Comput Mech

DOI 10.1007/s00466-012-0688-8

ORIGINAL PAPER

Upper bound limit analysis of plates utilizing the C1 natural

element method
Shutao Zhou Yinghua Liu Shenshen Chen

Received: 16 June 2011 / Accepted: 2 February 2012

Springer-Verlag 2012

Abstract In this article, a novel numerical solution procedure is proposed to evaluate the upper bound limit load
multipliers for thin plate problems, which incorporates the
C1 natural element method (C1 NEM) with a direct iteration algorithm. Due to its remarkable interpolation property
to the nodal function and the nodal gradient values, the C1
NEM with the C1 - continuous trial function is used here to
deal with the upper bound limit analysis problem of perfectly
rigid-plastic plates. The relevant discrete mathematical programming formulation is established based on the kinematic
theorem of plastic limit analysis, and a direct iteration algorithm with the advantages of simple solution formula and
easy procedure implementation is presented to solve it. Several representative examples governed by the von Mises yield
criterion are investigated. The numerical solutions obtained
in this paper are reasonable and satisfactory, and are in good
agreement with the previously reported results.
Keywords Plastic limit analysis Kinematic theorem
Thin plate C1 natural element method Direct iteration
algorithm

S. Zhou Y. Liu (B)

Department of Engineering Mechanics, AML,
Tsinghua University, Beijing, China
e-mail: yhliu@mail.tsinghua.edu.cn
S. Zhou
e-mail: zhoust07@mails.tsinghua.edu.cn
S. Chen
College of Civil Engineering, Hunan University of Technology,
Zhuzhou, China
e-mail: chenshenshen@tsinghua.org.cn

1 Introduction
The plate is a very important and common structural style,
and its representative geometric feature is that the thickness
is much smaller than the dimensions in other two directions. Plate structures are widely used in numerous practical engineering fields, such as mechanical, civil, marine
and aeronautic engineering, etc. Reliable and accurate estimation of the load-carrying capacity of plate structures is
very crucial and meaningful in fulfilling optimum design of
engineering structures. Plastic limit analysis offers a very
effective and straightforward means to evaluate the limit
loads of plate structures. Up to now, many scholars have
devoted their efforts to this research field and a lot of fruits
have been achieved. On the one hand, many scholars investigated the limit loads of plates and their attempts were focused
on the derivation of the analytical solutions by using different
yield criteria. The commonly used yield criteria include maximum principal stress criterion, Tresca criterion, von Mises
criterion and unified strength theory, etc. and systematic and
comprehensive summaries were reported in the monographs
of Hodge [1], Save and Massonnet [2], Zyczkowski [3], Xu
and Liu [4], Lubliner [5] and Yu et al. [6], etc. On the other
hand, with the rapid development of the computer technology, the researchers more and more tend to concentrate their
attentions on the development of high-efficiency optimization algorithms and numerical solution procedures mainly
with the help of the finite element method (FEM), and a lot
of significant contributions have been published by Hodge
and Belytschko [7], Nguyen [8], Capsoni and Corradi [9],
Turco and Caracciolo [10], Corradi and Vena [11], Corradi
and Panzeri [12], Tran TN et al. [13], Le CV et al. [14]
and Capsoni and Vicente da Silva [15], etc. The meshless
method[1622] is a recently developed numerical method,
which only needs the node information and can effectively

