Available online at www.sciencedirect.

com

International Journal of Non-Linear Mechanics 38 (2003) 1149 – 1162

On the nite displacement analysis of quadrangular plates

A.V. Singh ∗ , Y. Elaghabash
Department of Mechanical and Materials Engineering, University of Western Ontario, London, Ont., Canada N6A 5B9

Abstract
In this paper, a numerical method for the linear and geometrically non-linear static analysis of thin plates is presented. The
method begins with the elasticity equations pertaining to strain components, stresses, displacement components, strain energy
and work due to externally applied loads. The plate geometry is de ned by a quadrangular boundary with four straight edges
and the natural coordinates in conjunction with the Cartesian coordinates are used to map the geometry. The matrix equation
of equilibrium is derived using the work–energy principle with the displacement elds expressed by algebraic polynomials,
the coe6cients of which are then manipulated to satisfy the kinematic boundary conditions. To validate the results from
the present method, square plates having all sides fully xed and all sides simply supported without in-plane movement
are analysed. Comparison is made for the uniformly loaded square plate with the results obtained by Levy who solved
the non-linear plate bending problem using the Th.von Ka’rma’n’s equations. Rhombic plates are examined and numerical
results corresponding to these cases are presented in this paper. Very good comparison of the results regarding de9ection
and bending stresses with other sources available in the literature is found.
? 2002 Elsevier Science Ltd. All rights reserved.

Keywords: Ritz formulation; Nonlinear elastic analysis; Square and rhombic plates

1. Introduction Here, w is the transverse displacement component of

the plate, h the thickness, q the distributed load normal
In 1910, Th. von Ka’rma’n derived the following to the plate, D=Eh3 =[12(1− 2 )] is the 9exural rigidity
two equations for the large de9ection analysis of plates of the plate, E and are the modulus of elasticity and
[1]: the Poisson’s ratio respectively of the plate material,

F the stress function and ∇2 = @2 =@x2 + @2 =@y2 .
2 2 h q @ 2 F @2 w @ 2 F @2 w
∇ ∇ w= + 2 + 2 During the 1930 and 1940, many researchers con-
D h @y @x2 @x @y2
sidered the problem of the bending of rectangular

@2 F @2 w plates undergoing large lateral de9ection. Levy [2,3]
−2 ; solved the above equation for rectangular plates with
@x@y @x@y

2  conditions: (a) simply supported on four sides (NACA
2 2 @2 w @ 2 w @2 w TR 737)) and (b) fully clamped plate under nor-
∇ ∇ F =E − 2 : mal pressure (NACA TR 740). He reported some nu-
@x@y @x @y2
merical results, which have become the standard for
comparison and are used even today by researchers
∗ Corresponding author. working on this problem. These results have been
E-mail address: [email protected] (A.V. Singh). cited literally in numerous papers published in the

0020-7462/03/$ - see front matter ? 2002 Elsevier Science Ltd. All rights reserved.
PII: S 0 0 2 0 - 7 4 6 2 ( 0 2 ) 0 0 0 6 0 - 4
1150 A.V. Singh, Y. Elaghabash / International Journal of Non-Linear Mechanics 38 (2003) 1149 – 1162

Nomenclature

a reference length parameter used for normalization

[dL ] diIerential operator relating {} and {}
E modulus of elasticity
[E] matrix to relate stress to strain
fj (); fk () algebraic polynomials used for de ning the displacements elds
h plate thickness
[k] stiIness matrix
p load per unit area
pa4 =Eh4 dimensionless load parameter
U strain energy
W work by mechanical forces
u; v; w displacement components at the reference surface of the plate
x; y; z cartesian coordinates
x ; y linear strain components at the middle surface
1 ; 2 components of rotation of the normal to the middle surface
xy ; yz ; zx linear shear strains at the middle surface
{ } vector made of the unknown coe6cients
!x ; !y ; !z linear curvature-like terms
"a2 =Eh2 dimensionless stress
#x ; #y ; #xy non-linear strain components
{}T {u v w 1 2 }, displacement vector
{}T {x y xy yz zx !x !y !xy }, strain-curvature vector
[$L ]; [$NL ] linear and non-linear energy functionals
;  parametric coordinates for mapping the geometry

open literature. Wang [4] solved von Karman’s equa- and plate bending behaviours were combined. As
tions numerically for the simply supported rectangular numerical examples, they investigated the de9ection
plates having aspect ratios 1.5 and 2.0. Berger [5] pre- and stresses for simply supported square plates and
sented an approximate method for the large de9ection plates undergoing cylindrical bending. The matrix
analysis of plates. An excellent bibliography on the incremental method was used by Yang [8] for the
non-linear static analysis of plates is given in the book analysis of rectangular plates and he successfully
by Sathyamoorthy [6]. In this book, the author has compared de9ections and stresses with the corre-
enlisted more than 400 journal papers on non-linear sponding results found in the work of Levy [3].
static analysis of circular, rectangular and other shaped Finite element method, using a non-linear quadrilat-
plates. Majority of the papers deals with the problem eral plate-bending-element, was applied by Bergan
solved through classical methods. and Clough [9] to solve stability problems and large
This problem also received considerable interest de9ection analysis of plates and shallow shells. They
from researchers working with the nite element developed the procedure based on the von Karman
methods. Murray and Wilson [7] studied large de9ec- strain expressions and Marguerre shallow shell theory.
tion behaviour of plates and used triangular element Pica et al. [10] presented a geometrically non-linear
based on the KirchhoI’s plate theory. Because of the (GNL) nite element formulation for the bending of
coupling of the in-plane and out-of-plane eIects in Mindlin plates under the assumption of small rota-
the large de9ection plate problem, the plane strain tions and reported numerical results for square, skew,
 A.V. Singh, Y. Elaghabash / International Journal of Non-Linear Mechanics 38 (2003) 1149 – 1162 1151

