Available online at www.sciencedirect.

com

ScienceDirect

Comput. Methods Appl. Mech. Engrg. 315 (2017) 1–24

www.elsevier.com/locate/cma

Geometrically non-linear static analysis of functionally graded

material shells with a discrete double directors shell element
A. Frikha ∗ , F. Dammak
Mechanical Modeling and Manufacturing Laboratory (LA2MP), National School of Engineers of Sfax, B.P 1173-3038, Sfax,
University of Sfax, Tunisia

Received 4 July 2016; received in revised form 5 October 2016; accepted 6 October 2016
Available online 26 October 2016

Highlights
• Geometrically nonlinear shell theory of functionally graded material.
• 3D-shell model based on a discrete double directors.
• The third-order shear deformation theory.
• Remove the shear correction factor.

Abstract

A general shell model, including both theories of thin and thick shells, Kirchhoff–Love and Reissner–Mindlin undergoing
finite rotations is presented. Based on Higher-order shear theory, where the fiber is cubic plane, the developed model does not
need any transverse shear coefficients. The implementation is applicable to the analysis of isotropic and functionally graded shells
undergoing fully geometrically nonlinear mechanical response. Material properties of the shells are assumed to be graded in the
thickness direction according to a simple power-law and sigmoid distribution. The accuracy and overall robustness of the developed
shell element are illustrated through the solution of several non trivial benchmark problems taken from the literature. The effect of
the material distribution on the deflections and stresses is analyzed.
⃝c 2016 Elsevier B.V. All rights reserved.

Keywords: Functionally graded shells; Finite element analysis; Geometrically nonlinear shell theory; Double director vectors; Third-order theory

1. Introduction

Shells are widely used in various mechanical structures, civil engineering, aerospace and naval. These structures
are more and more replaced by composites because of their superior mechanical properties. First, the abrupt change
of the properties across the interface between different materials in conventional composite material is the source of
cracks. Second, the presence of residual stresses due to the difference in coefficient of thermal expansion of different
materials in conventional composite generates a decrease of the lifetime. Third, conventional composites are made to

∗ Corresponding author.
E-mail address: frikhaahmed@yahoo.fr (A. Frikha).

http://dx.doi.org/10.1016/j.cma.2016.10.017
0045-7825/⃝ c 2016 Elsevier B.V. All rights reserved.
2 A. Frikha, F. Dammak / Comput. Methods Appl. Mech. Engrg. 315 (2017) 1–24

support a moderate temperature. To overcome these thermo-mechanical disadvantages, special kind of composites,
known as functionally graded materials (FGMs) with a gradual transition of material properties from one material to
another, are made.
The analysis of classical shell structures is based on four kinematic assumptions, which are membrane, Kirchhoff–
Love, Reissner–Mindlin and the refined model. Membrane theory, where the bending and shear strain are neglected, is
not applicable to thin flexible structures. Kirchhoff–Love theory, where the shear strain are assumed zero, is not accept-
able for composite shell [1,2]. Reissner–Mindlin theory gives a correct overall assessment. Nevertheless, it does not
allow a good analysis through the thickness (constant shear). Another disadvantage of the Reissner–Mindlin theory is
the need to introduce the transverse shear coefficients. Shear coefficients are easy to obtain for linear isotropic material
(5/6). Nevertheless, it appears complex for composites mainly for non-linear behavior law. Some evaluation methods
of shear correction coefficients were presented in [3–5]. To expand the application domain of the first-order models and
avoid the shear correction coefficients associated to Reissner–Mindlin theory, higher-order theories were developed.
The linear mechanical behavior of FGM plates was analyzed by several researchers. Based on the third-order shear
deformation of plate theory, the static behavior of FGM plate was investigated in [6]. The static and free vibration of
plates were studied using the classical plate theory and Fourier series expansion [7]. By using the Reddy’s third order
plate theory, the buckling of FGM plates was performed in [8]. The vibration of FGM plates was investigated by using
the ordinary differential equations coupled with the power series method [9,10].
More general models were developed to analyze the linear mechanical behavior of shell structures. Typically
functionally graded shell structures were presented with shear deformation theories by using the first-order shear
deformation theory (FSDT) [11,12]. Higher-order shear deformation theory through the FGM shell thickness is
developed by numerous researchers [13,14]. Two approaches can exist in a refined kinematic model of plate and
shell, which are layer-wise theory [15,16] and single layer theory [17]. The first model is solid, nevertheless it needs
2N + 3 independent kinematic variables for N layers, which is expensive in computation time. Single layer theory
consists of an approximation of the displacement field by a series expansion in the vicinity of the middle surface.
Recently, a unified formulation developed by Carrera and his co-workers (CUF), which can generate any refined
theory, is developed in static and free vibration for laminate composites and FGM shells [18,19]. A sampling surfaces
method is developed by Kulikov and his co-workers to analyze FGM shells [20–22].
Because of the high modulus and high strength properties that FGM have, FGM shells undergo large deflection
and rotation before the inelastic behavior. Therefore, an accurate prediction of geometrically non-linear behavior is
required. Even if theories and formulations of geometrically non-linear analysis are numerous, a good representation
of finite rotation of shells is a dominant factor.
The geometrically non-linear of shells is largely used for isotropic and composite laminates materials [23,24].
Applied to FGM, the studies of non-linear behavior of plates and shells were limited generally to a Von-Karman
assumption that only includes membrane forces, which is restricted to moderately small deformations [6,25]. Recently,
a good representation of finite rotation in FGM shells is presented in [26,27].
The present work is based on a double directors shell model, in which the fiber is cubic plane [28–30].
This shell model, taking into account the exact geometric assumptions, includes the classical models: membrane,
Kirchhoff–Love and Reissner–Mindlin. This refined theory is known for plates as high-order shear theory. In this
work, a single layer theory is proposed based on curved shell model with a Cartesian description.
This paper is organized as follows. First, kinematic strain of a double directors shell model is described in Section 2.
The weak form is introduced in Section 3. The details of the finite element formulation are given in Section 4. The
materials properties of FGM shell used in the present paper are presented in Section 5. Numerical results are illustrated
in Section 6. Finally in Section 7, conclusions are drawn.

2. Kinematics of a double directors shell model

To distinguish the initial configuration C0 from the deformed Ct , capital letters (resp. lowercase letters) are used
for quantities relative to the configuration C0 (resp. Ct ).

2.1. Parametrization of the reference surface

The reference surface in an arbitrary configuration, initial or deformed, is described by the same parametric
coordinates ξ 1 and ξ 2 . The position vector of an arbitrary point ( p) of the mean reference surface, in the initial
 A. Frikha, F. Dammak / Comput. Methods Appl. Mech. Engrg. 315 (2017) 1–24 3

configuration C0 is defined by its components in the Cartesian global basis (Ei ):

X p = X i (ξ 1 , ξ 2 )Ei , (ξ 1 , ξ 2 ) ∈ Aξ ⊂ R 2 , (1)
where Aξ is the parametric surface. A differential element at point p is given by:

dX p = Aα dξ α , Aα = X p,α , α = 1, 2. (2)
The vectors of the covariant basis A1 and A2 are tangent at coordinate lines ξ 1 and ξ 2 . The square of the incremental
element length is given by:

d S 2p = dX p · dX p = Aαβ dξ α dξ β , Aαβ = Aα · Aβ . (3)

d S 2p is known as the first fundamental form (quadratic form) and the coefficients Aαβ are the components of the
covariant metric tensor A. The unit normal vector at the mean reference surface, in the initial configuration C0 is given
by:
A1 ∧ A2
N= , Aα · N = 0, ∥N∥ = 1. (4)
∥A1 ∧ A2 ∥
The parametric basis (A1 , A2 , N) is covariant at the point ( p) of the reference surface in the initial configuration. Two
unit vectors (T1 , T2 ), located in the tangent plane can be defined at any point on the middle surface. These vectors,
with the normal vector N, make an orthonormal basis (T1 , T2 , N). This basis can be obtained from the unit vectors N
and E3 :
N = Q · E3 , (5)
where Q = Tα ⊗ Eα + N ⊗ E3 .
The position vector at any point ( p), of the reference mean surface in the deformed configuration Ct , is defined by
its components in the Cartesian basis (Ei ):

x p = x p (ξ 1 , ξ 2 ) = xi (ξ 1 , ξ 2 )Ei . (6)
A differential element at point ( p) is given by:
dx p = aα dξ α , aα = x p,α , α = 1, 2, (7)
where a1 and a2 are the covariant vectors. The square of the incremental element length is given by:

ds 2p = dx p · dx p = aαβ dξ α dξ β , aαβ = aα · aβ . (8)

The coefficients aαβ are the covariant components of metric tensor a.

