o
W ¼ ½Nu  U ðeÞ ; (15.a)
h in o
B ¼ Nb U ðeÞ ; (15.b)

where, the shape functions are given by:

½Nu  ¼ ½ N1 ðxÞ N2 ðxÞ 0 N3 ðxÞ N4 ðxÞ 0 ; (16.a)

h i
Nb ¼ ½ 0 0 N5 ðxÞ 0 0 N6 ðxÞ ; (16.b)

with,

ð1  xÞ2 ð2 þ xÞ
N1 ðxÞ ¼ ; (16.c)
4

Le ð1  xÞ2 ð1 þ xÞ
N2 ðxÞ ¼ ; (16.d)
8
2
ð1 þ xÞ ð2  xÞ
N3 ðxÞ ¼ ; (16.e)
4
2
Le ð1 þ xÞ ð1  xÞ
N4 ðxÞ ¼  ; (16.f)
8
1x
N5 ðxÞ ¼ ; (16.g)
2
1þx
N6 ðxÞ ¼ : (16.h)
2

The above expressions use a reduced coordinate: x ¼ 2x e e

Le  1, with L the element length, x2½0L  and x2½  1 1. Inserting
Eqs. (16) and (15) into (13) gives an elementary eigenvalue equation:
 n o
½K e ðuÞ  u2 ½M e  U ðeÞ ¼ 0: (17)

where ½M e  and ½K e ðuÞ are respectively mass and stiffness elementary matrixes. The assembly of these elementary matrixes
gives non-linear eigenvalue equation:
h i
KðuÞ  u2 M fUg ¼ 0: (18)

In this equation, ½M and ½KðuÞ are respectively global mass and stiffness matrixes, and fUg is the complex mode of vi-
bration. This matrix formulation is well known in free vibrations analysis of viscoelastic structures [32e37]. Eq. (18) cannot be
solved by classical eigenvalue procedures when ½KðuÞ depends nonlinearly on u. Many attempts have been proposed in the
literature to solve this problem. The complex eigenvalue method [38] and the asymptotic approach [33] are limited to the case
of constant stiffness viscoelastic structures. The modal strain energy method [36] is applicable for frequency dependent
structures but in some practical cases the error can be relatively large [39]. The direct frequency response method was applied
in Ref. [31], but it is of high computational cost. Other numerical methods for nonlinear eigenvalue problems that account for
frequency dependence have been proved efficient to solve Eq. (18): the inverse iteration method [40], the asymptotic nu-
merical method [33,35] (ANM), the order-reduction iterative algorithm [36] and the Arnoldi/Jacobi-Davidson iterative pro-
jection methods [41,42]. Hamdaoui et al. [43] provided a comparison of these techniques in terms of accuracy and
computational times. A toolbox (Diamant Toolbox) has been designed with the help of automatic differentiation techniques to
make the asymptotic numerical method generic and easy of use [44e46]. The solution can be performed easily and in an exact
way whatever the dependence on frequency of the matrix [K] is. The exact complex eigenmodes are investigated based on
this procedure.
 K.S. Ledi et al. / Journal of Sound and Vibration 428 (2018) 13e25 19

3.2. Asymptotic numerical method

In this section, we will discuss the Asymptotic Numerical Method (ANM). The latter allows to solve the nonlinear
eigenvalue equation (Eq. (18)), and it will be used in particular in section 3.3 within our inverse identification method. In
general, the unknowns of the nonlinear equation (Eq. (18)) are the eigenmodes U and the eigenvalues l with l ¼ u2 . The ANM
expresses them in terms of power series with a path parameter p. So we have:

X
N
U¼ p i Ui (19)
i¼0

X
N
l¼ pi li (20)
i¼0

By inserting Equations (19) and (20) in the nonlinear eigenvalue Equation (18) and performing a high order Taylor
expansion, we obtain a collection of linear problems. The combination of ANM and automatic differentiation technique for
high order derivatives computation is implemented within the DIAMANT toolbox [44,45] which will be used in the following
for the calculation of the eigenmodes. Elastic and viscoelastic layers of our sandwich structure are isotropic, then Eq. (18) is
rewritten as:

½Kð0Þ þ EðlÞKv  lMfUg ¼ 0 (21)

The residue can be split into two parts, one linear S and the other nonlinear T, as follows:

RðU; lÞ ¼ ðKð0Þ  lMÞ fUg þ EðlÞKv fUg ¼ 0 (22)

