DOI 10.1007/s00158-009-0390-4
RESEARCH PAPER
Received: 30 October 2008 / Revised: 16 March 2009 / Accepted: 7 April 2009 / Published online: 6 May 2009
Springer-Verlag 2009
Abstract Recent developments on the optimization of applications in sandwich structures are presented and
passive damping for vibration reduction in sandwich discussed.
structures are presented in this paper, showing the
importance of appropriate finite element models as- Keywords Sandwich structures Layerwise finite
sociated with gradient based optimizers for compu- element model Passive damping Multiobjective
tationally efficient damping maximization programs. optimization Gradient optimization
A new finite element model for anisotropic lami- Genetic algorithms
nated plate structures with viscoelastic core and lam-
inated anisotropic face layers has been formulated,
using a mixed layerwise approach. The complex mod- 1 Introduction
ulus approach is used for the viscoelastic material
behavior, and the dynamic problem is solved in the Structural damping can be defined as the process by
frequency domain. Constrained optimization is con- which a structure or structural component dissipates
ducted for the maximization of modal loss factors, using mechanical energy or transfers it to connected struc-
gradient based optimization associated with the devel- tures or ambient media. These mechanisms have the
oped model, and single and multiobjective optimiza- effect of controlling the amplitude of resonant vibra-
tion based on genetic algorithms using an alternative tions and modifying wave attenuation and sound trans-
ABAQUS finite element model. The model has been mission properties, increasing structural life through
applied successfully and comparative optimal design reduction in structural fatigue.
Passive damping treatments are widely used in en-
gineering applications in order to reduce vibration and
noise radiation (Nashif et al. 1985; Sun and Lu 1995).
Some parts of this paper were presented at EngOpt 2008,
International Conference on Engineering Optimization, held Passive layer damping, usually implemented as con-
at Rio de Janeiro, Brazil. strained layer damping, is the most common form of
damping treatment, where the damping layer deforms
A. L. Arajo (B)
ESTIG, Polytechnic Institute of Bragana, Campus de Sta. in shear, thus dissipating energy in a more efficient
Apolnia, Apartado 1134, 5301-857 Bragana, Portugal way.
e-mail: aaraujo@ipb.pt The theoretical work on constrained layer damp-
ing can be traced to DiTaranto (1965) and Mead and
P. Martins C. M. Mota Soares C. A. Mota Soares
IDMEC/IST, Technical University of Lisbon, Markus (1969) for the axial and bending vibration
Av. Rovisco Pais, 1049-001 Lisbon, Portugal of sandwich beams. Since then, different formulations
and techniques have been reported for modeling and
J. Herskovits
COPPE, Federal University of Rio de Janeiro,
predicting the energy dissipation of the viscoelastic
Caixa Postal 68503, 21945-970, core layer in a vibrating passive constrained layer
Rio de Janeiro, Brazil damping structure (Rao 1978; Yan and Dowell 1972;
570 A.L. Arajo et al.
Rao and He 1993). Other proposed formulations in- In this work, maximization of modal loss factors
clude thickness deformation of the core layer dealing of sandwich plates with elastic laminated constraining
with the cases where only a portion of the base structure layers and a viscoelastic core is conducted with layer
receives treatment (Douglas and Yang 1978). thicknesses and laminate layer ply orientation angles
Sandwich plates with viscoelastic core are very ef- as design variables. The problem is solved through gra-
fective in reducing and controlling vibration response dient based optimization using a new mixed layerwise
of lightweight and flexible structures, such as those finite element model, developed by the authors for
frequently encountered in aerospace industries, where the analysis and optimization of laminated sandwich
the soft core is strongly deformed in shear, due to the plates (Arajo et al. 2009). To the authors knowl-
adjacent stiff layers. Due to this high shear developed edge, few plate/shell sandwich models with viscoelastic
inside the core, equivalent single layer plate theories, core exist, and are mostly limited to isotropic materi-
even those based on higher order deformations, are not als (Boudaoud et al. 2008). In the paper the authors
adequate to describe the behavior of these sandwiches, show that with such appropriate models associated to
mainly due to the high deformation discontinuities that gradient based optimization, computationally efficient
arise at the interfaces between the viscoelastic core ma- strategies for passive damping maximization can be ac-
terial and the surrounding elastic constraining layers. complished. An alternative approach is also presented,
The usual approach to analyze the dynamic response which associates ABAQUS commercial finite element
of sandwich plates uses a layered scheme of plate and models with an implementation of genetic algorithms
brick elements with nodal linkage, leading to a time for single and multiobjective optimization.