123

Comput Mech

overcome the difficulties brought about by the mesh. As an

alternative and complementary approach to the FEM, it is
gradually widely used in the field of computational mechanics, such as explosion, large deformation, crack propagation, metallic plastic forming, etc. Under this background,
the meshless method is also successfully applied to investigate the limit analysis of plate bending problems in the last
several years. For example, Le et al. [23,24], respectively,
adopted the element-free Galerkin (EFG) method to carry
out the upper bound and lower bound limit analyses of plates
and slabs with the combination help of stabilized conforming
nodal integration technique (SCNI) and second-order cone
programming (SOCP) optimization algorithm, and Le et al.
[25] implemented a kinematic upper bound limit analysis for
plates using the h-adaptive EFG method.
In the plastic limit analysis of thin plates, the Kirchhoff
assumption is adopted and the rotations of the mid-plane
normal line are represented by the derivatives of deflection
according to the classical theory of thin plate bending problems. This demands that the first derivations of deflection are
continuous, in other words, the C1 -continuous trial functions
are required like in the elastic analysis of thin plates. The construction of the C1 -continuous trial functions for plate problems is far from a negligible and easy work, and great efforts
have been devoted to this study. Until now, many investigators have proposed some types of plate elements satisfying the C1 -continuous properties in the framework of the
FEM [2630]. However, though some of the proposed plate
elements have many excellent performances, they also have
some more or less drawbacks, such as restricted-use in the
specific conditions, sensitivity to mesh geometry and difficulty in implementation. Thus, it is necessary to further
explore some alternative methods to construct the C1 -continuous trail functions for plate problems. Compared with
the FEM, the higher-order consistency of the trial function
can relatively easily be achieved in the meshless method. For
example, as a widely adopted technique in the construction of
the trail functions in the meshless method, the high-order and
even the C continuity of the moving least squares (MLS)
approximation can conveniently be fulfilled. However, the
calculations of the MLS approximation involve a large number of matrix products and inversions, and need some userdefined parameters and artificial selected weight functions.
Moreover, the obtained shape functions by using the MLS
approximation do not satisfy the property of Kronecker delta,
which leads to the fact that the additional efforts should be
made to deal with the imposition of essential boundary conditions. Over a decade ago, an effective and accurate numerical method named as the C1 NEM for the calculation of
partial differential equations (PDEs) was proposed and demonstrated by Sukumar and Moran [31]. The C1 NEM is based
on the Voronoi diagram and Delaunay triangulation of the
discrete points and is a variant form of the NEM. This method

123

does not need the node connected information and is not sensitive to the non-uniform distributed nodes, which has a strict
interpolation property to the nodal functions and nodal gradient values and can accurately impose the essential boundary conditions. Additionally, as pointed out by Fischer et
al. [32], the degrees of freedom (DOF) needed in the C1
NEM are fewer than those in some C1 FEMs, which are only
limited to the nodal function values and their first derivatives and do not require their second derivatives and boundary normal derivatives. Some researchers have successfully
employed the C1 NEM to analyse the problems of non-linear
gradient elasticity [32] and CahnHilliard phase-field model
[33]. It can be seen that the C1 NEM offers an alternative
effective numerical method for the treatment of problems
involved higher-order differential equations, and has great
developing potentialities owing to its notable and favorable
advantages.
In this article, a novel numerical solution procedure is proposed for estimating the upper bound limit load multipliers of
perfectly rigid-plastic thin plates obeyed the von Mises criterion, which is implemented in the framework of the C1 NEM.
Based on the kinematic theorem of plastic limit analysis, the
relevant discrete mathematical programming formulation for
upper bound analysis of plate problems is constructed. The
C1 shape functions built on a transformation of the C1 natural neighbour interpolant are utilized to approximate the displacement fields of thin plates. A direct iterative algorithm is
adopted to solve this formulation constrained by an equality.
In this algorithm, the plastic incompressibility condition is
fulfilled in a simple and skillful way by introducing a constant
matrix to modify the goal function, and the nondifferentiable
and nonsmooth characteristic of goal function is overcome
by identifying the rigid zones from the plastic zones at each
iteration step. The implementation of solution procedure provides a simple and effective way to evaluate the upper bound
limit load multipliers of plates because each iterative process
is equivalent to solving a set of associated linear algebraic
equations. The recovery of nodal strain rate is accomplished
by utilizing the MLS approximation technique, and accordingly, the calculation and exhibition of smooth plastic dissipation work rate in the whole plate zones are implemented
and visually depicted, respectively.

