JID:CRAS2B AID:3561 /FLA [m3G; v1.231; Prn:23/02/2018; 12:40] P.

1 (1-12)
C. R. Mecanique ••• (••••) •••–•••

Contents lists available at ScienceDirect

Comptes Rendus Mecanique

www.sciencedirect.com

Non-linear vibrations of sandwich viscoelastic shells

Lahcen Benchouaf a,∗ , El Hassan Boutyour a , El Mostafa Daya b ,
Michel Potier-Ferry b
a
Laboratory of Mechanics, Energetics, Electronics & Telecommunications, Department of Applied Physics, Faculty of Sciences and Technology,
Hassan 1st University, PO Box 577, Settat, Morocco
b
Laboratoire d’étude des microstructures et de mécanique des matériaux (LEM3), UMR CNRS 7239, Université de Lorraine, Metz, France

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

Article history: This paper deals with the non-linear vibration of sandwich viscoelastic shell structures.
Received 9 October 2017 Coupling a harmonic balance method with the Galerkin’s procedure, one obtains an
Accepted 4 December 2017 amplitude equation depending on two complex coefficients. The latter are determined by
Available online xxxx
solving a classical eigenvalue problem and two linear ones. This permits to get the non-
Keywords:
linear frequency and the non-linear loss factor as functions of the displacement amplitude.
Non-linear vibration To validate our approach, these relationships are illustrated in the case of a circular
Viscoelastic sandwich ring.
Sandwich © 2017 Académie des sciences. Published by Elsevier Masson SAS. All rights reserved.
Harmonic balance method
Eigenvalue
Loss factor

1. Introduction

In the mechanical structures field, the viscoelastic material is widely used to reduce vibration and noise in many domains
(e.g., aerospace industry). Indeed, it can induce an effective damping especially when it is sandwiched between two elastic
hard layers. Generally, the damping properties are characterized by two modal parameters that are the frequency and the
loss factor. Many investigations have been carried out on the linear dynamic analysis of viscoelastic structures. A major
difficulty in their study is that the stiffness matrix is complex and depends non-linearly on the vibration frequency. The
solution yield complex modes and complex eigenvalues whose real and imaginary parts are associated respectively with
the frequencies and with the loss factors. Several procedures have been developed to determine these quantities. Analytical
methods were devoted to simple structures [1–10], and numerical ones using finite element simulations were introduced to
design structures with complex geometries and generic boundary conditions [11–22]. The simplest technique is the modal
strain energy method used by Ma and He [12], which defines a rather good estimate of the loss factor from a sort of
one-mode Galerkin approximation. One notes that from an engineering viewpoint, the most relevant quantity is the loss
factor, which is associated with any mode.
In the case of non-linear viscoelastic structures, only a few investigations have been devoted to take into account the
non-linear geometrical effects. For instance, these studies concern sandwich viscoelastic structures with simple geometry
as beams or plates [23–26]. As it is well known, the non-linear geometrical effects induce some dependence between the
frequencies and the loss factors with respect to the amplitude [25,27]. Recently, Boumediene et al. [28] developed a reduc-
tion method based on a high-order Newton algorithm and reductions techniques to determine the modal characteristics of

* Corresponding author.
E-mail address: [email protected] (L. Benchouaf).

https://doi.org/10.1016/j.crme.2017.12.013
1631-0721/© 2017 Académie des sciences. Published by Elsevier Masson SAS. All rights reserved.
JID:CRAS2B AID:3561 /FLA [m3G; v1.231; Prn:23/02/2018; 12:40] P.2 (1-12)
2 L. Benchouaf et al. / C. R. Mecanique ••• (••••) •••–•••

Fig. 1. Geometry of a 3D sandwich structure with two elastic layers and a central viscoelastic one.

viscoelastic sandwich structures. The forced harmonic response of viscoelastic sandwich structures with a reasonable com-
putational cost was also studied, employing a reduction technique and the asymptotic numerical method [29]. Based on
von Kármán’s theory and taking into account geometric imperfections, the nonlinear vibrations of viscoelastic thin rectan-
gular plates subjected to normal harmonic excitation are investigated by Amabili [30]. Lougou [31] proposed a double-scale
asymptotic method for the vibration modeling of large repetitive sandwich structures with a viscoelastic core. In his work
[32], Lampoh computes the sensitivity of eigensolutions using a homotopy-based asymptotic numerical method, then a
first-order automatic differentiation to study the modeling of the linear free vibration of a sandwich structure including vis-
coelastic layers yields a complex nonlinear eigenvalue problem. The work of El Khaldi [33] presents a gradient method for
viscoelastic behavior identification of damped sandwich structures devoted to the passive control of mechanical vibration.
The aim of this paper is to establish a much simple methodology for the non-linear vibration analysis of viscoelastic shell
structures. The approach is based on a coupling of an approximated harmonic balance method with a Galerkin’s procedure
with one mode. The non-linear modal relationship giving the frequency (free and forced) and the loss factor, with respect
to the displacement, are obtained by solving a classical eigenvalue problem and two linear ones [24,27]. To validate our
approach, one gives an application to a sandwich viscoelastic ring.