2.2. General kinematic shell assumptions

The 3D position of any point (q) of the structure is obtained by a power series expansion in the vicinity of the
reference mean surface:
n

xq (ξ 1 , ξ 2 , z) = x p (ξ 1 , ξ 2 ) + z k dk (ξ 1 , ξ 2 ), (9)
k=1

where dk (k = 1, 2, . . . , n) are the n unit vectors field at each point of the surface, referred to as the directors field. For
n = 0, corresponding to membrane theory, the shell is represented by the mean surface. The case n = 1, corresponding
to single director shell model or first order theory, includes the Kirchhoff–Love and Reissner–Mindlin assumptions.
The case n = 2 leads to two director vectors, which allow to have a better representation of transverse shear strain
and remove the correction factor associated to Reissner–Mindlin theory.
4 A. Frikha, F. Dammak / Comput. Methods Appl. Mech. Engrg. 315 (2017) 1–24

Fig. 1. Kinematic description of the initial configuration.

2.3. Initial configuration

In the initial configuration, the shell geometry description is done with a single vector. The position of any point
(q) of shell domain is given by (show Fig. 1):

Xq (ξ 1 , ξ 2 , z) = X p (ξ 1 , ξ 2 ) + zD(ξ 1 , ξ 2 ). (10)
The covariant basis at point (q) is obtained from the position vector by (G1 , G2 , G3 ) = (∂Xq /∂ξ 1 , ∂Xq /∂ξ 2 , ∂Xq /
∂z), which yields in base vectors relative to the initial state:

Gα = Aα + zD,α , α = 1, 2

(11)
G3 = D.
In the initial configuration, the covariant components of metric tensor become:

G αβ = Aαβ + z Bαβ + z 2 Cαβ


(12)
G 3α = ωα0 + z Bα , G 33 = λ20 ,

where the kinematic variables Aαβ , Bαβ , Cαβ , ωα0 , Bα and λ20 are given by:

Aαβ = Aα · Aβ , Bαβ = Aα · D,β + Aβ · D,α , Cαβ = D,α · D,β


(13)
ωα0 = Aα · D, Bα = D · D,α = λ0 λ0,α , λ20 = D · D,

where Aαβ , Bαβ and Cαβ are the curvature tensors. ωα0 is initial shear term. Bα and λ20 are the variation of the thickness
in the initial configuration. With the assumption of a constant thickness along the element, the metric tensor becomes:

G αβ = Aαβ + z Bαβ + z 2 Cαβ


(14)
G 3α = ωα0 , G 33 = 1.

2.4. Deformed configuration

To force a particular behavior of the shell along the thickness, some assumptions must be introduced. The first
assumption, known as the plane curve fiber, reflects a higher-order distribution of the position vector in the deformed
configuration. This assumption is based on double director shell model. In the second assumption, known as cubic
fiber, it is assumed that the position vector is cubic in function of the thickness variable z. The third assumption
consists on imposing a zero shear stress at the top and bottom surfaces.
The fiber, which is assumed initially straight, becomes curve after deformation. This plane curve is defined in a
plane formed by two director vectors d1 and d2 . These vectors are initially identical and equal to D. Vectors d1 , d2
and D are unit if the thickness is assumed constant. The angle between the both director vectors d1 and d2 , noted by
γ = (d1 , d2 ), is assumed small. This assumption makes neglect the thickness variation viewed through the both fibers
 A. Frikha, F. Dammak / Comput. Methods Appl. Mech. Engrg. 315 (2017) 1–24 5

Fig. 2. Curved fiber assumptions.

(d1 , d2 ). The curve fiber is tangent to one from both director vectors (d1 ), in the vicinity of the reference surface. With
these geometric assumptions, the real position (q) after deformation is given by (show Fig. 2):
xq = x p + zd3 , (15)
where d3 is located in the fiber plane defined by d1 and d2 . The vector d3 , which has the same modulus as d1 and
d2 , can be obtained by a rigid rotation of the vector d1 of an angle α(z) around the unit vector e, normal to the plane
(d1 , d2 ):
d3 = Q(α) · d1 , (16)
where Q(α) is given by:
d1 ∧ d2
Q(α) = ex p(α(z) · e), e = . (17)
∥d1 ∧ d2 ∥
e is the skew-symmetric tensor such that e · e = 0.
The rotating matrix Q is written as:

Q = I + sin(α)e + (1 − cos(α))e2 . (18)

From Eqs. (16)–(18), d3 becomes:
d3 = cos(α)d1 + sin(α)y, (19)
where the vector y is defined by:
(d1 ∧ d2 ) ∧ d1
y = e ∧ d1 = . (20)
∥d1 ∧ d2 ∥
According to double vector product formula, Eq. (20) becomes:
1 ∥d1 ∥
y= [r12 d2 − cos(γ )d1 ] , r12 = . (21)
sin(γ ) ∥d2 ∥
With the assumption ∥d1 ∥ ≈ ∥d2 ∥, the ratio r12 is equal to 1. Using Eqs. (16), (19) and (21), the position vector xq ,
Eq. (15), can be written in the following form:
xq = x p + f 1 (z)d1 + f 2 (z)d2 , (22)
6 A. Frikha, F. Dammak / Comput. Methods Appl. Mech. Engrg. 315 (2017) 1–24

where the form functions f 1 (z) and f 2 (z) are defined by:

f 1 (z) = z cos(α(z)) − cos(γ ) f 2 (z)


(23)
f 2 (z) = z sin(α(z))/ sin(γ ).
The angle γ is defined by:

d1 · d2

cos(γ ) = , sin(γ ) = 1 − cos2 (γ ). (24)
∥d1 ∥ · ∥d2 ∥
Assuming that the variation of the angle γ with the parametric coordinates ξ α is neglected (γ,α ≈ 0), the vectors
of the covariant basis associated to the geometric description becomes:

gα = aα + f 1 (z)d1,α + f 2 (z)d2,α , g3 = f 1′ (z)d1 + f 2′ (z)d2 . (25)

The angle γ is not dependent on the thickness variable z. Using Eq. (23), the derivatives of f 1 and f 2 are given by:

f 1 (z) = cos(α) − zα ′ sin(α) − f 2′ (z) cos(γ )

 ′
(26)
f 2′ (z) = (sin(α) + zα ′ cos(α))/ sin(γ ).

Assuming the director vector d1 is tangent to the fiber at the mean surface (g3 (0) = d1 ), f 1′ (0) and f 2′ (0) become:

f 1′ (0) = 1, f 2′ (0) = 0. (27)

From Eq. (26), α(z) is zero at the origin (z = 0). Under the assumption of small angle α, the functions f 1 and f 2 and
those derivatives can be simplified as:

f 1 (z) = z − cos(γ ) f 2 (z) f 1 (z) = 1 − cos(γ ) f 2′ (z)

  ′
(28)
f 2 (z) = zα/ sin(γ ), f 2′ (z) = (α + zα ′ )/ sin(γ ).

