ARTICLE IN PRESS

JOURNAL OF SOUND AND VIBRATION

Two-dimensional sublamination theory for analysis of functionally graded plates

Department of Mechanical and Aerospace Engineering, University of Missouri-Columbia, Columbia, MO 65211, USA Department of Aeronautics and Astronautics, Air Force Institute of Technology, Wright-Patterson Air Force Base, OH 45433, USA Received 11 April 2006; received in revised form 19 January 2007; accepted 18 July 2007 Available online 4 September 2007

Abstract Presented is a two-dimensional sublamination theory for analysis of plates made of functionally graded materials. The sublamination plate theory extends a layer-wise higher-order shear-deformation theory for laminated plates by considering material properties being continuous functions of the thickness coordinate and using a new concept of sublamination to increase the degrees of freedom. The theory accommodates free shear stress conditions on the bonding surfaces, accounts for non-uniform deformation-dependent distributions of transverse shear stresses through the thickness, can be used for evaluating boundary restraint effects, and can be used for analyzing thin and thick plates with any boundary conditions. A sublamination plate element based on this theory is developed and validated for static and dynamic analysis. The degrees of freedom of the element is adaptable. For an element away from boundaries, its degrees of freedom can be reduced at the elemental level without loss of accuracy. Analytical shear warping functions are presented. Moreover, modal analysis of functionally graded plates with different boundary conditions is performed to show the capability and accuracy of the theory. r 2007 Published by Elsevier Ltd.

1. Introduction Some structural components serve under conditions that require material properties vary within the component. For example, a kitchen knife is required to be hard at its cutting edge and strong and tough elsewhere. A gear needs to have a hard, wear-resistant surface but a tough body. A turbine blade needs to have a strong, tough and creep-resistant body but a refractory and oxidation-resistant outer surface. A composite made of ceramic on one side and metal on the other side is ideal for wear-resistant linings in the mineral processing industry; the hard ceramic face is for the exposed side and the tough metal face is for the rear side that is bolted or welded to a support frame. Other examples include the heat shield of rockets and human implants (e.g., bone and dental implants), which demands a heat or corrosion resistant outer layer and a tough metallic base material in order to increase the life time of the component. Composite materials manufactured by traditional methods may fulll such requirements, but a traditional composite is a man-made geometrical
Corresponding author.

E-mail address: [email protected] (P. Frank Pai). 0022-460X/$ - see front matter r 2007 Published by Elsevier Ltd. doi:10.1016/j.jsv.2007.07.027

ARTICLE IN PRESS
P. Frank Pai, A.N. Palazotto / Journal of Sound and Vibration 308 (2007) 164189 165

combination of different pre-existing phases and is a compromise between the properties of the component materials. Unfortunately, abrupt transition in material composition and properties within a conventional composite component often results in severe local stress concentration due to manufacturing (i.e., residual stresses) or external mechanical/thermal loading. For a composite structure formed by mixing different materials, temperature changes during fabrication and subsequent service can generate high internal stresses and/or unwanted pre-curvatures as a result of unequal thermal expansion/contraction between the constituent phases. Similarly, different local deformation elds are induced in the two materials during imposed mechanical loading because of their different mechanical properties. This incompatibility in thermomechanical deformation results in internal stresses and strains that play a key role in the creation of threading dislocations, localized yielding around interfaces, interfacial debonding, microcracking and brittle failure of the hard phase, or separation by void nucleation and growth in the ductile phase. Hence, composite materials with smooth gradation of material properties had been occasionally suggested in the literature [1], and those were the precursors of functionally graded materials (FGMs). However, as a formulated concept studied under a large organized research program, the term FGM originated in Japan in 1984 during a spaceplane project that required a thermal barrier material capable of withstanding a surface temperature of 2000 K in an oxidizing environment and a temperature gradient of 1000 K across a thickness less than 10 mm and being tough and thermally conductive on the lower-temperature side [2]. A functionally graded material is a composite material whose composition and microstructure are locally varied during manufacturing so that certain variations of local material (thermal and mechanical) properties are achieved. An FGM is very often a two-component composite characterized by a compositional gradient from one component to the other, and hence the properties of both components can be fully utilized. One important group of low-cost FGMs is ceramicmetal FGMs. The smooth gradation of material composition from ceramic to metal would enhance the toughness of the ceramic face and also provide a ceramicmetal bonding better than that produced by sintering a ceramic and then coating it with metal. FGMs are ideal candidates for applications involving severe thermal gradients, including thermal structures in advanced aircraft and aerospace engines and computer circuit boards. Comparing to traditional composites merits of FGMs include: (1) small thermal stresses and a designed peak thermal stress location; (2) small stress concentrations at material interfaces in an FGM; (3) strong interfacial bonding; (4) reduced free-edge effects; (5) delayed plastic yielding; and (6) increased fracture toughness by reducing driving force for crack growth along/across material interfaces [3,4]. A functionally graded material can be manufactured by chemical or physical vapor deposition methods, plasma spraying methods, powder metallurgy methods, self-propagating high-temperature combustion synthesis methods, surface chemical reaction methods, sintering methods, and many other techniques [2,3]. A common approach toward the fabrication of an FGM is to vary material composition through the thickness of a thin substrate. However, if the manufacturing facility is versatile enough, gradient material composition along two dimensions is possible. The concept of FGMs allows full integration of material fabrication and structural engineering into the nal design of structural components. Because the static and dynamic behavior of an FG structure is determined by the mechanical and thermal properties of the FGM and the actual boundary conditions and mechanical/thermal loads, structural analysis and design of material fabrication process need to be performed before actual manufacturing of the FGM, and it is a multi-disciplinary task and a reverse engineering problem. Because of the variation of material composition, many problems exist, including design of fabrication process in order to have an FGM with desired material properties, manufacturing-induced residual stresses, difculties in material property characterization, negative effects of stress concentrations at free surfaces and edges, and difculties in the modelling and analysis of structures made of FGMs. Research areas related to FGMs include process metallurgy, composite synthesis, mechanics and micromechanics of composites, fracture mechanics, and design methodologies. The science and technology of FGMs has been dramatically advanced during the last decade. Comprehensive reviews of current FGM research can be found in Refs. [25]. To extend conventional structural analysis techniques for functionally graded structures certain advances in macrostructural property characterization and thermomechanical modelling are needed. Because of nonuniform material distribution, design of functionally graded structures faces challenging modelling and analysis problems that exist in traditional composite structures. Thermomechanical analysis and engineering

ARTICLE IN PRESS
166 P. Frank Pai, A.N. Palazotto / Journal of Sound and Vibration 308 (2007) 164189

of functionally graded structures require effective properties of each FGM made of multiple materials. Hence, accurate estimation of effective properties of FGMs is the key to the eventual success in the design of FGMs. However, because precise information about the size, shape, and distribution of material particles in an FGM is not available and the effective moduli of the graded composite must be evaluated based only on the volume fraction distributions and the approximate shape of the dispersed phase. When particles are small, the material properties Pi of a mixture are difcult to be estimated using the volume fractions V i [2]. Micromechanics models and analysis methods have been developed by researchers to infer the effective properties of an equivalent macroscopically homogeneous composite material, including the Voigt method (i.e., P P1 V 1 P2 V 2 , an arithmetic average), the Reuss method (i.e., 1=P V 1 =P1 V 2 =P2 , a harmonic average), a generalized averaging method (P P1 V 1 P2 V 2 P12 V 1 V 2 , where P12 depends on P1 , P2 , V 1 , V 2 , and microstructure-dependent quantities), self-consistent methods [6,7], the MoriTanaka method [8], differential schemes, bounding techniques, nite-element unit cell models, and numerical modelling of a bulk FGM with boundary conditions [3,9]. Similar to traditional composite laminates, a functionally graded plate (FGP) may experience severe deformation-dependent transverse shear deformations, free-edge effects, and boundary constraint effects due to non-uniform material distribution and thick thickness. Detailed nite-element analysis using small threedimensional (3D) solid elements can be used to study such problems, but it is cumbersome and timeconsuming. However, some exact 3D elasticity solutions that account for deformation-dependent shears exist in the literature [1014], but they are all for plates with special boundary conditions, especially simply supported boundary conditions. A 2D plate theory that can treat arbitrary boundary conditions and account for deformation-dependent shear deformations and free-edge and boundary constraint effects does not exist. This work contributes to the modelling and analysis of FGPs by deriving a 2D sublamination plate theory (SPT) that can account for higher-order deformation-dependent transverse shear deformations and boundary effects caused by free edges and warping restraints at boundaries. A method of computing shear warping functions is derived and numerically demonstrated, and a sublamination plate element (SPE) is also formulated and validated. 2. Sublamination plate theory An FGP can be fabricated by a stepwise or a continuous grading method. Shear warping functions of an FGP fabricated using a stepwise grading method can be derived by layerwise analysis, and shear warping functions of an FGP fabricated using a continuous grading method can also be derived by layerwise analysis with many assumed layers, as shown next. 2.1. Shear warping functions by layerwise analysis Fig. 1 shows an FGP fabricated using a stepwise grading method and the xyz coordinate system for modeling, where the xy represents the reference plane of the plate, the x and y are in-plane (or reference plane) coordinates, and the z is the thickness coordinate. To include transverse shear deformations in the modeling of a general FGP, each layer fabricated during one step of the stepwise grading method needs an assumed displacement eld because each layer has different material properties. To account for transverse shear stresses, we generalize a layerwise third-order shear-deformation theory [15] by using a new concept of
y z h

b
a
x

Fig. 1. The plate geometry and the coordinate system.

