Composite Structures 106 (2013) 537–548

Contents lists available at SciVerse ScienceDirect

Composite Structures
journal homepage: www.elsevier.com/locate/compstruct

Interlaminar stress calculation in composite and sandwich plates

in NURBS Isogeometric finite element analysis
Hitesh Kapoor a,⇑, Rakesh K. Kapania b, Som R. Soni c
a
Aerospace & Ocean Engineering Department, Virginia Tech, Blacksburg, VA 24060, United States
b
Aerospace & Ocean Engineering Department, 213E Randolph Hall, Virginia Tech, Blacksburg, VA 24061, United States
c
AdTech Systems Research Inc., Dayton, Ohio, United States

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

Article history:
Available online 10 June 2013
Keywords:
NURBS Isogeometric analysis
Laminated composite plate
First-order shear deformation theory
Interlaminar stress calculation
Direct integration
k-Refined NURBS higher order element
This research paper describes the development of NURBS Isogeometric finite element analysis and post-
processor for interlaminar stress calculation in composite and sandwich plates. First-order, shear-
deformable laminate composite plate theory is utilized in deriving the governing equations using a var-
iational formulation. Linear, quadratic, higher order and k-refined NURBS elements are constructed and
numerical validation is performed for laminated composite and sandwich plates. Lagrange finite element
suffers from higher order stress gradient oscillations due to Gibbs phenomenon and require alternative
stress recovery procedures for accurate interlaminar stress calculations, especially interlaminar normal
stress. In this paper, direct post-processing is performed which computes interlaminar shear and normal
stresses from higher order gradients of NURBS basis in a single step procedure. Interlaminar stresses are
found to be in an excellent agreement with 3D elasticity solution and FSDT along with k-refinement pro-
cedure of NURBS basis is found to compute equivalent or better interlaminar normal stress than higher-
order shear deformation theory.
Ó 2013 Elsevier Ltd. All rights reserved.

1. Introduction
Inspite of extensive use of finite element methods, the barriers
between engineering design and analysis still exist and the way to
bridge gap is to reconstitute the entire process. Idea is to use one
model and use it as an analysis model which require a change from
classical finite element to an analysis procedure based CAD repre-
sentation, formally known as Isogeometric analysis. There are sev-
eral CAD functions which can be used for CAD representation of
analysis module. Of most widely used CAD basis in engineering de-
sign process are NURBS (non-uniform rational B-splines) as pre-
sented by Piegle and Tiller [1], Farin [2], Cohen et al. [3] and
Rogers [4]. NURBS basis are useful for analysis purposes because
they possess useful mathematical property of refinement through
knot insertion and variational diminishing property of convex hull.
There are other computational geometry technologies that can be
utilized as the basis for Isogeometric analysis such as sub-division
surface by Peters and Reif [5] and Warren and Weimer [6] Gordon
patches [7], Gregory patch [8], S-patch [9] and A patch [10], etc.
Hughes et al. [11] introduced the idea of Isogeometric analysis
using NURBS (non-uniform rational B-spline). They used NURBS
to exactly represent the CAD geometry and then, constructed a
⇑ Corresponding author.
E-mail address: hkapoor@vt.edu (H. Kapoor).
coarse mesh for the analysis. The idea behind Isogeometric analysis
is to model the geometry exactly which also serves the basis for the
solution space i.e. invoking isoparametric concept. Kagan et al. [12]
developed B-Spline finite element analysis and integrated with
geometric design. Same authors [13] included adaptive refinement
like hp and h-refinement techniques in their finite element code.
Advance multi-layered composite and sandwich plate/shell
structures are being increasingly used in aerospace, shipbuilding,
bridges and other industries. These structures have smaller thick-
ness as compared to other dimensions and therefore, are often sub-
jected to large deformation behavior under external loads. The
plate theories for laminated composites can be categorized into
equivalent single-layer theories (ELS) and layerwise theories.
Equivalent single layer theories namely, classical, first-order and
higher-order shear deformation theories, are derived from their
3D counterpart (i.e. layerwise and 3D elasticity) by making appro-
priate assumptions to the state of strain/stress in the thickness
direction, reducing 3D continuum problem to a 2D problem.
Several articles are available in literature on shear deformation
theories. Reissner–Mindlin theory [14–16] (first-order shear defor-
mation theory, FSDT) assumes constant state of through the thick-
ness strain. This theory was further extended for anisotropic plates
by Whitney and Pagano [17]. Urthaler and Reddy [18] developed a
mixed finite element for bending analysis of laminated composite
plates using FSDT. They treated bending moment as a field variable