The angle α, used in the expression of f 1 , f 2 , f 1′ and f 2′ , can be expressed as a power series expansion of the
thickness variable z. The second hypothesis assumes a cubic function of the deformed fiber. Since the angle α is zero
at the origin, α(z) becomes:
α(z) = z(α1 + zα2 ). (29)

To reduce or remove the kinematic variables α1 and α2 from Eq. (29), some kinematic assumptions must be
imposed. Contrary to Başar et al. [24], which a zero strain is assumed in the thickness (E 33 = 0), the transverse shear
strain on the top and bottom shell edges is assumed zero. Using Eq. (25), the shear components of the metric tensor
can be written in the deformed configuration:

gα3 = gα · g3 = f 1′ aα · d1 + f 2′ aα · d2 + ( f 1 d1,α + f 2 d2,α ) · ( f 1′ d1 + f 2′ d2 ). (30)

The transverse shear strain is given by:
γα = 2E α3 = gα3 − G α3 = f 1′ aα · d1 + f 2′ aα · d2 − Aα · D
+ ( f 1 d1,α + f 2 d2,α ) · ( f 1′ d1 + f 2′ d2 ) − zD,α · D. (31)

These strains can be written in the third order as follows ( f 1′ = 1 − cos(γ ) f 2′ ):

γα ≈ f 1′ γ1α + f 2′ γ2α + zχα + O(z 3 ), (32)

where γ1α , γ2α and χα are given by:

γ1α = aα · d1 − Aα · D


γ2α = aα · d2 − cos(γ )Aα · D ≈ aα · d2 − Aα · D (33)

χα = d1,α · d1 − D,α · D,

 A. Frikha, F. Dammak / Comput. Methods Appl. Mech. Engrg. 315 (2017) 1–24 7

where the assumption cos(γ ) ≈ 1. It is assumed that the director vector d2 does not induce shear strain (Kirchhoff–
Love), the strain γα and γ2α can be written as:
γα = f 1′ γ1α + zχα , γ2α = 0. (34)
Replacing Eqs. (28)–(29) into Eq. (34), γα becomes:

γα = (1 − (2α1 z + 3α2 z 2 ) cos(γ )/ sin(γ ))γ1α + zχα . (35)

The zero shear in the top and bottom of the shell can be written as:
   
h h
γα z = − = γα z = = 0. (36)
2 2
From Eq. (35), α1 , α2 and α become:

χα tg(γ ) 4tg(γ ) 4z 2 4z 2 tg(γ )

 
zχα
α1 = , α2 = , α= + 2 tg(γ ) ≈ . (37)
2γ1α 3h 2 2γ1α 3h 3h 2
Since χα measures a small strain. (χα is zero for a constant thickness), the term zχα /(2γ1α ) is neglected in front
4z 2 /(3h 2 ). The functions f 1 and f 2 become:
4 3 4
f 1 (z) = z − z , f 2 (z) = 2 z3. (38)
3h 2 3h cos(γ )
The shear strain angle γ , formed by d1 and d2 , is assumed small and the term cos(γ ) in f 2 (z) is equal to 1. The
position vector of any shell point, defined by Eq. (22), will be considered with the functions f 1 (z) and f 2 (z) given
by Table 1. The double directors model is an expansion of Reissner–Mindlin model, which include Kirchhoff–Love
model. It is noticed that Reissner–Mindlin, Kirchhoff–Love and double director models have the same relations:
f 1 (z) + f 2 (z) = z, f 1′ (z) + f 2′ (z) = 1. (39)
The general model with functions f 1 and f 2 , defined in Table 1, can be reduced in linear case to obtain plate model
by projection in the tangent plane. The following displacement field is obtained:

z3 4z 3
  
u = u 0 + z − 4
 β 1 + α1
3h 2 3h 2



z3 4z 3 αk = −w,k , k = 1, 2, (40)
 
 v = v 0 + z − 4 2
β 2 + 2
α2
3h 3h



w = w0 ,


where u 0 , v0 and w0 are the mean surface displacement. β1 and β2 are the rotations and α1 and α2 are the derivative
of transverse displacement w. This displacement field is initially proposed in linear case [31–33]. Using an energy
approach, other functions f 1 (z) and f 2 (z) were investigated in [34,35] (Table 2).

2.5. Strain measure

Using the kinematic assumptions, the Lagrangian strain E can be written as follows [28,29]:

E = eαβ + f 1 (z)χαβ 1
+ f 2 (z)χαβ
2
 αβ


1 1 2E α3 = f 1 (z)γα + f 2 (z)γα
′ 1 ′ 2
E = (g − G), E i j = (gi j − G i j ), (41)
2 2 1  ′
f 1 + f 2′ d − 1 ,
  
 E 33 =

2
where eαβ , χαβ
k and γ k denote the membrane, the bending and the shear strains, which can be computed as:
α

1
δeαβ = aαβ − Aαβ , γαk = cαk − Cαk
 
2 k = 1, 2, (42)
1 k 

 χαβ
k
= k
bαβ − Bαβ ,
2
8 A. Frikha, F. Dammak / Comput. Methods Appl. Mech. Engrg. 315 (2017) 1–24

Table 1
Shell theories functions f 1 and f 2 .

Theory Membrane Kirchhoff–Love Reissner–Mindlin Double director model

f 1 (z) 0 0 z z − 4z 3 /(3h 2 )
f 2 (z) 0 z 0 4z 3 /(3h 2 )

Table 2
Shell theories functions f 1 and f 2 .

Theory f 1 (z) f 2 (z)

Levinson [31], Bickford [32], Reddy [33] z − f 2 (z) 4z 3 /3h 2
Murthy [34], Shi [35] z − f 2 (z) −z/4 + 5z 3 /3h 2

k , B k , ck , C k and d are defined by:

where aαβ , bαβ αβ α α

aαβ = aα · aβ , bαβ k k k
, Bαβ
k
= Aα · Dk,β + Aβ · Dk,α

= aα · d,β + aβ · d,α
(43)
cαk = aα · dk , Cαk = Aα · Dk , d = d1 · d1 , k = 1, 2.
In matrix notation, the membrane, bending and shear strains vectors are given by:

χ11
   k
e11  
γ1k
e =  e22  , χ =  χ22
k k
, γ =
k
, k = 1, 2. (44)
 
2e12 k γ2k
2χ12
Vanishing of the transverse shear stress on the top and bottom shell faces, σ13 (±h/2) = σ23 (±h/2) = 0, the shear
strain can be obtained as follows:
γα2 = 0, 2E α3 = f 1′ (z)γα1 . (45)
This kinematic constraint will be imposed in a discrete form in the finite element approximation.
The generalized strain and virtual strain vectors Σ and δΣ are defined by:

δe
   
e
χ 1  δχ 1 
 
0
Σ =  2 δΣ =  2  γ =
2
, (46)
χ δχ
   
0
γ 1 11×1 δγ 1 11×1
where the virtual strains can be written as:
δeαβ = 1/2(aα · δx,β + aβ · δx,α ), δγαk = aα · δdk + δx,α · dk

k = 1, 2. (47)
δχαβ
k
= 1/2(aα · δdk,β + aβ · δdk,α + δx,α · dk,β + δx,β · dk,α ),

3. Weak form and linearization

Using the total Lagrangian formulation, the weak form of equilibrium equations is given by:

G= S i j δ E i j d V − G ext = 0, (48)
V

where d V is the shell volume element in the initial configuration, δ E i j are the covariant components of the virtual
Green–Lagrange strain tensor, S i j are the contravariant components of the second Piola–Kirchhoff stress tensor and
G ext is the external virtual work. Performing the integration through the thickness of the shell, and using Eqs. (41),
 A. Frikha, F. Dammak / Comput. Methods Appl. Mech. Engrg. 315 (2017) 1–24 9

(44) and (45), the weak form becomes:

  2  

G= N · δe + Mk · δχ + T1 · δγ d A − G ext = 0,
k 1
(49)
A k=1

where δe, δχ k and δγ 1 are the variations of shell strains. N, Mk and T1 are the membrane, bending and shear stress
resultants, which can be written in matrix form as:
 11   11 
N Mk  
T1
22 , Mk =  Mk  , T1 = 12 , k = 1, 2.
 22 
N= N   (50)
T1
N 12 Mk 12

These components are defined as follows:

 h/2  h/2  h/2
αβ αβ αβ αβ
N = σ dz, Mk = f k (z)σ dz, T1α = f 1′ (z)σ α3 dz. (51)
−h/2 −h/2 −h/2

The generalized resultant of stress R is defined by:

T
.