circular and elliptical plates. They compared de9ec- example: u; v; w are the displacement components; 1
tion and stresses with the results of Levy [3] and and 2 are the components of the rotation of the nor-
Rushton [11]. mal to the middle plane of the plate; z is the distance
Reported in this paper is a numerical (Ritz type) measured from the reference plane along its normal.
formulation procedure for the analysis of non-linear The strain–displacement relations for the geomet-
plate bending problems. The geometry, bounded by rically non-linear analysis of plates described in the
four straight edges, is de ned in Cartesian coordi- Cartesian co-ordinate system is very well established
nate system. Further, the plate geometry is mapped and the reader is referred to the book of Novozilov [12]
using the natural coordinates, because of which the for further details. With some simplifying assump-
mathematics involved in the procedure becomes very tions and also addition of the non-linear terms, one can
simple and straightforward. Plate equations are based obtain the following for the strain components.
on the rst-order shear deformation theory known as


the “Reissner–Mindlin plate theory”. Non-linearity xx = x + za!
R x + #x ; y
= y + za!
R y + #y ;
here is associated with the in-plane strain components
only, whereas the linear form of the transverse shear 
xy = xy + za!
R xy + #xy ; 
yz = yz ;
strain components are used. The displacement elds

zx = zx ; (2)
are de ned by multiplying simple polynomials of the
two natural coordinates and the boundary conditions
where, zR = z=a; a= reference length and
are manipulated by controlling the coe6cients of the
polynomials. Square plates having (a) all sides sim- @u @v
ply supported without in-plane movement and (b) all x = ; y = ;
@x @y
sides fully clamped, are analysed rst and the results
are then compared with the published data by other @u @v @w
xy = + yz = + 2 ;
researchers. Using symmetry about the diagonals, uni- @y @x @y
formly loaded rhombic plates are also analysed and
@w @1
numerical results are presented in this paper with com- zx = + 1 ; !x = ;
parison with the data presented by other sources. @x @x
@2 @1 @2
!y = ; !xy = + ;
@y @y @x
2. First-order shear deformation plate theory
2
2
@w @w
#x = (1=2) ; #y = (1=2) ;
The middle plane bounded by the top and the bot- @x @y
tom planes is used as the reference plane on which
@w @w
the thickness (h) is considered to be very small com- #xy = : (3)
@x @y
pared to the other two dimensions. The displacement
components along the Cartesian axes at a point in the
The component z
does not appear in Eq. (2) because
plate are denoted by u
; v
and w
. Furthermore, the dis-
of the fact that the transverse displacement component
placement components can be expressed in terms of
w is assumed to remain constant over the thickness of
their counter parts at the reference plane plus addi-
the plate. In this formulation, the non-linear strains are
tional variables.
associated with the in-plane (membrane) strain com-
u
= u + z1 ; ponents only. The transverse shear strains and curva-
ture terms retain the form of the linear case. Also, the
v
= v + z2 ; material is assumed to remain elastic during the defor-
mation. Eq. (2) can be assigned the following matrix
w
= w: (1)
form:
In the above, all terms on the right-hand side
are referred from the middle plane of the plate. For {
} = [Z]{} + {#}; (4)
1152 A.V. Singh, Y. Elaghabash / International Journal of Non-Linear Mechanics 38 (2003) 1149 – 1162

where The fth-order matrix [E] is composed of the elas-

tic modulus (E) and the Poisson’s ratio ( ) and its
{
}T = {x
y

xy 
yz 
xz }; non-zero terms are as follows.
{}T = {x y xy yz zx a!x a!y a!xy }; E11 = E22 = E=(1 − 2
); E12 = E21 = E11 ;
T
{#} = {#x #y #xy 0 0} (5) E33 = 0:5E=(1 + ) and
and E44 = E55 = kE33 : (10)
 
1 0 0 0 0 zR 0 0
  Note in the above that a shear correction factor (k) is
0 1 0 0 0 0 zR 0  introduced to accommodate the parabolic distribution
 
  of the transverse shear stresses over the thickness of
[Z] = 
0 0 1 0 0 0 0 zR 
: (6)
  the plate. The value of this factor is (5=6 = 0:8333)
0 0 0 1 0 0 0 0 for the rst-order shear deformation theory of plates.
 