2 Kinematic theorem for plastic limit analysis of plates

The theory of structural limit analysis mainly investigates the
problems as follows: what does the maximum applied external load achieve when a structure plastically collapses? For
plastic limit analysis of plates, when one or several points
in a plate enter into the plastic state as the increase of the
applied external load, the plate does not go into the plastic
limit state. Only when the plastic yield-line is produced in

Comput Mech

the plate and leads to the result that the plate is turned into
a geometrically deformable mechanism, the plate just enters
into the plastic limit state [4]. In this case, the deformation
of plate will infinitely develop, and consequently, the loadcarrying capacity of plate will lose. The upper bound limit
load multiplier of plates can be evaluated on the basis of
kinematic theorem of plastic limit analysis. The kinematic
theorem can be represented as follows [4,34]: suppose that
a structure meets geometrical restriction conditions and can
turn into a geometrically deformable mechanism under the
applied external load, and the power rate of the applied external load is not smaller than the internal plastic dissipation
work rate, the displacement velocity field is referred to as the
kinematically admissible velocity field in this case. The upper
bound limit loads are the external loads associated with the
kinematically admissible velocity fields, and the minimum
of which is called as the true limit load or the collapse load.
The upper bound limit analysis of plates can be represented
as the following mathematical programming formulation:

 
(1)
D i j d
= min :


u i ti d = 1 on t

s.t.

(2)

t

u i,i = 0 in

(3)

u i = 0 on u

(4)

where denotes the limit load multiplier of plate, u i is the

kinematically
admissible displacement velocity field, and
 
D i j stands for the plastic dissipation work rate in terms of
the kinematically admissible strain rate
 i j . The plastic dissipation work rate is formulated as D i j = i j i j , where
i j is the stress associated with the strain rate i j according
to the plasticity yield condition. Assume the analysed plates
in this paper obey the von Mises yield criterion, the plastic
dissipation work rate in Eq. 1 can be written as:

 
2 
s i j i j
(5)
D i j = i j i j =
3
where s denotes the yield stress of material.
In Eqs. 14, the following hypotheses are usually adopted
[34]: (1) the structural material is perfectly rigid-plastic, and
elasticity property and hardening effect are ignored; (2) the
structural deformation is small, and the geometrical relationship of deformation is linear; (3) the structure keeps stability
before it reaches the limit state; (4) all the applied external
loads satisfy the proportional loading condition. Equation 2
is the normalization condition and indicates that only the
surface tractions ti are applied on the stress boundary t
and the body forces are ignored. For a thin plate, the kinematically admissible displacement velocity field in Eq. 2 is
T

x = w ,x and y = w ,y , respecu = w x y , where w,

tively, represent the rate of deflection, the rate of x-rotation

and the rate of y-rotation. Equation 3 denotes that the plastic
incompressibility condition should be satisfied at every point
in the entire plate.

3 C1 natural element method

The C1 natural element method (C1 NEM) is a variant of the
NEM, which is also based on Voronoi diagram and Delaunay triangulation of the scattered points. In the C1 NEM,
the trial function is constructed on the basis of a transformation of C1 natural neighbour interpolation and the global
system equation is discretized by employing the Galerkin
method. In the original NEM, there exist several different
natural neighbourhood interpolation schemes, and the Sibson interpolant and the Laplace interpolant (non-Sibson interpolant) are the two widespread schemes currently adopted
to construct the shape function. The obtained shape functions
have many distinct merits, for example, their calculations are
relatively simple in which no complicated matrix operations
are involved and no any artificial parameters are required,
and more importantly, they satisfy the property of Kronecker
delta and the essential boundary conditions can be conveniently imposed as in the FEM. However, as it is pointed
out, respectively, by Sukumar et al. [35,36], and Cueto et al.
[22], the Sibson shape functions are C everywhere except at
nodal locations and the support boundary where they respectively are C0 and C1 , and the Laplace shape functions are
C0 everywhere in their whole support domain. These features limit the application of the NEM in the solution of
problems referred to the C1 -continuous trial functions. Farin
[37] embedded the Sibsons natural neighbour coordinates in
the BernsteinBzier surface representation of a cubic simplex, and constructed the C1 natural neighbour interpolant.
According to the transformation of Farins work, Sukumar
and Moran [31] proposed a computational methodology to
obtain the C1 NEM shape functions which can be adopted to
interpolate the nodal function and nodal gradient values, and
used them to carry out the numerical solution of the PDEs. In
this section, the constructions of the C1 NEM shape functions
will be concisely reviewed.
Assume a point x R2 has n natural neighbour nodes
(n = 5 in Fig. 1), and the Sibson shape
functions associated to


these nodes are denoted as  = 1 (x) , 2 (x) n (x) .
Considering  is the natural neighbour coordinate of x, and
embedding it in the BernsteinBzier representation of a
cubic simplex, the C1 natural neighbour interpolant adopted
in the C1 NEM can be constructed as [31,37]:
u h () =