R= N M1 M2 T1 (52)

The weak form of the equilibrium equation can then be rewritten as:

G(Φ, δΦ) = δΣ T · Rd A − G ext (Φ, δΦ) = 0 (53)
A

where Φ = (u, d1 , d2 ) is the displacement and directors vectors. G ext consists of the conservative boundary traction
and surface pressure load. The details of the finite element approximation of G ext , which exists in the literature of
shell element, is not detailed in the present manuscript.
Eq. (53) describes the nonlinear shell model, which can be solved by the Newton iterative algorithm. The consistent
tangent operator for the Newton solution procedure can be constructed by the directional derivative of the weak form
in the direction of the increment ∆Φ = (∆u, ∆d1 , ∆d2 ). To analyze large displacement, the tangent stiffness should
be composed of material and geometric stiffness matrices, denoted by D M G · ∆φ and DG G · ∆φ, respectively:
DG · ∆Φ = DG G · ∆Φ + D M G · ∆Φ. (54)

3.1. Material part

The material part of the tangent operator results from the variation in the stress resultants and thus takes the form:
  
D M G · ∆Φ = δΣ T · ∆R d A. (55)
A

The material tangent modulus is expressed as:

 
H11 H12 H13 0
H12 H22 H23 0 
∆R = HT ∆Σ , HT =  , (56)
H13 H23 H33 0 
0 0 0 H44
where:
 h/2  
(H11 , H12 , H13 , H22 , H23 , H33 ) = 1, f 1 , f 2 , f 12 , f 1 f 2 , f 22 Hdz. (57)
−h/2
10 A. Frikha, F. Dammak / Comput. Methods Appl. Mech. Engrg. 315 (2017) 1–24

In Eq. (56), H44 is defined by:

 h/2  ′ 2
H44 = f 1 Hτ dz, (58)
−h/2

where H and Hτ , in Eqs. (57) and (58), are in plane and out-of-plane linear elastic sub-matrices, which can be
expressed in a Cartesian system as:

ν(z)
 
1 0  
E(z)  E(z) 1 0
H= ν(z) 1 0  , Hτ = , (59)
1 − ν 2 (z) 0 0 (1 − ν(z))/2 2(1 + ν(z)) 0 1

where E(z) and ν(z) are the Young’s modulus and the Poisson’s ratio, respectively.

3.2. Geometrical part

The geometrical part results from the variation of the virtual strain by maintaining constant the stress resultants:
  
DG G · ∆Φ = ∆δΣ T · R d A. (60)
A

This expression can be decomposed in membrane, bending and shear terms as follows:
DG G · ∆Φ = DG G m · ∆Φ + DG G b1 · ∆Φ + DG G b2 · ∆Φ + DG G s1 · ∆Φ. (61)

The weak form Eq. (53), the material part Eq. (55) and the geometrical part Eq. (61) are developed after the finite
element approximation.

4. Finite element approximation

In this section, the numerical implementation of the presented shell theoretical formulation based upon a four node
shell element is established. Using the isoparametric concept, the mid-surface (X) and surface displacement field
(u = x − X) are approximated by:
4
 4

u= N I u I , ∆u = N I ∆u I , (62)
I =1 I =1

where N I are the standard bilinear isoparametric shape functions. The first director vector d1 is approximated with
the same functions:
4
 4

δd1 = N I δd1I , ∆d1 = N I ∆d1I . (63)
I =1 I =1

For the variation and increment of the second director vector d2 , we choose a quadratic interpolation as the same one
proposed in [28,36]:
4
 8
 4
 8

δd2 = N I δd2I + PK δα K t K , ∆d2 = N I ∆d2I + PK ∆α K t K , (64)
I =1 K =5 I =1 K =5

where (I ) and (K ) represent a node of the element and the mid-point of the element boundaries, respectively. δα K
are variables associated to δd2 on the element boundaries and are given as in [28,36]. The vector t K is unit and its
direction is defined by the position of the nodes couple (I, J ) [28]. The variables δα K will be replaced by introducing
the constraint Eq. (45) over the element boundaries under integral form. The shape functions PK are quadratic and are
given in Table 3.
 A. Frikha, F. Dammak / Comput. Methods Appl. Mech. Engrg. 315 (2017) 1–24 11

Table 3
Shape functions PK for quadri-
lateral element.
P5 = 0.5(1 − ξ 2 )(1 − η)
P6 = 0.5(1 − η2 )(1 + ξ )
P7 = 0.5(1 − ξ 2 )(1 + η)
P8 = 0.5(1 − η2 )(1 − ξ )

Table 4
Strain–displacement matrices.
– Membrane
 I 
n1T N ,1
 
I
I = BI 0 , Bmm
I
 
Bm mm 0 =
 n2T N ,2 

I I
n1T N ,2 + n2T N ,1
– First bending  
T NI  I 
d1,1 ,1 n1T N ,1
   
B1I = B1m
I I I = T NI I = I 

B1b 0 , B1m d1,2 , B1b n2T N ,2 


 ,2   
T N I + dT N I n T N I + nT N I
d1,1 ,2 1,2 ,1 1 ,2 2 ,1
– Shear
Bs = J0−1 Bsξ
 1 T 2 aT 2 dT 2 aT 3 dT 3 aT 4 dT 3 aT

N,1 d1B N,1 1B 0 N,1 1B N,1 1B 0 N,1 1D N,1 1D 0 N,1 1D N,1 1D 0
Bsξ = 1 dT 4 aT 2 dT 3 aT 3 dT 3 aT 4 dT 4 aT
N,2 1A N,2 2A 0 N,2 1C N,2 2C 0 N,2 1C N,2 2C 0 N,2 1A N,2 2A 0
– Second
 bending [28,36]

B2I = B2m
I 0 I
B2b
T N , I +nT M I n1T · Mr,1
I
   
d2,1 1 1 d,1
I = T N , I +nT M I n2T · Mr,2
I
  I  
B2m  d2,2 2 2 d,2
, B = 
 2b 


T N , I +nT M I + d T N , I +nT M I
d2,1 n1T · Mr,2
I + nT · M I
2 1 d,2 2,2 1 2 d,1 2 r,1
I + P td I , td I = 3 t ⊗ d , M I = N I I + P tt I + P tt I , tt I = 3 t ⊗ t .
MdI = PK td K M M K 2L K K K r K K M M K 4 K K

Taking into account the finite element approximation, the membrane, bending and shear strain components
become:

δe = Bm δΦ n , δχ k = Bk · δΦ n , δγ 1 = Bs δΦ n , k = 1, 2, (65)

where Bm , Bk and Bs are the discrete strain–displacement matrices given in Table 4. For the shear strain, the assumed
natural transverse shear strain method of Bathe and Dvorkin [37] is considered. It is important to mention that in
Table 4 the shear matrix is given for all nodes. Nevertheless, the membrane and bending matrices are given at node I .
A typical isoparametric finite element is considered as depicted in Fig. 3. A, B, C and D denote the mid-points of the
element boundaries set.
The virtual and incremental generalized strains are expressed as follows:

δΣ = BδΦ, ∆Σ = B∆Φ, (66)

where B is expressed as:

 
Bm
B f 1 
B= B f 2  . (67)


Bs
12 A. Frikha, F. Dammak / Comput. Methods Appl. Mech. Engrg. 315 (2017) 1–24

Fig. 3. Assumed strain construction of the isoparametric shell element.

The weak form Eq. (53) leads to the following internal forces vector:

r= BT Rd A. (68)
A
The material tangent operator is deduced from Eqs. (55) and (56):

Km = BT HT Bd A. (69)

In matrix form, the geometric tangent operator, based on Eq. (61), becomes:

DG G · ∆Φ = δΦnT KGm + KGf1 + KGf2 + KGc1 ∆Φn .