consuming spatial modeling task. To overcome these
difficulties, layerwise theories have been considered
for constrained viscoelastic treatments, and most re- 2 Mixed layerwise sandwich finite element model
cently (Moreira et al. 2006; Moreira and Rodrigues
2006), among others, presented generalized layerwise The development of the new layerwise finite element
formulations in this scope. model (Arajo et al. 2009) is briefly presented here, to
Optimal design of constrained layer damping treat- analyze sandwich laminated plates with a viscoelastic
ments of vibrating structures has been a main subject of (v) core and laminated anisotropic face layers (e1 , e2 ),
research, aiming at the maximization of modal damping and top and bottom piezoelectric sensor (s) and actua-
ratios and modal strain energies, by determining the tor (a) layers, as shown in Fig. 1.
optimal material and geometric parameters of the treat- Although sensor and actuator layers are included in
ments, or minimizing weight by selecting their optimal the formulation presented here, the piezoelectric effect
length and location. For example, (Baz and Ro 1995) will not be considered in this work, as it deals only with
optimized performance of constrained layer damping passive damping.
treatments by selecting the optimal thickness and shear The basic assumptions in the development of the
modulus of the viscoelastic layer, and (Marcelin et al. sandwich plate model are:
1992, 1995) used a genetic algorithm and beam finite
elements to maximize the damping factor for partially 1. All points on a normal to the plate have the same
treated beams, using as design variables the dimensions transverse displacement w(x, y, t), where t denotes
and locations of the patches. As verified by Nokes time, and the origin of the z axis is the medium
and Nelson (1968), this layout optimization can lead plane of the core layer;
to significant saving in the amount of material used. 2. No slip occurs at the interfaces between layers;
Recently, optimal number, position and shape of sur-
face passive constrained layer damping treatments has
been addressed (Alvelid 2008) in automotive indus-
trial applications. For fully covered sandwich beams,
(Lifshitz and Leibowitz 1987) determined the optimal
passive constrained layer damping, with layer thick-
nesses as design variables. The vibration damping of
fully covered constrained layer damping structures is
determined by a large number of parameters which
include material properties and thicknesses of both the
constraining layers and the viscoelastic layer (Arajo
et al. 2008). Fig. 1 Sandwich plate
Damping optimization of viscoelastic laminated sandwich. . . 571
3. The displacement is C0 along the interfaces; The displacement continuity at the layer interfaces
4. Elastic layers (a, e2 , e1 , s) are modeled with first can be written as:
order shear deformation theory (FSDT) and vis-
hs he
coelastic core (v) with a higher order shear defor- us x, y, zs , t = ue1 x, y, ze1 + 1 , t
2 2
mation theory (HSDT);
5. All materials are linear, homogeneous and or- hs he1
v x, y, zs , t = v
s e1
x, y, ze1 + ,t
thotropic and the elastic layers (e1 ) and (e2 ) are 2 2
made of laminated composite materials;
he1 v hv
6. For the viscoelastic core, material properties are u e1
x, y, ze1 , t = u x, y, , t
2 2
complex and frequency dependent.