0 0 0 0 1 0 0 0 Substituting Eq. (4) into Eq. (9) and making use of
the symmetry of [E] matrix, we obtain the following:
Use of the above into Eq. (3) gives the following linear
form of the strain components. dU = (1=2)({}T [Z]T [E][Z]{} + 2{}T [Z]T [E]{#}

{} = [dL ]{}; (7) + {#}T [E]{#}) d x dy d z: (11)

where, {}T ={u v w 1 2 } and [dL ], where the sub- Integrating Eq. (11) over the thickness, we can show
script L stands for the linear case, is a matrix with that the strain–energy expression has the following
diIerential operators as given below. form:
 
@
0 0 0 0 U = (1=2) ({}T [D]{} + $NL ) d x dy; (12)
 @x  Area
 
 @ 
 0 0 0 0  where
 @y 
 
 
 @
 @ 
 $L = {}T [D]{};
 @y @x 0 0 0 
 
 
 @  $NL = 2[(D11 #x + D12 #y )x + (D21 #x + D22 #y )y
 0 0 1 0 
 @x 
[dL ] = 

:
 (8) + D33 xy #xy ] + D11 #2x + 2D12 #x #y
 0 @ 
 0 0 1 
 @y  + D22 #2y + D33 #2xy (13a)
 
 @ 
 0 0 0 0 
 @x  and [D]8×8 is composed of the thickness and elastic
 
 properties of the plate.
 @ 
 0 0 0 0 
 @y  R
+h=2
  [D] = [Z]T [E][Z] d z: (13b)
 @ @  R
−h=2
0 0 0
@y @x
Since the majority of the terms in [D]8×8 is zero, only
Value of the strain energy under a given state of stress the non-zero terms are given below.
(or strain) for an in nitesimal volume dV = d x dy d z
is given by D11 = D22 = K0 ; D12 = D21 = D11 ;
dU = (1=2){
}T [E]{
} d x dy d z: (9) D33 = 0:5(1 − )K0 ; D44 = D55 = kD33 ;
 A.V. Singh, Y. Elaghabash / International Journal of Non-Linear Mechanics 38 (2003) 1149 – 1162 1153

D66 = D77 = D0 ; D67 = D76 = D0 ; derivatives with respect to  and  and inverse of the
Jacobian matrix.
D88 = 0:5(1 − )D0 ; The solution procedure is based on the Ritz method
3
in which the process begins with the properly de ned
R −
K0 = E h=(1 2
); D0 = (1=12)E hR =(1 − 2
); displacement elds, also known as the admissible dis-
placement elds. These are so chosen that the es-
hR = h=a: (14)
sential boundary conditions are fully satis ed. In this
investigation, the displacement elds, corresponding
3. Solution procedure to u; v; w; 1 and 2 , are created by the product of
two simple polynomials in  and , respectively. These
The solution procedure is developed here for a plate are
having its middle plane bounded by a quadrilateral n+1 n+1

region in the x–y plane. The co-ordinates of the four u= aRjk fk ()fj ();
corner points are expressed by x(j); y(j) where j = j=1 k=1

1; 2; 3 and 4 in a sequential counter clockwise sense. n+1 n+1

Let us assume that terms x; y and z represent the v= bRjk fk ()fj ();
co-ordinates at an arbitrary point in the body of the j=1 k=1
plate (Fig. 1). n+1 n+1
The procedure developed in this study makes w= cRjk fk ()fj ();
use of the parametric co-ordinates  and  which j=1 k=1
map the quadrangle into a square bounded by n+1 n+1
−1 6 ( or ) 6 + 1. The Jacobian matrix [J (; )] 1 = dR jk fk ()fj ();
and its determinant |J (; )| are used while integrat- j=1 k=1
ing the energy expressions over the surface area of the
n+1 n+1
plate. For example, the in nitesimal area is given by
2 = eR jk fk ()fj (): (16)
d x dy = |J (; )| d d: (15) j=1 k=1

From here on, all the functional forms of the plate The values of j and k are assigned to be 1; 2; : : : ; n +
quantities are expressed in terms of the parametric 1; n= the order of the polynomials fk () and fj ()
co-ordinates  and . The derivatives of the variables in each of  and  directions, respectively; and
such as the displacement and rotation components with aRjk ; bRjk : : : . are two-dimensional arrays of the coef-
respect to x and y can be obtained in terms of their cients representing displacement elds u; v; w; 1


4
3

y

Fig. 1.
1154 A.V. Singh, Y. Elaghabash / International Journal of Non-Linear Mechanics 38 (2003) 1149 – 1162

and 2 , respectively. In all there are (n + 1)2 unknown where { }T = {aRjk bRjk cRjk dR jk eR jk }. There are: m =
coe6cients in each of the above displacement=rotation 5(n + 1)(n + 1) unknown coe6cients in this vector.
components. The parametric function, in this case, is Matrix [F(; )]5×m is made of the product of fk ()
taken to be and fj () as explained below.

fj () = (1 + )j−1 (1 − )n−j+1 [F(; )]

 
for j = 1; 2; 3; : : : ; (n + 1): (17) Fjk (; ) 0 0 0 0
 
 0 Fjk (; ) 0 0 0 
The coe6cients aRjk ; bRjk ; : : : are the unknown param-  
 
eters at this point of the solution stage and have 
= 0 0 Fjk (; ) 0 0 :

the same unit as the displacement components they  
 0 0 0 Fjk (; ) 0 
are associated with. Their values govern the actual  
displacement elds of the plate problem under con- 0 0 0 0 Fjk (; )
sideration. In order to explain this form of the trial
displacement elds, let us consider the expression The rst column basically represents (n + 1)(n + 1)
for w from Eq. (16) as an example. Here, the expan- columns associated with the same number of coe6-
sion is carried over k. cients aRjk for j=1; 2; : : : ; (n+1) and k=1; 2; : : : ; (n+1).
n+1 Similarly the remaining four columns correspond to
w= (cRj1 f1 () + cRj2 f2 () + · · · v; w; 1 and 2 . By substituting Eq. (19) into Eq. (7)
j=1 we get