 
(70)
Membrane, bending and shear contributions can be grouped to form the geometric tangent operator:

DG G · ∆Φ = δΦnT (KG ) ∆Φn . (71)

The global geometric tangent operator is detailed in the Appendix. Note that KG is symmetric.

4.1. Local cartesian coordinates

The Jacobian transformation J0 from basis n01 , n02 to (A1 , A2 ) is expressed as:
 

 
n01 · A1 n02 · A1
J0 = 0 . (72)
n1 · A2 n02 · A2
I I
The derivatives of shape functions N ,1 and N ,2 (Table 4) are given by:
 I  
I
N ,1 −1 N,ξ
I = [J 0 ] . (73)
N ,2 N,ηI
The unit vectors of the actual basis (n1 , n2 ) in Table 4 are expressed as:
 I
n1 = N ,1 (X + u) I ,
 I (74)
n2 = N ,2 (X + u) I .

4.2. Nodal transformation

The relationship between

 the  incremental director vectors ∆dk and the incremental rotation is given by ∆dk =
∆Θk1
Λk ∆Θ k , where ∆Θ k = 2 and k = 1, 2. For more details, the reader can refer to [23]. The nodal transformation
∆Θk
 A. Frikha, F. Dammak / Comput. Methods Appl. Mech. Engrg. 315 (2017) 1–24 13

at a particular node (I ) is expressed as:

 
∆u
∆Φ I = ∆d1  = T I ∆U I , (75)
∆d2 I

where T I and ∆U I are expressed as:

   
I 0 0 ∆u
T I = 0 Λ1I 0 , ∆U I = ∆Θ 1  . (76)
I ∆Θ 2 7×1
0 0 Λ2 9×7

This transformation leads to an element with seven-degrees-of-Freedom per node. The matrices Λ1I and director
vectors d1I at nodes, at integration points and at mid-side for the ANS method, are done following [23]. However, the
matrices Λ2I and director vectors d2I at nodes, at integration point and mid-sides for the discrete constrain are done
following [36].

5. Functionally graded materials

The functionally graded materials can be produced by continuously varying the constituents of multi-phase
materials in a predetermined profile. These material properties are assumed to vary through the shell thickness
according to the rule of mixture. Most researchers use the power-law (P-FGM) or sigmoid functions (S-FGM) to
describe the volume fractions. This article also uses FGM plates and shells with power-law or sigmoid function.
The material properties of the P-FGM are assumed to obey to a power-law function written as:
P(z) = (Pc − Pm )Vc + Pm , (77)
where P(z), Pm and Pc denote respectively the effective material property, the properties of the metal and ceramic.
Vc is the volume fraction, which varies according to the two general four-parameter power-laws distribution [38].
 p
z c
   
1 z 1
FGM I (a, b, c, p) : Vc (z) = 1 − a + +b + , (78)
2 h 2 h
 p
z c
   
1 z 1
FGM II (a, b, c, p) : Vc (z) = 1 − a − +b − , (79)
2 h 2 h
where p is the power-law index. The parameters a, b and c determine the material variation profile through the
functionally graded shell thickness. Instead of the rule of mixture, Eqs. (77)–(79), another analytical method for
estimating the effective properties can be used as Mori–Tanaka or self-consistent methods [10].
The volume fraction of sigmoid function (S-FGM) is defined by:
1 h/2 − z p
 
h
V f (z) = 1 −
1
, 0≤z≤ ,
2 h/2 2
 p (80)
1 h/2 + z −h
V f2 (z) = 1 − , ≤ z ≤ 0.
2 h/2 2
The material properties of S-FGM can be calculated using the rule mixture as:
  h
P(z) = V f1 (z)Pm + 1 − V f1 (z) Pc , 0 ≤ z ≤ ,
2 (81)
  −h
P(z) = V f (z)Pm + 1 − V f (z) Pc ,
2 2
≤ z ≤ 0.
2

6. Numerical examples

The applicability of the proposed theory and the performance of the finite element implementation are assessed by
numerical simulation including large rotation. Numerical examples are divided into two classes: two isotropic shell
14 A. Frikha, F. Dammak / Comput. Methods Appl. Mech. Engrg. 315 (2017) 1–24

Fig. 4. Description of the spherical shell geometry.

Fig. 5. Center deflection vs. load parameter for isotropic hemisphere shell under point load.

examples and three functionally graded structures. In isotropic structure examples, we focus on applications with
warped elements and on the finite rotation capability. The FGM structures are simulated to analyze its mechanical
behavior. The results are compared with several formulations from the literature. The used laws distribution of the
FGM shells are the classical power-law P-FGM and sigmoid S-FGM. The P-FGM can be obtained by the general
four-parameter power-law distribution Eq. (78): F G M I (1, 0, 0, p).
According to the choice of the functions f 1 and f 2 in Table 1, three elements are implemented. These elements are
SQAD4, S4 and SHO4 corresponding to the discrete Kirchhoff quadrilateral [36], the first order shear deformation
shell element (Reissner–Mindlin [23]) and the present double directors shell element, respectively.

6.1. Pinched hemisphere shell

The pinched hemisphere problem, a popular benchmark problem for shell analysis, is considered with an 18◦ hole
at the top subjected to two inward and two outward forces F 90◦ apart. From the geometrical symmetry of the sphere,
the analysis is reduced to one quadrant (show Fig. 4).
The spherical geometry is defined by the radius R = 10 and the thickness t = 0.04. The material properties are the
Young’s modulus E = 6.82510 107 and the Poisson’s ratio ν = 0.3.
The loads F are increased to 450 to compare with results by Saleeb et al. [39], Simo et al. [23], Kim et al. [40].
Fig. 5 shows a good performance for SQAD4, S4 and SHO4 in large displacement and finite rotation analysis. The
present results based on SHO4, are located between S4 and SQAD4 results. The initial and deformed configurations,
subjected to F = 450, obtained using SHO4 element mesh are shown in Fig. 6.
 A. Frikha, F. Dammak / Comput. Methods Appl. Mech. Engrg. 315 (2017) 1–24 15

Fig. 6. Hemispherical shell (a) Initial geometry (b) Deformed geometry.

Fig. 7. Initial and deformed configurations of the ring plate (a) Initial geometry (b) Deformed geometry under 60 p.

Fig. 8. Load–deflection curve of the ring plate.

6.2. Ring plate loaded at free edge

The ring plate geometry, which is shown in Fig. 7(a), was analyzed in [41]. The inner and the outer diameters are
denoted by r = 6 m and R = 10 m, respectively. The ring is considered thin with a thickness h = 0.03 m. The
material is considered isotropic with the Young’s modulus E = 2.1 × 108 kN/m2 and the Poisson’s coefficient ν = 0.
The applied load at the free edge is equal to p = 0.1 f kN/m. The analysis is based on a 8 × 48 mesh consisting of
SHO4 element. The deformed geometry under a load factor f = 60 is plotted in Fig. 7(b). The displacements W at
points A, B and C versus the load factor f are plotted in Fig. 8. The W displacement is located along the normal to
the non-deformed ring plate.
Two types of elements are used in this example, which are S4 and SHO4. The displacements using SHO4 element
are in good agreement with the S4 element solutions. These displacements at points A, B and C agree very well with
results reported in [41,42].
16 A. Frikha, F. Dammak / Comput. Methods Appl. Mech. Engrg. 315 (2017) 1–24

Fig. 9. Cantilevered plate.

Fig. 10. Tip deflection vs. shear load F for an isotropic cantilevered plate.

Fig. 11. Tip deflection vs. shear load F for FGM cantilevered plate.