he hv
v e1 x, y, ze1 1 , t = v v x, y, , t
2 2
2.1 Displacement fields (3)
v hv he2
u x, y, , t = u e2
x, y, ze2 + ,t
The FSDT displacement field of the face layers may be 2 2
written in the general form: v hv he2
v x, y, , t = v e2
x, y, ze2 + ,t
2 2
ui (x, y, z, t) = ui0 (x, y, t) + (z zi )xi (x, y, t)
he2 ha
u e2
x, y, ze2 , t = u x, y, za + , t
a
v i (x, y, z, t) = v0i (x, y, t) + (z zi ) yi (x, y, t) (1) 2 2
he ha
wi (x, y, z, t) = w0 (x, y, t) v e2 x, y, ze2 2 , t = v a x, y, za + , t
2 2
where ui0 and v0i are the in-plane displacements of the where the coordinates of layer mid-planes are:
mid-plane of the layer, xi and yi are rotations of nor- hv hs
mals to the mid-plane about the y axis (anticlockwise) zs = + he1 +
2 2
and x axis (clockwise), respectively, w0 is the transverse
hv he1
displacement of the layer (same for all layers in the ze1 = +
2 2
sandwich), zi is the z coordinate of the mid-plane of
each layer, with reference to the core layer mid-plane zv = 0 (4)
(z = 0), and i = e1 , e2 is the layer index. hv he
For the viscoelastic core layer, the HSDT displace- ze2 = 2
2 2
ment field is written as a second order Taylor series ex-
hv ha
pansion of the in-plane displacements in the thickness za = he2
coordinate, with constant transverse displacement: 2 2
and ha , he2 , hv , he1 and hs are the thicknesses of each
v
u (x, y, z, t) =uv0 (x, y, t) + zxv (x, y, t) layer, as shown in Fig. 1.
Applying displacement continuity conditions at the
+ z2 u0 v (x, y, t) + z3 x v (x, y, t) layer interfaces, one obtains:
of interest, respectively, and is the convergence respectively, and FT H is the Tsai-Hill failure criteria
tolerance. parameter for the elastic composite material layers,
defined as:
2 2 2
11 22 33
3 Optimal design formulations FT H = + +
X Y Z
The objective of this study is to maximize damping in 1 1 1
+ 2 2 11 22
sandwich plate structures with respect to layer thick- X2 Y Z
nesses and fiber orientation angles. If the structure is
1 1 1
subjected to a given load or load set, this maximization + 22 33
Y2 Z2 X2
of damping must be conducted with design constraints
on maximum displacement, total mass, failure criteria, 1 1 1
+ 11 33
as well as physical constraints on design variables and Z2 X2 Y2
objective function. 2 2 2
23 13 12
Two different approaches are used, namely: the + + + <1 (15)
R S T
Layerwise/FAIPA model, associating the finite element
model described in the previous section with a gradient where the stress components are calculated for each
based optimization algorithm and continuous design elastic layer ply and refer to the principal material
variables, and the ABAQUS/GA model which uses 3D directions of the ply, X, Y and Z are lamina failure
finite elements from commercial software (ABAQUS stresses in the associated principal directions, which
2005) associated with a specifically developed imple- must respect the sign of the stresses, and one must con-
mentation of a genetic algorithm and discrete design sider different values in traction and compression. R,
variables. Both approaches are described next. S and T are failure stresses in shear for the associated
planes in (15). It should be mentioned here that the
3.1 Layerwise/FAIPA approach transverse normal stress component 33 is assumed to
be zero for the Layerwise/FAIPA approach.