2. Formulation

2.1. Kinematics and constitutive law of the model

Let us consider a thin symmetric sandwich shell having three layers, as shown in Fig. 1; the central layer is viscoelastic
and the external ones are elastic. The shear deformation is neglected in the elastic layers, but, it is taken into account in
the viscoelastic one; it is induced by the difference between the tangential displacements at the interfaces. For each layer,
one denotes by u i (i = 1, 2, 3) the components of the displacement vector in the z direction and given by:

u i (x, y , z, t ) = v i (x, y , t ) + ( z − zi )βi (x, y , t ) i = 1, 3

(1)
u 2 (x, y , z, t ) = v (x, y , t ) + zψ(x, y , t )
where t is the time parameter, (x, y, z) is a coordinate system (z denotes the variation through the thickness). Because of
h +h
the symmetry, one puts z1 = c 2 f = − z3 , hc and hf being the thicknesses of the central and external layers, respectively.
The subscript i indicates the layer variation, starting from the internal layer; 1 and 3 represent the elastic layers, while 2 is
associated with the viscoelastic one. βi and ψ denote the rotations of the cross-section, v i (i = 1, 3) and v denote tangential
components of the displacement vector of the middle planes corresponding to the external and central layers, respectively.
The displacement continuity conditions at the interfaces between the central layer and the external ones permit to get:
hc hf
v1 = v + ψ+ β1
2 2R 1
(2)
hc hf
v3 = v − ψ− β3
2 2R 3
The Green–Lagrange strain in each layer can be decomposed into a linear part and a quadratic one:

γi = γi (u i ) + γnl (u i , u i ) (3)
For the elastic layers, the behavior is described by the classical Hook law, and it is given, for the viscoelastic one, by the
classical convolution product ⊗ of the relaxation function D (t ) by the time derivative of the deformation:

S i = D (0)γ̇i i = 1, 3
S 2 = D ⊗ γ̇2 (4)
where S i is the second Piola–Kirchhoff stress tensor corresponding to the layer i and D (0) is the delayed elasticity modulus.
JID:CRAS2B AID:3561 /FLA [m3G; v1.231; Prn:23/02/2018; 12:40] P.3 (1-12)
L. Benchouaf et al. / C. R. Mecanique ••• (••••) •••–••• 3

2.2. Governing equations

Using the principal of virtual work, the equations describing the non-linear forced vibrations of a 3D sandwich viscoelas-
tic structure can be written in the following general form:

L (U ) + Q (U , U ) + M (Ü ) = f (t ) (5)

where U = (u 1 , S 1 , u 2 , S 2 , u 3 , S 3 ) is a mixed vector, its components are the generalized displacements and the stress cor-
responding to the three layers. L (·) is a linear operator, Q (·, ·) is a bilinear and symmetric one and M (·) is the inertial
operator and f (t ) is the external applied load.
  
 
L (U ), δ U = S 1 : γ1 (δ u 1 ) dv 1 + S 2 : γ1 (δ u 2 ) dv 2 + S 3 : γ1 (δ u 3 ) dv 3 (6)
v1 v2 v1
 
     
Q (U , U ), δ U = δ S 1 : γnl (u 1 , u 1 ) + S 1 2γnl (δ u 1 , u 1 ) dv 1 + δ S 2 : γnl (u 2 , u 2 ) + S 2 2γnl (δ u 2 , u 2 ) dv 2
v1 v2
 (7)
 
+ δ S 3 : γnl (u 3 , u 3 ) + S 3 2γnl (δ u 3 , u 3 ) dv 3
v3
  
 
M (Ü ), δ U = ρ1 ü 1 δ u 1 dv 1 + ρ2 ü 2 δ u 2 dv 2 + ρ3 ü 3 δ u 3 dv 3 (8)
v1 v2 v3

where ρi and v i are respectively the mass densities and the reference configuration of the layer i.

3. Non-linear free vibration by an approximated harmonic balance method

The aim of this section is to get approximate solutions to the non-linear problem (5) and (4), assuming that f (t ) = 0. As
a first approximation, the solution is assumed to be harmonic in time and almost parallel to a single mode in space with
arbitrary complex amplitude. This approximation assumes that the frequency is near the frequency of an associated linear
elastic structure. As in non-linear elastodynamics, the harmonic response has to be corrected to balance the quadratic terms
in (5) and (4). Thus, a non-linear complex frequency–amplitude relationship is obtained by using the one-mode Galerkin
procedure.

3.1. First-order modal approximation

Let us consider a first approximated solution U h to the problem (5) and (4), which is supposed harmonic and propor-
tional to the linear mode:
1  
Uh = U n aeiωt + CC (9)
2
where CC denotes the conjugate complex of the preceding term, a is an unknown complex amplitude, ω the frequency, U n
is the n-th linear vibration mode of the associated elastic system, defined by a classical real eigenvalue problem:

L (U n ) − ωn2 M (U n ) = 0
(10)
S n = D (0)ε (un )

One notes that this first approximation of the non-linear and complex problem is more valid when the damping is small,
and it is used in the modal strain energy to determine the loss factor.

3.2. Computation of the correction term

Let us consider a second-order approximated solution to (5)–(4) by adding a corrective term U c to the linear re-
sponse (9):

U = Uh + Uc (11)

The correction term is assumed to be small with respect to the main term. That is why the equations defining the
correction are linearized with respect to U c . This U c balances the quadratic terms in (5)–(4):

L (U c ) + M Ü c = − Q (U h , U h ) (12)
JID:CRAS2B AID:3561 /FLA [m3G; v1.231; Prn:23/02/2018; 12:40] P.4 (1-12)
4 L. Benchouaf et al. / C. R. Mecanique ••• (••••) •••–•••

The correction term U c combines a time-independent term and a harmonic term with a double frequency:
1 2 
U c = |a|2 U 0 + a U 2 e2iωt + CC (13)
2
When restricted to the elastic case, the approximations (9)–(12) correspond to the two first terms of a Poincaré–Lindstedt
expansion [34], which yields a parabolic approximation of the backbone curve. It holds for moderately large amplitude: the
first harmonic term (9) is small (O(a)) and the correction term is smaller than the first one (O(a2 )). This way, the coupling
term Q (U h , U c ) can be neglected in (12) (O(a3 )), as well as the quadratic term Q (U c , U c ) (O(a4 )).
The substitution of (13) into (12) leads to two linear time-independent problems satisfied by the amplitudes U 0 and U 2 .
1
L (U 0 ) = − Q (U n , U n )