Bi3 () bi

(6)

|i|=3

123

Comput Mech

y3,0,0,0

y2,0,0,1 y
2,0,1,0

y2,1,0,0
4

y1,0,0,2

y1,2,0,0

y1,1,0,1

y1,0,1,1 y
1,0,2,0
y1,1,1,0

y0,0,0,3
y0,1,0,2

y0,3,0,0

where multi-index i stand for n-tuples of zero-started nonnegative integers. By using the components of i, the norm |i|
and factorial i! defined for a cubic simplex can be written as:
|i| = i 1 + i 2 + + i n = 3 and i! = i 1 !i 2 ! i n !, respectively.Bi3 ( ) denotes the cubic BernsteinBzier basis function with respect to n variables in the Sibsons coordinates
and is defined as [33,37]:
3! i1
(x) 2i2 (x) nin (x)
i! 1

(7)

In addition, bi in Eq. 6 imply the Bzier ordinates associated with the control points yi R2 , where yi represent the
projection of the control points of the cubic Bzier polynomial associated with the (n 1) dimensional simplex onto
the plane [31,33,37]:

Bj1 (i/3)xj
(8)
yi =
|j|=1

herein, xj denote the natural neighbour nodes. Figure 2 illustrates the projection of the control net of a cubic tetrahedron onto the plane. The Bzier ordinates can be divided
into vertex, tangent and center Bzier ordinates, where vertex
Bzier ordinates are equal to the nodal function values, tangent Bzier ordinates are the linear combinations of the nodal
function and the nodal gradient values, and center Bzier
ordinates are related to the centroid of vertex and tangent
Bzier ordinates in order to ensure quadratic precision.
With the purpose of convenient implementation for the
numerical solutions of PDEs based on the Galerkin scheme,
Eq. 6 can be further written as the following matrix form
[31]:
u h () = BT () b = BT () TU =  T () U
with the following defined matrices:

123

y0,1,1,1

y0,0,2,1

y0,1,2,0

y0,0,3,0

Fig. 2 Projection of the control net of a cubic tetrahedron onto the

plane

Fig. 1 Computation of Sibson interpolant shape functions

Bi3 () =

y0,2,0,1
y0,2,1,0

y0,0,1,2

(9)

b = TU

(10)

 () = B () T
T

U = w1 1x 1y wn nx ny
T

(11)
(12)

 () = [
1 ()
2 ()
3 ()
T

3n2 ()
3n1 ()
3n ()]

(13)

2
where

B () and b indicate the column vectors with n +
n
dimensions of the BernsteinBzier basis function and
3
the Bzier ordinate, respectively. U denotes the vector with
regard to the nodal DOF, wn = w(xn ) stands for the nodal
displacement of node n, and nx = w,x (xn ) and ny =
w,y (xn ) represent respectively the corresponding nodal rotations in the thin plate problem.  T () is the column vector
T denotes a
with 3n dimensions of the C1 shape
 function.

n
3n dimensions,
transformation matrix with n 2 +
3
which builds a one-to-one relationship between the Bzier
ordinates and the nodal DOF including both the nodal function value and the gradient values, and can be calculated
by implementing the procedure proposed by Sukumar and
Moran [31].
As illustrated by Sukumar and Moran [31], the shape
function  () used in the C1 NEM has many important
properties, such as positivity, partition of unity, quadratic
completeness, and the remarkable one is the interpolation to
the nodal function and the nodal rotations values, that is:

3I 2 (x J ) = I J ,
3I 1 (x J ) = 0,
3I (x J ) = 0

(14)

3I 2,x (x J ) = 0,
3I 1,x (x J ) = I J ,
3I,x (x J ) = 0 (15)

3I 2,y (x J ) = 0,
3I 1,y (x J ) = 0,
3I,y (x J ) = I J (16)
This property indicates that shape functions
3I 2 ,
3I 1,x
and
3I,y of node I are respectively associated with three

Comput Mech

y
0.5
1

0.75

0.5

0.5

0.5

0.25

-0.5

0
-0.5
-0.25

-0.25

Y0

0.25

0.25
0.5 0.5

0.5

0.0000 0.1111 0.2222 0.3333 0.4444 0.5556 0.6667 0.7778 0.8889 1.0000

(b)