+ cRjn fn ())fj (): (18)

{} = [dL ][F(; )]{ }: (20)
The above series has control coe6cients cRj1 through
cRjn . The values of w at the two ends are the same
Substituting Eq. (20) into (12), we obtain the stiIness
as the rst and the last control coe6cients cR11 and
matrix for the linear part.
cR1n , respectively, for j = 1. The ends, just mentioned
here, are located at  = −1 and  = +1, respectively. +1 +1
Therefore, to achieve the condition of w =0 at =−1, UL = (1=2) $L |J (; )| d d
−1 −1
the condition required is cRj1 = 0. Also, if the rst two
control points are made equal in value, i.e. cRj1 = cRj2 =0, = (1=2){ }T [kL ]{ }; (21)
the slope of the curve is zero at  = −1. If the last
two control coe6cients are made equal to zero, the +1 +1
where[kL ]= −1 −1 [[dL ][F(; )]]T [D][[dL ][F(; )]]
de9ection w and its derivative at the other end are also |J (; )| d d.
zero. By taking the variation of the energy given in Eq.
As said earlier that Eq. (16) contains the trial (21), one can easily show that
functions for the Ritz solution, in the sense that by
controlling the coe6cients aRjk ; bRjk , etc., the C (0) and
1UL = {1 }T [kL ]{ }: (22)
C (1) conditions can be achieved, it is now possible to
proceed towards the derivation of the stiIness matrix.
Now, in order to obtain the non-linear part of the stiI-
Also, the functions in this equation can be diIeren-
ness matrix, let us introduce a vector as given below.
tiated or integrated without di6culties in terms of 
and  and the Jacobian matrix that relates the two
{ }T = {ux vy xy wx wy }: (23)
co-ordinate systems (x and y) and ( and ).
Eq. (16) can be written in a matrix form as
The second term $NL in Eq. (12) have been dealt with
in full details in Appendix A. By making use of Eq.
{} = [F(; )]{ }; (19) (16), it is possible to write the above equation in the
 A.V. Singh, Y. Elaghabash / International Journal of Non-Linear Mechanics 38 (2003) 1149 – 1162 1155

following form: In vector=matrix form, the work expression can be

written as follows:
{ } = [dNL ][F(; )]{ } = [BNL ]{ }: (24)
W = {cRjk }T {pR jk } = {{aRjk }; {bRjk }; {cRjk }; : : : ; {eR jk }}
The variation of the above yields  

 {0}  
1$NL = {1 }T [DNL ]{ } 
 


 

 
 {0} 

+1 +1  
= {1 }T [BNL ]T [DNL ][BNL ] × {pR jk } : (28)

 

−1 −1 
 

 {0}  

 
 

 
{0}
× |J (; )| d d { }
The form of the above after considering the variation
T is
= {1 } [kNL ]{ }: (25)
+1 +1 1W = {1 }T {p}:
R (29)
In the above, [kNL ] = −1 −1 [BNL ]T [DNL ][BNL ]
|J (; )| d d.
3.2. Point load
We have derived the linear and non-linear parts of
the stiIness matrix for the plate-bending problem. In-
Let us consider a point load (P0 ) applied at an arbi-
tegration here can be carried out numerically. In the
trary point (0 ; 0 ) to the plate in the transverse direc-
present circumstances, we shall use Gaussian quadra-
tion. The transverse displacement component at this
ture for which the number of integration points se-
point under the given load and displacement eld is
lected will depend upon the order of the polynomial
given by
used in the displacement elds.
n+1 n+1
w0 = cRjk fk (0 )fj (0 ): (30)
3.1. Uniformly distributed transverse load j=1 k=1

Consider a uniformly distributed load p applied in The work done by the applied force against the plate
the transverse direction of the plate. The work done by is, therefore,
this force on the plate under the assumed displacement n+1 n+1

eld is W = P 0 w0 = P 0 cRjk fk (0 )fj (0 ): (31)

j=1 k=1
n+1 n+1
W= pw d x dy = p cRjk Here also, the load components associated with each
j=1 k=1 of cRjk is given by the following:
+1 +1 P0 fk (0 )fj (0 ): (32)
fk ()fj ()|J (; )| d d: (26)
−1 −1 At the geometric centre of the plate i.e. at  =  = 0,
The load components associated with each of cRjk is the values of the function fj () de ned in Eq. (17)
given by the following: is unity. The work done by the force on the plate is
given by
+1 +1
{pR jk } = p fk ()fj ()|J (; )| d d: (27) W = {c11 ; c12 ; c13 ; : : :}{p}: (33)
−1 −1

The double integration can be evaluated numerically 3.3. Equilibrium equation

using the Gaussian quadrature with the number of in-
tegration points depending upon the order of the poly- With the stiIness matrix and load vector de ned
nomials fj () and fk (). above, we now work towards the derivation of the
1156 A.V. Singh, Y. Elaghabash / International Journal of Non-Linear Mechanics 38 (2003) 1149 – 1162