For damping maximization with passive treatments in Assuming a uniform sandwich plate structure made
sandwich type structures, the overall goal will be to of a given set of materials, with fixed in-plane dimen-
maximize the modal loss factor of a particular mode sions, the natural choice for the design variables xi in
of interest, or of a particular set of modes of interest (14) are the thicknesses of the constituent layer plies
within some frequency range. Thus, a weighted sum and the orientation angles of the laminated elastic com-
of reciprocal loss factors was chosen as the objective posite material plies. In (14), xli and xiu are the lower and
function to be minimized in this framework, subjected upper bounds on the design variables.
to design constraints: Calculation of the objective function is done by
solving the eigenvalue problem of (11) iteratively, for
N
1 a frequency dependent complex stiffness matrix and
min f = wk
xi k real mass matrix. The derivatives of the objective func-
k=1
tion with respect to the design variables are calculated
s.t. g j : j 0, j = 1, . . . , N through the following expression:
m
g N+1 : 10
mmax 1 wn n
wn = 2 (16)
w xi n n xi
g N+2 : 10
wmax
where:
g N+3 : FT H 1 0
xli xi xiu , i = 1, . . . , n (14) n 1 n n
= n (17)
where wk are weighting factors associated with each xi n xi xi
modal loss factor k , N is the total number of modes of
interest, m and mmax are the overall mass and maximum with n and n being the real and imaginary
allowable mass of the structure, respectively, w and parts of the complex eigenvalue n , respectively.
wmax are the maximum displacement of the structure The derivative of the complex eigenvalue with re-
and the maximum allowable value of the displacement, spect to design variables, for the particular case of
574 A.L. Arajo et al.
hysteretic damping, can be obtained using the following where the 1.1 factor in penalty functions is intended to
expression (Haftka and Gurdal 1992): keep solutions feasible. For this reason the associated
constraints might never be active.
K
n unT xi
n M
xi
un Design variables are again the layer thicknesses of
= (18) viscoelastic core and elastic plies and fiber orientation
xi unT Mun
angles of the elastic plies, but they are discrete in
where the derivatives of the stiffness and mass matrix this approach, instead of continuous as in Layerwise/
are evaluated by forward finite difference at the ele- FAIPA.
ment level. Multiobjective function optimization was also con-
Calculation of response quantities such as displace- ducted, where the objective was to maximize modal loss
ments and stresses is done after the eigenvalue problem factors, while simultaneously minimizing the weight of
has been solved, and is conducted in the frequency the structure. In this case the two conflicting objectives
domain, by first making a forward Fourier transform of are:
the applied load time history, and then solving (10) in
N
1
order to the displacement vector, for the resonant fre- f1 = wi P1 P2
i=1
i (21)
quency of interest. Afterwards, stresses in each elastic
material layer ply are calculated and the Tsai-Hill factor f2 = mP1 P2
FT H in (15) is evaluated.
As for sensitivities of displacement and stress However, in this case of conflicting objectives, there
quantities, they can be calculated analytically, semi- is no unique optimal solution, and one obtains a series
analytically, or using a global finite difference ap- of non dominant solutions (Pareto frontier) for which
proach (Haftka and Gurdal 1992). no decrease can be obtained in any of the objectives
In this approach, the problem is solved using FAIPA, without simultaneously increasing at least one of the
the Feasible Arc Interior Point Algorithm (Herskovits remaining objectives.
1986, 1998; Herskovits et al. 2005), along with the A Genetic Algorithm (Goldberg 1989) with binary
developed finite element sandwich model. FAIPA has encoding is used to solve the problem (Yang et al.
proven to be robust and efficient in a wide variety of 1998). The algorithm initializes a random sample of
applications (Herskovits and Mazorche 2009; Canelas individuals with different parameters to be optimist
et al. 2009). using evolution via survival of the fittest. The selection
scheme used is tournament selection with a shuffling
technique for choosing random pairs for mating.