2  (14)
1
S i0 + D (0) γ1 (u i0 ) + γnl (u in , u in ) , i = 1, 2, 3
2
1
L (U 2 ) − 4ωn2 M (U 2 ) = − Q (U n , U n )

2 
1
S i2 = D (0) γ1 (u i2 ) + γnl (u in , u in ) , i = 1, 3 (15)
2


1
S 22 = D (2iωn ) γ1 (u 22 ) + γnl (u 2n , u 2n )
2
where D (0) is the tensor of the delayed elasticity of the viscoelastic material and D (2ω) is the viscoelastic tensor at
frequency 2ω . Thus, the general solution to (5) and (4) induces a principal harmonic and two secondary ones.
1   1 2 
U= U n aeiωt + cc + |a|2 U 0 + a U 2 e2iωt + cc (16)
2 2
As previously said, the approximation (9) assumes that the structure oscillates with a frequency ω near the linear one ωn .
So, the tensor D (2ω) in (15) will be replaced by D (2ωn ).

3.3. Amplitude equation

To get the non-linear frequency–amplitude relationship, one applies the one-mode Galerkin procedure, which consists in
projecting the equation (5) on U n e−iωt , the displacement being given by (16).
2π/ω
 
L (U ) + Q (U , U ) + M (Ü ), U n e−iωt dt = 0 (17)
0

The equation (17) leads to an equation for the complex amplitude in the following form:
 
a kl − ω2 m + a|a|2 knl = 0 (18)
where kl and knl are complex constants, which correspond, respectively, to the linear and non-linear modal stiffness; m is
the modal mass.
     
kl = L (U n ), U n , knl = 2Q (U n , U 0 ) + Q (U n , U 2 ), U n , m = M (U n ), U n (19)
The amplitude equation can be considered as a generic bifurcation equation, which holds for any form of the non-
linearity. It has been first derived in [24], but with a procedure that can only be applied in specific cases, as straight beams
or flat plates. When it is restricted to an elastic material, the amplitude equation (15) coincides with the parabolic approx-
imation of the backbone curve, which can be deduced, for instance, through the Poincaré–Lindstedt asymptotic procedure.
The linearized form of (15) permits to recover the results of the modal strain energy method [3], which is a classical ap-
k
proach in the analysis of viscoelastic linear structures. The ratio ml permits to define the damped linear frequency Ωl and
the linear loss factor ηl .

kl
= Ωn2 (1 + iηn ), Ωn2 = kRl /m, ηn = kIl /kRl (20)
m
where (kRl , kIl ) are, respectively, the real and imaginary parts of kl . Equation (18) establishes that the non-linear complex
frequency is a function of the amplitude |a|.
kl knl
ω2 = + |a|2 (21)
m m
JID:CRAS2B AID:3561 /FLA [m3G; v1.231; Prn:23/02/2018; 12:40] P.5 (1-12)
L. Benchouaf et al. / C. R. Mecanique ••• (••••) •••–••• 5

Fig. 2. Circular sandwich ring with two elastic external layers and a viscoelastic central one.

2
As in the modal strain energy, the non-linear modal frequency Ωnl and the non-linear modal loss factor ηnl are deduced
from the complex frequency in the same way as in the linear case.

2
  1 + C I |a|2
Ωnl = Ωn2 1 + C R |a|2 , ηnl = ηn (22)
1 + C R |a|2

where C R = kRnl /kRl and C I = kInl /kIl .

4. Forced non-linear vibrations

The analysis is limited here to harmonic excitation f (t ) = f 0 eiωt , ω being a real number corresponding to the frequency
of the excitation and f 0 its amplitude. Applying to Equation (5), the harmonic balance method and Galerkin method with
one mode, one gets:

−ω2 Ma + K l a + a|a|2 K nl = F (23)


F represents the projection of f 0 on the mode (F = υ f 0 (θ)U n (θ) dθ ).
The amplitude a is searched in the following form:

a = reiΘ (24)

where r is the real amplitude and Θ is the phase.

The solution to (23) permits to get the frequency and the phase versus the amplitude:

α ± α2 − β 1 + r2 C I
ω2 = , tg(Θ) = ηn (25)
M −(ω/Ωn )2 + 1 + r 2 C R
F
where α = | K l | cos(ϕ ) + r 2 | K nl | cos(ψ), β = | K l |2 + r 4 | K nl |2 + 2r 2 | K l K nl | cos(ϕ − ψ) − r2
, ϕ and ψ are the arguments of K l
and K nl , respectively.