(a) nodal grid

0.025

0.025

3I 2

-0.025

-0.025
0.5

0.5
0.25

0.25
0

Y 0

0.25

0.25
Y

-0.25

-0.25

0.5

0.5
0

-0.5 -0.5
-0.0332 -0.0258 -0.0185 -0.0111 -0.0037 0.0037

-0.25

-0.25
-0.5 -0.5

0.0111

0.0185

0.0258

0.0332

(c) 3 I 1

-0.0332 -0.0258 -0.0185 -0.0111 -0.0037 0.0037

(d)

0.0111

0.0185

0.0258

0.0332

3I

Fig. 3 The shape function of C1 NEM

nodal DOF w I , I,x and I,y . The shape functions of the

C1 NEM are illustrated in Fig. 3. Meanwhile, though the evaluation of C1 shape function involves the basic linear algebra
matrix-vector computations, its computational costs are not
observable more than its C 0 variant. Thus applying the C1
NEM to investigate the plastic limit analysis of plate bending
problems is an appealing and interesting choice. In the next
section, we will construct the mathematical programming
formulation for the upper bound limit analysis of plates by
using the C1 NEM and solve it by utilizing a direct iteration
algorithm.

4 Upper bound limit analysis of plates based

on the C1 NEM
4.1 The kinematic formulation of plates using the C1 NEM
Owing to its appealing features presented previously, the
C1 NEM is adopted to investigate upper bound limit analysis problem of thin plates specified in Eqs. 14. In the plastic limit analysis of thin plates, the plates are made up of
perfectly rigid-plastic material and the similar assumptions
are used as in the elastic analysis of plates, namely [4]: (a)

123

Comput Mech

the perpendicular lines of the mid-plane in undeformed state

are still perpendicular to the mid-plane after the deformation, that means the Kirchhoff assumption is obeyed; (b) the
deflection of mid-plane is far less than the thickness of plate;
(c) the stress z in the thickness direction is much smaller than
other stress components and can be ignored, which states that
thin plate can be approximately regarded as in the plane stress
state. According to these three assumptions, thin plate bending problem can be simplified as two-dimensional case, and
the strain can be represented by the deflection of mid-plane.
In the following discussion, we will establish the kinematic
formulation of plates based on the C1 NEM.
Assume the x y plane is set to coincide with the mid-plane
of plate, and the z axis is perpendicular to the mid-plane.
In the C1 NEM, the problem domain used to describe the
plate structure is discretized by a series of scattered nodes,
and the C1 shape function  can be easily calculated by utilizing computational methodology stated above. If n nodes
are identified as the natural neighbouring nodes of an integral
point by using the empty circumcircle criterion and their dis the generalized
placement velocity vector is denoted as U,

strain rate of this integral point can be evaluated as = BU

where B represents the strain-displacement relation matrix.
Based on the relationship = z between the strain rate
the expression of i j i j in
and the generalized strain rate ,
Eq. 5 for thin plate problem can be deduced as:
= z 2 U

T BT DBU
T K U
i j i j = D = z D = z U
T

2 T

the following matrix form can be obtained:

i j i j = 2x + 2y + 2z + 2 2x y + 2 2yz + 2 2zx


1 2
2
2
= 2 x + y + x y + x y
4


1 2
2
2
2
= 2z x + y + x y + x y
4
2 T
= z U KU

have respectively been given

where the two matrices K and U
in Eqs. 18 and 21, and the matrix D is written as follows:

210
D = 1 2 0
(24)
0 0 0.5
By introducing a constant matrix D, the goal function term
has already contained the plastic incompressibility condition
of thin plate problem.
and K into the
Respectively extending and assembling U
and the global
global nodal displacement velocity vector U
matrix K, and using the traditional quadrature rule to implement the numerical integration, the goal function can be discretized as follows:


2
s
3


i j i j d =

(17)
K = B DB

(18)

= [ x y x y ]T

T
= x y x y
=
w w T
U
1 1x 1y
n nx ny

(19)
(20)
(21)


=

where
T

(23)