0263-8223/$ - see front matter Ó 2013 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.compstruct.2013.05.028
538 H. Kapoor et al. / Composite Structures 106 (2013) 537–548
along with displacement and rotation. Similarly, Cen et al. [19],
Kim et al. [20] and Mai-Duy et al. [21] have previously developed
elements based on Mindlin–Reissner theory (FSDT).
In higher-order shear deformation theories (HSDT), Putcha and
Reddy [22] developed a mixed finite element approach with 9-
node Lagrangian quadrilateral element, Kant and Kommineni [23]
formulated refined HSDT with 9-node quadrilateral element for
linear and nonlinear finite element analysis of laminated compos-
ite and sandwich plates, Polit and Touratier [24] studied large
deflection behavior using HSDT, Phan and Reddy [25] developed
special third order theory (STTR) and Ren and Hinton [26] devel-
oped a third order plate theory for the analysis of laminated com-
posite plates. Reddy and Robbins [27]’s provide extensive literature
review on ELS and layerwise theories for laminated composite
plates. Carrera [28] provide detailed review of zig–zag theories
for multi layered plates and shells. Comparing higher-order theo-
ries with FSDT, higher-order plate theories enhance the accuracy
of the solution slightly but are computationally more expensive.
On the other hand, FSDT has often been used due to its simplicity
and provide best compromise between economy and accuracy in
predicting the response of thin to moderately thick laminates
[29,30]. Kapoor and Kapania [31] developed NURBS Isogeometric
finite element approach to study geometrically nonlinear behavior
of composite plates. In this analysis, they were able to remove
locking in thin plates and also were able to formulate a stable
hourglass behavior using modified NURBS elements in parent
domain.
It is a well known fact that composites are prone to damage un-
der low transverse loads due to comparatively low transverse
modulus and have higher probability of failure. The failure or dam-
age is generally, initiated due to a variety of failure modes like
delamination, matrix cracking, fiber failure, etc.; delamination
being the primary mode of failure. Delamination is initiated when
the interlaminar stresses attain the maximum interfacial strength.
Therefore, predicting through-the-thickness stresses accurately is
essential. This requires calculating higher-order derivatives of in-
plane stress accurately which in turn requires higher-order dis-
placement derivatives.
First order shear deformation theory with the use of constitu-
tive relation produces highly inaccurate interlaminar stresses
exhibiting non-physical discontinuity at the ply interface of com-
posite laminate. The accurate of interlaminar stresses using 3D
equilibrium equations in the context of finite element analysis de-
pends upon the computation of higher-order in-plane strain gradi-
ents, which is generally poor as Lagrange polynomial gradient
oscillates. Reddy [32] calculated transverse shear stresses using
derivative of in-plane stresses that were obtained by differentiat-
ing interpolation functions in finite element approximation. In or-
der to accurately predict delamination, interlaminar normal stress
is as important as the shear stress. The computation of interlami-
nar normal stress requires an additional derivative of the basis
function. These higher-order derivatives are not obtained directly
in the finite element code. Tessler [33] developed smoothing vari-
ational formulation which combined discrete least square and pen-
alty constraints functional in a single variational form and
recovered the stress gradients more accurately. Byun and Kapania
[34] developed a post-processing technique to overcome this
drawback. They interpolated the finite element displacement data
using polynomial functions like Chebyshev and orthogonal polyno-
mials in a global domain. Use of Chebyshev polynomials require
nodal displacement data to be available at some specific points
i.e. Chebyshev points and cannot interpolate boundary edge nodal
data while orthogonal polynomials use the arbitrarily distributed
data points but are more difficult to obtain. Lee and Lee [35] intro-
duced the non-iterative post processing procedure for the recovery
of transverse stresses. They followed the equilibrium based stress
recovery method, using the one dimensional, least square finite
element in the thickness direction.
Park and Kim [36] presented a predictor–corrector post-pro-
cessing procedure for the accurate recovery of stresses and dis-
placement in the multi-layered composite panels. The predictor
only predicts the transverse shear stress while the corrector meth-
od enhances the accuracy of the displacement, in-plane and trans-
verse normal stress in the thickness direction using the results of
the predictor and finite element analysis. Park et al. [37], later,
used the nonlinear predictor–corrector method to obtain the stres-
ses and displacement in composite panels in geometrically nonlin-
ear formulation. Noor et al. [38] developed a computational
procedure to get the transverse shear stresses in multi-layered
composite panels. They first used the super convergent recovery
technique to evaluate the in-plane stresses and then, used the
piecewise integration in the thickness direction to obtain the inter-
laminar stresses.
Matsunaga [39] analyzed the displacement and stresses in com-
posite laminated beams using global higher-order beam theory.
Author expanded the displacement field variables with power ser-
ies of z-coordinate. Makeev and Armanios [40] presented an itera-
tive method to approximate analytical solution of elasticity
problems in composite laminates. Rolfes and Rohwer [41] devel-
oped a method for calculating the improved transverse shear stres-
ses in laminated composites using first-order shear-deformation
theory. Kant and Manjunatha [42] developed numerical algorithms
for accurate calculations of transverse stresses using higher-order
shear-deformation theories. They used direct integration, finite dif-
ference and exact surface fitting approach. Recently, Kant et al. [43]
proposed a semi analytical model for the accurate estimation of
stresses and displacement in composite and sandwich structures.
The two-point boundary value problem governed by a set of linear
first-order differential equations through the thickness is solved
using fourth-order Runge–Kutta method. Nosier and Bahrami
[44] developed the analytical solution to study the edge effects
in anti-symmetric angle-ply laminates using first order shear
deformation and Reddy’s layerwise theory.
Stress recovery is highly dependent on the structure of a partic-
ular element and on the formulation used in deriving elements.
Stress recovery methods include interpolation-extrapolation from
super-convergent points [46], L2 projection [47], stress smoothing
[48] and integral stress techniques [49]. Zienkeiwicz and Zhu
[50] developed an efficient post-processing technique in terms of
super-convergent patch recovery (SPR) procedure. A modified ver-
sion of this technique was developed to obtain in-plane stresses at
nodes and interlaminar stresses using equilibrium equations [52].
Stress recovery procedures suffer from extraneous stress oscilla-
tions [53].
This research present development of NURBS Isogeometric fi-
nite element and interlaminar stress calculations in laminated
composite plates under sinusoidal load. Geometry is modeled ex-
actly in Isogeometric framework and isoparametric and sub-para-
metric finite element representation is invoked for solution space.
First-order, shear-deformable laminated composite plate theory is
utilized in deriving the governing equations in variational formula-
tion. Linear, quadratic, p and k-refined NURBS elements in parent
domain are constructed. Interlaminar stresses are computed by di-
rect integration of 3D equilibrium equations, utilizing non-oscilla-
tory nature of higher order NURBS basis and their gradients. In-
plane stresses are computed using constitutive equations and
strain gradients for interlaminar stress computations are com-
puted from higher-order gradients of NURBS basis. Most accurate
interlaminar stresses, equivalent to 3D elasticity solution, are ob-
tained by k-refined NURBS elements with coarsest meshes.
 H. Kapoor et al. / Composite Structures 106 (2013) 537–548 539

2. First-order shear deformation plate theory for laminated where N = {Nxx, Nyy, Nxy} are the force resultants, M = {Mxx, Mxy, Myy}
composite plates are the moment resultants and Q = {Qx, Qy} are the shear force resul-
tants. These resultants acting per unit length can be written as:
2.1. Governing equations 8 9 8 9
< Nxx >
> = Z h=2 > < rx >=
The origin of material coordinate system is considered to be the
Nyy ¼ ry dz
>
: >
; h=2 >: >
mid-plane of the laminate and Kirchhoff assumption that the Nxy rxy ;
8 9 8 9
transverse normal remain perpendicular to the mid-surface after < M xx >
> = Z h=2 > < rx > = ð9Þ
deformation is relaxed. However, transverse strain, z, and out-of M yy ¼ ry zdz
plane normal stress, rz are considered negligible. Layers are >
: >
; h=2 >
: >
M xy rxy ;
assumed to be perfectly bonded. The displacement field is   Z h=2  
defined as: Qx sx
¼ dz
Qy h=2 s y
uðx; y; zÞ ¼ u0 ðx; yÞ þ z/x ðx; yÞ
The laminate constitutive equations relate the force and mo-
v ðx; y; zÞ ¼ v 0 ðx; yÞ þ z/y ðx; yÞ ð1Þ
ment resultant to strains in laminate coordinate system through
wðx; y; zÞ ¼ w0 ðx; yÞ ABD matrix.
8 9 2 38 m 9 2 38 b 9
where u0, v0, w0, are the displacement along the x, y and z-axis and < Nxx >
> = A11 A12 A16 >< x >= B11 B12 B16 > < x >=
6 7 6 7
/x, /y are the rotation of transverse normal of the mid-plane about Nyy ¼ 4 A12 A22 A26 5 my þ 4 B12 B22 B26 5 by
>
: >
; >
: m >; >
: b >
the y and x-axis respectively. Nxy A16 A26 A66 xy B16 B26 B66 xy ;
Total in-plane strain  is decomposed into membrane and bend-
ing strains. ð10Þ

 ¼ m þ zb ð2Þ 8 9 2 38 m 9 2 38 b 9
< M xx >
> = B11 B12 B16 <> x >= D11 D12 D16 < > x >=
6 7 m 6 7 b
m, the membrane strain can be written as follows: M yy ¼ 4 B12 B22 B26 5 y þ 4 D12 D22 D26 5 y
8 9 >
: >
; >
: m >; >
: b >
8 m9
> @u0
> M xy B16 B26 B66 D16 D26 D66 xy ;
< x >
> > > xy
= < @x =
@v 0
m
¼  m
y ¼ @y ð3Þ ð11Þ
>
:c >
; >
> >
m
xy
: @u0 þ @ v 0 >
;     
@y @x Qy A44 A45 cyz
¼K ð12Þ
b, the bending strain can be written as follows: Qx A45 A55 cxz
8 9
8 9 > @/y
>
>
< bx >
=
>
>
< @y >
>
= XN 
1   1  
b ¼ > by ¼
@/y
ð4Þ ðAij ; Bij ; Dij Þ ¼ Q kij ðzkþ1  zk Þ; Q kij z2kþ1  z2k ; Q kij z3kþ1  z3k
: >
; >
> >
>
@y
k¼1
2 3
cbxy >
: @/x þ @/y >
;
@y @x N