5. Application

In this section, the presented approach is applied to study the in-plane free non-linear vibrations of a sandwich vis-
coelastic circular ring shown in Fig. 2. In this analysis, the rotations are assumed to be moderate, the rotary inertia terms
of the kinetic energy are neglected, and the shear deformation is taken into account for the viscoelastic layer and neglected
for the elastic ones. The displacement field is given by:

u i1 = v i + ( z − zi )βi , u i2 = w , i = 1, 3 (26)

u 21 = v + zψ, u 22 = w (27)

The continuity condition of the displacements at the interfaces between the central layer and the external ones gets:
   
hc hf hc hf 2R 1 2R 3
v 1 = α1 v + ψ− w , v 3 = α3 v − ψ+ w , α1 = , α3 = (28)
2 2R 1 2 2R 3 2R 1 − hf 2R 3 + hf
JID:CRAS2B AID:3561 /FLA [m3G; v1.231; Prn:23/02/2018; 12:40] P.6 (1-12)
6 L. Benchouaf et al. / C. R. Mecanique ••• (••••) •••–•••

The Green–Lagrange deformations in each layer i are given by:

v i + w 1 vi − w βi
γi = εi + (z − zi )ki , εi = + βi2 , βi = , ki = , i = 1, 3 (29)
R2 2 Ri Ri
v + w 1 v − w ψ w − v
γ2 = ε2 + zk2 , ε2 = + β22 , β2 = , k2 = , 2τ = +ψ (30)
R2 2 R2 R2 r2
where v and w denote respectively the radial and tangential displacements of the central layer, βi the rotation of the
cross-section relative to the layer i and () = dd()
θ.
The behavior law is given by:

N i = E i A i εi , M i = E i I i ki , i = 1, 3 (31)
∗ ∗ ∗
N 2 = A 2 E 2 ⊗ ε̇2 , M 2 = I 2 E 2 ⊗ k̇2 , T 2 = A 2 G 2 ⊗ τ̇ (32)
where N i , M i , I i , A i are respectively the normal force, the bending moment, the inertia moment, and the cross-sectional
area corresponding to the layer i (i = 1, 2, 3), T 2 is the shear transverse force relative to the layer 2. To simplify the analysis,
one assumes that the complex Young and shear modulus are constants that do not depend on the frequency:

E 2 (αω) = E 20 (1 + iη E ), G 2 (αω) = G 20 (1 + iηG ), α = 0, 2 (33)

where η E and ηG are the material loss factor in extension and shear, in this analysis, one assumes that η E = ηG = η v , E 20
and G 20 are the delayed Young and shear delayed elasticity moduli, respectively.
The motion equations describing the non-linear free vibrations are given by:
   
 M 1  M 3
−α1 N 1 + − α3 N 3 + − N 2 − T 2 + α1 N 1 β1 + α3 N 3 β3 + N 2 β2
R1 R3
+ m11 v̈ + m12 ẅ  + m13 ψ̈ = f 1 (t ) (34)
   
M  α1 hf  M 1 M  α3 hf  M 3
N 1 − 21 − N1 + + N3 − 3 + N3 + + N 2 − T 2
R1 2R 1 R1 R3 2R 3 R3
   
α1 hf α3 hf
+ 1+ ( N 1 β1 ) + 1 − ( N 3 β3 ) + ( N 2 β2 ) + m21 v̈  + m22 ẅ  + m23 ẅ + m24 ψ̈  = f 2 (t ) (35)
2R 1 2R 3

   
hc  M 1  M 3 hc
− α1 N 1 + − α3 N 3 + + R 2 T 2 − M 2 + (α1 N 1 β1 − α3 N 3 β3 )
2 R1 R3 2
 

+ m31 v̈ + m32 ẅ + m33 ψ̈ = f 3 (t ) (36)
Neglecting the non-linear parts, assuming that f (t ) = 0, and using the behavior law (22) with a real Young and shear
moduli in (33), one gets a linear real eigenvalue problem, its solution gives the linear mode un and the corresponding
eigenfrequency ωn . The details are given in Appendix A.
⎧ ⎫
⎨ v n = V cos(nθ) ⎪
⎪ ⎬
un (θ) = w n = W sin(nθ) (37)
⎪
⎩ ⎪
⎭
ψn = Ψ cos(nθ)
where n is the circumferential wave number, V , W and Ψ are arbitrary constants determined by a normalization condition;
here one assumes:

V2 + W2 +Ψ2 = 1 (38)
The obtained linear eigenvalues are in good agreement with those obtained by Patel et al. [25].
The correction term is obtained in the same way as in the general case. For the ring, the linear problems (14) and (15)
give linear differential equations (see Appendix B), whose resolution gets u 0 = ( v 0 , w 0 , ψ0 ) and u 2 = ( v 2 , w 2 , ψ2 ) in the
following form:
⎧ ⎫ ⎧
⎧ ⎫ ⎫ ⎧ ⎫ ⎧ ⎧ ⎫ ⎫
⎪
⎨v 0 ⎪
⎬ ⎨ v 01 sin(2nθ) ⎪
⎨ v 00 ⎪
⎪
⎪ ⎬ ⎬ ⎨v 2 ⎪
⎪ ⎬ ⎨ v 21 sin(2nθ) ⎪
⎪ ⎨ v 20 ⎪
⎪ ⎬ ⎬
w 0 = w 00 + w 01 cos(2nθ) , w 2 = w 20 + w 21 cos(2nθ) (39)
⎪
⎩ ⎪
⎭ ⎪
⎩ ⎪
⎭ ⎪
⎩ ⎪
⎭ ⎪
⎩ ⎪
⎭ ⎪
⎩ ⎪
⎭ ⎪
⎩ ⎪
⎭
ψ0 ψ00 ψ01 sin(2nθ) ψ2 ψ20 ψ21 sin(2nθ)
where u 0 = ( v 0 j , w 0 j , ψ0 j ) are real constants and u 2 = ( v 2 j , w 2 j , ψ2 j ) are complex ones.
Inserting (39) in the constitutive laws (14)–(15) and using (19), one gets the constants of the amplitude equation (18),
for details, see Appendix C.
JID:CRAS2B AID:3561 /FLA [m3G; v1.231; Prn:23/02/2018; 12:40] P.7 (1-12)
L. Benchouaf et al. / C. R. Mecanique ••• (••••) •••–••• 7