ARTICLE IN PRESS
P. Frank Pai, A.N. Palazotto / Journal of Sound and Vibration 308 (2007) 164189 167

sublamination to improve the versatility and accuracy in predicting transverse shear stresses and boundary effects. For the ith layer with zi pzpzi1 , the displacement eld is assumed to be
i ui u wx z gi z a0 ai z2 ai z3 , 5 1 1 2 i ui v wy z g4 z bi bi z2 bi z3 , 2 0 1 2

ui w, 3

where i 1; . . . ; I and I is the total number of layers. Here, ux; y; t, vx; y; t, and wx; y; t are the displacements of the point on the xy plane (i.e., z 0), t denotes the time, and gi and gi are shear rotation 5 4 angles at the xy plane for the displacement eld of the ith layer, as shown in Fig. 2. Moreover, x  q =qx and y  q =qy. ai x; y; t account for the displacement along the x-direction due to shear warping, k bi x; y; t account for the displacement along the y-direction due to shear warping, and they are to be k determined by the continuity conditions of in-plane displacements and interlaminar shear stresses and the free shear stress conditions on the top and bottom surfaces of the plate as shown next. The transverse normal strain 33 can be accounted for by assuming ui w Zi z Zi z2 . However, as shown in Ref. [15], 33 is 3 1 2 mainly caused by Poissons effect and the external normal loads on the top and/or bottom surfaces, 33 and the transverse normal stress s33 can be accurately obtained by integrating the 3D equilibrium equations (see Eq. (41b)) after the reference plane deformations u, v, w, gi , and gi are obtained from a 2D plate theory. 5 4 Pagano [11] also pointed out that even the classical plate theory can predict s33 well. Because inclusion of 33 and s33 increases the number of dependent variables and formulation and computation difculties without signicant accuracy increase, it is neglected here. It follows from Eq. (1) that the transverse shear strains of the ith layer are i 13 qui qu3 1 gi 2zai 3z2 ai , 5 1 2 qz qx qui qu3 2 gi 2zbi 3z2 bi . 4 1 2 qz qy 3 5,
i i

i 23

(2)

The stressstrain relation for the ith layer is given by [16,17] 8 9 8 9 2 < si = < i = Qi 23 23 44 i Q2 i ; Qi 4 i 2 : si ; : 13 ; Q45 13 Qi cos2 yGi sin2 yG i ; 44 23 13
(a) z

Qi 45 Qi 55

Qi sin2 yG i cos2 yG i ; 55 23 13
(b) z

Qi sin y cos yGi G i , 45 13 23

(3)

dx

dy

undeformed

h/2 h/2 dx

deformed y x 5
(i)

(i)

y t2

dy

t1

Fig. 2. Shear deformation of a differential plate element: (a) undeformed and deformed congurations, and (b) shear warping of cross sections. The t1 and t2 are stress vectors.

ARTICLE IN PRESS
168 P. Frank Pai, A.N. Palazotto / Journal of Sound and Vibration 308 (2007) 164189

^ where y is the rotation angle of bers (if composite laminates) from the material axis x to the structural axis x, i i and G i are shear moduli. If the ith layer is an isotropic one, Q44 Qi Gi and Q45 0. 55 kl If there is no delamination, the in-plane displacements u1 and u2 and interlaminar shear stresses s13 and s23 are continuous across the interface of any two adjacent layers. Moreover, it is assumed that there is no shear loadings on the top and bottom surfaces of the plate, i.e., s23 s13 0 at z z1 and z zI1 . Consequently, we have sI x; y; zI1 ; t 0, 23 sI x; y; zI1 ; t 0, 13 si x; y; zi1 ; t si1 x; y; zi1 ; t 0 23 23 si x; y; zi1 ; t si1 x; y; zi1 ; t 0 13 13
i u2 x; y; zi1 ; t ui1 x; y; zi1 ; t 0 2 i u1 x; y; zi1 ; t ui1 x; y; zi1 ; t 0 1

for i 1; . . . ; I 1, for i 1; . . . ; I 1, for i 1; . . . ; I 1, for i 1; . . . ; I 1,

s1 x; y; z1 ; t 0, 23 s1 x; y; z1 ; t 0; ai bi 0, 13 0 0 (4) ^ ^ ^ ^ layer is the layer that contains the reference plane xy, and ai bi 0 for the ith layer because u The ith 0 0 and v represent the displacements of the point on the reference plane (see Eq. (1)). These 6I algebraic equations can be used to determine the 6I unknowns ai and bi (k 0; 1; 2 and i 1; . . . ; I) in terms of gi 4 k k and gi . Substituting Eqs. (1)(3) into Eq. (4) yields 5 Afag Bfgg; fag Afgg,
1 ^ ^

ai bi 0 1 1

^ for i 1; . . . ; I; iai.

(5)

where A is a 6I 6I constant matrix, B is a 6I 2I constant matrix, A A B is a 6I 2I constant matrix, and fag  fa1 ; a1 ; a1 ; b1 ; b1 ; b1 ; . . . ; aI ; aI ; aI ; bI ; bI ; bI gT , 0 1 2 0 1 2 0 1 2 0 1 2 fgg  fg1 ; g1 ; . . . ; gI ; gI gT . 5 5 4 4 To reduce the total number of dependent variables, an FGP of I layers can be grouped into J pI sublaminates. If the whole plate is treated as one sublaminate, g4 gi and g5 gi for i 1; . . . ; I. If the rst 5 4 10 layers is grouped as the rst sublaminate, the second 10 layers as the second sublaminate, and the rest as the third sublaminate, we have g2 gi and g2 gi for i 11; . . . ; 20, as shown in Fig. 3. Hence, B can be 5 5 4 4 reduced to a 6I 2J constant matrix, A A1 B is a 6I 2J constant matrix, and fag  fa1 ; a1 ; a1 ; b1 ; b1 ; b1 ; . . . ; aI ; aI ; aI ; bI ; bI ; bI gT , 0 1 2 0 1 2 0 1 2 0 1 2 fgg  fg1 ; g1 ; . . . ; gJ ; gJ gT . 5 5 4 4
J X j1 J X i;j j b04 g4 bi;j gj ; 05 5 j1 J X j1 J X i;j j b14 g4 bi;j gj ; 15 5 j1 J X j1 J X i;j j b24 g4 bi;j gj , 25 5 j1

With the sublamination shown in Fig. 3 and Eq. (6), one can obtain from the second equation of Eq. (5) that
i a0

ai;j gj ai;j gj ; 04 4 05 5

ai 1

i;j a14 gj ai;j gj ; 4 15 5

ai 2

ai;j gj ai;j gj , 24 4 25 5

bi 0

bi 1

bi 2

(7)

ARTICLE IN PRESS
P. Frank Pai, A.N. Palazotto / Journal of Sound and Vibration 308 (2007) 164189 169

sublaminate [3] z sublaminate [2] sublaminate [1]

x
Fig. 3. A functionally graded plate divided into I layers and 3 sublaminates.
i;j where i 1; . . . ; I. Moreover, akl and bi;j are entries of A and are constants determined by material kl properties and thickness coordinates of layers. Hence it follows from Eq. (1) that the displacement eld of the ith layer can be presented in the following form: i ui u wx z g5 z 1 J X j i;j g4 g14 gj gi;j , 5 15 j1 J X j i;j i;j g4 g24 gj g25 , ui v wy z gi z 5 2 4 j1

ui 3 where

w,

gi;j ai;j ai;j z2 ai;j z3 ; 14 04 14 24 gi;j bi;j bi;j z2 bi;j z3 ; 24 04 14 24 and are shear warping functions, and Eqs. (8) and (2) that the strains are given by i gi 5 13 i gi 23 4 i 12
J X j1 J X j1

gi;j ai;j ai;j z2 ai;j z3 , 15 05 15 25 gi;j bi;j bi;j z2 bi;j z3 , 25 05 15 25 gi;j 25 (9)

gi;j 15

gi;j 24

gi;j 14

and

represent shear coupling functions. It follows from

gj gi;j gj gi;j , 5 15z 4 14z gj gi;j gj gi;j , 5 25z 4 24z

uy vx 2zwxy zgi zgi 5y 4x

gj gi;j gj gi;j gj gi;j gj gi;j , 5x 25 5y 15 4x 24 4y 14

i ux zwxx zgi 5x 11 i vy zwyy zgi 22 4y i 33 0.

gj gi;j gj gi;j , 5x 15 4x 14

J X j i;j g4y g24 gj gi;j , 5y 25 j1

10

The analysis will be based on a linear, uncoupled thermoelasticity theory, where the heat generated by elastic straining is assumed to be a negligible second-order effect and the inuences of structural deformation on the thermal domain and thermal properties are also neglected. Hence, for an FGP subjected to

ARTICLE IN PRESS
170 P. Frank Pai, A.N. Palazotto / Journal of Sound and Vibration 308 (2007) 164189

thermomechanical loads, strains due to thermal expansion can be linearly added to the mechanical strains, and the mechanical stressstrain relation of the ith layer can be written as [1618] fsi g Qi fi g fLi gt
J X i;j j S 3 fc3 g fLi gt Q S 2 fc2 g i j1