6.3. A cantilevered plate strip

A cantilever plate strip subjected to a distributed end shear load F as shown in Fig. 9, where L = 10, b = 1 and
h = 0.1. First, an isotropic material is considered. This problem was studied in many works [43–45]. The material
properties are E = 1.2 × 106 and ν = 0.3.
Fig. 10 and Table 5 depict the axial and vertical deflections of the plate tip −U and W for isotropic case. Results
obtained by the present model (SHO4) are compared to analytical and numerical models. Results obtained by S4 and
SQAD4 are close and in excellent agreement with analytical solution in [46].
The analysis of FGM cantilevered plate was conducted using the same geometry of the isotropic plate and a power
law variation of Aluminum and Zirconia (P-FGM). The material properties of Aluminum and Zirconia are given in
Table 6. The lower surface of the plate is assumed to be metal rich and the top surface is assumed to be ceramic.
Fig. 11 shows Axial and vertical deflection −U and W of FGM plate using double director shell element of P-FGM,
under mechanical loading. The deflection of metallic plate is greater than ceramic due to the high bending stiffness of
ceramic plate. It is shown that the deflection decreases with the power index p from metal to ceramic.
 A. Frikha, F. Dammak / Comput. Methods Appl. Mech. Engrg. 315 (2017) 1–24 17

Fig. 12. Non-dimensional shear stress τ13 for a cantilevered plate.

Table 5
Deflection of a cantilevered plate.

F W U
Theory [46] SQAD4 S4 SHO4 Theory [46] SQAD4 S4 SHO4
2 4.94 4.21 4.78 4.78 1.60 1.06 1.49 1.47
4 6.70 6.40 6.60 6.61 3.29 2.73 3.15 3.14
6 7.44 7.40 7.39 7.41 4.34 3.95 4.23 4.22
8 7.85 7.94 7.82 7.84 5.04 4.79 4.97 4.96
10 8.11 8.27 8.10 8.11 5.54 5.41 5.50 5.48

Table 6
Material properties of FGM plate [6].

Parameter Young’s modulus (GPa) Poisson’s ratio Density (kg/m3 )

Aluminum 70 0.3 2707
Zirconia 151 0.3 3000
Alumina 380 0.3 3800

The distribution across the plate thickness of the non-dimensional shear stress τ13 hb/F for various values of the
power index at the fixed end is illustrated in Fig. 12. It is shown that the zero-transverse condition shear stress on top
and bottom faces of the FGM plate is verified using SHO4 element. The shear stress field of isotropic material (metal
and ceramic) is symmetric with a parabolic distribution along the thickness.

6.4. Simply supported square FGM plate

A static analysis was performed on a S-FGM square plate of side length a = 100 cm and thickness h = 2 cm,
simply supported on all its edges. Due to the symmetry, one quarter of the square plate is modeled with 8 × 8 mesh.
The material properties are E 1 = 2.1 × 106 kg/cm2 and ν = 0.3. The deflection of S-FGM plate under uniform
load q0 = 1.0 kg/cm2 for p = 2 is computed. The results of S-FGM plates using Reissner–Mindlin solution with a
constant transverse shear correction factor of 5/6 [40] are compared with the present solutions using double directors
vectors in Fig. 13. The present results agree very well with Kim et al. [40]. It is shown that the deflection increases
with E 1 /E 2 , when the stiffness of FGM plates decreases. The small difference between the present and the reference
results in the case of E 1 /E 2 = 10 can be explained by the constant shear correction factor used in the reference
model [40,47].
A second square plate of side length a = 0.2 m and thickness h = 0.01 m, simply supported on all its
edges is analyzed here. As the previous plate, the same mesh is considered. The results are presented in terms of
18 A. Frikha, F. Dammak / Comput. Methods Appl. Mech. Engrg. 315 (2017) 1–24

Fig. 13. The deflection of S-FGM plate along the x direction for different E 1 /E 2 .

Fig. 14. Non-dimensional center deflection w vs. load parameter P for a simply supported, Aluminum–Zirconia FGM square plate under uniform
transverse load.

non-dimensionalized parameters which are: center deflection, w/ h; load parameter, P = q0 a 4 /(E m h 4 ); thickness
co-ordinate, z = z/ h; and stress, σ = σ h 2 /(|q0 | a 2 ). q0 and E m denote the intensity of the applied mechanical
load and the Young modulus of metal, respectively. The analysis was conducted using Aluminum/Zirconia and
Aluminum/Alumina. The material properties of Aluminum, Zirconia and Alumina are given in Table 6. In all cases,
the lower surface of the plate is assumed to be metal rich and the top surface is assumed to be ceramic. This analysis
is based on P-FGM.
Fig. 14 shows the comparison between the linear and non-linear analysis using double director shell element of
P-FGM, under mechanical loading. The deflection magnitudes using linear analysis are always overpredicted. The
effect of nonlinearity is more pronounced for metallic plate, having the less important stiffness. The difference between
linear and non-linear deflection decreases as the plate becomes more and more ceramic. The present results agree very
well with those given in [48,6].
Fig. 15 depicts the non-dimensional axial stress through the non-dimensional thickness of the plate under uniform
loading applied on the top surface made of Aluminum and Zirconia. The present results agree very well with those of
Praveen and Reddy [48]. The axial stress is compressive at the top surface and tensile at the bottom surface. For the
power index chosen, the plate corresponding to n = 2.0 yielded the maximum compressive stress at the top surface.
The plate corresponding to n = 0.2 yielded the minimum tensile stress at the bottom surface.
Fig. 16 shows similar behavior to Fig. 15, except for the magnitude by replacing the Zirconia by the Alumina.
The present results are in good agreement with those of Praveen and Reddy [48]. In both cases, the FGM plate
corresponding to n = 2.0 yields the maximum compressive stress at the top surface, the metallic and the ceramic
plates experience the maximum tensile stress at the bottom surface and the FGM plate corresponding to n = 0.2
yields the minimum tensile stress at the bottom surface.
 A. Frikha, F. Dammak / Comput. Methods Appl. Mech. Engrg. 315 (2017) 1–24 19

Fig. 15. Non-dimensional axial stress in simply supported square FGM plate under uniform loading of −1 × 104 N/m2 , 10 loadsteps (Aluminum–
Zirconia).

Fig. 16. Non-dimensional axial stress in simply supported square FGM plate under uniform loading of −1 × 104 N/m2 , 10 loadsteps (Aluminum–
Alumina).

6.5. Cylindrical panel

The non linear mechanical response of cylindrical panel subjected to a concentrated load is examined (show
Fig. 17). Variants of this problem are found throughout the literature [49–51]. In this example, the geometrical
parameters are α = 0.1 rad, L = 508 mm and R = 2540 mm, where R is the radius of the undeformed mid-
surface. Isotropic and FGM cylindrical panel are examined with a shell thickness of h = 25.4 mm and h = 12.7 mm
respectively. The material properties of isotropic case are E = 3102.75 MPa and ν = 0.3. Functionally graded
cylindrical panel is compound of aluminum and zirconia (show Table 6). Numerical simulation are conducted using
one quarter of the physical domain. The complete model is obtained using appropriate symmetry boundary conditions.
a uniform 2 × 2 mesh for the quarter model and a 4 × 4 discretization for the full domain.
Fig. 18 depicts the deflection of the isotropic shell at the loaded point versus the applied load P for cylindrical
panel where h = 25.4 mm using 2 × 2 and 20 × 32 mesh for the quarter model. Results obtained using 2 × 2 mesh
are similar to those obtained from the refined mesh. The present results agree strongly with those reported in [27].
The reference model is based on 81 nodes by element and the present model is based on 2 × 2 nodes by element.
This comparison proves the performance and accuracy of our element. Later, the cylindrical panel is modeled using
2 × 2 mesh. For isotropic cylindrical panel, the applied load increases monotonic with the deflection at the center of
the panel.
Numerical results for metal–ceramic functionally graded panels, for h = 12.7 mm, are shown in Fig. 19. Displace-
ment control method is used in this example. The metal is taken as the bottom material and the ceramic (zirconia)
as the top constituent. The power law is used in this model and the index of FGM is varied between 0.2 and 5 by
considering the extremely cases of ceramic and metal. For ceramic panel, it is shown that the load increases until a
deflection of 10 mm. Then the applied load decreases until a deflection of 20 mm. Next, the load increases monotonic
20 A. Frikha, F. Dammak / Comput. Methods Appl. Mech. Engrg. 315 (2017) 1–24

Fig. 17. Cylindrical panel.