Table 1
Linear eigenfrequencies for circular ring with R /h = 100.

n Present Maple Patel [25] Belvins [29] [30]

2 7.1622 7.1617 7.2000 7.2000 7.2001
3 57.2977 57.2967 57.6000 57.6000 57.6001
4 210.5630 210.6513 211.7651 211.7647 211.7656

Table 2
Vibration modal coefficients versus the circumferential wave number n.

n Ωn2 ηn CR CI
2 1720.6795 1.2493 10−2 −2.9408 −0.1507
4 49637.2070 3.3344 10−3 −152.3605 −1.9052 10−2
8 911401.0331 8.4770 10−4 −3802.8828 0.3211

Fig. 3. Variation of the non-linear modal frequencies ratio (Backbone curves) versus the radial displacement near the linear frequencies Ωn (n = 2, 4, 8).
η v = 0.5.

2π 2π
kl = {l v v n + l w w n + lψ ψn } dθ, knl = {q v v n + q w w n + qψ ψn } dθ
0 0

2π
m= {m v v n + m w w n + mψ ψn } dθ (40)
0

In this application, the geometrical data are: radii R 1 = 0.9997, R 2 = 1 and R 3 = 1.003, thicknesses hc = hf /2 = 0.002
and a width b = 0.012. The structure is described by one degree of freedom and by the angle θ (0 ≤ θ ≤ 2π). In Table 1, one
gives the first linear frequencies. In Table 2, one presents the constants C R and C I for various vibration modes, the same
results are given by Maple and a Fortran program. One notes that C R is a negative number and that |C R | is greater than |C I |.
In Figs. 3 and 4, one presents the backbone curves corresponding to non-linear modal frequencies and the modal loss factors
with respect to the adimensionalized radial displacement when the ring vibrates near the linear frequencies associated with
n = 2, 4 and 8. It is clearly seen that the frequencies decrease (non-linearity of soft type) while the loss factor increases
with the displacement. The increase and decrease in frequencies and loss factor, respectively, are more important for higher
vibration modes. In Fig. 5 and Fig. 6, one gives the forced non-linear response for various excitation amplitudes and material
loss factors. The forced response parts, tangent to the non-linear free response (F = 0) and situated below it, are instable,
so the structure can jump between several equilibrium positions. In Fig. 7, the variation of the non-linear phase versus the
excitation frequency is presented for various material losses factor.

6. Conclusion

In this study, an amplitude equation has been presented for the nonlinear vibrations analysis of viscoelastic shells struc-
tures. This amplitude equation is obtained by coupling an approximated harmonic balance method with the one-mode
Galerkin procedure. It involves two modal parameters C R and C I , which account for the non-linear effects. These constants
JID:CRAS2B AID:3561 /FLA [m3G; v1.231; Prn:23/02/2018; 12:40] P.8 (1-12)
8 L. Benchouaf et al. / C. R. Mecanique ••• (••••) •••–•••

Fig. 4. Variation of the non-linear modal loss factor ratio versus the radial displacement near the linear frequencies Ωn (n = 2, 4, 8) for various wave
circumferential numbers n. η v = 0.5.

Fig. 5. Variation of the non-linear response with the load amplitude (n = 2, η v = 0.5).

Fig. 6. Variation of the non-linear response with the loss factor (n = 2, F = 5).
JID:CRAS2B AID:3561 /FLA [m3G; v1.231; Prn:23/02/2018; 12:40] P.9 (1-12)
L. Benchouaf et al. / C. R. Mecanique ••• (••••) •••–••• 9

Fig. 7. Variation of the phase versus the excitation frequency for various loss factors η v (F = 5).

are determined by solving three classical problems. The first one is a real eigenvalue problem that allows one to define the
linear frequency and the linear loss factor. The two others are linear problems. In the case of free vibrations, the backbone
corresponding to the non-linear modal frequency and the modal non-linear loss factor, with respect to the displacement,
are obtained. The non-linear forced response and the corresponding non-linear phase are also determined. This approach
will be extended to harmonic forcing vibration problems and to others viscoelastic sandwich or composite shell structures
such as the cylinder.

Appendix A. Computation of linear vibration modes

The linear part of (37) gets:

i
L 11 v n + L 12
i
w n + L 13
i
ψn = M 11 v̈ n + M 12 ẅ n + M 13 ψ̈n
i
L 21 v n + L 22
i
w n + L 23
i
ψn = M 21 v̈ n + M 22 ẅ n + M 24 ψ̈n (41)
i
L 31 vn + i
L 32 wn + i
L 33 ψn = M 31 v̈ n + M 32 ẅ n + M 33 ψ̈n
with the following operators:

i i d2 i i i d3 i d i i d2 i i i d3 i d
L 11 = A 11 + A 12 , L 12 = A 13 + A 14 , L 13 = A 15 + A 16 , L 21 = A 21 + A 22
dθ 2 dθ 3 dθ dθ 2 dθ 3 dθ
i i d4 i d2 i i i d3 i d i i d2 i
L 22 = A 23 + A 24 + A 25 , L 23 = A 26 + A 27 , L 31 = A 31 + A 32
dθ 4 dθ 2 dθ 3 dθ dθ 2
i i d3 i d i i d2 i d
L 32 = A 33 + A 34 , L 33 = A 35 + A 36 , M 11 = m11 , M 12 = m12 , M 13 = m13
dθ 3 dθ dθ 2 dθ
d d2 d d
M 21 = m21 , M 22 = m22 + m23 , M 23 = m24 , M 31 = m31 , M 32 = m32 , M 33 = m33
dθ dθ 2 dθ dθ
The constants A ijk (i = 0, 2, j = 1, 2, 3, and k = 1, . . . , 7) are given by:
   
i α12 E 1 I1 α32 E 3 I1 E 2 (i j ω ) A 2 i k  G (i j ω ) A 2
A 11 = A1 + + A3 + + , A 12 =−
R1 R 21 R3 R 23 R1 R2