RðU; lÞ ¼ SðU; lÞ þ TðU; lÞ ¼ 0 (23)

SðU; lÞ ¼ ðKð0Þ  lMÞfUg (24)

TðU; lÞ ¼ EðlÞKv fUg (25)

The original problem (22) is transformed by the homotopy technique [46]; so that the path parameter p is introduced in
order to drive the solution ðU; lÞ of the problem from, an initial step corresponding to a problem with real eigenvalues SðU;
lÞ ¼ 0, with ðU0 ; l0 Þ as solution, to a final step corresponding to the problem with complex values RðU; lÞ ¼ 0. The residue is
then rewritten as follows:

RðU; l; pÞ ¼ SðU; lÞ þ pTðU; lÞ ¼ 0; pε½0; 1 (26)

By writing S and T as a Taylor series, the DIAMANT toolbox allows us thanks to the automatic differentiation technique to
solve the problem as a linear system; thus for each initial solution ðU0 ; l0 Þ, and for a given truncation order and tolerance the
solution (U, l) is obtained.

3.3. Inverse identification method

In this section an inverse method is proposed for viscoelastic material sandwiched between two elastic layers. This
method is based on the nonlinear eigenvalue equation (Eq. (18)). As mentioned before, the elastic and viscoelastic materials
are assumed to be isotropic. Thus, the frequency dependent stiffness matrix KðuÞ of the sandwich structure can be rewritten
as [46]

KðuÞ ¼ K 0 þ GðuÞKv ¼ K 0 þ Gð1 þ ihÞKv (27)

Let ður ; Ur Þ be a solution of Eq. (18), thus this equation becomes:

h i
K 0 þ Gð1 þ ihÞKv  u2r M Ur ¼ 0 (28)

By multiplying Equation (28) by UrH (where :H stands for Hermitian transposition), one obtains the expression of the
complex shear modulus of the viscoelastic core:
20 K.S. Ledi et al. / Journal of Sound and Vibration 428 (2018) 13e25

UrH M Ur UrH K 0 Ur
G* ¼ Gð1 þ ihÞ ¼ u2r  (29)
UrH Kv Ur UrH Kv Ur

The values of the shear modulus G and the loss factor h appear as function of the frequency u and of the fixed mode U.
Their identification can be obtained from an iterative procedure.
Equation (29) establishes a relationship between the shear modulus G and the loss factor h from one part and, the pul-
sation ur and the mode shape Ur from the other part. The problem is that if ur can be easily determined experimentally, the
mode shape is harder to get, but it is still needed to identify the shear modulus and loss factor. The proposed method, in this
work, relies on an iterative resolution of Eq. (29). First of all, ur is determined experimentally and is set equal to uexp ðu2exp ¼
U2exp ð1 þ ihexp ÞÞ with Uexp the measured real pulsation and hexp the measured loss factor of the sandwich beam.
At iteration k, we suppose that we know the shear modulus Gk and the loss factor hk, then we solve for Equation (18) to
determine the mode shape Urk using the ANM with automatic differentiation (DIAMANT Toolbox). We can determine the new
values of the shear modulus Gkþ1 and the loss factor hkþ1, at iteration kþ1 using Eq. (29) by taking real and imaginary part of

and Dh ¼ h h
kþ1
hk
the right hand side. A convergence test is performed by computing DG ¼ Gkþ1 Gk
Gk k . The iterations are

stopped if (maxðDG; DhÞ < tolerance) or if the number of iterations exceeds Niter .
The initial values of the shear modulus G and loss factor h of the viscoelastic core are chosen closer to those given by the
RKU relationships (Equations (31) and (32)). Different starting points are tested and our inverse method gives good results for
example with G0 ¼ 2:5±2 MPa and h0 ¼ 0:5±0:25. In these cases, the maximal relative errors are respectively less than 8e-
05% and 8.52e-02% whatever the mode for the shear modulus and the loss factor.
The flowchart of the whole identification process is shown in Fig. 4. The process is repeated for each resonant mode to
obtain a frequency dependent description of the shear modulus G and loss factor h. In this work a tolerance of 104 and a
maximal number of iterations of 1500 have been considered.