X 
 ðA44 ; A45 ; A55 Þ ¼ Q k44 ; Q k45 ; Q k55 ðzkþ1  zk Þ ð13Þ
The transverse shear strain can be written as follows: k¼1
  ( @w )
cyz @x
0
þ /x Here, Aij, Bij and Dij represent extensional and shear, extensional-
¼ ð5Þ
cxz @w0
@y
þ /y bending coupling and bending stiffness matrices and Q kij s are
plane-stress reduced stiffness terms. K is the shear correction factor.
The stress strain relation w.r.t material co-ordinate system of a Weak form of the governing equations is obtained by pre-mul-
lamina can be written as: tiplying the equation of motion with du0, dv0, dw0, d /x and d/y
8 9ðkÞ 2 3ðkÞ 8 9
respectively and integrating by parts over the element domain.
< rxx >
> = Q 11 Q 12 Q 16 < xx >
> =
6 7 yy Substituting force, moment and shear force resultant:
ryy ¼ 4 Q 12 Q 22 Q 26 5 ð6Þ
Z  I
>
: >
; >
:c > ;
rxy Q 16 Q 26 Q 66 @du0 @du0
xy 0¼ Nxx þ Nyy dxdy  Px du0 ds
Xe @x @y Ce
 ðkÞ " #ðkÞ   Z  I
syz Q 44 Q 45 cyz @dv 0 @dv 0
¼ ð7Þ 0¼ Nxy þ N yy dxdy  Py du0 ds
sxz Q 45 Q 55 cxz Xe @x @y Ce
Z  
@dw0 @dW 0 @dw0 b @dw0 b @dw0
The equilibrium equations are developed from stress resultants 0¼ Qx þ Q y  dw0 q þ N xx þ N xy
Xe @x @y @x @x @y
by considering balance of force and moment on an infinitesimal  
area of the laminate. The strong form of the governing equation @dw0 b @dw0 b @dw0
þ N xy þ N yy dxdy
of motion can, then, be written as: @y @x @y
I 
@Nxx @N xy b xx @dw0 þ N b xy @dw0 nx
þ ¼0  Qx þ N
@x @y Ce @x @y
@Nxy @N yy  
þ ¼0 b xy @dw 0 b yy @dw 0
@x @y þ Qy þ N þN ny dw0 ds
@x @y
@Q x @Q y Z  I
þ þq¼0 ð8Þ @d/x @d/x
@x @y 0¼ M xx þ Mxy þ d/x Q x dxdy  T x d/x ds
Xe @x @y Ce
@M xx @M xy Z  I
þ  Qx ¼ 0 @d/y @d/y
@x @y 0¼ M xy þ Myy þ d/y Q y dxdy  T y d/y ds
@M xy @M yy Xe @x @y Ce
þ  Qy ¼ 0 ð14Þ
@x @y
540 H. Kapoor et al. / Composite Structures 106 (2013) 537–548

And, the secondary variables are given as: is the knot index where i = 1, 22, . . . , n + p + 1, p is the order of
the polynomial and n is the number of basis functions. The order
P x ¼ Nxx nx þ Nxy ny ; Py ¼ N xy nx þ Nyy ny
of the polynomial, p = 0, 1, 2, 3 . . . , refer to the constant, linear,
T x ¼ M xx nx þ Mxy ny ; T y ¼ Mxy nx þ M yy ny quadratic, cubic piecewise polynomials, respectively. If more than
 
Qn ¼ Qx þ N b xx @w0 þ N
b xy @w0 nx þ Q y þ N b xy @w0 þ N
b yy @w0 ny one knot is located at the same parametric co-ordinate, these are
@x @y @x @y termed as repeated knots. In case of the open knot vector, first
ð15Þ and last knots are repeated p + 1 times. Basis function formed using
the open knot vector are interpolatory at the beginning and end of
where nx and ny are the directional cosines of the outward normal the parametric space interval, i.e. [n1, nn+p+1]. This distinguishes the
vector. knots from the nodes in the finite element analysis as all the nodes
are interpolatory. As a starting point, B-spline basis functions are
3. Linear NURBS Isogeometric finite element formulation defined recursively using Cox–DeBoor algorithm, starting with
piecewise constants (p = 0).
3.1. Displacement field approximation 
1 if ni 6 n < niþ1 ;
Npi ðnÞ ¼ ð21Þ
0 otherwise
The dependent displacement field variables are approximated
by NURBS basis as follows: And, subsequently, basis functions for orders p = 1, 2, 3, . . . , are
X
nCP X
nCP defined as follows:
u0 ðx; yÞ ¼ uj Rj ðxðn; gÞ; yðn; gÞÞ; v 0 ðx; yÞ ¼ v j Rj ðxðn; gÞ; yðn; gÞÞ n  ni p1 niþpþ1  n p1
j¼1 j¼1 Npi ðnÞ ¼ N ðnÞ þ N ðnÞ ð22Þ
niþp  ni i niþpþ1  niþ1 iþ1
X
nCP
w0 ðx; yÞ ¼ wj Rj ðxðn; gÞ; yðn; gÞÞ
j¼1
3.2.2. Rational B-spline (NURBS) basis
X
nCP X
nCP
/x ðx; yÞ ¼ /xj Rj ðxðn; gÞ; yðn; gÞÞ; /y ðx; yÞ ¼ /yj Rj ðxðn; gÞ; yðn; gÞÞ 2D NURBS basis are formed by tensor product of B-spline basis
j¼1 j¼1 in n and g direction and using projective weights associated with
ð16Þ the control points. Rational basis are very similar to B-spline basis
and derive the continuity and support for the function from the
where nCP are the number of control points per element and Rj are knot vector. Also, NURBS basis form a partition of unity and are
2D NURBS basis functions. pointwise non-negative. Rational basis are formed as follows:
Substituting the displacement field approximation into the
weak form, we obtain element stiffness matrix and load vector. Npi ðnÞMqj ðgÞW i;j
Rp;q
i;j ðn; gÞ ¼ PCPw PCPu p
 Rpk fk ¼ 1 : nCPg ð23Þ
For instance, the size of a sub-matrix, K11, in element stiffness i¼1 N
i¼1 i ðnÞM q
j ð gÞW i;j
matrix, is (nCP)  (nCP) and the size of element load vector, F, is
(nCP  dof, 1). nCP stands for control points per element, dof are NURBS surface is defined as,
the no. of degrees of freedom/control point and (a, b = 1, 2, . . . , 5). X
CPw X
CPu
2 11 12 13 14 15
3 Sðx; yÞ ¼ Rp;q
i;j ðn; gÞBi;j ð24Þ
½K  ½K  ½K  ½K  ½K  j¼1 i¼1
6 21 7
6 ½K 
6 ½K 22  ½K 23  ½K 24  ½K 25  7
7
6 7 where Npi ðnÞ and Mqj ðgÞ are the B-Spline basis in n and g direction
½K ab  ¼ 6 ½K 31  ½K 32  ½K 33  ½K 34  ½K 35  7 ð17Þ and Rp;q
6 7 i;j denotes a 2D NURBS basis function. CPw and CPu are the
6 ½K 41  ½K 43  ½K 43  ½K 44  ½K 45  7 number of control points in n and g directions respectively and
4 5
½K 51  ½K 52  ½K 53  ½K 54  ½K 55  nCP = CPu  CPw is the total number of control points per element.
Wi,js are the weights associated with the control points, Bi,j and
Z weights are the vertical coordinates of the corresponding control
F ia¼3 ¼ qRpi dxdy ð18Þ points. Here, weights, Ws considered are equal to one.
Xe

From the element stiffness matrix, the expression for K11, a sub- 3.3. Numerical integration
matrix of element stiffness matrix, K, is as follows:
Z " p p p p
!#
A multi-element patch, a tensor product of knot vectors in n and
11 @Rpi @Rj @Rpi @Rj @Rpi @Rj @Rpi @Rj
K ij ¼ A11 þ A66 þ A16 þ dxdy g directions, i.e. {0 0 0.5 1 1}  {0 0 0.5 1 1}, creates a 2  2 plate
Xe @x @x @y @y @x @y @y @x
mesh in physical domain. NURBS basis multiplied with respective
ð19Þ
control points constructs a geometric model of plate. In order to
The equation of motion can be written in the matrix form as: perform numerical integration, Guass quadrature is employed.
The integration on physical domain is performed by transforma-
½KfDg ¼ fFg ð20Þ
tion from physical to parametric and from parametric to parent do-
main. Mapping between physical and parametric domain is
3.2. NURBS basis performed as follows:
  X
nCP  
x Bxk
This subsection details B-spline and rational basis generation ¼ R1k ðn; gÞ
y k¼1
Byk
and describes the transformation from physical to parametric to
parent domain. The p and k-refined NURBS elements are con- 8 p9 8 @Rp ðn;gÞ 9
structed by modifying the knot vector by knot-insertion. Last part < @Rk = < k =
@x @n
of this subsection describes the construction of various elements. ¼ ½J 1 
: @Rpk ; : @Rpk ðn;gÞ ;
@y @g ð25Þ
" @x @y
#
3.2.1. B-spline basis
@n @n
A knot vector in one dimension is a set of co-ordinates in the ½J ¼ @x @y
parametric space, N = {n1, n2, . . . , nn+p+1}, where ni is the ith knot, i @g @g
 H. Kapoor et al. / Composite Structures 106 (2013) 537–548 541