 
 
i α1 E 1 α1 A 1 hf I1 α1 hf α3 E 3 α3 A 3 hf I3 α3 hf
A 13 =− + 1+ + − 1−
R 21 2 R1 2R 1 R 23 2 R3 2R 3

   
i α1 E 1 A 1 α3 E 3 A 3 E (i j ω ) A 2 k  G 2 (i j ω ) A 2 i hc α12 E 1 I1 α32 E 3 I1
A 14 = + + + , A 15 = A1 + − A3 +
R1 R3 R2 R2 2 R1 R 21 R3 R 23

 
 
α1 E 1 α1 A 1 hf I1 α1 hf α3 E 3 α3 A 3 hf I3 α3 hf
i
A 16 = k A 2 G 2 (i j ω), i
A 21 = + 1+ − − 1−
R 21 2 R1 2R 1 R 23 2 R3 2R 3

 2 
 2 
i i i E1 α12 A 1 h2f I1 α1 hf E3 α32 A 3 h2f I3 α3 hf
A 22 = − A 14 , A 23 =− + 1+ − + 1−
R 31 4 R1 2R 1 R 33 4 R3 2R 3
JID:CRAS2B AID:3561 /FLA [m3G; v1.231; Prn:23/02/2018; 12:40] P.10 (1-12)
10 L. Benchouaf et al. / C. R. Mecanique ••• (••••) •••–•••

   
i α1 E 1 A 1 α3 E 3 A 3 k  A 2 G 2 (i j ω ) i E 1 A1 E 2 (i j ω ) A 2 E 3 A3
A 24 = hf − + , A 25 =− + +
R 21 R 23 R2 R1 R2 R3

 
 
i hc α1 E 1 α1 A 1 hf I1 α1 hf α3 E 3 α3 A 3 hf I3 α3 hf
A 26 = + 1+ + − 1−
2 R 21 2 R1 2R 1 R 23 2 R3 2R 3
 
i hc α1 E 1 A 1 α3 E 3 A 3 k G 2 (i j ω) A 2 i i i i i i
A 27 =− − + , A 31 = A 15 , A 32 = A 16 , A 33 = − A 26
2 R1 R3 R2

   
i i i h2c α12 E 1 I1 α32 E 3 I1 E 2 (i j ω ) I 2 i i
A 34 = − A 27 , A 35 = A1 + + A3 + + , A 36 = − R 2 A 32
4 R1 R 21 R3 R 23 R2
hf  
m11 = ρ1 A 1 R 1 α12 + ρ2 A 2 R 2 + ρ3 A 3 R 3 α32 , m12 = − ρ1 A 1 α12 − ρ3 A 3 α32 = −m21
2
 
hc  2 2
 h2f ρ1 A 1 α12 ρ3 A 3 α32
m13 = ρ1 A1 R 1 1 α − ρ3 A 3 R 3 α3 = m31 , m22 = − +
2 4 R1 R3
hc hf  
m23 = ρ1 A 1 R 1 + ρ2 A 2 R 2 + ρ3 A 3 R 3 , m24 = ρ1 A 1 α12 + ρ3 A 3 α3 = −m32 2
4
h2c  
m11 = ρ1 A 1 R 1 α12 + ρ3 A 3 R 3 α32
4
The general solution is given by:
⎧ ⎫
⎨ v n = V cos(nθ) ⎪
⎪ ⎬
un (θ) = w n = W sin(nθ) eiωn t (42)
⎪
⎩ ⎪
⎭
ψn = Ψ cos(nθ)
where ωn is a real number corresponding to the real frequency.
Injecting (42) in (41), one gets a real linear eigenvalue problem allowing one to have the real linear mode and the
associated linear frequencies.
⎡ ⎤⎧ ⎫
n2 A 011 − A 012 n3 A 013 − n A 014 ⎪
⎨V n2 A 015 − A 016 ⎪
⎬
⎢ 3 0 0 ⎥
⎣ −n A 21 + n A 022 −n A 23 + n A 24 − A 25 −n A 26 + n A 27 ⎦ W
4 0 2 0 0 3 0
⎪
⎩ ⎪
⎭
n2 A 031 − A 032 n3 A 033 − n A 034 n2 A 035 − A 036 Ψ
⎡ ⎤⎧ ⎫
m11 nm12 m13 ⎪
⎨V ⎪ ⎬
⎢ ⎥
= ωn2 ⎣ −nm21 −n2m22 + m23 −nm24 ⎦ W (43)
⎪
⎩ ⎪
⎭
m31 nm32 m33 Ψ

Appendix B. Computation of U 2 and U 0

In the case of the ring, equation (15) gives:

   


M 12  M 
α1 N 12 + + α3 N 32 + 32 + N 22 + T 22 + 4ω02 (m11 v 2 + m12 w 2 + m13 ψ2 )
R1 R3
1
= α1 ( N 1m β1m ) + α3 ( N 3m β3m ) + N 2m β2m
2    

M 12  
M 32 
M 32
α1 hf  M 12 α3 hf  
− N 12 + + N 12 + − N 32 + − N 32 + − N 22 + T 22
R 21 2R 1 R1 R3 2R 3 R3
 