! Qi Si fcg fLi gt, 11

where fsi g  fsi ; si ; si ; si ; si gT , 11 22 23 13 12

fLi g  fLi ; Li ; 0; 0; Li gT , 1 2 12 ^ ^ Li a1 cos2 y a2 sin2 y; 1 ^ ^ Li a1 sin2 y a2 cos2 y; 2 ^ Li 2^ 1 a2 cos y sin y, a 12

fc2 g  fux ; vy ; uy vx ; wxx ; wyy ; wxy gT , fcj g  fgj ; gj ; gj ; gj ; gj ; gj gT , 5 3 4x 4y 5x 5y 4 fcg  ffc2 gT ; fc1 gT ; . . . ; fcJ gT gT , 3 3 1 60 6 6 S 2  6 0 6 6 40 0 2 6 6 6 S i;j  6 3 6 6 4 g14 0 0 0 zdij g24 0 zdij g24 0 0 g14 2 0 1 0 0 0 0 z 0 0 0 0 1 0 0 0 0 z 0 0 0 0 g25 0 0 zdij g15 3 0 0 7 7 7 0 7, 7 7 0 5 2z 0 0 dij g24z g14z 0 0 0 3i;j . (12)

zdij g15 0 0 0 g25

7 7 7 g25z 7 7 dij g15z 7 5 0

Here t is the temperature increase beyond the reference temperature of the stress-free reference ^ ^ ^ ^ conguration, a1 and a2 are thermal expansion coefcients with respect to the material axes x and y, ^ respectively, y is the rotation angle from the x-axis to the x-axis, and Qi is the transformed and reduced (because s33 0) material stiffness matrix. Moreover, dij 1 if the ith layer is within the jth sublaminate (i.e., gi gj ), and dij 0 if the ith layer is not within the jth sublaminate. Note that the dij used here 4 4 is not the regular Dirac delta function. For example, if the second layer is within the rst sublaminate, d21 1 and d22 0. We note that, even if an FGP is fabricated by a continuous grading method, this method also works because the plate can be articially divided into I layers with material properties being those at the mid-point of each layer. The transverse shears due to non-uniform thermal expansion through the plate thickness are assumed to be negligible, and the temperature distribution can be assumed in the following form: tx; y; z; t t1 x; y; tt2 z. (13)

ARTICLE IN PRESS
P. Frank Pai, A.N. Palazotto / Journal of Sound and Vibration 308 (2007) 164189 171

2.2. Effective mechanical and thermal properties Mechanical properties of an FGP include Youngs modulus E, shear modulus G, Poissons ratio n, and mass ^ density r, and thermal properties include the coefcient of thermal expansion a, the thermal conductivity k, and the specic heat capacity c. r and n are usually linear functions of material volume ratios, but others are nonlinear functions of material volume ratios because they depend on material microstructures [9]. The FGP theory presented here can work with effective material properties estimated by using any method. For the purpose of demonstration, the effective properties P (e.g., E and a) will be estimated using the simple, Voigt method (i.e., an arithmetic method) as P P1 V 1 P2 V 2 ; V 1 V 2 1, (14) where P1 and P2 are the properties of the rst and second constituent materials (e.g., metal and ceramic), and V 1 and V 2 are the volume fractions of the constituent materials. The distributions of volume fractions through the plate thickness are assumed to follow the following simple power law:   z z1 n V2 , (15) zI1 z1 where n can be any non-negative real number, and z1 pz1 . Note that V 2 0 at z z1 if z1 z1 , and V 2 40 at z z1 if z1 oz1 . Hence, we have   z z1 n P P1 P2 P1 . (16) zI1 z1 Most of mechanical and thermal properties are linear or nonlinear functions of temperature, but r and n are usually independent of temperature. If a property P is a linear function of temperature, one can assume that ^ P P0 1 Pt, (17) ^ where t is the temperature beyond the reference temperature and P and P are constant. For detailed modeling of effective material properties of FGMs the reader is referred to Refs. [3,9]. 3. Finite-element formulation The extended Hamilton principle states [19] Z t dK e dP dW nc dt 0,
0

(18)

where K e is the kinetic energy, P the elastic energy, and W nc the non-conservative energy due to external distributed and/or concentrated loads and dampings. 3.1. Elastic energy It follows from Eqs. (11)(13) that Z I X Z Z zi1 ^ dP fdi gT fsi g dz dA fdcgT Ffcg fFgt1 dA,
i1 A zi A

(19)

where A denotes the area of the reference plane, and zi and zi1 indicate the locations of the lower and upper ^ surfaces of the ith layer. Moreover, F is a 6 6J 6 6J symmetric matrix and fFg is a 6 6J 1 vector, which are dened as I I X Z zi1 X Z zi1 i T i i ^ S Q S dz; fFg S i T Qi fLi gt2 z dz. (20) F
i1 zi i1 zi

ARTICLE IN PRESS
172 P. Frank Pai, A.N. Palazotto / Journal of Sound and Vibration 308 (2007) 164189

Because the heat generated by elastic straining is negligible, the inuences of elastic deformations on plates geometry and thermal properties are neglected, and the variation frequency of temperature is assumed to be much lower than the mechanical vibration frequency, the temperature t can be solved separately using the heat transfer equation. Consequently, t can be treated as a known dependent variable for the mechanical problem here and hence dt 0 is used. The integration of Eq. (20) can be obtained using Gauss quadratures [20]. Using the nite-element discretization scheme, we discretize the displacements as fu; v; w; g1 ; g1 ; . . . ; gJ ; gJ gT Nfqn g, 5 5 4 4
n

(21)

where fq g is the displacement vector of the nth element and N is a matrix of 2D shape functions. Here fournode quadrilateral elements are assumed to be used. Each node has 6 2J degrees of freedom (dof) (i.e., u; v; w, wy ; wx ; wxy , g1 ; g1 ; . . . ; gJ ; gJ ). Hence N is a 3 2J 24 8J matrix. Moreover, 5 5 4 4 fcg Dfqn g; D  qN, (22) where q is a 6 6J 3 2J matrix consisting of differential operators, and D is a 6 6J 24 8J matrix. Substituting Eq. (22) into Eq. (19) yields Ne Ne XZ X ^ ^ n dP fdqn gT DT FDfqn g fFgt1 dA fdqn gT K n fqn g fF g
n1 An n1

^ fdqgT Kfqg fF g, where K 

23 Z
A
n

D FD dA;

^ n fF g 

^ DT fFgt1 dA,

(24)

N e is the total number of elements, An is the area of the nth element, K n is the stiffness matrix of the nth ^ n element, K is the structural stiffness matrix, fF g is the thermal-induced mechanical loading, and fqg is the structural displacement vector. 3.2. Kinetic energy The displacement eld shown in Eq. (8) can be rewritten as fui g S 2 fc2 g where fui g  fui ; ui ; ui gT , 1 2 3 fc2 g  fu; v; w; wx ; wy gT ; 1 0 6 S2  4 0 1 0 0 2 0 0 1 fc3 g  fgj ; gj gT ; 5 4 z 0 0 3 0 7 z 5; 0
j J X j1

S 3 fc3 g S fcg,

i;j

(25)

fcg  ffc2 gT ; fc3 gT ; . . . ; fc3 gT gT , 2 3i;j zdij g15 7 g25 5 . 0

g14 i;j 6 S 3  4 zdij g24 0

(26)

Hence, the variation of kinetic energy is given by I XZ Z dK e Z

zi1

fdui gT ri fui g dz dA 27

fdcgT Ffcg dA,

ARTICLE IN PRESS
P. Frank Pai, A.N. Palazotto / Journal of Sound and Vibration 308 (2007) 164189 173

where ri is the mass density of the ith layer, and F is a 5 2J 5 2J symmetric matrix given by I X Z zi1 i i F S T ri S dz. (28)
i1 zi

Using the same discretization scheme used in Eqs. (21) and (22), we obtain that fcg Dfqn g; D  qN, (29)

where q is a 5 2J 3 2J matrix consisting of differential operators, and D is a 5 2J 24 8J matrix. Substituting Eq. (29) into Eq. (27) yields Ne Ne XZ X dK e fdqn gT DT FDfqn g dA fdqn gT M n fqn g
n1 An n1

fdqg Mfqg, where M n  Z

30

DT FD dA,

(31)

M n is the mass matrix of the nth element and M is the structural mass matrix. 3.3. External loads For the non-conservative virtual work we only consider the external transverse distributed load f 3 and obtain that Z Ne Ne XZ X f 3 dw dA fdqn gT fN 3 gf 3 dA fdqn gT fF n g fdqgT fF g, (32) dW nc
A n1 An n1

where fN 3 g is the transpose of the third row of the N shown in Eq. (21) and Z n fF g  fN 3 gf 3 dA.
An

(33)

3.4. Equation of motion Substituting Eqs. (23), (30), and (32) into Eq. (18) yields the equation of motion as ^ _ Mfqg Cfqg Kfqg fF g fF g, (34)

where a damping matrix C is added. Note that this SPE is a 2D one but it accounts for 3D effects caused by transverse shear stresses. This linear plate element is a new addition to the nite-element code geometrically exact structural analysis (GESA), which is written using MATLAB syntax and is mainly for analysis of highly exible structures. 4. Numerical results 4.1. Shear warping functions To show shear warping functions we consider an orthotropic laminate studied by Pagano [10,11]. The properties of each graphite/epoxy layer are E 11 172:38 GPa; E 22 6:90 GPa; E 33 6:90 GPa; n12 0:25; n13 0:25; n23 0:25,

ARTICLE IN PRESS
174 P. Frank Pai, A.N. Palazotto / Journal of Sound and Vibration 308 (2007) 164189

(35) ^ is the layer thickness. If three sublaminates are assumed (i.e., J 3), it follows from Eq. (8) that shear where h warpings W i and W i are given by 1 2 W i  ui u wx z 1 1 gi z 5 g1 gi;1 4 14
J X j i;j g4 g14 gj dij z gi;j 5 15 j1 i;3 g1 gi;1 g2 gi;2 g2 gi;2 g3 g14 g3 gi;3 , 5 15 5 15 5 15 4 14 4 J X j1

G 12 3:45 GPa;