Bxk and Byk are control point coordinates and Jxnyg is the mapping Similarly, the shape functions in g
^ direction are obtained. The
from physical to parametric domain. The Mapping from parametric product rule yields the following bi-quadratic shape functions:
to parent domain is the standard mapping as is done in finite ele-
ment. Fig. 1 shows schematic of Isogeometric framework. R21 ð^n; g
^ Þ ¼ M 21 ð^nÞN21 ðg
^ Þ R22 ð^n; g
^ Þ ¼ M 21 ð^nÞN22 ðg
^Þ

3.4. NURBS parent elements R23 ð^n; g

^ Þ ¼ M 21 ð^nÞN23 ðg
^Þ

Various lower and higher order NURBS elements are con- R24 ð^n; g
^ Þ ¼ M 22 ð^nÞN21 ðg
^ Þ R25 ð^n; g
^ Þ ¼ M 22 ð^nÞN22 ðg
^Þ
structed by utilizing various refinement techniques. Linear NURBS
R26 ð^n; g
^ Þ ¼ M 22 ð^nÞN23 ðg
^Þ
element, 9LinNURBS1,2,3/F with 9 control point, constructed by a
knot insertion at {0}, {0.25} and {0.5} locations respectively, qua-
dratic NURBS element, 9QuadNURBS, with 9 control points, k-re- R27 ð^n; g
^ Þ ¼ M 23 ð^nÞN21 ðg
^ Þ R28 ð^n; g
^ Þ ¼ M 23 ð^nÞN22 ðg
^Þ
fined quadratic NURBS element, 16QuadNURBSKR and cubic R29 ð^n; g
^ Þ ¼ M 23 ð^nÞN23 ðg
^Þ
NURBS element, 16CubicNURBS, with 16 control point, k-refined
cubic NURBS element, 25CubicNURBSKR, and quartic NURBS ele- Fig. 2 shows NURBS basis in ^
n and g
^ directions along with the parent
ment, 25QuarticNURBS, with 25 control points and k-refined quar- element.
tic NURBS element, 36QuarticNURBSKR, are developed for
interlaminar stress calculations. In element notation, number in 3.4.2. k-refined quadratic NURBS element (16QuadNURBSKR)
front denotes the number of control points/element and KR de- A tensor product of knot vector {1 1 1 0 1 1 1} in ^
n and g
^
notes k-refinement procedure. A few NURBS elements construction direction results in a bi-quadratic NURBS element, 16Quad-
in parent domain are detailed here. NURBSKR1(F/R). The knot vector is divided into two intervals
and shape functions have unique value over each interval. The
3.4.1. Quadratic NURBS element (9QuadNURBS)) shape functions in ^n direction over each interval are derived as
A tensor product of knot vector {1 1 1 1 1 1} in ^ n and g^ follows:
direction results in a bi-quadratic NURBS element in the parent do- ! !
main. The element consists of 9 control points. The shape function ^n4  ^n ^n4  ^n
in ^
n direction can be written as follows: M 21 ð^nÞ ¼
^n4  ^n2 ^n4  ^n3
! !
^n4  ^n ^n4  ^n ! ! ! !
M21 ð^nÞ ¼ ^n  ^n2 ^n4  ^n ^n5  ^n ^n  ^n3
^n4  ^n2 ^n4  ^n3 M 22 ð^nÞ ¼ þ ^n3 6 ^n < ^n4
! ! ! ! ^n4  ^n2 ^n4  ^n3 ^n5  ^n3 ^n4  ^n3
n^  n^2 n^4  n^ ^n5  n^ ^n  ^n3
! !
M22 ð^nÞ ¼ þ ^n3 6 ^n < ^n4
^n  ^n3 ^n  ^n
^n4  ^n2 ^n4  ^n3 ^n5  ^n3 ^n4  ^n3
! ! M 23 ð^nÞ ¼
n^  n^3 ^n  n^ n^5  n^3 ^
^n4  n3
M23 ð^nÞ ¼
^n5  ^n3 ^n4  ^n3 M 24 ð^nÞ ¼ 0
ð26Þ ð27Þ

Fig. 1. Mapping between physical and parent domain: a framework for Isogeometric finite element analysis.
542 H. Kapoor et al. / Composite Structures 106 (2013) 537–548

results in q  1 continuous derivatives instead of p  1 continuous

derivatives, where p and q are the order of the knot vector before
and after order elevation, respectively.
For interlaminar stress calculation, two other modifications of
k-refined quadratic NURBS elements are formulated by a single
knot insertion in the knot vector at different locations. 16Quad-
NURBSKR2 and 16QuudNUUBSKR3 elements are formed by a tensor
product of similar knot vector, {1 1 1 0.25 1 1 1} and
{1 1 1 0.5 1 1 1} in ^n and g
^ directions, respectively. Fig. 3 shows
k-refined, quadratic NURBS element in the parent domain and
NURBS basis in each direction.

4. Calculation of interlaminar stresses

Fig. 2. Quadratic NURBS element with basis functions in each direction.
The evaluation of transverse stresses, sxz, syz and rzz from the
stress–strain constitutive relations leads to discontinuity at the
interface of the two adjacent layers of the laminate. Three-
dimensional analysis is very complex due to variation in
constitutive laws in the thickness direction and continuity
requirement of transverse stresses and displacement across the
interface. Therefore, 3D elasticity equilibrium equations are used
to calculate the transverse stresses in the kth lamina and the
equations are given as:

@ rkxx @ rkxy @ skxz

þ þ ¼0
@x @y @z
@ rkxyx @ rkyy @ skyz
þ þ ¼0 ð29Þ
@x @y @z
@ skxz @ skyz @ rkzz
þ þ ¼0
@x @y @z
The stresses are calculated using the direct integration meth-
od. It can be seen that the equations for transverse shear stress,
sxz and syz are a first-order equation and solving these equations
result in one constant of integration. However, this problem is
complax as the transverse shear stresses are known at the top
and bottom of the laminate. Therefore, one can only obtain a
non-unique solution as both the traction boundary conditions
cannot be enforced simultaneously. For transverse normal stress,
a second-order equation is derived and require double integration
through the thickness. The transverse stresses are computed di-
rectly as follows,
Z z
@ rxx @ rxy
sxz ¼  þ dz þ C 1
Fig. 3. k-refined Quadratic NURBS element with basis functions in each direction. zk @x @y
Z z
@ rxy @ ryy
syz ¼  þ dz þ C 2 ð30Þ
@x @y
M 21 ð^nÞ ¼ 0 zk
! !
Z z Z
! ! @ 2 rxx @ 2 ryy @ 2 sxy
^n5  ^n ^n5  ^n rzz ¼ þ þ2 dz dz þ zC 3 þ C 4
M 22 ð^nÞ ¼ zk z @x2 @y2 @x@y
^n5  ^n3 ^n5  ^n4
! ! ! ! C1, C2, C3 and C4 are the constants of integration. The constants, C1
^n  ^n3 ^n  ^n ^n  ^n3 n^5  ^n and C2 are obtained from the known values of sxz and syz at either
M 23 ð^nÞ ¼ þ ^n4 6 ^n < ^n5
^n5  ^n3 ^n4  ^n3 ^n5  ^n3 ^n5  ^n3 of the two boundaries at z = ± h/2 while C3 and C4 are obtained from
! ! the known values of rz at z = ± h/2. Also, noteeworthy is that the
^n  ^n4 ^n4  ^n calculation of rz requires the use of third derivative of
M 24 ð^nÞ ¼ displacement.
^n6  ^n4 ^n5  ^n4
ð28Þ
5. Numerical testing
Similarly, the shape functions in g
^ direction are obtained. The
product rule yields 16 bi-quadratic shape functions, R2k , requiring Different boundary conditions, plate to thickness ratios and
4  4 Gauss points. In k-refinement process, the order of the exist- symmetric and anti-symmetric, cross-ply and angle ply laminated
ing knot vector from base model is elevated first and then, a knot is composite and sandwich plate are studied for numerical valida-
inserted instead of elevating the order after knot insertion. This tion. Due to bi-axial symmetry, only a quadrant of the plate is
 H. Kapoor et al. / Composite Structures 106 (2013) 537–548 543