3.2 ABAQUS/GA approach
layers will have equal thickness he1 = he2 , considered For the layerwise model, a 6 1 finite element model
as a design variable. As for the viscoelastic core, its was used, with a total of 543 degrees of freedom. As for
thickness hv will also be a design variable. Thickness the ABAQUS model, each lamina was discretised with
design variables are allowed to vary between 0.5 mm 16 elements (8 along the length, 2 along the width, and
and 10 mm, and for the ABAQUS/GA approach in 1 along the thickness), with a total of 80 elements and
increments of 0.5 mm. The orientation angles of the 1627 degrees of freedom. For the GA, the population
composite layers are considered to be fixed at 0 w.r.t. size was 25, the mutation probability 0.03, and a total
the longitudinal axis. of 100 and 150 evaluations where made for single and
The fundamental flexural modal loss factor of the multiobjective function approaches, respectively.
beam will be maximized, with a maximum allow- Results are presented in Table 1, where the initial
able mass mmax = 0.005 kg and a maximum allowable or intermediate and final designs are presented for
displacement wmax = h/5, where h = ha + he1 + hv + both approaches, using a single objective function. For
he2 + hs is the total thickness of the beam. The failure the final design the only active constraint is the mass
stresses in the Tsai-Hill expression were considered constraint. Figure 2 shows the magnitude of the fre-
to be, for the elastic layers, X = 820 MPa, Y = Z = quency response for the initial and final sandwich beam
45 MPa, both in tension and compression, and R = S = designs, and Fig. 3 shows the decrease of the objective
T = 45 MPa. The excitation consisted of a 1 N force function with iteration number.
applied at the mid-point of the beam at t = 0. As for the multiobjective optimization (ABAQUS/
GA approach), results for the Pareto front solution
and 175 , with increments of 15 for the ABAQUS/GA with 24 elements (6 along the length, 4 along the width,
approach. and 1 along the thickness), with a total of 216 elements
The fundamental flexural modal loss factor of the and 3636 degrees of freedom. For the GA, the popu-
beam will be maximized, with a maximum allowable lation size was 10 individuals, the mutation probability
mass mmax = 0.5 kg and a maximum allowable displace- 0.01, and a total of 500 evaluations were made for both
ment wmax = h/5, where h = ha + he1 + hv + he2 + hs is single and multiobjective optimization.
the total thickness of the plate. The failure stresses Results are presented in Table 3, where the initial,
in (15) are, for the elastic layers, X = 820 MPa, Y = intermediate and final designs are presented for both
Z = 45 MPa, both in tension and compression, and approaches, using a single objective function. For the
R = S = T = 45 MPa. The excitation consisted of a final design, the only active constraints are the mass
10 N force applied at the mid-point of the plate at t = 0. constraint, and the lower bounds on elastic layer thick-
For the layerwise model, a 6 4 finite element mesh nesses. Figure 6 shows the magnitude of the frequency
was used, with a total of 1505 degrees of freedom. As response for the initial and final sandwich beam de-
for the ABAQUS model, each lamina was discretised signs, and Fig. 7 shows the decrease of the objective
function with iteration number.
5 Conclusions
of mechanical and piezoelectric properties through gradi- Herskovits J, Mappa P, Goulart E, Mota Soares CM (2005) Math-
ent optimisation and experimental vibration data. Compos ematical programming models and algorithms for engineer-
Struct 58:307318 ing design optimization. Comput Methods Appl Mech Eng
Arajo AL, Lopes HMR, Vaz MAP, Mota Soares CM, 194:32443268
Herskovits J, Pedersen P (2006) Parameter estimation in Lifshitz JM, Leibowitz M (1987) Optimal sandwich beam de-
active plate structures. Comput Struct 84:14711479 sign for maximum viscoelastic damping. Int J Solids Struct
Arajo AL, Mota Soares CM, Mota Soares CA (2008) Optimal 23:10271034
design of active, passive, and hybrid sandwich structures. Marcelin JL, Trompette P, Smati A (1992) Optimal constrained
In: Lindner DK (ed) Modeling, signal processing, and con- layer damping with partial coverage. Finite Elem Anal Des
trol for smart structures, proc. of SPIE, vol 6926. SPIE, 12:273280
Bellingham, p 69260T Marcelin JL, Shakhesi S, Pourroy F (1995) Optimal constrained
Arajo AL, Mota Soares CM, Mota Soares CA (2009) Finite layer damping of beams: experimental numerical studies.
element model for hybrid active-passive damping analysis of Shock Vib 2:445450
anisotropic laminated sandwich structures. J Sandw Struct Mead DJ, Markus S (1969) The forced vibration of a three-layer,
Mater doi:10.1177/1099636208104534 damped sandwich beam with arbitrary boundary conditions.