equilibrium equation using the work–energy princi- D C

ple. For this let us de ne the potential energy by
5 = U − W . The strain energy expression and work
done by the applied load are both functions of the un-
known coe6cients ({ }) of the displacement elds.
The potential energy at the stable equilibrium of the
O
plate must be at a minimum. This gives us the condi- 
tion of 15 = 0 and the following equation of equilib-
rium.
{1 }T ([k]{ } − {p})
R =0
or A B
(a)
[k]{ } = {p}:
R (34)
In the above, [k]=[kl ]+[kNL ]. It should be kept under y
notice that the stiIness matrix [k] is dependent upon
C(4) η
the deformed con guration of the plate. This makes
the problem non-linear and the numerical computation
procedure requires some careful considerations. Begin
with by assuming f1 = f2 = 0. Calculate stiIness
matrix [k] and for a given load obtain the displacement
components. This solution represents the linear case. (3)
Use the calculated displacement components to obtain
the stiIness matrix [kNL ] and the same given load to
obtain the new displacement components. Repeat the
above procedure, until convergence is established. ξ

x
O(1) B(2)
4. Numerical results and discussions
(b)
The transverse de9ection (w=h) normalised with the
Fig. 2. Geometry of the rhombic plate.
thickness (h) of the plate and the dimensionless form
of the load, pa4 =Eh4 are used in presenting numerical
results for the uniformly loaded plate. Normal stress is geometry is de ned by the side of the rhombus and
also made dimensionless in the form of "a2 =Eh2 . Here, the angle (7) made by the slant side with the vertical
a is one of the prominent length parameters for the axis as shown in Fig. 2a. Assume that the length of
plate, p is the distributed load over the middle surface the side AB is a. The intersecting diagonals for a
and E is the modulus of elasticity for the plate mate- rhombus are orthogonal and the quarters are equal.
rial. The matrix equation of equilibrium is modi ed The quarter section OBC of the rhombus is shown in
accordingly so that the normalized load–de9ection as Fig. 2b in which the corner points are assigned num-
well as the load–stress results can be calculated di- bers from 1 through 4 in the parentheses as shown.
rectly. We can use only one of the quarter sections of the
rhombus for the analysis, if the load is symmetri-
4.1. Analysis of rhombic plates cally distributed. For the uniformly loaded plate, the
geometric conditions used along the axes OB and
This plate con guration is chosen so that the ge- OC of symmetry are (v = 2 = 0) and (u = 1 = 0),
ometric symmetry can be applied conveniently. The respectively. Along the side BC, the following are
 A.V. Singh, Y. Elaghabash / International Journal of Non-Linear Mechanics 38 (2003) 1149 – 1162 1157

Table 1
◦
De9ection and maximum stress at the centre point of a square plate with all sides pinned (7 = 0 )

Load (Q) Centre de9ection (w=h) Centre stress ("a2 =Eh2 )

Ref. [11] Ref. [10] Present ANSYS Ref. [11] Ref. [10] Present ANSYS

0 0 0 0 0 0 0 0 0
9.16 0.335 0.3478 0.3473 0.3488 2.46 2.6016 2.5510 2.5895
36.6 0.818 0.8184 0.8154 0.8168 6.90 6.9787 6.8938 6.9136
146.5 1.47 1.4655 1.4630 1.4643 14.5 14.615 14.7641 14.5439
586.0 2.40 2.3928 2.3970 2.3924 30.0 30.130 30.3603 30.0398
2344.0 3.83 3.8128 3.8240 3.8228 65.2 65.661 62.9723 65.4340
9377.0 6.06 6.0530 6.0560 6.1030 148.3 149.87 140.000 149.720

applied. Along the edge OB (1–2): v = 2 = 0.

u = v = w = 1 = 2 = 0; the clamped condition and 50 57 64 71 78 85 92
u = v = w = 0; the pinned condition: 197 204 211 218 225 232 239

Along the edge BC (2–3– 4): u = v = w = 0.

The displacement elds, given in Eq. (16), are repre-
sented by sixth-order polynomials in each of  and , 14 21 28 35 42 43 44 45 46 47 48 49 56
respectively. This gives us seven coe6cients along  63 70 77 84 91 92 93 94 95 96 97 98
direction and seven along  direction. Consequently, 105 112 119 126 133 140 141 142 143 144 145 146 147
each of the ve displacement elds corresponding to
u; v; w; 1 and 2 , respectively, possesses 49 coef- Linear and geometrically non-linear analyses have
cients. Out of these coe6cients, some are associ- been performed by many researchers in the past on
ated with the edges and when either the boundary rectangular plates. A good amount of numerical re-
or the symmetry conditions are employed, they be- sults is available in the literature, but only a limited
come known (generally equal to zero). The majority quantity of results is truly reliable. The works by
of the coe6cients still remains unknown. To under- Levy [3] and Rushton [11] are widely cited by other
stand the procedure for enforcing the boundary condi- researchers on this problem. Some results are also
tions, let us examine the displacement elds with dou- available due to Pica et al. [10] on this problem. In
ble summations in Eq. (16). The rst set of coe6cients addition, the problem is solved using the ANSYS, a
(1– 49) is assigned to the displacement component u. commercially available general-purpose nite element
The coe6cients (1–7) correspond to (k = 1 and j = 1– computer code. The ANSYS model used the full plate
7), i.e. along the  direction. Similarly, the coe6cients geometry and consisted of (24 × 24) eight-node plate
(8–14) correspond to (k = 2 and j = 1–7) and nally elements. In this paper, comparison of the present
the coe6cients (43– 49) correspond to (k = 7 and j = results with previously published values of de9ection
1–7). For v; w; 1 and 2 , we proceed accordingly and stress is presented and discussed in the following.
and there is a total of (5 × 7 × 7 = 245) coe6cients in Table 1 includes results from four sources, namely;
the model. Rushton [11], Pica et al. [10], the present analysis and
Case I: All edges pinned. With reference to the ANSYS, respectively. The values of the de9ections
model described above, the following boundary con- and maximum stress are compared here. One can no-
ditions are used. tice that the agreement is quite good. De9ection agrees
Along the edge OC (1– 4): u = 1 = 0. better than the stress and the present analysis predicts
stress values somewhat lower than those of the others.
◦ ◦ ◦
1 2 3 4 5 6 7 The rhombic plate with angles 7 = 0 ; 30 and 45
148 149 150 151 152 153 154 and pinned at all four sides are analyzed. Results for
1158 A.V. Singh, Y. Elaghabash / International Journal of Non-Linear Mechanics 38 (2003) 1149 – 1162