G13 3:45 GPa;

G 23 1:38 GPa;

^ h 0:127 mm,

W i  ui v wy z 2 2 W i 1 gi z 4 g1 gi;1 4 24

gj dij z gi;j gj gi;j 5 25 4 24 36

i;3 g1 gi;1 g2 gi;2 g2 gi;2 g3 g24 g3 gi;3 , 5 25 5 25 5 25 4 24 4

where is the shear warping of the yz cross section and is the shear warping of the xz cross section. Note that the dij is not the regular Dirac delta function. dij 1 if the ith layer is within the jth sublaminate even if iaj, and dij 0 if the ith layer is not within the jth sublaminate even if i j. Also, it follows from Eqs. (3) and (10) that 8 9 2 9 38 PJ j i;j j i;j i < si = Qi Q45 < j1 g4 dij g24z g5 g25z = 23 44 5 P 4 i . (37) i : si ; Q45 Q55 : J gj gi;j gj dij gi;j ; j1 4 14z 5 13 15z For a 0 =90 =0 laminate, numerical results conrm that Qi 0 and shear coupling functions gi;j 45 14 and gi;j are always zero for orthotropic cross-ply laminates [21]. Fig. 4 shows the shear warping functions 25 gij and their contributions to the transverse shear stresses (see Eq. (37)). For this case, the ith layer is dened kl to be the ith sublaminate. The bottom and top layers are the rst and third layers, respectively. Note that each of the shear warping functions of every sublaminate is continuous and its contribution to sk3 is also continuous although dij gi;j and dij gi;j are discontinuous. Because three sublaminates are used, gi and 24z 15z 4 gi (i 1; 2; 3) are six independent variables, and their relations depend on boundary and loading conditions. 5 If g1 ; g2 ; g3 1; 1; 1g4 and g1 ; g2 ; g3 1; 1; 1g5 are assumed, Fig. 5 shows the shear warpings W i and 5 5 5 4 4 4 transverse shear stresses si3 , where si =g5 Qi W i =g5 ; 55 13 1z si =g4 Qi W i =g4 . 23 44 2z (38)

W i 2

The broken lines represent the exact distributions [10,11], which are functions of specied plate dimensions and boundary and loading conditions. We note that the predicted shear stresses do not match well with the exact ones because incomplete cubic polynomials are used to describe the cross-section warping, as shown in Eq. (1). Moreover, because the warping functions and shear stress distributions obtained by Pagano [10] are dependent on the boundary and loading conditions and the thickness-to-span ratio (see Figs. 4c and e of Ref. [10]), this discrepancy is also problem dependent. If the answers obtained from a set of specied plate dimensions and boundary and loading conditions are gj 2; 1; 2g4 and gj 0:5; 1; 0:5g5 , Fig. 6 shows the W i and si3 . Note that they are close to those of 5 4 Pagano [10,11]. If gj 0:58; 1; 0:58g4 and gj 0:375; 1; 0:375g5 , Fig. 7 shows the W i and si3 . At a clamped edge 5 4 parallel to the y-axis, because s13 0 at the bottom (z z1 ) and top (z zI1 ) surfaces and the reference plane should be parallel to the bottom and top surfaces, wx 0. Moreover, because u1 z1 uI zI1 0 at 1 1 the clamped edge,
1;1 1;3 W 1 z1 0 g1 z1 g15 z1 g2 g1;2 z1 g3 g15 z1 , 5 5 15 5 1 I;1 I;2 I;3 W I zI1 0 g1 g15 zI1 g2 g15 zI1 g3 zI1 g15 zI1 . 5 5 5 1

Hence, Fig. 7a can be used to simulate the restraint warping of a clamped plate edge parallel to the y-axis. This condition can be implemented in analysis by using two multiple point constraints. Similarly, wy W 2 0 on the bottom and top surfaces at a clamped edge parallel to the x-axis, and Fig. 7b can be used to simulate the

ARTICLE IN PRESS
P. Frank Pai, A.N. Palazotto / Journal of Sound and Vibration 308 (2007) 164189 175

(a) 0.2 j=3 0.1 z(mm)

(b) 0.2 j=3 0.1 z(mm) j=2 0

0.1 j=1 0.2 0.2 0.1 j=2 0 ijz+g15 (mm) (c) 0.2 j=3 0.1 z(mm) z(mm) (d)
(i,j)

0.1 j=1 0.1 0.2 0.2 0.2 0.1 0 ijz+g24 (mm) 0.2 j=3 0.1
(i,j)

0.1

0.2

j=2

0 j=2 0.1

0.1 j=1 0.2 2

j=1 0.2 4 2 0
(i) (i,j)

0
(i) (i,j)

Q55 (ij+g15z) (GPa)

Q44 (ij+g24z) (GPa)

i;j Fig. 4. Shear warping functions of a 0 =90 =0 laminate: (a) dij z gi;j , (b) dij z g24 , (c) Qi dij gi;j , and (d) Qi dij gi;j . The 55 15 15z 44 24z solid lines are of the rst sublaminate j 1, the dashed lines j 2, and the dash-dot lines j 3.

restraint warping of a clamped plate edge parallel to the x-axis. If the rst-order shear-deformation theory is used, one needs to assign wy g4 a0 in order to have a non-zero shear force at a clamped plate edge parallel to the x-axis, which violates the actual boundary condition. It is apparent that the use of sublamination increases the freedom for accurate modeling of different boundary conditions. It has been shown in the literature that shear warping functions may vary with the spatial location, vibration frequency, loading and boundary conditions, and structural dimensions [13,22]. The shear warping functions shown in Figs. 57a,b are very different, but each of them is continuous and has continuous sk3 . The actual cross-sectional warpings depend on the ratios between gj and gj , and gj and gj are determined by loading 5 5 4 4 and boundary conditions. Apparently, the use of sublamination and more dependent variables enables the modeling of such deformation-dependent shear warping functions. If each layer is treated as a sublaminate, the number of dependent variables will be 3 2I (i.e., u, v, w, g1 , g1 ,y, gI , gI ) and the theory is equivalent 5 5 4 4 to that in Ref. [23]. One can reduce the number of dependent variables by reducing the number of sublaminates by using specic ratios between gj and gj , as shown in Figs. 5 and 6. 5 4 Next, we consider an orthotropic laminate studied by Noor and Burton [24]. The material properties of each layer are E 11 103:43 GPa; E 22 6:90 GPa; E 33 6:90 GPa; n12 0:3; n13 0:3; n23 0:49, (39)

G 12 3:45 GPa;

G 13 3:45 GPa;

G 23 2:42 GPa;

^ h 0:127 mm.

ARTICLE IN PRESS
176 P. Frank Pai, A.N. Palazotto / Journal of Sound and Vibration 308 (2007) 164189

(a) 0.2 0.1 z (mm) 0 0.1 0.2 0.2 (c) 0.2 0.1 z (mm)

(b) 0.2 0.1 z (mm) 0.1 0 0.1 0.2 0.2 (d) 0.2 0.1 z (mm) 0 0.1 0.2 0 2 13/5 (GPa) 4 0 1 2 23/4 (GPa) 3 4

0 0.1 W1/5 (mm)

0.2

0.1

0 0.1 W2/4 (mm)

0.2

0 0.1 0.2

Fig. 5. A 0 =90 =0 laminate: (a) shear warping W 1 =g5 , (b) shear warping W 2 =g4 , (c) shear stress s13 =g5 , and (d) shear stress s23 =g4 , where gj 1; 1; 1g4 and gj 1; 1; 1g5 are assumed and the broken lines are exact solutions. 5 4

For a 0 =90 =0 =90 laminate, Fig. 8 shows that the obtained distribution of s13 =g5 (solid lines) is similar to the exact one (broken lines) obtained by Noor and Burton [24, Fig. 3], where the reference plane is chosen to be on the neutral axis of the yz cross section. Because the neutral axis of the xz cross section is different from that of the yz cross section, the obtained s23 =g4 is not as accurate as s13 =g5 . For an anisotropic laminate, the neutral axis of the xz cross section may not be on the midplane, the neutral axis of the yz cross section may not be on the midplane, and these two neutral axes may not be on the same plane. These cause difculties in and reveal complexity of the analysis of anisotropic laminates. However, for symmetric and skew-symmetric laminates, the neutral axes are always on the midplane. Because gj (j 1; . . . ; 4) are four independent variables, using different 5 ratios between gj may result in more accurate s13 =g5 , as shown in Fig. 8b. To improve the accuracy 5 one can even divide the four layers into eight (or more) sublaminates, as shown in Fig. 8c. Similarly, this approach can be used to improve the accuracy of s13 and s23 in Figs. 6c and d by using more sublaminates. For a 21-layer 0 =90 10 =0 laminate, the shear warping functions and transverse shear stresses are shown in Fig. 9. We note that, although the shear warping functions and shear stresses are zigzag, their global distributions are very similar to those of an isotropic plate (i.e., a parabolic function 1 4z2 =h2 ). This is expected because the number of layers is high. It reveals that, for a laminate consisting of many layers, one can treat it just like an isotropic one, except that the reference plane is better to be on the neutral plane. For general anisotropic laminates, shear coupling functions g14 and g25 are non-zero [21] and hence the distributions of s13 and s23 depend on the values of g4 and g5 , which are determined by loading and boundary conditions and plate dimensions. However, FGPs are usually isotropic on the xy plane.