3.4. Error minimization method

The classical method consisting in minimizing the error between the finite element model and the experimental mea-
surements is also considered for validation of the obtained results. Given a set of experimentally measured frequencies ðf1 ;
…; fp Þ and loss factors ðh1 ; …; hp Þ and a numerical model (here our finite element model described in section 3.1) that gives us
finum and hnum
i for each mode i and couple ðG; hÞ of shear modulus and material loss factor, the error

fi  finum hi  hnum
i
εi ¼ þ ; (30.a)
fi hi

is minimized for each mode i. The optimization problem that is solved for each mode i reads as follows:

minð0GGmax ;0h1Þ εi (30.b)

Fig. 4. Flowchart of the identification process.

 K.S. Ledi et al. / Journal of Sound and Vibration 428 (2018) 13e25 21

This procedure allows us to retrieve the shear modulus Gi and material loss factor hi for each mode resulting in a
description of ðGi ; hi Þ in function of the frequency. A value of Gmax ¼ 109 Pa is used. The error is minimized for each mode
versus ðGi ; hi Þ using the global search strategy of Matlab implemented via the function gs [47]. Sequential quadratic algorithm
implemented in the function fmincon is used for local optimization with central finite differences for gradient computation.
The starting point is chosen at random between lower and upper bounds.

3.5. Ross-Kerwin-Ungar relationships (ASTM norm)

In the following lines, Ross-Kerwin-Ungar (RKU) relationships [48] commonly used to identify viscoelastic modulus and
loss factor based on vibration tests are described:

ðA  BÞ  2ðA  BÞ2  2ðAhn Þ2 E1 H1 H2 an

E2n ¼ 2ð1 þ nÞ (31)
ð1  2A þ 2BÞ2 þ 4ðAhn Þ2 L2

Ahn
h2n ¼ (32)
ðA  BÞ  2ðA  BÞ2  2ðAhn Þ2

 2  
fn B
A¼ ð2 þ rr h2 Þ (33)
f1n 2

1
B¼ (34)
6ð1 þ h2 Þ2
!
hc1
where h2 is the ratio between the viscoelastic core thickness and the elastic layer thickness h2 ¼ hf1
, fn and hn are the nth

resonant frequency and loss factor of the sandwich beam set (Table 2) and f1n is the nth Euler Bernoulli beam resonant
frequency. The quantity rr is the ratio between viscoelastic and elastic material densities. L is the beam length and, n is the
Poisson's ratio of the viscoelastic material which is equal to 0:3.
The viscoelastic core shear modulus G and the viscoelastic core loss factor h are deduced by:

E2n
G¼ (35)
2ð1 þ nÞ

h ¼ h2n (36)

3.6. Validation of the inverse method

Our method is compared with the RKU relationships (ASTM norm) and the error minimization method for beam 1. In this
case, the useful length of beam 1 is considered equal to 47.5 cm. We adopted the following convention for naming the three
methods (method 1 ¼ ASTM Norm, method 2 ¼ Inverse method, method 3 ¼ Global search method).
The results are displayed in Figs. 5 and 6 for the shear modulus and the loss factor, respectively. The three methods give an
increasing variation as a function of the frequency for the core shear modulus (Fig. 5). The core loss factor is a decreasing
function of the frequency as displayed on Fig. 6. The maximal relative error between method 2 and method 3 is 1.8% for the
core shear modulus and 3.07% for the core loss factor. Thus, we can conclude that method 2 and method 3 give the same
results. The maximal relative error between method 2 and method 1 is 28.16% for the core shear modulus and 4.3% for the
core loss factor. This shows us that method 2 and method 3 give more accurate results than method 1 which emphasizes that
using the ASTM Norm can be misleading and stresses out the efficiency of global optimization methods and inverse models
for viscoelastic material identification. These findings validate our inverse identification method as can be clearly noticed on
Figs. 5 and 6.

Table 2
Resonant frequency and loss factor per mode for the sandwich beam 1 with L ¼ 47.5 cm.

Mode 1 2 3 4 5 6 7
fn (Hz) 10.79 67.18 186.74 363.98 598.15 885.37 1214.99
hn 0.061 0.070 0.082 0.041 0.041 0.049 0.015
22 K.S. Ledi et al. / Journal of Sound and Vibration 428 (2018) 13e25

Fig. 5. Comparison of calculated core shear modulus between method 1, 2 and 3.

Fig. 6. Comparison of calculated core loss factors between method 1, 2 and 3.

Fig. 7. Spectrum of vibration amplitude displacement at the free extremity for different lengths for beam 1.