+ 4ω02 m21 v 2 + m22 w 2 + m23 w 2 + m24 ψ2

    
1 α1 hf α3 hf
= 1+ ( N 1m β1m ) + 1 − ( N 3m β3m ) + ( N 2m β2m )
2 2R 1 2R 3

    
hc  M 12  M 
 
α1 N 12 + − α3 N 32 + 32 − R 2 T 22 + M 22 + 4ω02 m31 v 2 + m32 w 2 + m33 ψ2
2 R1 R3
hc
= (α1 N 1m β1m − α3 N 3m β3m ) (44)
4
The corresponding behavior law are given by:
JID:CRAS2B AID:3561 /FLA [m3G; v1.231; Prn:23/02/2018; 12:40] P.11 (1-12)
L. Benchouaf et al. / C. R. Mecanique ••• (••••) •••–••• 11


   
1 hf β2
hc
N 12 = E 1 S 1 ψ2 − α1 v 2 +
w 2 + w 2 + 1m
R1 2 2R 1 4

    
    
E1 I1  h  α h  1 hc α3 hf
w 2
c 1 f
M 12 = 2
α1 v 2 + ψ2 − 1 + w2 , β32 = α3 v 2 − ψ2 − 1 −
R1 2 2R 1 R3 2 2R 3

   2 
1 hc hf β3m
N 32 = E 3 A 3 ψ2 + α3 v 2 −
w  + w 2 +
R3 2R 3 2 42

     (45)
E3 I3 hc α3 hf v 2 − w 2
M 32 = 2
α3 v 2 − ψ2 − 1 − w 2 , β22 =
R3 2 2R 3 R2
 2 
v2 + w2 β2m
N 22 = E 2 (2iω) A 2 +
R2 4
 
E 2 (2iω) I 2 w 2 − v 2
M 22 = ψ2 , T 22 = k G 2 (2iω) A 2 + ψ2
R2 R2

Injecting the last behavior law in the equation, one gets the following linear complex system:
⎡ ⎤
−4n2 A 211 + A 212 + 4ωn2m11 8n3 A 213 − 2n A 214 − 8nωn2 m12 −4n2 A 215 + A 216 + 4ωn2m13
⎢ ! ⎥
⎢ 16n4 A 223 − 4n2 A 224 + A 225 + ⎥
⎢ −8n3 A 2 + 2n A 2 + 8nω2m21 −8n3 A 226 + 2n A 227 + 8nωn2 m24 ⎥
⎢ 21 22 n
4ωn2 (−4n2 m22 + m23 ) ⎥
⎣ ⎦
−4n2 A 231 + A 232 + 4ωn2m31 8n3 A 233 − 2n A 234 − 8nωn2 m32 −4n2 A 235 + A 236 + 4ωn2m33
⎧ ⎫ ⎧ 2 ⎫
⎨ v 21 ⎪
⎪ ⎬ ⎪ ⎨ c 11 ⎪ ⎬
× w 21 = c 22 2
(46)
⎪
⎩ ⎪
⎭ ⎪ ⎩ 2 ⎪ ⎭
ψ21 c 31

U 0 is obtained as a particular case from U 2 by putting ωn = 0 and using real Young and shear moduli in the behavior
corresponding to layer 2.

Appendix C

     
 M 1n 
 M 3n
l v = −α1 N 1n + − α3 N 3n + − N 2n + T 2n
R1 R3
   

M 1n α1 hf  M  M  α3 hf  M  
l w = N 1n − − N 1n + 1n + N 3n − 3n + N 3n + 3n − N 2n − T 2n
R 21 2R 1 R1 R3 2R 3 R3

     
hc  M 1n  M 3n 
lψ = − α1 N 1n + − α3 N 3n + + R 2 T 2n − M 2n
2 R1 R3
m v = m11 v n + m12 w n + m13 ψn
m w = m21 v n + m22 w n + m23 w n + m24 ψn
mψ = m31 v n + m32 w n + m33 ψn

   
   
1 1 1 1
q v = α1 N 1n β10 + β12 + β1n N 10 + N 12 + α3 N 3n β30 + β32 + β3n N 30 + N 32 (47)
 2   2  2 2
1 1
+ N 2n β20 + β22 + β2n N 20 + N 22
2 2
 
   
α1 hf   1    1 
qw = 1 + N 1n β10 + β12 + β1n N 10 + N 12
2R 1 2 2
 
   
α3 hf   1    1 
+ 1− N 3n β30 + β32 + β3n N 30 + N 32
2R 3 2 2
   
  1    1 
+ N 2n β20 + β22 + β2n N 20 + N 22
2 2

   
   
hc 1 1 1 1
qψ = α1 N 1n β10 + β12 + β1n N 10 + N 12 − N 3n β30 + β32 + β3n N 30 + N 32
2 2 2 2 2
JID:CRAS2B AID:3561 /FLA [m3G; v1.231; Prn:23/02/2018; 12:40] P.12 (1-12)
12 L. Benchouaf et al. / C. R. Mecanique ••• (••••) •••–•••