ARTICLE IN PRESS
P. Frank Pai, A.N. Palazotto / Journal of Sound and Vibration 308 (2007) 164189 177

(a) 0.2 0.1 z (mm) 0 0.1 0.2 0.2 (c) 0.2 0.1 z (mm)

(b) 0.2 0.1 z (mm) 0.1 0 0.1 0.2 0.2 (d) 0.2 0.1 z (mm) 0 0.1 0.2 0 1 13/5 (GPa) 2 0 1 2 23/4 (GPa) 3 4

0 0.1 W1/5 (mm)

0.2

0.1

0 0.1 W2/4 (mm)

0.2

0 0.1 0.2

Fig. 6. A 0 =90 =0 laminate: (a) W 1 =g5 , (b) W 2 =g4 , (c) s13 =g5 , and (d) s23 =g4 , where gj 2; 1; 2g4 and gj 0:5; 1; 0:5g5 and the 5 4 broken lines are exact solutions.

4.2. Theory validation To validate this 2D SPT in predicting all plate vibration modes we consider the free undamped vibration of a simply supported plate that was studied in Refs. [12,14]. For any mechanical system without body forces, the dynamic version of the principle of virtual work can be derived using the extended Hamilton principle to be [19] Z XX I 3 k si si si ri ui dui dV 0, (40) 1kx 2ky 3kz k
V i1 k1

where V is the system volume and s1kx  qs1k =qx, etc. Hence, the equilibrium equations of a differential material particle are k si si si ri ui ; 1kx 2ky 3kz k 1; 2; 3. (41)

For a simply supported plate under free vibration the boundary conditions are given by [10] ui ui si 0 at x 0; a; 2 3 11 ui ui si 0 at y 0; b; 1 3 22 i 1; . . . ; I, i 1; . . . ; I, i 1; I. (42)

i si si s33 0 at z h=2; 13 23

ARTICLE IN PRESS
178 P. Frank Pai, A.N. Palazotto / Journal of Sound and Vibration 308 (2007) 164189

(a) 0.2 0.1 z (mm) 0 0.1 0.2 0.2 (c) 0.2 0.1 z (mm) 0 0.1 0.2

(b) 0.2 0.1 z (mm) 0.1 0 0.1 0.2 0.2 (d) 0.2 0.1 z (mm) 0 0.1 0.2 0 2 13/5 (GPa) 4

0 0.1 W1/5 (mm)

0.2

0.1

0 0.1 W2/4 (mm)

0.2

0 2 23/4 (GPa)

Fig. 7. A 0 =90 =0 laminate: (a) W 1 =g5 , (b) W 2 =g4 , (c) s13 =g5 , and (d) s23 =g4 , where gj 0:58; 1; 0:58g4 and gj 0:375; 1; 5 4 0:375g5 .

(a) 0.3 0.2 z (mm)

(b) 0.3 0.2 z (mm)

(c) 0.3 0.2 z (mm) 0 1 2 13/5 (GPa) 3 0.1 0 0.1 0.2 0 1 2 13/5 (GPa) 3

0.1 0 0.1 0.2 0 1 2 3 13/5 (GPa) 4

0.1 0 0.1 0.2

Fig. 8. The s13 =g5 of a 0 =90 =0 =90 laminate: (a) gj 1; 1; 1; 1g5 , (b) gj 0:76; 1; 0:76; 0:76g5 , and (c) gj 0:9; 0:76; 1; 1; 0:77; 5 5 5 0:76; 0:5; 0:4g5 . The broken lines are exact solutions.

An FGP is usually transversely isotropic. For an orthotropic laminate or a transversely isotropic plate its material stiffness matrix can be reduced by using s33 0 (see Eq. (11)) to be 3 2 i 0 0 0 Q11 Qi 12 7 6 i i 6Q 0 0 0 7 7 6 21 Q22 7 6 i i 6 0 0 Q44 0 0 7. (43) Q 6 7 7 6 i 6 0 0 0 Q55 0 7 5 4 0 0 0 0 Qi 66

ARTICLE IN PRESS
P. Frank Pai, A.N. Palazotto / Journal of Sound and Vibration 308 (2007) 164189 179

(a) 1 0.5 z (mm) 0 0.5 1 1 0.5 0 0.5 W1/5 (mm) 1

(b) 1 0.5 z (mm)

(c) 1 0.5 z (mm) 2 0 0.5 1 0 2 13/5 (GPa) 4 2 0 2 23/4 (GPa) 4

0 0.5 1

Fig. 9. A 0 =90 10 =0 laminate: (a) W 1 =g5 , (b) s13 =g5 , and (c) s23 =g4 , where gj g4 and gj g5 are assumed. 5 4

The solutions of dependent variables that satisfy Eqs. (41) and (42) have the following forms: u U cos mx sin ny sin ot; gi G i cos mx sin ny sin ot; 5 5 v V sin mx cos ny sin ot; w W sin mx sin ny sin ot, m  mp=a; n  np=b (44)

gi G i sin mx cos ny sin ot; 4 4

which can be proved by substituting Eq. (44) into Eqs. (8) and (10), then into Eq. (11) with t 0 and Qi i shown in Eq. (43), and then into Eq. (41), as shown next. Here the U, V, W, G4 , and Gi are unknown 5 i;j i;j constants, and o is the unknown, undamped natural frequency. Because g14 g25 0 for transversely isotropic plates, substituting Eq. (44) into Eq. (8) yields ui U 1 cos mx sin ny sin ot; 1 ui V 1 sin mx cos ny sin ot; 2
J X j1

ui W sin mx sin ny sin ot, 3

U 1 z  U W mz G i z 5

G j gi;j ; 5 15

V 1 z  V W nz G i z 4

G j gi;j . 4 24

(45)

Substituting Eq. (44) into Eq. (10) yields i E i cos mx sin ny sin ot; 13 13 i E i cos mx cos ny sin ot; 12 12 i E i sin mx sin ny sin ot; 22 22 E i  G i 5 13
J X j1

i E i sin mx cos ny sin ot, 23 23 i E i sin mx sin ny sin ot, 11 11 i E i sin mx sin ny sin ot, 33 33 E i  G i 23 4
J X j1 J X j1

G j gi;j ; 5 15z

G j gi;j , 4 24z G j mgi;j Gj ngi;j , 5 4 24 15

E i  Un V m 2zW mn zG i n zG i m 5 12 4 E i  Um zW m2 zGi m 5 11 E i  V n zW n2 zG i n 22 4
J X j1 J X j1

G j mgi;j , 5 15 46

G j ngi;j . 4 24

ARTICLE IN PRESS
180 P. Frank Pai, A.N. Palazotto / Journal of Sound and Vibration 308 (2007) 164189

The E i are known functions of z, U, V, W, G i , and G i . Substituting Eq. (46) into Eq. (11) with t 0 and ij 5 4 Qi given by Eq. (43) yields
i s11 Si sin mx sin ny sin ot; 11 i s33 Si sin mx sin ny sin ot; 33 i s13 Si cos mx sin ny sin ot; 13

si S i sin mx sin ny sin ot, 22 22 si S i sin mx cos ny sin ot, 23 23 si S i cos mx cos ny sin ot, 12 12 S i 0, 33 47 S 12  Qi E i . 66 12

S i  Qi E i Qi E i ; 11 11 11 12 22 S i  Qi E i ; 23 44 23

i i S i  Q21 E i Qi E 22 ; 22 11 22

S 13  Qi E i ; 55 13

The S i are known functions of z, U, V, W, Gi , and G i . Note that S i 0 for the proposed theory. If ij 5 4 33
i ui w Zi z Z2 z2 is assumed in Eq. (1), E i and Si will be also known functions of z, U, V, W, G i , and 3 1 33 33 4 G i . However, E i and S i are neglected from the beginning because they are mainly caused by normal loads 5 33 33 on the top and/or bottom surfaces and Poissons effect, and they do not have signicant inuences on other stresses and strains [15]. i Eq. (44) shows that the xy-plane distribution of gi (g5 ) of every layer is the same and hence the ratios 4 j j among g4 (g5 ) at any x; y location is the same. In other words, the shear warpings W i and W i are 1 2 independent of x and y, and hence there is no boundary restraint effect in a simply supported plate. Substituting Eqs. (47) and (45) into Eq. (40) yields I X Z zi1 i i 0 mS 11 nS i S i ro2 U 1 C s dU 1 mS i nS22 S i ro2 V 1 S c dV 1 12 13z 12 23z i1 zi

mS i nS i ro2 W Ss dW dz sin ot 13 23 ~ ~ fdpgT A o2 Bfpg sin ot, Z fC s ; S c ; S s g 

fcos2 mx sin2 ny; sin2 mx cos2 ny; sin2 mx sin2 nyg dx dy, 48

fpg  fU; V ; W ; G 1 ; G1 ; . . . ; G J ; G J gT , 5 5 4 4

~ ~ where A and B are 3 2J 3 2J constant matrices. Note that the u, v, and w in Eq. (44) represent the in-plane (or reference plane) deformation mode shape and the U 1 z and V 1 z in Eq. (45) represent the thickness deformation mode shape. For a known in-plane mode shape, the thickness mode shapes and natural frequencies can be obtained by solving the following eigenvalue problem from Eq. (48) ~ ~ A o2 Bfpg f0g. The transverse normal stress S i 33 can be obtained by integrating the third equation of Eq. (41), i.e. S i mS i nS i ro2 W 33z 13 23 (50) (49)

and setting S1 z1 0 at the bottom surface. 33 For a 5 cm 5 cm 1 cm aluminum plate with E 70 GPa, n 0:3, and r 2702 kg=m3 , Table 1 compares the natural frequencies o11;k (i.e., m; n; k 1; 1; k) of vibration modes obtained from the SPT with the 3D elasticity solutions obtained by Vel and Batra [14] using p series method. The natural a power frequencies o11;k are normalized to be non-dimensional as o11k  o11k a2 r=E =h. With the use of 3 layers and 3 sublaminates (i.e., I J 3), there are only 9 variables in the vector fpg of Eqs. (48) and (49) and hence it can give only 9 natural frequencies. It is clear that all natural frequencies are close to the exact ones except o11;9 . When the number of layers increases beyond 10, all the frequencies agree well with the exact ones. Fig. 10 shows the thickness-direction distributions of stresses S13 , S11 , and S 33 of 6 modes obtained using 15 layers and 15 sublaminates (i.e., I J 15) and setting the maximum of the 15 Gj to be one. All the S ij are 5 plotted using the same scale, but the S 13 and S33 in Fig. 10a are small and hence they are magnied by 5 and 10 times, respectively. By scrutinizing the in-plane and thickness-direction distributions of U, V, W, U 1 , V 1 , W i ,