Fig. 8. Spectrum of vibration amplitude displacement at the free extremity for different lengths for beam 2.
 K.S. Ledi et al. / Journal of Sound and Vibration 428 (2018) 13e25 23

4. Identification of the properties of viscoelastic core

In order to have several points, vibration trials are performed on two beams with clamped-free boundary conditions. For
each beam, we have four useful lengths, which corresponds to four different trials. Thus, in Figs. 7 and 8, the spectrum
amplitude of the displacement of the vibration on the free extremity of the beam is shown for the beam 1 and beam 2,
respectively.
On each beam, the more the useful length increases during trials, the more the resonant frequency decreases and more
modes can be measured. Based on these experimental data for beam 1 and beam 2 (resonant frequency fi and loss factor hi at
each mode of the beam) available on Table 3 and our inverse method, we calculated the shear modulus G and the loss factor h

Table 3
Frequencies and loss factors for the two beams with variable lengths.

Trials Length in cm Frequencies in Hz (Loss factors)

f1 f2 f3 f4 f5 f6 f7
(h1) (h2) (h3) (h4) (h5) (h6) (h7)
Beam 1 31.8 22.87 140.26 387.76 754.75
(0.043) (0.151) (0.055) (0.038)
35.4 18.74 114.66 319.77 624.98
(0.042) (0.122) (0.039) (0.041)
40 14.62 89.69 248.62 485.93 796.36
(0.045) (0.093) (0.056) (0.037) (0.053)
47.5 10.79 67.18 186.74 363.98 598.15 885.37 1214.99
(0.061) (0.070) (0.082) (0.041) (0.041) (0.049) (0.015)
Beam 2 31.8 21.95 136.45 381.94 739.33
(0.042) (0.146) (0.047) (0.058)
35.4 17.88 111.46 311.99 603.94
(0.041) (0.116) (0.044) (0.052)
40 13.65 85.94 240.54 470.14 767.32
(0.053) (0.091) (0.056) (0.052) (0.065)
47.5 9.82 63.14 176.99 344.01 563.78 833.36 1156.3
(0.096) (0.062) (0.082) (0.074) (0.069) (0.078) (0.103)

Fig. 9. Numerical results and fitted curve for the identified viscoelastic shear modulus.

Fig. 10. Numerical results and fitted curve for the identified viscoelastic loss factor.
24 K.S. Ledi et al. / Journal of Sound and Vibration 428 (2018) 13e25

of the viscoelastic core. We then performed two fits (Figs. 9 and 10) in order to find the shear modulus and loss factor ex-
pressions in function of frequency (Hz). These expressions are given by Equations (37) and (38).

G ¼ a1 þ b1 u þ c1 u2 (37)

where a1 ¼ 1*106 ; b1 ¼ 2:9907*104 ; c1 ¼ 0:637.

b2 c2
h ¼ a2 þ þ (38)
u u0:25
where a2 ¼ 0:1912 ; b2 ¼ 50:2153 ; c2 ¼ 3:2341.
One can notice that the fits are quite near the data points providing a good description of the behavior of the identified
quantities over a wide frequency range which is not possible directly with traditional techniques such as DMA. Now, thanks to
our method, the identified material properties can be subsequently used in the finite element analysis, for instance.

5. Conclusion

An inverse numerical method based on the Rayleigh quotient is proposed to identify the frequency dependent properties
of viscoelastic sandwich beams. It relies on solving a non-linear eigenvalue problem by finite elements using the ANM and on
vibration tests to get modal characteristics of the beams (resonant frequencies and damping ratios). The method is imple-
mented and is successfully validated. The shear modulus and loss factor of the viscoelastic material is determined for each
mode. By varying the length of the beams several data points are generated and a fit of the viscoelastic material properties
over a wide frequency range is proposed. The application of this method to determine the viscoelastic property of a visco-
elastic layer embedded into a sandwich beam allows investigating the damping effect of the viscoelastic layer in situ.

Acknowledgements

This work was supported by the French State through the program “Investment in the future” operated by the National
Research Agency (ANR) and referenced by ANR-11-LABX-0008-01 (LabEx DAMAS).