Baz A, Ro J (1995) Optimum design and control of active con- AIAA J 10:163175
strained layer damping. J Mech Eng Des 117:135144 Moreira RAS, Rodrigues JD (2006) A layerwise model for thin
Boudaoud H, Belouettar S, Daya EM, Potier-Ferry M (2008) soft core sandwich plates. Comput Struct 84:12561263
A shell finite element for active-passive vibration control Moreira RAS, Rodrigues JD, Ferreira AJM (2006) A generalized
of composite structures with piezoelectric and viscoelastic layerwise finite element for multi-layer damping treatments.
layers. Mech Adv Mater Struct 15:208219 Comput Mech 37:426444
Canelas A, Roche JR, Herskovits J (2009) The inverse elec- Nagendra S, Justin D, Gurdal Z, Haftka RT, Watson LT (1996)
tromagnetic shaping problem. Struct Multidiscipl Optim Improved genetic algorithm for the design of stiffened com-
doi:10.1007/s00158-008-0285-9 posite panels. Comput Struct 58:543555
DiTaranto RA (1965) Theory of vibratory bending for elastic and Nashif AD, Jones DIG, Henderson JP (1985) Vibration damping.
viscoelastic layered finite-length beams. ASME J Appl Mech Wiley, New York
32:881886 Nokes DS, Nelson FC (1968) Constrained layer damping with
Douglas BE, Yang JCS (1978) Transverse compressional damp- partial coverage. Shock Vib Bull 38:510
ing in the vibratory response of elastic-viscoelastic beams. Rao DK (1978) Frequency and loss factors of sandwich beams
AIAA J 16:925930 under various boundary conditions. Int J Mech Eng Sci 20:
Goel T, Vaidyanathan R, Haftka RT, Shyy W, Queipo NV, 271278
Tucker K (2007) Response surface approximation of pareto Rao MD, He S (1993) Dynamic analysis and design of laminated
optimal front in a multi-objective optimization. Comput composite beams with multiple damping layers. AIAA J
Methods Appl Mech Eng 196:879893 31:736745
Goldberg DE (1989) Genetic algorithms in search, optimization Sorensen DC (1995) Implicitly restarted arnoldi/lanczos methods
and machine learning. Addison-Wesley, Boston for large scale eigenvalue calculations. Tech. Rep. Technical
Haftka RT, Gurdal Z (1992) Elements of structural optimization. Report TR95-13, Department of Computational and Ap-
Kluwer, Dordrecht plied Mathematics, Rice University, Houston, Texas
Herskovits J (1986) A two-stage feasible directions algorithm for Sun CT, Lu YP (1995) Vibration damping of structural elements.
nonlinear constrained optimization. Math Program 36:1938 Prentice Hall PTR, Englewood Cliffs
Herskovits J (1998) A feasible directions interior point technique Yan MJ, Dowell EH (1972) Governing equations of vibrat-
for nonlinear optimization. J Optim Theory Appl 99:121 ing constrained-layer damping sandwich plates and beams.
146 ASME J Appl Mech 39:10411046
Herskovits J, Mazorche JR (2009) A feasible directions algorithm Yang G, Reinstein LE, Pai S, Carroll DL (1998) A new ge-
for nonlinear complementarity problems and applications in netic algorithm technique in optimization of permanent 125-i
mechanics. Struct Multidiscipl Optim 37:435446 prostate implants. Med Phys 25:23082315