The computed stresses are compared with 3D elasticity solu-

tionPagano and 4Q8Reddy/FE i.e. 4  4 mesh of 8 node quadratic La-
grange element. Interlaminar stresses are found to be in close
proximity to 3D elasticity solution for the k-refined quadratic
NURBS elements. The k-refinement provide smoother in-plane
stress gradients and thus, resulting in smoother interlaminar stres-
ses computation for the coarsest mesh. Figs. 4 and 5 shows com-
parison of interlaminar shear stresses s xz and syz for family of
NURBS elements with 3D elasticity solution. Table 1 shows the
comparison of in-plane and interlaminar stresses computed using
NURBS post-processor with 3D elasticity solution.

Fig. 4. Interlaminar shear stress r

 xz (a/2, 0, z) in, a simply supported (SS1), cross- 5.2. A simply-supported, three layer, cross-ply (0/90/0) square plate
ply (0/90/90/0) square plate under sinusoidal loading.

In this analysis, a simply-supported, square, cross-ply (0/90/0)

Fig. 5. Interlaminar shear stress r
 yz ð0; b=2; zÞ in a simply supported (SS1), cross-
ply (0/90/90/0) square plate under sinusoidal loading.
modeled and interlaminar stresses are computed at nearest Gauss
points of specific location. The boundary conditions are as follows:
SS1 : v 0 ¼ w0 ¼ /y ¼ 0; at x ¼ a=2; u0 ¼ w0 ¼ /x ¼ 0; at y ¼ b=2
SS2 : u0 ¼ w0 ¼ /y ¼ 0; at x ¼ a=2; v 0 ¼ w0 ¼ /x ¼ 0;
symmetry B:C:; x ¼ 0; u0 ¼ /x ¼ 0 y ¼ 0; v 0 ¼ /y ¼ 0
5.1. Simply supported, four layer, cross-ply (0/90/90/0) square plate
In this analysis, a simply-supported, square, cross-ply (0/90/90/
0) laminated composite plate with a/h = 10 ratio, under sinusoidal
loading condition is studied. Material for each lamina is considered
as homogeneous, elastic and orthotropic. Material properties used
are as follows: E1 = 25E2, G12 = G13 = 0.5E2, G13 = 0.2E2, m12 = 0.25,
and Ks = 5/6. 9QuadNURBS with 3  3 mesh and 2  2 Gauss inte-
gration scheme and family of k-refined quadratic NURBS elements,
16QuadNURBSKR, with 2  2 mesh and 4  4 Gauss integration
scheme are used for computing in-plane and interlaminar stresses.
The in-plane stresses are obtained using constitutive equations
with smoother displacement gradients. Interlaminar stresses are
computed from 3D equilibrium equations by direct integration,
without any recovery procedure and are computed from strain
gradients.
laminated composite plate with various length to thickness ratios,
a/h = 10, 20, 100, subjected to sinusoidal loading is studied. Mate-
rial for each lamina is considered as homogeneous, elastic and
orthotropic. Material properties used are as follows: E1 =
1.72369  1011 N/m2, E2 = 6.894757  109 N/m2, G12 = G13 =
3.447378  109 N/m2, G23 = 1.378951  109 N/m2, m12 = 0.25, and
Ks = 5/6. A family of k-refined quadratic NURBS elements, 16Quad-
NURBSKR with 2  2 mesh and two Gauss integration schemes i.e.
3  3 and 4  4, are used.
Interlaminar stresses are computed from direct integration of
3D equilibrium equations, without any recovery procedure. Com-
puted stresses are compared with 3D elasticity solutionPagano, Byun
and Kapania (7  7 mesh of 9 node quadratic Lagrange element
with stress recovery procedure) and 4Q8Reddy/FE (4  4 mesh of
8 node quadratic Lagrange element).
Stresses are found to be in an excellent agreement with 3D elas-
ticity solution for at least one of the elements in the family of k-re-
fined quadratic NURBS elements when compared with 3D
elasticity solution and Byun and Kapania’s results. However, to ob-
tain the most accurate interlaminar stresses, a single knot in the
at y ¼ b=2
knot vector of a particular element of the family of k-refined qua-
dratic NURBS elements, for which the most accurate interlaminar
stress is obtained, is replaced by a new knot. This procedure pro-
duces the most accurate interlaminar stresses which are almost ex-
act to 3D elasticity solution. This new knot produces more
smoother in-plane stress gradients and thus, interlaminar stresses.
This trend is seen for all the length to thickness ratios and for both
integration schemes. Tables 2–4 provide the comparison of in-
plane and interlaminar stresses with 3D elasticity solution, Byun
and Kapania’s and Reddy’s results.
5.3. A simply-supported, two layer, cross-ply (0/90), square laminate
A simply supported, two layered (0/90), square laminate with a/
h ratio = 4, subjected to sinusoidal loading is considered. The mate-
rial properties used here are as follows:
E1 =E2 ¼ 25; G12 =E2 ¼ 0:50; G23 =E2 ¼ 0:20; G13 ¼ G12 ; E2
¼ E3 ; m12 ¼ m23 ¼ m13 ¼ 0:25 ð31Þ

Table 1
 of a simply supported (SS1), cross-ply (0/90/90/0) square plate under sinusoidal loading with 3D elasticity solutionPagano.
Comparison of non-dimensionalized stresses r

a/h Source ð0;0;h=2Þ

r xx r ð0;0;h=4Þ ða=2;b=2;h=2Þ
r xy r ða=2;0;0Þ
xz r ð0;b=2;0Þ
yy yz

10 ELS/Pagano 0.5590 0.4010 0.0276 0.301 0.196

9QuadReddyFE 0.4954 0.3589 0.0240 0.318 0.181
9QuadNURBS 0.5197 0.4029⁄ 0.0263 0.2980⁄ 0.1812
16QuadNURBSKR1 0.5703 0.3959 0.0307 0.4211 0.2606
16QuadNURBSKR2 0.5365 0.4142 0.0283 0.3194 0.2127
16QuadNURBSKR3 0.5478⁄ 0.4193 0.0274⁄ 0.2976⁄ 0.2057⁄
544 H. Kapoor et al. / Composite Structures 106 (2013) 537–548

Table 2
Non-dimensionalized stress r
 in a simply supported (SS1), cross-ply (0/90/0) square plate under sinusoidal loading for a/h = 10.

a/h Source r xx r yy r xy r xz r yz
1
10 16QuadNURBSKR 0.4658 0.2466 0.0237 0.5109 0.1564 44
16QuadNURBSKR1 0.5928⁄ 0.3118 0.0319 0.1755 0.0559 33
16QuadNURBSKR2 0.5571 0.2859 0.0295 0.3995 0.1345⁄ 44
16QuadNURBSKR2 0.5294 0.2738 0.0257 0.3204⁄ 0.1107⁄ 33
16QuadNURBSKR3 0.5664 0.2873⁄ 0.0287⁄ 0.3587⁄ 0.1378 44
16QuadNURBSKR3 0.4816 0.2470 0.0227 0.7397 0.2348 33
16QuadNURBSKROP – – – 0.3571(0.4985) 0.1246(0.275) 44
16QuadNURBSKROP – – – 0.3575(0.29) 0.1259(0.30) 33
ElasticityPagano 0.590 0.288 0.0290 0.357 0.1228
Byun/Kapania 0.513 0.254 0.0252 0.380 0.1106
Reddy/FE 0.5098 0.2518 0.0250 0.4060 0.0908