Table 2
De9ection and stress at the centre point (O) of rhombic plates with all sides pinned

Load (Q) Centre de9ection (w=h) Centre stress ("a2 =Eh2 )

◦ ◦ ◦ ◦ ◦ ◦
7=0 7 = 30 7 = 45 7=0 7 = 30 7 = 45

0 0 0 0 0 0 0
17.79 0.5494 0.4007 0.2363 4.3041 3.8083 2.9311
38.3 0.8336 0.6494 0.4241 7.0788 6.7236 5.7377
63.4 1.048 0.8393 0.5850 9.4580 9.2178 8.3226
95.0 1.238 1.0070 0.7261 11.7584 11.6475 10.8471
134.9 1.418 1.1650 0.8588 14.1514 14.1098 13.4116
184.0 1.592 1.3170 0.9846 16.6633 16.6684 16.0195
245.0 1.766 1.4670 1.1090 19.3118 19.3944 18.7702
318.0 1.938 1.6150 1.2300 22.1120 22.2542 21.6281
402.0 2.106 1.7570 1.3460 24.9542 25.1545 24.5566

this case regarding de9ection at the centre of the plate Table 3

◦
and the stress are presented in Table 2. The case with De9ection (w=h) at the centre point of a square plate (7 = 0 )
◦ with all sides xed
7 = 0 fully corresponds to the square plate. For this
◦
set, the values of the load parameter (Q) considered Load (Q) 7=0
are 0.0, 17.79, 38.3, 63.4, 95.0, 134.9, 184, 245, 318 Ref. [3] Ref. [10] Present ANSYS
and 402. The de9ection (w=h) decreases as the angle
7 increases for all values of the load parameter (Q). 0 0 0 0 0
17.8 0.237 0.2291 0.2388 0.2391
The stress pattern is similar to that of the de9ection 38.3 0.471 0.4576 0.4734 0.4739
up to Q = 134:9, but for the load parameter (Q) higher 63.4 0.695 0.6781 0.6963 0.6973
than 134.9, the stress values are higher for the case 95.0 0.912 0.8900 0.9089 0.9101
◦
with 7 = 30 than the other two angles. 134.9 1.121 1.0947 1.1140 1.1150
Case II: All edges fully xed. Conditions at edges 184.0 1.323 1.2906 1.3110 1.3111
245.0 1.521 1.4838 1.5030 1.5045
OC, OB and BC of the pinned case are applied. Ad- 318.0 1.714 1.6708 1.6900 1.6920
ditional condition along the edge BC (2–3– 4) used is 402.0 1.902 1.8488 1.8690 1.8706
1 = 2 = 0. In terms of the coe6cients, the condition
is
161 168 175 182 189 190 191 192 193 194 195 196 by node (3). The displacement results from the present
203 210 217 224 231 238 240 241 242 243 244 245 method agree quite well to those obtained by ANSYS.
Found in Table 6 is that the stress values at the centre
Table 3 presents the centre point de9ection for the point of the plate are in very good agreement with the
◦
fully xed square plate for which 7 = 0 . De9ection results of Pica et al. [10], within 3%. But the agreement
results from the present investigation match closely is not that good for the stress values at the mid-side
◦
with those of Levy [3]. The stress results correspond- of the plate, when 7 = 30 .
ing to de9ections of Table 3 are presented in Table 4 Numerical results are also plotted in Figs. 3 and
for the fully xed rectangular plate. In this table, we 4 to examine the eIects of non-linearity on the de-
observe very good agreement in the results from var- 9ection and stresses of pinned and fully xed square
ious sources. The present results do show, however, plates subjected uniformly distributed load. From
better match with the results of Levy [3] and ANSYS Fig. 3, it is seen that the initial rate of de9ection for
than Pica et al. [10]. Results for the fully xed rhom- the pinned boundary condition is much higher than
◦ ◦
bic plates with 7 = 30 and 7 = 45 are presented in that of the xed plate. In both cases, the de9ection rate
Tables 5 for displacements and in Table 6 for stresses decreases with the load and becomes nearly constant
at the centre (point O) and mid side point represented beyond the value of Q = 250. As expected, the xed
 A.V. Singh, Y. Elaghabash / International Journal of Non-Linear Mechanics 38 (2003) 1149 – 1162 1159

Table 4
◦
Stresses at the centre (O) point and the mid-edge point, node (3), of a square plate (7 = 0 ) with all sides xed

Load (Q) Centre stress ("a2 =Eh2 ) Midside stress ("a2 =Eh2 )