References

[1] F. Fahy, J. Walker, Fundamentals of Noise and Vibration, E&FN Spon, New York, 1998.
[2] M.P. Norton, D.G. Karczub, Fundamentals of Noise and Vibration Analysis for Engineers, second ed., Cambridge University Press, Cambridge, 2003.
[3] A. Thumann, R.K. Miller, Fundamentals of Noise Control Engineering, Fairmont Press, Atlanta, GA, 1986.
[4] F.J. Plantema, Sandwich construction: the Bending and Buckling of Sandwich Beams, Plates and Shells, Wiley, New York, 1966.
[5] H.G. Allen, Z.N. Feng, Classification of structural sandwich panel behaviour, in: Proceeding of the EuroMech 360 Colloquium, Saint-Etienne, France,
1997, pp. 1e12.
[6] E. Nilsson, A.C. Nilsson, Prediction and measurement of some dynamic properties of sandwich structures with honeycomb and foam cores, J. Sound
Vib. 251 (3) (2002) 409e430.
[7] S.S. Tavalaelly, Wave Propagation in Sandwich Structures, PhD Thesis, Department of Vehicle Engineering, the Marcus Wallenberg Laboratory for
Sound and Vibration Research, Stockholm, 2001.
[8] I.M. Daniel, J.L. Abot, Fabrication, testing and analysis of composite sandwich beams, Compos. Sci. Technol. 60 (2000) 2455e2463.
[9] A.D. Nashif, D.I.G. Jones, Vibration Damping, Wiley, New York, 1985.
[10] P.R. Cunningham, R.G. White, A new measurement technique for the estimation of core shear strain in closed sandwich structures, Compos. Struct. 51
(2001) 319e334.
[11] T. Saito, R.D. Parbery, Parameter identification for aluminium honeycomb sandwich panels based on orthotropic Timoshenko beam theory, J. Sound
Vib. 208 (2) (1997) 271e287.
[12] Y. Shi, H. Sol, H. Hua, Material parameter identification of sandwich beams by an inverse method, J. Sound Vib. 290 (2006) 1234e1255.
[13] F.M. Guillot, D.H. Trivett, Complete elastic characterization of viscoelastic materials by dynamic measurements of the complex bulk and Young's
moduli as a function of temperature and hydrostatic pressure, J. Sound Vib. 330 (2011) 3334e3351.
[14] T. Pritz, Measurement methods of complex Poisson's ratio of viscoelastic materials, Appl. Acoust. 60 (3) (2000) 279e292.
[15] D. Backstro €m, Vibration of Sandwich Beams, PhD thesis, KTH Stockholm, 2006.
[16] C.A. Gallimore, Passive Viscoelastic Constrained Layer Damping Application for a Small Aircraft Landing Gear System, Master's thesis, Virginia
Polytechnic Institute and State University, 2008.
[17] J.-L. Wojtowicki, L. Jaouen, R. Panneton, A new approach for the measurement of damping properties of materials using the Oberst beam, Rev. Sci.
Instrum. 75 (8) (2004) 2569e2574.
[18] ASTM International, ASTM E756-05-standard Test Method for Measuring Vibration-damping Properties of Materials, 2010.
[19] D. Ross, E.E. Ungar, E.M. Kerwin, Damping of Plate Flexural Vibrations by Means of Viscoelastic Laminae, American Society of Mechanical Engineering,
1959, pp. 49e87.
[20] A.D. Nashif, D.I.G. Jones, Vibration Damping, Wiley-Interscience, 1985, pp. 203e212.
[21] F.S. Barbosa, M.C.R. Farage, A finite element model for sandwich viscoelastic beams: experimental and numerical assessment, J. Sound Vib. 317 (2008)
91e111.
[22] E. Barkanov, E. Skukis, B. Petitjean, Characterisation of viscoelastic layers in sandwich panels via an inverse technique, J. Sound Vib. 327 (2009)
402e412.
[23] I. Elkhaldi, I. Charpentier, E.M. Daya, A gradient method for viscoelastic behaviour identification of damped sandwich structures, C. R. Mecanique 340
(2012) 619e623.
[24] Manex Martinez-Agirre, Maria Jesus Elejabarrieta, Dynamic characterization of high damping viscoelastic materials from vibration test data, J. Sound
Vib. 330 (2011) 3930e3943.
 K.S. Ledi et al. / Journal of Sound and Vibration 428 (2018) 13e25 25

s, M.J. Elejabarrieta, An approximate numerical method for the complex eigenproblem in systems characterised by a structural damping
[25] F. Corte
matrix, J. Sound Vib. 296 (2006) 166e182.
[26] Thibault Wassereau, Fre
de
ric Ablitzer, Charles Pe