Fig. 18. Vertical deflection of a shallow isotropic cylindrical panel under point loading, h = 25.4 mm.

Fig. 19. Vertical deflection of functionally graded metal–ceramic shallow cylindrical panels under point loading, h = 12.7 mm.

with deflection. Due to the high stiffness of ceramic compared to metal, the applied load to ceramic panel is more
important to metal. All the cylindrical panels with intermediate properties undergo corresponding intermediate values
of loads and deflection. These results are visually in unison with the deflection curves provided in [52,27].
 A. Frikha, F. Dammak / Comput. Methods Appl. Mech. Engrg. 315 (2017) 1–24 21

7. Conclusion

The geometrically non-linear theory of 3D-FGM-shell structures undergoing the finite rotation was analyzed using
a finite element model based on a discrete double director shell element. The present model, based on higher-order
shear deformation theory, does not need any shear coefficients. The grading function is assumed along the thickness
direction. A pinched hemisphere shell and a ring plate with isotropic material are used to validate the model for shells
subjected to finite rotation. Functionally graded shells as a cantilevered plate strip, a simply supported square and a
cylindrical panel are computed by the present model. To validate the results of the present study and demonstrate its
accuracy, the results are compared with the literature. A very good agreement among the results confirms the high
accuracy of the current non-linear model. Deflection and axial stress are analyzed by varying the grading index.

Appendix. Geometrical tangent matrix

For a couple of nodes (i, j), the matrix [KG ] becomes:

    
UU Mi j I + UUF2i j U B F1i j + U BC1i j I UBF2i j
,
     
KGij =  BU F1i j + BU C1i j I B B F1i j + B BC1i j I 0 (A.1)

   
BUF2i j 0 BBF2i j
where UU M, (U B F1 , B B F1 , UUF2 , UBF2 , BBF2 ) and (U BC1 , B BC1 ) correspond to membrane, bending and
shear respectively. The expression of the membrane term is written as:
     
UU Mi j = Ni,1 N 11 N j,1 + N 12 N j,2 + Ni,2 N 12 N j,1 + N 22 N j,2 d A. (A.2)
A
The bending terms of the first director vector d1 are given by:
     
U B F1i j = Ni,1 M111 N j,1 + M112 N j,2 + Ni,2 M112 N j,1 + M122 N j,2 d A, (A.3)
A
     
B B F1ii = − Ni,1 (a1 · d1i ) M111 + (a2 · d1i ) M112 + Ni,2 (a1 · d1i ) M112 + (a2 · d2i ) M122 d A, (A.4)
A
B B F1i j = 0, i ̸= j; BU F1i j = U B F1 ji . (A.5)
The bending term of the second director vector d2 is given by:
     
UUF2i j = N ju,1 Ni,1 M211 + Ni,2 M212 + N ju,2 Ni,1 M212 + Ni,2 M222 d A
A
     
+ T
Niu,1 N j,1 M211 + N j,2 M212 + Niu,2
T
N j,1 M212 + N j,2 M222 d A. (A.6)
A
The expression of Niu is given by the following relation:
2
Niu = Pki tdki − Pmi tdm
i
, tdm
i
= tm ⊗ d m , (A.7)
2L k
where tm and dm are the unit vectors defined by the segment of the element with center m and the normal in this point,
respectively.
      
UBF =
2i j N jd,1 Ni,1 M211 + Ni,2 M212 + N jd,2 Ni,1 M212 + Ni,2 M222 d A
A (A.8)
BUF2i j = UBF2 ji .


The expression of Nid is given by the following relation:

3
Nid = Ni I − Pki ttik − Pmi ttim , ttim = tm ⊗ tm . (A.9)
4
22 A. Frikha, F. Dammak / Comput. Methods Appl. Mech. Engrg. 315 (2017) 1–24

From Eq. (A.1), the rotating term BBF2i j is zero for i ̸= j:


BBF2i j = 0, i ̸= j
(A.10)
BBF2i j = −B B F2i j I, i = j,
where B B F2i j is defined by:
     
B B F2i j = M211 a1 + M212 a2 · Nid,1 + M212 a1 + M222 a2 · Nid,2 di d A. (A.11)
A

The measurement of transverse shear strain relatively to substitution strain model allows to deduce U BC1 and
B BC1 . The shear Terms U BC1i j can be grouped in matrix form:

−α − β
 
−β 0 −α
1 β β −γ −γ 0 ,
[UBC1 ] =  (A.12)
8  0 γ γ +δ δ 
α 0 −δ α−δ
where the coefficients α, β, γ and δ are defined by the following expressions:
  
 α = (1 − ξ )T2 d A, β = (1 − η)T1 d A


A A (A.13)
γ = (1 + ξ )T2 d A, δ = (1 + η)T1 d A,


A A
BU C1i j = U BC1 ji . (A.14)
The shear terms B BC1i j are zero for i ̸= j (B BC1i j = 0). The non-zero terms can form the column vector:
  

 αaA2 + βaB1 · d1 

   

 
βa B
γ C
 
1

 1 + a2 · d 2


{BBC1 } = −   . (A.15)
4 γ a C
+ δa D
· d3

2 1

 


 

 D
  
δa1 + αa2 · d4 
A
 

References

[1] M.M. Hrabok, T.M. Hrudey, A review and catalogue of plate bending finite elements, Comput. Struct. 19 (3) (1983) 479–495.
[2] H.T.Y. Yang, S. Saigal, D.G. Liaw, Advances of thin shell finite elements and some applications - version i, Comput. Struct. 35 (4) (1990)
481–504.
[3] P. Lardeur, J.L. Batoz, Composite plate analysis using a new discrete shear triangular finite element, Internat. J. Numer. Methods Engrg. 27
(1989) 343–359.
[4] S. Vlachoutsis, Shear correction factors for plates and shells, Internat. J. Numer. Methods Engrg. 33 (1992) 1537–1552.
[5] A. Hajlaoui, A. Jarraya, K. El Bikri, F. Dammak, Buckling analysis of functionally graded materials structures with enhanced solid-shell
elements and transverse shear correction, Compos. Struct. 132 (2015) 87–97.
[6] J.N. Reddy, Analysis of functionally graded plates, Internat. J. Numer. Methods Engrg. 47 (2000) 663–684.
[7] S.-H. Chi, Y.-L. Chung, Mechanical behavior of functionally graded material plates under transverse load. Part 1: Analysis, Int. J. Solids
Struct. 43 (2006) 3657–3674.
[8] Z.Q. Cheng, R.C. Batra, Exact correspondence between eigenvalues of membranes and functionally graded simply supported polygonal plates,
J. Sound Vib. 229 (4) (2000) 879–895.
[9] H. Matsunaga, Free vibration and stability of functionally graded plates according to a 2-D higher-order deformation theory, Compos. Struct.
82 (2008) 499–512.
[10] S.S. Vel, R.C. Batra, Three-dimensional exact solution for the vibration of functionally graded rectangular plates, J. Sound Vib. 272 (2004)
703–730.
[11] J.C. Simo, D.D. Fox, M.S. Rifai, On a stress resultant geometrically exact shell model. Part ii: The linear theory, Comput. Methods Appl.
Mech. Engrg. 73 (1989) 53–92.
 A. Frikha, F. Dammak / Comput. Methods Appl. Mech. Engrg. 315 (2017) 1–24 23