Ref. [3] Ref. [10] Present ANSYS Ref. [3] Ref. [10] Present ANSYS

0 0 0 0 0 0 0 0 0
17.79 2.6 2.8929 2.4142 2.5511 5.48 4.8114 5.6836 5.4960
38.30 5.2 6.0767 5.0065 5.2973 11.52 10.199 11.957 11.5223
63.4 8.0 9.3357 7.6223 8.0390 18.03 16.168 18.839 18.1470
95.0 11.1 12.590 10.239 10.7260 25.32 22.778 26.468 25.5290
134.9 13.3 15.840 13.587 13.3757 33.5 30.079 34.452 33.8067
184.0 15.9 19.074 15.687 15.9887 42.4 37.966 44.498 42.9367
245.0 19.2 22.418 18.688 18.6700 52.8 46.615 54.925 53.1900
318.0 21.9 25.845 21.884 21.4020 63.9 55.804 66.460 64.3633
402.0 25.1 29.315 25.261 24.1627 75.8 65.267 78.425 76.1733

Table 5
De9ection at the centre point (O) of rhombic plates with all sides xed
◦ ◦
Load (Q) 7 = 30 7 = 45

Ref. [10] Present ANSYS Ref. [10] Present ANSYS

0 0 0 0 0 0 0
17.8 0.1412 0.1479 0.1480 0.0690 0.0732 0.7326
38.3 0.2940 0.3067 0.3070 0.1473 0.1560 0.1561
63.4 0.4585 0.4754 0.4758 0.2394 0.2529 0.2530
95.0 0.6309 0.6499 0.6506 0.3476 0.3654 0.3656
134.9 0.8061 0.8258 0.8270 0.4703 0.4914 0.4916
184.0 0.9779 0.9976 0.9996 0.6013 0.6245 0.6247
245.0 1.1487 1.1680 1.1709 0.7388 0.7627 0.7633
318.0 1.3140 1.3340 1.3367 0.8762 0.9001 0.9013
402.0 1.4708 1.4910 1.4944 1.0083 1.0320 1.0336

Table 6
Stresses at the centre point (O) and mid-edge point, node (3), of rhombic plates with all sides xed

Load (Q) Centre stress ("a2 =Eh2 ) Midside stress ("a2 =Eh2 )
◦ ◦ ◦ ◦
7 = 30 7 = 30 7 = 30 7 = 45

Ref. [10] Present Ref. [10] Present Ref. [10] Present Ref. [10] Present

0 0 0 0 0 0 0 0 0
17.79 2.2334 2.2334 1.5930 1.6595 3.0991 3.3504 1.6819 1.6420
38.3 4.8374 4.8374 3.4775 3.6143 6.6972 7.2076 3.6455 3.5554
63.4 7.7986 7.7986 5.7852 5.9887 10.974 11.712 6.0640 5.8909
95.0 11.021 11.021 8.5907 8.8485 16.038 17.177 9.1004 8.8295
134.9 14.382 14.382 11.863 12.160 21.925 23.445 12.875 12.408
184.0 17.751 17.751 15.425 15.752 28.524 30.742 17.386 16.798
245.0 21.186 21.186 19.209 19.562 35.950 39.019 22.761 22.086
318.0 24.630 24.630 23.016 23.440 44.020 48.161 28.868 28.212
402.0 28.040 28.040 26.708 27.244 52.452 57.751 35.504 34.955
1160 A.V. Singh, Y. Elaghabash / International Journal of Non-Linear Mechanics 38 (2003) 1149 – 1162

2.5 5. Concluding remarks

2 In this paper, a detailed Ritz-type numerical

Pinned Square Plate procedure for the non-linear static analysis of a
quadrangular-shaped plate is discussed. The geometry
Deflection (w/h)

1.5
is de ned by a plane-region bounded by four straight
edges and subsequently the region is mapped using
1 Fully Fixed Square Plate the natural coordinates (; ). The stiIness matrix and
load vector are derived using the work–energy-based
0.5 variational method with displacement elds de ned
by simple algebraic polynomials. Results from the
present method and nite element methods ([10]
0
0 50 100 150 200 250 300 350 400 450 and ANSYS) compare really well for the square and
Load (Q = pa4/Eh4) rhombic plates. For the square plates fully clamped
and or pinned at the four sides, results from the
Fig. 3. Load versus de9ection curves for the pinned and fully present investigation are found to be in good match
xed square plates. with the published data from the well known paper
by Levy [3]. The method described in this paper is
versatile and can produce reliable and accurate results
for any plate with four straight edges.
30

25
Appendix A. Variation of the non-linear strain
Stress (sigma a2/Eh2)

Pinned Square Plate

20 energy

15
Fully Fixed Square Plate
We begin with Eq. (13a).

10 $NL = 2[(D11 #x + D12 #y )x + (D21 #x + D22 #y )y

5 + D33 xy #xy ] + D11 #2x + 2D12 #x #y

+ D22 #2y + D33 #2xy :

0
0 50 100 150 200 250 300 350 400 450
Load (Q = pa4/Eh4) Here, ux = @u=@x; uy = @u=@y and so on.
The other form of the above
Fig. 4. Load versus stress curves for the pinned and fully xed
square plates.
$NL = D11 [wx2 ux + (1=4)wx4 ]