ARTICLE IN PRESS
P. Frank Pai, A.N. Palazotto / Journal of Sound and Vibration 308 (2007) 164189 Table 1 The natural frequencies o11;k of thickness modes of a 5 cm 5 cm 1 cm aluminum plate (the # indicates the mode number) Exact a 5h SPT J I 15 layers 5.2813 J I 11 layers 5.2813 JI 3 layers 5.2813 SPE JI 3 layers 5.3606 (#1) (#1) 13.887 (#6) (#4) 23.351 (#14) (#9) 50.642 (#52) (#29) 55.170 (#67) (#42) 99.458 (#214) (#151) 101.21 (#228) (#158) 163.64 (#532) (#412) J I 11 layers 5.3605 (#1) (#1) 13.887 (#6) (#4) 23.351 (#14) (#9) 50.628 (#52) (#29) 55.155 (#67) (#42) 98.426 (#211) (#147) 100.20 (#225) (#158) 148.09 (#473) (#366) U V 0, U 1 V 1 , W 1 W 2 , W a0, S 11 S22 , S13 S23 (BS) U V U 1 V 1 ; W W 1 W 2 0, S 11 S22 const., S13 S23 S33 0 (TC) U V U 1 V 1 ; W W 1 W 2 0, S 11 S22 const., S13 S23 0 (TT) U V W 0; U 1 W 1 V 1 W 2 , S 11 S22 ; S13 S23 , S33 0 (S) U V 0, U 1 V 1 , W 1 W 2 , W a0, S 11 S22 , S13 S23 (SB) U V; W 0, U 1 V 1 , W 1 W 2 , S 11 S22 ,S13 S23 , S33 0 (STC) U V , W 0, U 1 V 1 , W 1 W 2 , S 11 S22 , S13 S23 (STT) U V 0, U 1 V 1 , W 1 W 2 , W 0, S 11 S22 , S13 S23 (SB) Characteristics 181

o11;1

5.3036

o11;2

13.777

13.777

13.777

13.777

o11;3

23.136

23.287

23.287

23.287

o11;4

50.619

50.619

50.619

50.634

o11;5

54.727

55.146

55.146

55.162

o11;7

98.386

98.395

98.411

99.442

o11;8

104.75

100.17

100.19

101.20

o11;9

146.01

148.04

148.08

163.63

and S ij the characteristics of these modes are determined and listed in the last column of Table 1. The 1; 1; 1 mode is a bending mode with small shear deformation and will be called a bendingshearing (BS) mode. Although the S13 of Mode 1; 1; 1 in Fig. 10a is small, it makes the classical plate theory cannot predict this frequency well. The 1; 1; 2 mode is a pure in-plane tensioncompression (TC) mode, as shown later in Fig. 12. Mode 1; 1; 3 is a pure in-plane tensiontension (TT) mode. Because of the extensionextension deformation (i.e., S11 S22 ), the dynamical inertial stretching along the thickness direction caused by Poissons effect results in a non-zero but small S33 in the 3D elasticity solution. This non-zero S 33 cannot be predicted by this theory because Eq. (50) shows that the S33 in this theory only accounts for those caused by shear stresses S13 and S23 . If ui w Zi z Zi z2 is assumed from the beginning in Eq. (1), this S 33 can be also obtained [15]. 3 1 2 Because there is no shear deformation in Modes 1; 1; 2 and 1; 1; 3, even the classical plate theory can predict these two frequencies well by post-processing the 3D elasticity equations after the reference-plane deformations u, v, and w are obtained from the plate theory. If u at x 0; a and v at y 0; b are xed to zero, these two modes will be prevented from happening. Mode 1; 1; 4 is a pure transverse S (shearing) mode, and Mode 1; 1; 5 is a transverse shear (see Fig. 2a and b) with small bending shearbending (SB) mode. Mode 1; 1; 6 (o11;6 86:840 in Ref. [14]) is missing because it is a thickness-stretching mode and it is excluded from the assumed displacement eld shown in Eq. (1). Again, if ui w Zi z Zi z2 is assumed from the beginning in Eq. (1), this mode would be also predicted 3 1 2 [15]. Mode 1; 1; 7 is an sheartensioncompression (STC) mode with a major transverse shear accompanied by a small in-plane tensioncompression deformation similar to Mode 1; 1; 2. Mode 1; 1; 8 is a sheartensiontension (STT) mode with a major transverse shear accompanied by a small in-plane tensiontension deformation similar to Mode 1; 1; 3 (see Fig. 2). Mode 1; 1; 9 is an SB mode having transverse shears accompanied by a very small bending deformation. The stresses in Fig. 10 agree well with the 3D elasticity solutions (Fig. 2 in Ref. [14]). Eq. (50) shows that S i za0 if S i z S i z and m n, as 33 13 23

ARTICLE IN PRESS
182 P. Frank Pai, A.N. Palazotto / Journal of Sound and Vibration 308 (2007) 164189

(a) 5
10S33

(b) 5

(c) 5

z (mm)

z (mm)

0
S11 5S13

z (mm) 10

5 200 (d) 5

100

0 100 Sij (GPa)

200 (e)

5 20

0 10 Sij (GPa)

20

30 (f)

5 20

10

0 10 Sij (GPa)

20

30

z (mm)

z (mm)

z (mm) 10

5 20

10

0 Sij (GPa)

10

20

5 20

0 10 Sij (GPa)

20

5 15 10

5 0 5 Sij (GPa)

10

15

Fig. 10. The stress distributions of different modes: (a) Mode 1; 1; 1, (b) Mode 1; 1; 4, (c) Mode 1; 1; 5, (d) Mode 1; 1; 7, (e) Mode 1; 1; 8, and (f) Mode 1; 1; 9, where the solid, broken, and thin lines represent S13 , S11 , and S33 , respectively.

shown in Table 1 and Fig. 10a, c, e and f. In other words, the S i z is mainly due to Poissons effect induced 33 by the in-plane tensiontension deformation caused by shear strains, as shown in Fig. 2b. On the other hand, i i S33 z 0 if S13 z Si z and m n, as shown in Table 1 and Fig. 10b and d. However, if ui 23 3 i w Zi z Z2 z2 is assumed from the beginning in Eq. (1), this would result in a non-zero but small S i z. All 1 33 the results also conrm that S33 can be accurately predicted by post-processing results obtained from a 2D plate theory using the 3D elasticity equations. The S 13 curves in Fig. 10df become smooth when the number of layers is increased beyond 20, as shown in Fig. 11a. Fig. 11b shows that, although o11;9 is not accurate when only 3 layers are used, the predicted distributions of shear stresses are fairly accurate. However, the inaccurate S 13 causes the integrated S 33 to be non-zero on the top surface. Note that Fig. 11a and b have different scales because each is individually scaled to have its maximum G j to be one. If only the deformation dynamics of a plates surfaces is the interest of 5 analysis and/or testing, Mode 1; 1; 1 is the most important one. However, other thickness modes are important for accurate stress analysis and wave propagation study. It is apparent that this method works well because the analytical reference-plane mode shapes of a simply supported isotropic (or orthotropic) plate are available and the reference-plane mode shapes are not sensitive to thickness vibrations even for thick plates. After a reference-plane mode shape is obtained using a 2D plate theory, the transverse normal stress si (and si and si if the classical plate theory is used) can be obtained by 33 13 23 integrating Eq. (41), and all other strains and stresses can be obtained using the straindisplacement relations and constitutive equations (e.g., Eqs. (10) and (11)). Hence, the original 3D problem becomes 1D, as shown by Eqs. (45) and (48). However, for structures with no analytical reference-plane mode shapes, one needs to use a numerical method (e.g., the nite-element method) to obtain the reference-plane and thickness mode shapes simultaneously. Unfortunately, if a 2D plate theory does not have enough dofs for different thickness vibration modes (e.g., the classical, the rst-order shear, and the third-order shear theories), the high-order modes shown in Figs. 10bf cannot be obtained and restraint boundary conditions cannot be treated.

ARTICLE IN PRESS
P. Frank Pai, A.N. Palazotto / Journal of Sound and Vibration 308 (2007) 164189 183

(a) 5 S33 z (mm) S11 S13

(b) 5

z (mm)

5 15

10

0 5 Sij (GPa)

10

15

5 30

20

10 0 Sij (GPa)

10

20

Fig. 11. The stress distributions of Mode 1; 1; 9 using: (a) 31 layers and (b) 3 layers, where the solid, broken, and thin lines represent S13 , S11 , and S33 , respectively.