Table 3
Non-dimensionalized stress r
 in a simply supported (SS1), cross-ply (0/90/0) square plate under sinusoidal loading for a/h = 20.

a/h Source r xx r yy r xy r xz r yz
1
20 16QuadNURBSKR 0.4883 0.1959 0.0221 0.5246 0.1164 44
16QuadNURBSKR1 0.6179 0.2390 0.0263 0.1764 0.0443 33
16QuadNURBSKR2 0.5786 0.2156 0.0250 0.4191 0.1160⁄ 44
16QuadNURBSKR2 0.5513⁄ 0.2088⁄ 0.0231⁄ 0.3305⁄ 0.0880⁄ 33
16QuadNURBSKR3 0.5006 0.2166 0.0253 0.3839⁄ 0.1333 44
16QuadNURBSKR3 0.5890 0.1864 0.0229 0.7641 0.1768 33
16QuadNURBSKROP – – – 0.3851(0.501) 0.0935(0.31) 44
16QuadNURBSKROP – – – 0.3847(0.304) 0.0935(0.275) 33
ElasticityPagano 0.552 0.210 0.0234 0.385 0.0938
Byun and Kapania 0.532 0.200 0.0223 0.390 0.0899
Reddy/FE 0.5281 0.198 0.0222 0.4176 0.0754

Table 4
Non-dimensionalized stress r
 in a simply supported (SS1), cross-ply (0/90/0) square plate under sinusoidal loading for a/h = 100.

a/h Source r xx r yy r xy r xz r yz
100 16QuadNURBSKR1 0.5371⁄ 0.1903 0.0217⁄ 0.4463⁄ 0.0886⁄ 44
16QuadNURBSKR1 0.6256 0.2118 0.0224 0.1750 0.0415 33
16QuadNURBSKR2 0.55668 0.1821⁄ 0.0232 0.5722 0.1459 44
16QuadNURBSKR2 0.5594 0.1853 0.0225 0.3346⁄ 0.0781⁄ 33
16QuadNURBSKR3 0.5788 0.1846 0.0246 0.6651 0.2055 44
16QuadNURBSKR3 0.5069 0.1647 0.0242 0.7754 0.1496 33
16QuadNURBSKROP – 0.181(0.268) – 0.3957(0.0407) 0.0827(0.02) 44
16QuadNURBSKROP – – – 0.3954(0.309) 0.0826(0.275) 33
ElasticityPagano 0.539 0.181 0.0213 0.395 0.0828
Chansup Byun 0.539 0.181 0.0213 0.394 0.0825
Reddy/FE 0.5346 0.1791 0.0212 0.4215 0.0699

Table 5 Computed interlaminar stresses are compared with Kant’s and 3D

Non-dimensionalized interlaminar shear stresses r  xz;yz in a simply-supported (SS1),
elasticity solution. Kant uses various techniques such as finite dif-
cross-ply (0/90) square plate under sinusoidal loading for a/h = 4.
ference and exact surface fitting to compute interlaminar stresses
a/h Source r xz r yz Mesh from in-plane stresses along with HSDT. Cubic and k-refined qua-
4 16QuadNURBSKR1 – – 22 dratic NURBS elements are developed to compute interlaminar
16QuadNURBSKR2 – – 22 shear stresses. In k-refinement procedure, a single knot is inserted
16QuadNURBSKR3 – 0.3541 22 between 1 and 1 and the analysis is run for different knot inser-
16QuadNURBSKROP 0.3157(0.515) 0.3161(0.515) 22
16CubicNURBS 0.3403 0.3688 22
tions to obtain the most accurate stresses. Tables 5 and 6 compares
16QuadNURBSKROP 0.3137(0.515) 0.3151(0.515) 33 the interlaminar shear and normal stresses with 3D elasticity,
16CubicNURBS 0.3316 0.3489 33 Kant’s and Ren’s solutions.
ElasticityPagano 0.3127 0.3188 From Table 6, for cubic NURBS element, the computed r  yz is
Ren 0.32450 0.32580
found to be within 9% and for k-refined NURBS element, com-
HOST7A (9N)(FD) 0.313515 0.294540
HOST12(16N)(FD) 0.316315 0.319463 puted r  yz is found to be within 1% of 3D elasticity solution. In
k-refinement procedure, a knot insertion at a location 0.515 in
 H. Kapoor et al. / Composite Structures 106 (2013) 537–548 545

Table 6
Non-dimensionalized interlaminar normal stress r
 zz in a simply-supported (SS1), cross-ply (0/90) square plate under sinusoidal loading for a/h = 4.

z/h 3D Elasticity HOST11(16 N)(DI) 16CubicNURBS 25CubicNURBSKR 25QuarticNURBS 36QuarticNURBSKR

0.5 1 1 1 1 1 1
0.4 0.973684 0.924730 0.9011 0.9725(0.775,0.775) 0.9051 0.9233
0.3 0.960526 0.817625 0.8023 0.9646(0.7765,0.7765) 0.8102 0.84659
0.2 0.921053 0.698580 0.7034 0.9260(0.7755,0.7755) 0.7152 0.76988
0.1 0.868421 0.58460 0.6046 0.8654(0.774,0.774) 0.6204 0.6932
0.0 0.789474 0.491027 0.5057 0.78832(0.7725,0.7725) 0.5255 0.6165
0.1 0.671050 0.399094 0.3954 0.67387(0.75,0.175) 0.3796 0.4427
0.2 0.460526 0.287604 0.2966 0.5053(0.75,0.175) 0.2847 0.3321
0.3 0.25 0.171356 0.1977 0.2568(0.75,0.045) 0.1898 0.2241
0.4 0.07899 0.072587 0.0989 0.07815(0.768,0.75) 0.0949 0.0767
0.5 0.0 0.0 0.0 0.0 0.0 0.0

Table 7
Non-dimensionalized displacement and in-plane stresses in a simply-supported
(SS2), cross-ply (45/45) square plate under sinusoidal loading.

a/h Source   102

w   102
w r xx r xy
(n = 2) (n = 10) (n = 10) (n = 10)
4 9QuadNURBS 1.1850 0.8645 0.14377 0.14016
9LinNURBS 1.18038 0.8654 0.1386 0.1349
Closed Form 1.1576 0.8450 0.1430 0.1392
10 9QuadNURBS 0.5907 0.27098 0.14377 0.1402
9LinNURBS 0.58395 0.2690 0.1386 0.1350
Closed Form 0.5773 0.2647 0.1430 0.1392
20 9QuadNURBS 0.5053 0.1860 0.14379 0.1403
9LinNURBS 0.4987 0.1838 0.13865 0.1350
Fig. 6. Interlaminar shear stress, r
 xz ða=2; 0; zÞ, in a simply-supported (SS1), cross-
Closed Form 0.4944 0.1818 0.1430 0.1392
ply (0/90) square plate under sinusoidal loading.
100 9QuadNURBS 0.4757 0.1565 0.14379 0.1403
9LinNURBS 0.4713 0.1581 0.13883 0.135368
Closed Form 0.4678 0.1553 0.1430 0.1392

the knot vector of k-refined quadratic NURBS element computes

value of r  yz as 0.3161 when compared to 0.3188 obtained
from 3D elasticity’s solution, 0.319233 obtained from Kant’s
HOST11(16N) (direct integration (DI)) element, which is a cubic,
16 node element with 11° of freedom per node and 0.32580
computed from Ren’s higher order shear deformation theory
model. Fig. 6 shows the comparison of through the thickness
distribution of r  xz with 3D elasticity solution. Similarly, r  yz is
computed using cubic and quadratic NURBS element. Fig. 7 shows
the comparison of interlaminar shear stress, r  yz , with 3D elastic-
Fig. 7. Interlaminar shear stress r
 yz ð0; b=2; zÞ in a simply-supported (SS1), cross-
ply (0/90) square plate under sinusoidal loading.
ity solution.
However, r  zz requires calculation of third-order displacement
derivatives and thus, require higher order polynomial approxima-
tion for derivative calculations. Cubic and quartic NURBS
elements are constructed for computing r  zz . From Table 6, it is
observed that NURBS elements produce better estimate of
through thickness distribution of r  zz as compared to Kant’s solu-
tion. k-refinement procedure enhances the smoothness of strain
gradients and thus, 3D elasticity equivalent interlaminar normal
stresses are obtained for various single knot insertion values.
Fig. 8 shows the comparison of through thickness distribution
of interlaminar normal stress with 3D elasticity and Kant’s
solution.