zerat, Jean-Louis Guyader, Experimental identification of flexural and shear complex moduli by
inverting the Timoshenko beam problem, J. Sound Vib. 399 (2017) 86e103.
[27] Wei Sun, Zhuo Wang, Xianfei Yan, Mingwei Zhu, Inverse identification of the frequency-dependent mechanical parameters of viscoelastic materials
based on the measured FRFs, Mech. Syst. Signal Process. 98 (2018) 816e833.
[28] A.K. Chopra, Dynamics of Structures, third ed., Prentice Hall, New Jersey, 2007.
[29] D.K. Rao, Frequency and loss factors of sandwich beams under various boundary conditions, Mech. Eng. Sci. 20 (1978) 271e282.
[30] H. Hu, S. Belouettar, M. Potier-Ferry, E.M. Daya, Review and assessment of various theories for modeling sandwich composites, Compos. Struct. 84 (3)
(July 2008) 282e292.
[31] M. Bilasse, E.M. Daya, L. Azrar, Linear and nonlinear vibrations analysis of viscoelastic sandwich beams, J. Sound Vib. 329 (2010) 4950e4969 (1992).
[32] M.L. Soni, Finite element analysis of viscoelastically damped sandwich structures, Shock Vib. Bull. 55 (1) (1981) 97e109.
[33] E.M. Daya, M. Potier-Ferry, A numerical method for nonlinear eigenvalue problems at application to vibrations of viscoelastic structures, Comput.
Struct. 79 (5) (2001) 533e541.
[34] B. Ma, J. He, A finite element analysis of viscoelastically damped sandwich plates, J. Sound Vib. 152 (1) (1992) 107e123.
[35] L. Duigou, E.M. Daya, M. Potier-Ferry, Iterative algorithms for nonlinear eigenvalue problems. Application to vibrations of viscoelastic shells, Comput.
Meth. Appl. Mech. Eng. 192 (2003) 1323e1335.
[36] X. Chen, H.L. Chen, X.L. Hu, Damping prediction of sandwich structures by order-reduction-iteration approach, J. Sound Vib. 222 (5) (1999) 803e812.
[37] E.E. Ungar, E.M. Kerwin, Loss factors of viscoelastic systems in terms of energy concepts, J. Acoust. Soc. Am. 34 (7) (1962) 954e957.
[38] J. Wilkinson, The Algebraic Eigenvalue Problem, Clarendon Press, Oxford, 1965.
[39] R. Rickards, A. Chate, E. Barkanov, Finite element analysis of damping the vibrations of laminated composites, Comput. Struct. 47 (6) (1993)
1005e1015.
[40] K. Schreiber, Nonlinear Eigenvalue Problems: Newton-type Methods and Nonlinear Rayleigh Functionals, Ph.D. thesis, Technischen Universitat Berlin,
2008.
[41] H. Voss, An Arnoldi method for nonlinear eigenvalue problems, BIT Numer. Math. 444 (2004) 387e401.
[42] H. Voss, A Jacobi-Davidson method for nonlinear and non symmetric eigenproblems, Comput. Struct. 85 (2007) 1284e1292.
[43] M. Hamdaoui, A. Komlan, E.M. Daya, Comparison of non-linear eigensolvers for modal analysis of frequency dependent laminated viscoelastic
sandwich plates, Finite Elem. Anal. Des. 121 (2016).

rentiation automatique de la me
[44] I. Charpentier, M. Potier-Ferry, Diffe
thode asymptotique nume
rique type

e: l'approche Diamant, C.R. Me
canique 336
(2008) 336e340.
[45] Y. Koutsawa, I. Charpentier, E.M. Daya, M. Cherkaoui, A generic approach for the solution of nonlinear residual equations. Part I: the Diamant toolbox,
Comput. Meth. Appl. Mech. Eng. 198 (3e4) (2008) 572e577.
[46] M. Bilasse, I. Charpentier, E.M. Daya, Y. Koutsawa, A generic approach for the solution of nonlinear residual equations. Part II: homotopy and complex
nonlinear eigenvalue method, Comput. Meth. Appl. Mech. Eng. 198 (2009) 3999e4004.
[47] Ugray Zsolt, Leon Lasdon, John C. Plummer, Fred Glover, James Kelly, Rafael Martí, Scatter search and local NLP solvers: a multistart framework for
global optimization, INFORMS J. Comput. 19 (3) (2007) 328e340.
[48] American Society for Testing and Materials, ASTM E 756e04e1, Standard Test Method for Measuring Vibration-Damping Properties of Materials, Book
of Standards, Vol. 04.06dThermal Insulation, Environmental Acoustics 2004.