4.3. Finite-element analysis and boundary restraint effects Here we demonstrate that the SPE derived from the proposed SPT can accurately predict thickness modes and handle different boundary restraint effects. Moreover, the number of dofs of an SPE is adaptable and it i can directly compute si and s23 without post-processing by integrating the 3D equilibrium equations. 13 It follows from Eqs. (42), (8), (10), (11), and (43) that, for a simply supported plate, the boundary conditions for using the SPT are v w wy gj N 1 M 1 M 1 0 4 u w wx gj N 2 M 2 5 s13 s23 0 where j 1; . . . ; J and fN 1 ; M 1 ; M 1 g 
j I XZ i1 zi zi1 zi1 j M2 j

at x 0; a, at y 0; b, 51

at z h=2,

si f1; z; zdij gi;j g dz, 11 15

fN 2 ; M 2 ; M 2 g 

I XZ i1 zi

si f1; z; zdij gi;j g dz. 22 24

(52)

Note that the boundary conditions on s13 and s23 are automatically satised by the derived shear warping functions (see Eq. (4)). The boundary conditions N 1 M 1 M 1 0 at nodes on the boundaries x 0; a and N 2 M 2 M 2 0 at y 0; b are only to satisfy the boundary conditions si 0 at x 0; a and si 0 at 11 22 y 0; b in an average sense. These boundary conditions can be enforced through the use of multiple-point constraints, but it is inconvenient for a displacement-based nite-element method because it involves ux , uy , vx , vy , wxx , wyy , g4x , g4y , g5x , and g5y that are not direct nodal dofs. If the stress boundary conditions are not appropriately implemented, accuracy may loose due to the free-edge effect. However, for a simply supported plate, s11 is close to zero around x 0; a and s22 is close to zero around y 0; b (see Eq. (47)), and hence the free-edge effect is not expected here. Using 10 10 SPE elements the natural frequencies of the simply supported aluminum plate are obtained and also listed in Table 1. The frequencies are almost the same as those obtained from the analytical method shown in Section 4.2 using 3 and 11 layers, respectively. However, o11;1 is 1.5% higher because it is a bendingdominant mode and the polynomial shape functions used in the nite-element formulation makes the plate stiffer. When the number of elements increases to 15 15, o11;1 , o11;2 , and o11;3 reduce to 5:3368, 13:825, and
j j

ARTICLE IN PRESS
184 P. Frank Pai, A.N. Palazotto / Journal of Sound and Vibration 308 (2007) 164189

(a)
0.05 0.04 Y (m)

(b)
0.05 0.04 Y (m) 0.03 0.02 0.01 0 0 0.01 0.02 0.03 x (m) 0.04 0.05 0 0.01 0.02 0.03 x (m) 0.04 0.05 12 (GPa)

(c)
20 0.06 0 0.04
y

0.03 0.02 0.01 0

20 0 0.02
x (m)

0.02 0.04 0.06 0

Fig. 12. Modes 1; 1; 2 and 1; 1; 3: (a) mode shape 1; 1; 2, (b) mode shape 1; 1; 3, and (c) s12 of Mode 1; 1; 3.

23:316, but other o11;k remain almost the same because they are not signicantly coupled with the referenceplane bending. The distributions of S ij are virtually the same as those in Fig. 10. With the use of 11 sublaminates and 11 layers, the total number of dofs is 2736 (after boundary conditions implemented). To reduce the number of dofs in modal analysis, one can use the mode shape of w of Mode 1; 1; 1 to reduce all the dofs of wx; y; t into just one modal coordinate. This reduction does not cause any change of the mode shapes and natural frequencies shown in Table 1. This conrms again that these modes are not sensitive to the reference plane deformation shape. One can even use just one modal coordinate for each of gj and gj to 5 4 reduce more dofs. With this reduction, the total number of dofs reduces from 2736 to 1785, and the obtained natural frequencies have a highest error only 0:6% at o11;9 . Modes 1; 1; 2 and 1; 1; 3 are in-plane TC and TT modes, as shown in Fig. 12. It follows from the equation of E i in Eq. (46) and Table 1 that E 12 S 12 0 for 12 Mode 1; 1; 2 because U V , m n, and W G i G i 0. On the other hand, E 12 and S12 are non-zero 5 4 for Mode 1; 1; 3 and they have large in-plane shears around the four corners, as shown in Fig. 12c. As Table 1 shows, Mode 1; 1; 9 is actually the #473 (or #532 if 3 layers) mode in the nite-element analysis because there are many other low-frequency modes, including many thickness modes. With the use of m; n 1; 1 in the analytical method shown in Section 4.2 and other analytical methods in the literature, many modes with W 0 (i.e., similar to Modes 1; 1; 2 and 1; 1; 3) and having natural frequencies lower than o11;9 are missed. All the nite-element results validate the SPE. Since it is common in high-temperature applications to employ a ceramic top layer as a thermal barrier to a metallic structure, we choose the constituent materials of the FGP for demonstration to be Aluminum (#1) and SiC (#2) having the following material properties: E 1 70 GPa; E 2 427 GPa; n1 0:3; n2 0:17; r1 2702 kg=m3 , r2 3100 kg=m3 . (53)

Many actual applications of plates are constructed in a way similar to a cantilevered plate, as shown in Fig. 1. Moreover, setting up a cantilevered plate for experimental testing is much easier than setting up a simply supported plate. Unfortunately analytical mode shapes of a cantilevered plate do not exist. Hence, we will consider a 10 cm 6 cm 1 cm cantilevered plate to show the modeling capability of the sublamination plate element, the inuences of boundary restraints, and the inuences of material variation through the plate thickness on structural characteristics and the local stress distribution around boundaries. Using a sheardeformable 2D plate theory to analyze a cantilevered 3D solid different sets of boundary conditions may be used by different researchers because actual boundary conditions are affected by the clamping device, the clamping force, and the stiffness ratio of the plate and xture materials. Here we assume the boundary conditions to be u v w wx wy u1 0; y; h=2; t uI 0; y; h=2; t 0 1 1
i g5 0;

at x 0,

i 1; . . . ; I i:e:; si 0 at x a, 13

(m )

ARTICLE IN PRESS
P. Frank Pai, A.N. Palazotto / Journal of Sound and Vibration 308 (2007) 164189 185

(b) 8 (a) 15 z (mm) 10 5 0 0 50 x(mm) (c) 23(x,y,0) (GPa) 100 0 50

6 11, 22 (GPa) 4 2 0 2 0 22(x,y,h/2) 50 x(mm) 0 100 50 ) mm y( 11(x,y,h/2)

100
) m

(m

100

(d) 2 0.2 0 0.2 0.4 0 50 x (mm) 100 0

13(x,y,0) (GPa)

1 0 1 2 0 50 x (mm) 100 0 50 100

100 50
y )

(m

Fig. 13. The rst bending mode: (a) reference-plane mode shape, (b) s11 x; y; h=2 (solid lines) and s22 x; y; h=2 (broken lines), (c) s13 x; y; 0, and (d) s23 x; y; 0.

gi 0; 4

i 1; . . . ; I i:e:; si 0 at y 0; b. 23

y(

(54)

The boundary conditions u1 0; y; h=2; t uI 0; y; h=2; t 0 are implemented through multiple-point 1 1 constraints on dofs of each node on the clamped edge. If the plate material is a uniform mixture of Aluminum and SiC with V 1 V 2 0:5 through the whole thickness (i.e., a pure isotropic plate), Fig. 13 shows the rst exural mode shape of the reference plane and the xy-plane distributions of stresses. Because of the use of shear warping functions, all transverse shear stresses obtained from SPEs are continuous at any point x; y; z in the plates 3D domain, as revealed by Eq. (37). However, just like any nite-element analysis of plates using 3D solid elements, s11 , s22 , and s12 may not be continuous at nodes, and hence the averaged value is presented at each node. We note that the boundary restraint at the clamped end changes the thickness-direction distribution of s13 , as shown in Fig. 14. The transition length can be used to quantify the boundary restraint effect. Note that the distribution of s13 around x 0 is similar to Fig. 7c due to the boundary restraints u1 0; y; h=2 uI 0; y; h=2 0. If the plate material is distributed according to Eq. (16) with n 1:0 and 1 1 z1 h=2, Fig. 15 shows the thickness-direction distributions of s13 x; b=2; z and s11 x; b=2; z at 16 different longitudinal locations. The distributions of s13 and s11 are asymmetric and the highest s13 exists close to the top surface (i.e., the SiC-dominant area). If the plate material is distributed according to Eq. (16) with n 2:0 and z1 h=2, Fig. 16 shows the thickness-direction distributions of s13 x; b=2; z and s11 x; b=2; z at 16 different longitudinal locations. We nd that the upper part takes more share of stresses when the power index n increases. Note that s11 x; b=2; 0 does not show signicant bendingextension coupling. It may be due to the free boundary at x a, s11 a; y; z 0 not being enforced, and/or the use of low-order, bilinear shape functions for ux; y in the nite-element modeling, and this problem requires more studies. Fig. 17 shows the thickness-direction distributions of s13 x; b=2; z and s11 x; b=2; z of the second bending mode. Note that s13 changes sign at a location different from that of s11 . Figs. 1417 show that, for an element away from boundaries and hence free from free-edge and boundary restraint effects, one can easily reduce its dofs without loss of accuracy by using specic ratios between gj and gj , as demonstrated by Figs. 6 and 8. 5 4

ARTICLE IN PRESS
186 P. Frank Pai, A.N. Palazotto / Journal of Sound and Vibration 308 (2007) 164189

(a) 5 z (mm)

5 10 (b) 5 z (mm)

10

20

30 40 50 60 x (mm), 1013(x,b/2,z) (GPa)

70

80

90

100

5 10

10

20

30 40 50 60 x (mm), 11(x,b/2,z) (GPa)

70

80

90

100

Fig. 14. The rst bending mode with V 1 V 2 0:5 through the thickness: (a) s13 x; b=2; z and (b) s11 x; b=2; z.