References

[1] R.A. Ditaranto, W. Blasingame, Composite damping of vibrating sandwich beams, J. Eng. Ind. 89 (4) (1967) 633–638.
[2] D.J. Mead, S. Markus, The forced vibration of three-layer damped sandwich beam with arbitrary boundary conditions, J. Sound Vib. 10 (1969) 163–175.
[3] M.J. Yan, E.H. Dowell, Governing equations for vibrating constrained-layer damping sandwich plates and beams, J. Appl. Mech. 94 (1972) 1041–1047.
[4] V. Oravsky, S. Markus, O. Simkova, New approximate method of finding the loss factors of a sandwich cantilever, J. Sound Vib. 33 (1974) 335–352.
[5] D.K. Rao, Frequency and loss factors of sandwich beams under various boundary conditions, J. Mech. Eng. Sci. 20 (5) (1978) 271–282.
[6] E.A. Sadek, Dynamic optimisation of a sandwich beam, Comput. Struct. 19 (4) (1984) 605–615.
[7] P. Cupial, J. Niziol, Vibration and damping analysis of three-layered composite plate with viscoelastic mid-layer, J. Sound Vib. 183 (1) (1995) 99–114.
[8] J.F. He, B.A. Ma, Vibration analysis of viscoelastically damped sandwich shells, Shock Vib. Bull. 3 (6) (1996) 403–417.
[9] Y.C. Hu, S.C. Huang, The frequency response and damping effect of three-layer thin shell with viscoelastic core, Comput. Struct. 76 (2000) 577–591.
[10] B. Kovâcs, Vibration analysis of a damped arch using iterative laminate model, J. Sound Vib. 254 (2) (2002) 367–378.
[11] M.L. Soni, Finite element analysis of viscoelastically damped sandwich structures, Shock Vib. Bull. 55 (1) (1981) 97–109.
[12] B.A. Ma, J.F. He, A finite element analysis of viscoelastically damped sandwich plates, J. Sound Vib. 152 (1992) 107–123.
[13] R. Rikards, A. Chate, E. Barkanov, Finite element analysis of damping the vibrations of laminated composites, Comput. Struct. 47 (6) (1993) 1005–1015.
[14] C.D. Johnson, D.A. Kienholz, L.C. Rogers, Finite element prediction of damping in beams with constrained viscoelastic layer, Shock Vib. Bull. 51 (1)
(1981) 71–81.
[15] Y.P. Lu, J.W. Killian, G.C. Everstine, Vibrations of three layered damped sandwich plate composites, J. Sound Vib. 64 (1) (1979) 63–71.
[16] M.G. Sainsbury, Q.J. Zhang, The Galerkin element method applied to the vibration of damped sandwich beams, Comput. Struct. 71 (1999) 239–256.
[17] T.C. Ramesh, N. Ganesan, Finite element analysis of conical shells with a constrained viscoelastic layer, J. Sound Vib. 171 (5) (1994) 577–601.
[18] T.T. Baber, R.A. Maddox, C.E. Orozco, A finite element model for harmonically excited viscoelastic sandwich beams, Comput. Struct. 66 (1) (1998)
105–113.
[19] N. Alam, N.T. Asnani, Vibration and damping of multi layered cylindrical shell. Part I and II, AIAA J. 22 (1984) 803–810, 975–981.
[20] E.M. Daya, M. Potier-Ferry, A numerical method for nonlinear eigenvalue problem, application to vibrations of viscoelastic structures, Comput. Struct.
79 (5) (2001) 533–541.
[21] X. Chen, H.L. Chen, H. Le, Damping prediction of sandwich structures by order-reduction-iteration approach, J. Sound Vib. 222 (5) (1999) 803–812.
[22] E.M. Daya, M. Potier-Ferry, A shell finite element for viscoelastically damped sandwich structures, Rev. Eur. Elém. Fin. 11 (1) (2002) 39–56.
[23] H.-H. Lee, Non-linear vibration of a multilayer sandwich beam with viscoelastic layers, J. Sound Vib. 216 (4) (1998) 601–621.
[24] E.M. Daya, L. Azrar, M. Potier-Ferry, An amplitude equation for the non-linear vibration of viscoelastically damped sandwich beams, J. Sound Vib.
271 (3) (2003) 789–813.
[25] B.P. Patel, M. Ganapathi, D.P. Makhecha, P. Shah, Large amplitude free flexural vibration of rings using finite element approach, Int. J. Non-Linear Mech.
37 (2003) 911–921.
[26] M. Ganapathi, B.P. Patel, P. Boisse, O. Polit, Flexural loss factors of sandwich and laminated beams using linear and nonlinear dynamic analysis,
Composites, Part B, Eng. 30 (1999) 245–256.
[27] E.H. Boutyour, E.M. Daya, M. Potier-Ferry, A harmonic balance method for the non-linear vibration of viscoelastic shell, C. R. Mecanique 334 (2006)
68–73.
[28] F. Boumediene, J.M. Cadou, L. Duigou, E.M. Daya, A reduction model for eigensolutions of damped viscoelastic sandwich structures, Mech. Res. Commun.
57 (2014) 74–81.
[29] F. Boumediene, E.M. Daya, J.M. Cadou, L. Duigou, Forced harmonic response of viscoelastic sandwich beams by a reduction method, Mech. Adv. Mat.
Struct. 23 (11) (2016) 1290–1299.
[30] M. Amabili, Nonlinear vibrations of viscoelastic rectangular plates, J. Sound Vib. 362 (3) (2016) 142–156.
[31] K.G. Lougou, H. Boudaoud, E.M. Daya, L. Azrar, Vibration modeling of large repetitive sandwich structures with viscoelastic core, Mech. Adv. Mat. Struct.
23 (4) (2016) 458–466.
[32] K. Lampoh, I. Charpentier, E.M. Daya, Erratum to “Eigenmode sensitivity of damped sandwich structures” [C. R. Mecanique 342 (2014) 700–705], C. R.
Mecanique 343 (3) (2015) 246.
[33] I. Elkhaldi, I. Charpentier, E.M. Daya, A gradient method for viscoelastic behaviour identification of damped sandwich structures, C. R. Mecanique
340 (8) (2012) 619–623.
[34] A.H. Nayfeh, D.T. Mook, Nonlinear Oscillations, Wiley, New York, 1979.