5.4. A simply-supported, anti-symmetric, angle-ply, square laminate

Fig. 8. Interlaminar normal stress r

 zz ð0; 0; zÞ in a simply-supported (SS1), cross-ply A simply supported (SS2), 2 and 10 layered, (45–45), anti-
(0/90) square plate under sinusoidal loading. symmetric square laminate for various a/h ratios, subjected to
546 H. Kapoor et al. / Composite Structures 106 (2013) 537–548
Fig. 9. Through-the-thickness distribution of interlaminar shear stress,
r xz ða=2; 0; zÞ, in a simply-supported (SS2), anti-symmetric (45/45) square
laminated plate, a/h = 4.
Fig. 10. Through-the-thickness distribution of interlaminar shea stress, r
 zz ð0; 0; zÞ,
in a simply-supported (SS2), antisymmetric (45/45) square laminated plate, a/
h = 4.
Fig. 11. Through-the-thickness distribution of interlaminar normal stress,
r zz ð0; 0; zÞ, in a simply-supported (SS2), antisymmetric (45/45) square laminated
plate, a/h = 10.
Fig. 12. Through-the-thickness distribution of interlaminar normal stress,
r zz ð0; 0; zÞ, in a simply-supported (SS2), antisymmetric (45/45) square laminated
plate, a/h = 50.
sinusoidal loading is analyzed here. The material properties used
here are as follows:
E1 =E2 ¼ 40; G12 =E2 ¼ 0:60; G23 =E2 ¼ 0:5; G13 ¼ G12 ;
E2 ¼ E3 ; m12 ¼ m23 ¼ m13 ¼ 0:25 ð32Þ
Center deflection and in-plane stresses are computed using lin-
ear and quadratic NURBS element and are compared with closed-
form solution. For this particular angle, in-plane stresses do not
vary with increasing a/h ratios. Table 7 shows comparison of center
deflection and in-plane stresses with closed form solution.
9QuadNURBS and 9LinNURBS elements with 2  2 mesh compares
well with the closed form solution.
Next, the interlaminar shear and normal stresses are computed
for antisymmetric, two layered, (45/45), angle-ply laminate using
quadratic and higher-order NURBS elements. In case of quadratic
NURBS element, 2  2 and 4  4 meshes are considered while for
higher-order NURBS elements, 2  2 mesh is considered. Fig. 9
shows the through thickness distribution of r  xz for quadratic and
higher order NURBS elements. Figs. 10–12 shows the through
thickness distribution of r zz for higher order NURBS elements for
various a/h ratios. Most of the elements converges to the same val-
ues of stresses.
5.5. A simply supported, sandwich plate
A simply supported (SS1) sandwich plate subjected to sinusoi-
dally distributed transverse loading is analyzed here. The face
sheet’s material (orthotropic) properties used here are as follows:
E1 ¼ 25E2 ; E2 ¼ 1E6; G12 ¼ 0:50E2 ; G23 ¼ 0:2;
G13 ¼ G12 ; m12 ¼ m13 ¼ 0:25 ð33Þ
Core material (transversely anisotropic) properties are as follows:
E1 ¼ E2 ¼ 1E6; G13 ¼ G23 ¼ 0:06E6; G12 ¼ 0:16E6;
m12 ¼ 0:25 ð34Þ
Each face-sheet is assumed to have one tenth of the total thickness
of the sandwich plate. In-plane stresses are computed using
9QuadNURBS and 9LinNURBS elements, 2  2 mesh and different
Gauss integration schemes (GIS). Linear NURBS element with {0}
as single knot insertion computes in-plane stresses which are very
close to 3D elasticity solution. While 9QuadNURBS element be-
haves similar to 9QuadLagrange finite element with slightly better
in-plane stresses. For example, 9LinNURBS element with {0} as
single knot insertion and with 3  3 Gauss integration scheme
computes value of r  xx ð0; 0; h=2Þ as 1.5896 as compared to 1.556
computed using 3D elasticity solution, 0.8918 computed using
Reddy’s 4  4 mesh of 9QuadLagrange finite element and 0.9522
computed using 9QuadNURBS element with 2  2 mesh and
3  3 Gauss integration scheme. Similar results are obtained for
r yy ð0; 0; h=2Þ as well. However, in case of r xy ða=2; b=2; h=2Þ,
LinearNURBS element with knot insertion at {0.2} and with 3  3
Gauss integration scheme produces 3D elasticity equivalent in-
plane shear stress.
Interlaminar shear stresses are computed using 9QuadNURBS,
16QuadNURBSKR and 16CubicNURBS elements with 2  2 mesh
and 3  3 and 4  4 Gauss integration schemes. k-refined NURBS
element produces interlaminar shear stresses which are closer to
3D elasticity solution than 9QuadNURBS and 16CubicNURBS ele-
ments. Table 8 shows the comparison of in-plane and interlaminar
shear stresses with 3D elasticity and 4  4 mesh of 9QuadLagrange
finite element.
 H. Kapoor et al. / Composite Structures 106 (2013) 537–548 547

Table 8
Non-dimensionalized in-plane and interlaminar shear stresses in a simply-supported (SS1), sandwich square plate under sinusoidal loading.

a/h Source r xx r yy r xy r xz r yz GIS

(0) (0) (0.5)
4 9LinNURBS 0.8714 ) 0.15587 0.1157 – – 22
9LinNURBS 1.5896(0) 0.2651(0) 0.14397(0.2) – 33
9QuadNURBS 0.9167 0.16191 0.0939 0.26094 0.0682 22
9QuadNURBS 0.9522 0.1679 0.0977 0.2645 0.0683 33
16QuadNURBSKR – – – 0.2396(0.25) 0.1078(0.4) 33
16QuadNURBSKR – – – 0.2343(0.45) 0.1078(0.625) 44
16CubicNURBSKR – – – 0.2929 0.0781 33
16CubicNURBSKR – – – 0.2127 0.0789 44
9QuadFEM/Reddy 0.8918 0.1551 0.0901 0.2808 0.0746 33
3D Elasticity 1.556 0.2595 0.1481 0.239 0.1072