(a) 5 z (mm)

5 10 (b) 5 z (mm)

10

20

30 40 50 60 x (mm), 1013(x,b/2,z) (GPa)

70

80

90

100

5 10

10

20

30 40 50 60 x (mm), 11(x,b/2,z) (GPa)

70

80

90

100

Fig. 15. The rst bending mode with n 1:0 and z1 h=2 in Eq. (16): (a) s13 x; b=2; z and (b) s11 x; b=2; z.

Moreover, Figs. 16 and 17 show that, away from boundaries, the distribution prole of s13 is the same and it is the one with all gj having the same value (see Eq. (37)). 5 The rst 12 modes of the cantilevered plate are the 1st By (796.38 Hz, Fig. 13a), 1st T x (2779.4 Hz), 1st Bz (3830.9 Hz), 2nd By (4775.1 Hz), 2nd T x (8950.1 Hz), 1st E x (12,320 Hz), 3rd By (12,545 Hz), 2nd Bz (13,402 Hz), 1st By =1st Bx (14,438 Hz), 3rd T x (16,934 Hz), 2nd By /1st Bx (19,605 Hz), and 4th By (23,427 Hz), where By means bending w.r.t. the y-axis, T x means torsion w.r.t. the x-axis, Bz means bending w.r.t. the z-axis, E x means extension along the x-axis, etc. Fig. 18 shows the second to tenth modes and natural

ARTICLE IN PRESS
P. Frank Pai, A.N. Palazotto / Journal of Sound and Vibration 308 (2007) 164189 187

(a) 5 z (mm)

5 10 (b) 5 z (mm)

10

20

30 40 50 60 x (mm), 1013(x,b/2,z) (GPa)

70

80

90

100

5 10

10

20

30 40 50 60 x (mm), 11(x,b/2,z) (GPa)

70

80

90

100

Fig. 16. The rst bending mode with n 2:0 and z1 h=2 in Eq. (16): (a) s13 x; b=2; z and (b) s11 x; b=2; z.

(a) 5 z (mm)

5 10 (b) 5 z (mm)

10

20

30 40 50 60 x (mm), 213(x,b/2,z) (GPa)

70

80

90

100

5 10

10

20

30 40 50 60 x (mm), 0.211(x,b/2,z) (GPa)

70

80

90

100

Fig. 17. The second bending mode with n 2:0 and z1 h=2 in Eq. (16): (a) s13 x; b=2; z and (b) s11 x; b=2; z.

frequencies when the bending mode shape is xed to be the rst bending mode by using one modal coordinate. We nd that all of them are in-plane tension, compression, and/or shearing modes, and the transverse shears, if exist, have the same prole as that shown in Figs. 16a and 17a at locations away from boundaries. Because high-order transverse shears similar to those shown in Fig. 10df only exist at very high frequencies, they can be important for stress analysis and wave propagation studies under impulsive loading, but they would not have signicant inuences on the global surface dynamics of a plate, especially when the plate thickness becomes thin (e.g., a=h; b=h410).

ARTICLE IN PRESS
188
100 80 3830.9Hz y (mm) y (mm) 60 40 20 0 20 0 20 40 60 x (mm) 100 80 y (mm) 60 40 20 0 20 0 20 40 60 x (mm) 100 80 y (mm) 60 40 20 0 20 0 20 40 60 x (mm) 80 100 120 37311Hz y (mm) 100 80 60 40 20 0 20 0 20 40 60 x (mm) 80 100 120 44267Hz y (mm) 80 100 120 27559Hz y (mm) 100 80 60 40 20 0 20 0 20 40 60 x (mm) 100 80 60 40 20 0 20 0 20 40 60 x (mm) 80 100 120 45131Hz 80 100 120 32824Hz y (mm) 80 100 120 60 40 20 0 20 0 20 40 60 x (mm) 100 80 60 40 20 0 20 0 20 40 60 x (mm) 80 100 120 35075Hz 80 100 120

P. Frank Pai, A.N. Palazotto / Journal of Sound and Vibration 308 (2007) 164189
100 80 12320Hz y (mm) 60 40 20 0 20 0 20 40 60 x (mm) 80 100 120 100 80 13402Hz

Fig. 18. The second to tenth modes under the rst bending mode of the cantilevered plate with n 2:0 and z1 h=2 in Eq. (16).

Any temperature variation through the plate thickness will affect the distribution of material properties and hence the dynamic characteristics of FGPs. To examine temperature effects one needs to use, for example, ^ n Eqs. (17) and (13) and consider the fF g in Eq. (23). 5. Concluding remarks We presented a 2D sublamination theory of varying degrees of freedom (dof) for functionally graded plates that can account for layerwise higher-order transverse shear strains, continuity of interlaminar shear stresses, free shear-stress conditions on the bonding surfaces, and deformation-dependent shear warpings. The shear warping functions of a functionally graded plate are shown to depend on the detailed distribution of materials through the thickness, the vibration frequency, and boundary conditions. Moreover, a sublamination plate element of varying dofs is derived based on the plate theory and is numerically validated for performing analysis of complex stress states inside and around the boundaries of a functionally graded plate. The plate element can be used for detailed analysis of thick and thin plates with any boundary conditions. For quick analysis, one can use the theory with less dofs to obtain results equivalent to those from a 2D sheardeformable plate theory. For detailed analysis of stress states one can use the theory with more dofs to obtain results equivalent to those from 3D equilibrium equations.

ARTICLE IN PRESS
P. Frank Pai, A.N. Palazotto / Journal of Sound and Vibration 308 (2007) 164189 189

Acknowledgments The authors appreciate the support provided by the Structural Sciences Center of the Air Vehicles Directorate, US Air Force Research Laboratory, Wright Patterson AFB, Ohio. References
[1] M.B. Bever, P.E. Duwez, Gradients in composite materials, Materials Science and Engineering 10 (1972) 18. [2] T. Hirai, Functionally graded materials, in: R.J. Brook (Ed.), Materials Science and Technology, Vol. 17B of Processing of Ceramics, Part 2, VCH Verlags-gesellschaft mbH, Weinheim, Germany, 1996, pp. 292341. [3] S. Suresh, A. Mortensen, Fundamentals of Functionally Graded Materials, The Institute of Materials, IOM Communications Ltd., London, 1998. [4] J. Aboudi, M.J. Pindera, S.M. Arnold, Higher-order theory for functionally graded materials, Composites Part B-Engineering 30 (1999) 777832. [5] A.J. Markworth, K.S. Ramesh, W.P. Parks Jr., Review modelling studies applied to functionally graded materials, Journal of Materials Science 30 (1995) 21832193. [6] R. Hill, A self-consistent mechanics of composite materials, Journal of the Mechanics and Physics of Solids 13 (1965) 213222. [7] T. Reiter, G.J. Dvorak, Micromechanical modelling of functionally graded materials, in: X. Bahei-El-Din (Ed.), IUTAM Symposium on Transformation Problems in Composite and Active Materials, Kluwer Academic Publishers, London, 1997, pp. 173184. [8] T. Mori, K. Tanaka, Average stress in matrix and average elastic energy of materials with mistting inclusions, Acta Metallurgica 21 (1973) 571574. [9] J. Aboudi, Mechanics of Composite MaterialsA Unied Micromechanical Approach, Elsevier, Amsterdam, 1991. [10] N.J. Pagano, Exact solutions for composite laminates in cylindrical bending, Journal of Composite Materials 3 (1969) 398411. [11] N.J. Pagano, Exact solutions for rectangular bidirectional composites and sandwich plates, Journal of Composite Materials 4 (1970) 2034. [12] S. Srinivas, A.K. Rao, C.V.J. Rao, Flexure of simply supported thick homogeneous and laminated rectangular plates, ZAMM-Journal of Applied Mathematics and Mechanics 49 (1969) 449458. [13] M. Savoia, J.N. Reddy, A variational approach to three-dimensional elasticity solutions of laminated composite plates, Journal of Applied Mechanics 59 (1992) 166175. [14] S.S. Vel, R.C. Batra, Three-dimensional exact solution for the vibration of functionally graded rectangular plates, Journal of Sound and Vibration 272 (2004) 703730. [15] P.F. Pai, A.N. Palazotto, A higher-order sandwich plate theory accounting for 3-D stresses, International Journal of Solids and Structures 38 (2001) 50455062. [16] J.M. Whitney, Structural Analysis of Laminated Anisotropic Plates, Technomic Publishing Company, Inc., Lancaster, Pennsylvania, 1987. [17] R.M. Jones, Mechanics of Composite Materials, second ed., Taylor & Francis, Inc., Philadelphia, PA, 1999. [18] A.H. Nayfeh, P.F. Pai, Linear and Nonlinear Structural Mechanics, Wiley Interscience, New York, 2004. [19] K. Washizu, Variational Methods in Elasticity & Plasticity, third ed., Pergamon Press Inc, New York, 1982. [20] P.J. Davis, P. Rabinowitz, Methods of Numerical Integration, second ed., Academic Press, New York, 1984. [21] P.F. Pai, A new look at shear correction factors and warping functions of anisotropic laminates, International Journal of Solids and Structures 32 (1995) 22952313. [22] J.D. Renton, Generalized beam theory applied to shear stiffness, International Journal of Solids and Structures 27 (1991) 19551967. [23] E.J. Barbero, J.N. Reddy, J.L. Teply, An accurate determination of stresses in thick laminates using a generalized plate theory, International Journal for Numerical Methods in Engineering 29 (1990) 114. [24] A.K. Noor, W.S. Burton, Assessment of shear deformation theories for multilayered composite plates, Applied Mechanics Reviews 42 (1989) 113.