+ D12 [wy2 ux + wx2 vy + (1=2)wx2 wy2 ]

boundary condition produces less de9ection than that + D22 [wy2 vy + (1=4)wy4 ]
for the pinned case. At relatively large values of the
+ D33 [2xy wx wy + wx2 wy2 ]:
load parameter (Q), the diIerence of the two sets of
de9ections is seen to be constant. With regards to the
The variation of the above functional is
stress, rate of variation shows the same pattern as the
de9ection. The stress values for the xed plate case
1$NL = D11 [wx2 1ux + (2ux wx + wx3 )1wx ]
are lower initially than those for the pinned plate, but
surpass at high values of Q. + D12 [wy2 1ux + (2ux wy + wx2 wy )1wy
 A.V. Singh, Y. Elaghabash / International Journal of Non-Linear Mechanics 38 (2003) 1149 – 1162 1161

+ wx2 1vy + (2vy wx + wy2 wx )1wx ] Knowing that the strain energy has a factor of half,
we can write the above variational form as follows:
 T
+ D22 [wy2 1vy + (2vy wy + wy3 )1wy ]
 ux 

 


 


 v 
y 

 

+ D33 [2wx wy 1xy + (2xy wy + 2wx wy2 )1wx
1$ = 1 xy

 

 wx 
 
+ (2xy wx + 2wx2 wy )1wy ] 
 


 

 
wy
= D11 [2f1 wx 1ux + (2ux + 4f12 )wx 1wx ]   
0 0 0 D11 f1 D12 f2   ux 
  
 

 0 
 vy 
+ D12 [2f2 wy 1ux + 2f1 wx 1vy  0 0 D12 f1 D22 f2  




  

× 0 0 
0 D33 f2 D33 f1  xy ;
+ (2vy + 4f22 )wx 1wx + (2ux + 4f12 )wy 1wy ]  




 D11 f1 D12 f1 D33 f2 d44 d45   wx 
 






  
+ D22 [2f2 wy 1vy + (2vy + 4f22 )wy 1wy ] D12 f2 D22 f2 D33 f1 d54 d55 wy

where
+ D33 [(2f1 wy + 2f2 wx )1xy
d44 = D11 (2f12 + ux =2) + D12 (f22 + vy =2) + 4D33 f22 ;
+ (2f2 xy + xy wy + 8f22 wx )1wx
d45 = d54 = D12 f1 f2 + (1=2)D33 xy ;
+ (2f1 xy + xy wx + 8f12 wy )1wy ] d55 = D12 (f12 + ux =2) + D22 (2f22 + vy =2) + 4D33 f12 ;
1 @w 1 @w
= 2(D11 f1 wx + D12 f2 wy )1ux + 2(D12 f1 wx f1 = ; f2 = :
2 @x 2 @y
+ D22 f2 wy )1uy + 2D33 (f2 wx + f1 wy )1xy The variational form of the non-linear component of
the energy expression can be written in the simpli ed
+ 2(D11 f1 ux + D12 f1 vy + D33 f2 xy form.

1$NL = {1 }T [DNL ]{ };
+ [D11 (2f12 + ux =2)
where
+ D12 (f22 + vy =2) + 4D33 f22 ]wx  
0 0 0 D11 f1 D12 f2
 
 0 0 0 D12 f1 D22 f2 
+ [D12 f1 f2 + (1=2)D33 xy ]wy )1wx  
 
[DNL ] = 
 0 0 0 D33 f2 D33 f1 
:
+ 2(D12 f2 ux + D22 f2 vy + D33 f1 xy  
 D11 f1 D12 f1 D33 f2 d44 d45 
 
+ [D12 f1 f2 + (1=2)D33 xy ]wx D12 f2 D22 f2 D33 f1 d54 d55

+ [D12 (f12 + ux =2) + D22 (2f22 + vy =2) References

[1] S.P. Timoshenko, S. Woinowsky-Krieger, Theory of Plates

+ 4D33 f12 ]wy )1wy : and Shells, 2nd Edition, McGraw-Hill, New York, 1959.
1162 A.V. Singh, Y. Elaghabash / International Journal of Non-Linear Mechanics 38 (2003) 1149 – 1162

[2] S. Levy, Bending of rectangular plates with large de9ections, [8] T.Y. Yang, Finite displacement plate 9exure by the use of
NACA TR 737, 1942. matrix incremental method, Int. J. Numer. Methods Eng. 4
[3] S. Levy, Square plate with clamped edges under normal (1972) 415.
pressure producing large de9ections, NACA TR 740, 1942. [9] P.G. Bergan, R.W. Clough, Large de9ection analysis of plates
[4] C.-T. Wang, Bending of rectangular plates with large and shallow shells using nite element method, Int. J. Numer.
de9ections, NACA TN 1462, 1948. Methods Eng. 5 (1973) 543.
[5] H.M. Berger, A new approach to the analysis of large [10] A. Pica, R.D. Wood, E. Hinton, Finite element analysis of
de9ections of plates, J. Appl. Mech. 22 (1955) 465. geometrically non-linear plate behaviour using a Mindlin
[6] M. Satyamoorthy, Nonlinear Analysis of Structures, CRC formulation, Comput. Struct. 11 (1980) 203.
Press, Boca Raton, FL, USA, 1997. [11] K.R. Rushton, Large de9exion of plates with initial curvature,
[7] D.W. Murray, E.L. Wilson, Finite element large de9ection Int. J. Mech. Sci. 12 (1970) 1037.
analysis of plates, ASCE J. Eng. Mech. Division 95 (1969) [12] V.V. Novozhilov, Foundations of the Nonlinear Theory of
143. Elasticity, Graylock Press, Rochester, NY, 1953.