6. Conclusions
This research paper describes the development of NURBS
Isogeometric finite element and post-processor for computation
of interlaminar stresses in composite and sandwich plates. First-
order, shear-deformable laminate composite plate theory is
utilized in deriving the governing equations using a variational
formulation. Linear, quadratic, higher order and k-refined NURBS
elements in parent domain are constructed. Lagrange finite
element suffers from higher order stress gradient oscillations
due to Gibbs phenomenon and require alternative stress recovery
procedures. In this paper, a NURBS based post-processor is devel-
oped which computes interlaminar shear and normal stresses
directly from higher order gradients of NURBS basis in a single
step procedure. It is observed that linear NURBS element with
single knot insertion and higher order k-refined NURBS elements
compute 3D elasticity equivalent in-plane and interlaminar stres-
ses respectively.
Acknowledgments
I would like to thank ICTAS at Virginia Tech for funding this re-
search endevour. I would also like to thank Dr. Kapania and Dr.
Soni for their valuable discussions.
References
[1] Piegl L, W Tiller. The NURBS book (Monograph in visual communication). 2nd
ed. Springer-Verleg; 1997.
[2] Farin G. Curves and surfaces. A practical guide. 5th ed. Morgan Kaufmann
Publishers; 1993.
[3] Cohen E, Lyche T, Reisenfield R. Discrete B-Splines and subdivision techniques
in computer aided geometric design and computer graphics. Comput Graph
Image Process 1980;14:87–111.
[4] Rogers DF. Introduction to NURBS with historic prospective. Academic Press;
1995.
[5] Peters J, Reif U. Subdivision surface. Springer-Verlag; 1973.
[6] Warren J, Weimer H. Introduction to NURBS with historic prospective:
subdivision methods for geometric design. Morgan Kaufmann Publishers;
1985.
[7] Gordon W. Spline-blended interpolation through curve networks. J Meth Mech
1969;18:931–52.
[8] Gregory JA. N-sided surface patches. In: Gregory JA, editor. Mathematics of
surfaces. Clarendon Press; 1965.
[9] Loop CT, DeRose TD. A multisided generalization of Bezier surfaces. ACM Trans
Graph 1989;8:204–34.
[10] Bajaj C, Chen J, Xu G. Modeling with cubic a-patches. ACT Trans Graph
1995;14:103–33.
[11] Hughes TJR, Cottrell JA, Bazilevs Y. Isogeometric analysis: CAD, finite elements,
NURBS, exact geometry and mesh refinement. Comput Meth Appl Mech Eng
2005;194:4135–95.
[12] Kagan P, Fischer A, Yoseph PB. New B-Spline finite element approach for
geometrical design and mechanical analysis. Int J Numer Methods Eng
1998;41:435–58.
[13] Kagan P, Fischer A, Yoseph PB. Mechanically based model: adaptive refinement
of B-Spline finite element. Int J Numer Methods Eng 2003;57:1145–75.
[14] Reissner E. On the theory of bending of elastic plates. J Math Phys
1944;23:184–91.
[15] Reissner E. Effect of transverse shear deformation on the bending of elastic
plates. J Appl Mech 1945;12:69–77.
[16] Mindlin R. Influence of rotatory inertia and shear on flexural motions of
isotropic, elastic plates. J Appl Mech 1951;18:31–8.
[17] Whitney J, Pagano N. Shear deformation in heterogeneous anisotropic plates. J
Appl Mech 1970;37:1031–6.
[18] Urthaler Y, Reddy JN. A mixed finite element for the nonlinear bending
analysis of laminated composite plates based on FSDT. Mech Adv Mater Struct
2008;15:335–54.
[19] Chen S, Long YQ, Yao ZH, Chew SP. Application of the quadrilateral area co-
ordinate method: a new element for Mindlin–Reissner plate. Int J Numer
Methods Eng 2006;66:1–45.
[20] Kim KD, Chang SL, Sung CH. A 4-node corotational ANS shell element for
laminated composite structures. Compos Struct 2007;80(2):234–52.
[21] Mia-Duy N, Khennane A, Tran-Cong T. Computation of laminated composite
plates using integrated radial basis functions network. CMC: Comput, Mater
Continua 2007;5:63–77.
[22] Putcha NS, Reddy JN. A refined mixed shear flexible finite element for the
nonlinear analysis of laminated plates. Comput Struct 1986;22:529–38.
[23] Kant T, Kommineni JR. C0 finite element geometrically nonlinear analysis of
fiber reinforced composite and sandwich laminates based on a higher-order
theory. Comput Struct 1992;45:511–20.
[24] Polit O, Touratier M. A multi layered/sandwich triangular finite element
applied to linear and non-linear analyses. Compos Struct 2002:121–8.
[25] Phan ND, Reddy JN. Analysis of laminated composite plate using a higher order
shear deformation theory. Int J Numer Methods Eng 1985;21(12):2201–19.
[26] Ren JG, Hinton E. The finite element analysis of homogeneous and laminated
composite plates using a simple higher order theory. Commun Appl Numer
Methods 1986;2(2):217–28.
[27] Reddy JN, Robbins Jr DH. Theories and computational models for composite
laminates. Appl Mech Rev 1994;47(6):147–69.
[28] Carrera E. Historical review of zig–zag theories for multi layered plates and
shells. Appl Mech Rev 2003;56:287–303.
[29] Wang CM, Reddy JN. Shear deformable beams and plates: relationship with
classical solutions. Oxford (UK): Elsevier; 2000.
[30] Reddy JN, Sandidge D. Mixed finite element models for laminated composite
plates. J Eng Ind 1986;109:39–45.
[31] Kapoor H, Kapania RK. Geometrically nonlinear NURBS Isogeometric finite
element analysis of laminated composite plates. Compos Struct
2012;94(12):3434–47.
[32] Reddy JN. A simple higher order theory for laminated composite plates. J Appl
Mech 1984;51:745–52.
[33] Tessler A. Accurate interlaminar stress recovery from finite element analysis.
NASA technical memorandum, NASA-TM-109149; September 1994.
[34] Byun C, Kapania RK. Prediction of interlaminar stresses in laminated plates
using the global orthogonal interpolation polynomials. AIAA J 1992;30(11).
[35] Lee K, Lee S. A post-processing approach to determine transverse stresses in
geometrically nonlinear composite and sandwich structures. J Compos Mater
1996;37:2207–24.
[36] Park JW, Kim YH. Displacement, stresses and their sensitivity coefficients in
composite panels. J Compos Mater 1999;33(13):1222–43.
[37] Park BP, Park JW, Kim YH. Stress recovery in laminated composite and
sandwich panels undergoing finite rotation. Compos Struct 2003;59:
227–35.
[38] Noor AK, Kim YH, Peters JM. Transverse shear stresses and their sensitivity
coefficients in multi-layered composite panels. AIAA J 1994;32(6):1259–69.
[39] Matsunaga H. Interlaminar stress analysis of laminated composite beams to
global higher-order deformation theories. Compos Struct 2002;55:105–14.
[40] Makeev A, Armanios EA. An iterative method for solving elasticity problems
for composite laminates. ASME J Appl Mech 2000;67(1).
[41] Rolfes R, Rohwer K. Improved transverse shear stresses in composite finite
elements based on the first order shear deformation theory. Int J Numer
Methods Eng 1997;40:51–60.
[42] Kant T, Manjunatha BS. On accurate estimation of transverse stresses in
multilayer composites. Comput Struct 1994;50(3):3365–451.
[43] Kant T, Gupta AB, Pendhari SS, Desai YM. Elasticity solution of cross-ply
composite and sandwich laminate. Comput Struct 2008;83(1):13–24.
548 H. Kapoor et al. / Composite Structures 106 (2013) 537–548
[44] Nosier A, Baharami A. Interlaminar stresses in antisymmetric angle-ply
laminates. Compos Struct 2007;78:18–33.
[46] Barlow J. Optimal stress locations in finite element models. Int J Numer
Methods 1976;10(2):243–51.
[47] Oden JT, Brachli HJ. On the calculation of consistent stress distribution in finite
element approximation. Int J Numer Methods 1971;10(2):243–51.
[48] Hinton JS, Cambell E. Local and global smoothing of discontinuous finite
element functions using a least square method. Int J Numer Methods
1974;8(3):461–80.
[49] Chen DJ, Shah DK, Chan WS. Interfacial stress estimation using least-square
extrapolation and local stress smoothing in a laminated composites. Comput
Struct 1996;58:765–74.
[50] Zienkiewicz OC, Zhu JZ. The super convergent patch recovery and a posteriori
error estimates. Part I: the recovery technique. Int J Numer Methods Eng
1992;33(7):1331–64.
[52] Yuan J, Saleeb A, Gendy A. Projection stress: layerwise-equivalent formulation
for accurate prediction of transverse stresses in laminated plate and shells. Int
J Comput Eng Sci 1982;1(1):91–138.
[53] Prathap G, Naganarayana BP. Consistent thermal stress evaluation in finite
elements. Comput Struct 1995;54(3):415–46.
Further reading
[45] Senthil S, Batra RC. Interlaminar stresses in antisymmetric angle-ply
laminates. AIAA J 1999;37(11):1464–73.
[51] Yuan J, Saleeb A, Gendy A. Projection stress: layerwise-equivalent
formulation for accurate prediction of transverse stresses in laminated
plate and shells. Int J Comput Eng Sci 1982;1(1):91–138.