[12] F. Tornabene, Free vibration analysis of functionally graded conical, cylindrical shell and annular plate structures with a four-parameter
power-law distribution, Comput. Methods Appl. Mech. Engrg. 198 (2009) 2911–2935.
[13] J.L. Mantari, A.S. Oktem, C.G. Soares, Static and dynamic analysis of laminated composite and sandwich plates and shells by using a new
higher-order shear deformation theory, Compos. Struct. 94 (2011) 37–49.
[14] J.L. Mantari, C.G. Soares, Optimized sinusoidal higher order shear deformation theory for the analysis of functionally graded plates and
shells, Composites B 56 (2014) 126–136.
[15] E.J. Barbero, J.N. Reddy, Modeling of delamination in composite laminates using a layer-wise plate theory, Int. J. Solids Struct. 28 (3) (1991)
373–388.
[16] S. Mulmule, A.K. Rath, Application of the multi-director displacement field approach for sandwich shell structure analysis, Compos. Struct.
48 (1993) 653–660.
[17] K.H. Lo, R.M. Christensen, E.M. Wu, Stress solution determination for high order plate theory, Int. J. Solids Struct. 14 (1978) 655–662.
[18] E. Carrera, S. Brischetto, M. Cinefra, M. Soave, Refined and advanced models for multilayered plates and shells embedding functionally
graded material layers, Mech. Adv. Mater. Struct. 17 (8) (2010) 603–621.
[19] M. Cinefra, E. Carrera, L. Della Croce, C. Chinosi, Refined shell elements for the analysis of functionally graded structures, Compos. Struct.
94 (2012) 415–422.
[20] G.M. Kulikov, S.V. Plotnikova, Non-linear exact geometry 12-node solid-shell element with three translational degrees of freedom per node,
Internat. J. Numer. Methods Engrg. 88 (2011) 1363–1389.
[21] G.M. Kulikov, S.V. Plotnikova, Three-dimensional exact analysis of piezoelectric laminates plates via a sampling surfaces method, Int. J.
Solids Struct. 50 (2013) 1916–1929.
[22] G.M. Kulikov, S.V. Plotnikova, A sampling surfaces method and its application to three-dimensional exact solution for piezoelectric laminated
shells, Int. J. Solids Struct. 50 (2013) 1930–1943.
[23] J.C. Simo, D.D. Fox, M.S. Rifai, Resultant geometrically exact shell model. Part iii: Computational aspects of the non-linear theory, Comput.
Methods Appl. Mech. Engrg. 79 (1990) 21–70.
[24] Y. Başar, Y. Ding, R. Schultz, Refined shear-deformation models for composite laminates with finite rotations, Int. J. Solids Struct. 30 (19)
(1993) 2611–2638.
[25] P. Phung-Van, T. Nguyen-Thoi, H. Luong-Van, Q. Lieu-Xuan, Geometrically nonlinear analysis of functionally graded plate using a cell-based
smoothed three-node plate element (CS-MIN3) based on the C0-HSDT, Comput. Methods Appl. Mech. Engrg. 270 (2014) 15–36.
[26] R.A. Arciniega, J.N. Reddy, Large deformation analysis of functionally graded shells, Int. J. Solids Struct. 44 (2006) 2036–2052.
[27] G.S. Payette, J.N. Reddy, A seven-parameter spectral/hp finite element formulation for isotropic, laminated composite and functionally graded
shell structures, Comput. Methods Appl. Mech. Engrg. 278 (2014) 664–704.
[28] M. Wali, A. Hajlaoui, F. Dammak, Discrete double directors shell element for the functionally graded material shell structures analysis,
Comput. Methods Appl. Mech. Engrg. 278 (2014) 388–403.
[29] M. Wali, T. Hentati, A. Jarraya, F. Dammak, Free vibration analysis of FGM shell structures with a discrete double directors shell element,
Compos. Struct. 125 (2015) 295–303.
[30] A. Frikha, M. Wali, A. Hajlaoui, F. Dammak, Dynamic response of functionally graded material shells with a discrete double directors shell
element, Compos. Struct. 154 (2016) 385–395.
[31] M. Levinson, An accurate simple theory of statics and dynamics of elastic plates, Mech. Res. Commun. 7 (1980) 340–350.
[32] W.B. Bickford, A consistent higher order beam theory, Dev. Theor. Appl. Mech. 11 (1982) 137–150.
[33] J.N. Reddy, A refined nonlinear theory of plates with transverse shear deformation, Int. J. Solids Struct. 20 (1984) 881–896.
[34] M.V.V. Murthy, An Improved Transverse Shear Deformation Theory for Laminated Anisotropic Plates, NASA Technical Paper 1903, 1981,
pp. 1–37.
[35] G. Shi, A new simple third-order shear deformation theory of plates, Int. J. Solids Struct. 44 (2007) 4399–4417.
[36] F. Dammak, S. Abid, A. Gakwaya, G. Dhatt, A formulation of the non linear discrete Kirchhoff quadrilateral shell element with finite rotations
and enhanced strains, Rev. Europeenne des Elements Finis 14 (2005) 7–31.
[37] K.J. Bathe, E. Dvorkin, A four-node plate bending element based on Mindlin/Reissner plate theory and a mixed interpolation, Internat. J.
Numer. Methods Engrg. 21 (1985) 367–383.
[38] Z. Su, G. Jin, S. Shi, T. Ye, X. Jia, A unified solution for vibration analysis of functionally graded cylindrical, conical shells and annular plates
with general boundary conditions, Int. J. Mech. Sci. 80 (2014) 62–80.
[39] A.F. Saleeb, T.Y. Chang, W. Graf, S. Yingyeunyong, A hybrid/mixed model for non-linear shell analysis and its applications to large-rotation
problem, Internat. J. Numer. Methods Engrg. 29 (1990) 407–446.
[40] K.-D. Kim, G.R. Lomboy, S.-C. Han, Geometrically nonlinear analysis of functionally graded material (FGM) plates and shells using a
four-node quasi-conforming shell element, J. Compos. Mater. 42 (2008) 485–511.
[41] N. Buechter, E. Ramm, Shell theory versus degeneration—a comparison in large rotation finite element analysis, Internat. J. Numer. Methods
Engrg. 34 (1992) 39–59.
[42] P. Wriggers, F. Gruttmann, Thin shells with finite rotations formulated in biot stressed: Theory and finite element formulation, Internat. J.
Numer. Methods Engrg. 36 (1993) 2049–2071.
[43] M. Fafard, G. Dhatt, J.L. Batoz, A new discrete Kirchhoff plate/shell element with updated procedures, 31 (4) 1988 591–606.
[44] J.C. Simo, M.S. Rifai, D.D. Fox, On a stress resultant geometrically exact shell model. Part iv: Variable thickness shells with through-the-
thickness stretching, Comput. Methods Appl. Mech. Engrg. 81 (1990) 91–126.
[45] K.Y. Sze, X.H. Liu, S.H. Lo, Popular benchmark problems for geometric nonlinear analysis of shells, Finite Elem. Anal. Des. 40 (11) (2004)
1551–1569.
[46] S.P. Timoshenko, J.M. Gere, Mechanics of Materials, Von Vostran Reinhold, New York, 1972.
24 A. Frikha, F. Dammak / Comput. Methods Appl. Mech. Engrg. 315 (2017) 1–24

[47] S.-H. Chi, Y.-L. Chung, Mechanical behavior of functionally graded material plates under transverse load. Part 2: Numerical results, Int. J.
Solids Struct. 43 (2006) 3675–3691.
[48] G.N. Praveen, J.N. Reddy, Nonlinear transient thermoelastic analysis of functionally graded ceramic–metal plates, Int. J. Solids Struct. 35
(1997) 4457–4476.
[49] H. Parisch, Large displacement of shells including material nonlinearities, Comput. Methods Appl. Mech. Engrg. 27 (1980) 183–214.
[50] C.W.S. To, B. Wang, Hybrid strain-based three-node flat triangular laminated composite shell elements for vibration analysis, J. Sound Vib.
211 (2) (1998) 277–291.
[51] B. Brank, F.B. Damjanić, D. Perić, On implementation of a nonlinear four node shell finite element for thin multilayered elastic shells,
Comput. Mech. 16 (1995) 341–359.
[52] R.A. Arciniega, J.N. Reddy, Tensor-based finite element formulation for geometrically nonlinear analysis of shell structures, Comput. Methods
Appl. Mech. Engrg. 196 (2007) 1048–1073.