=

2
s
3

2 

h2

i j i j dSdz


2 s h 2
3 4

T K Ud
S
U

S S

2
MP
i |J|i
3


TKi U
(25)
U

iI G

where S stands for the domain of plate mid-plane, S denotes

the integral domain which is obtained by the discretization
2
of S, MP = s4h is the plastic limit bending moment of plate

1,x x
2,x x
3,x x

3n2,x x
3n1,x x
3n,x x
B =
1,yy
2,yy
3,yy . . .
3n2,yy
3n1,yy
3n,yy
2
1,x y
2,x y 2
3,x y
2
3n2,x y 2
3n1,x y 2
3n,x y
K is a matrix corresponding to the natural neighbouring nodes
of an integral point. D is a constant matrix which will be given
in the following derivation. w n , nx = w n,x and ny = w n,y
respectively denote the rate of deflection, the rate of x-rotation and the rate of y-rotation for the nth node.
As stated above, the stress state of thin plate is similar to
plane stress state. This means z = x z = yz = 0,x z =
yz = 0 and z = 0. The plastic incompressibility condition
indicated in Eq. 3 can be written as z = ( x + y ), which is
subsequently substituted into the expression of i j i j . Then

123

(22)

with the thickness h, I G denotes the total integral points of

the entire plate, and i and |J|i respectively stand for the integration weight and the Jacobian determinant at the integral
point i.
Substituting the discrete displacement Eq. 9 into the normalization condition Eq. 2, we have:
=1
FT U

F =  T tdS
S

(26)
(27)

Comput Mech

where F represents the equivalent nodal load vector associated with the prescribed transverse load t and  is the C1
shape function matrix.
To sum up, the discrete mathematical programming formulation for the upper bound limit analysis of plate problems
by using the C1 NEM can be constructed as follows:



2
TKi U

i |J|i U
= min :
(28)
MP
3
iI G

=1
s.t. FT U

(29)

The goal function Eq. 28 is subjected to an equality constraint and the plastic incompressibility condition has been
introduced into it. How to present a simple and direct numerical algorithm to overcome the nonlinear and nondifferentiable characteristic of the goal function is crucial for upper
bound limit analysis of plates. In the following discussion,
our attention will be concentrated on the establishment and
implementation of numerical iterative algorithm.
4.2 The establishment of numerical iterative algorithm
In this section, a direct iterative algorithm, which is initially recommended by Zhang et al. [38], is adopted to solve
the mathematical programming formulation in Eqs. 2829.
According to the necessary minimization condition of the
computational formulation, the normalization condition is
substituted into the goal function by utilizing the Lagrange
multiplier , and the following minimization problem can be
constructed:







2
TKi U
FT U
1

MP
i |J|i U
L U, =
3
iI G

(30)
First assume the strain rate at every integral point is nonzero,
that implies:
= 0, i I G
TKi U
U

(31)

L U,
L U,
In the light of ( U ) = 0 and ( ) = 0, we obtain:


2
Ki U
MP
= F
(32)
i |J|i 
3
TK U

U
iI G

=1
FT U

(33)

It is clearly seen that Eq. 32 is nonlinear and hard to be solved

directly. Considering this fact, we linearize Eqs. 3233 by
constructing the numerical iterative formula as follows:


k
2
Ki U
i |J|i 
= k F
(34)
MP
T
3

iI G
U k1 K i U k1
k =1
FT U

(35)

k and k respectively denote the nodal displacement

herein, U
velocity vector and the Lagrange multiplier which are the two
k1 is the nodal
unknown quantities at iterative step k, and U
displacement velocity vector at iterative step (k 1) and is
known before the kth iteration. That means Eqs. 34 and 35
k and k .
are just the linear equations of U
However, when Eq. 31 is not always satisfied and the strain
rates at some integral points are zero, the numerical iterative
formula in Eqs. 3435 is meaningless. So the priority task
before the kth iteration is that the strain rate at each integral
point should be inspected to ensure each iteration
to be per
T
k
k1 K i U
k1
formed successfully. In this way, let X i = U
and divide the total integral points I G into the rigid zone subset Rk and the plastic zone subset Pk before the kth iteration
by using the following partition criterion:
I G = Pk Rk



Tk1 K i U k1
= 0
Pk = i I G, X ik = U



Tk1 K i U
k1 = 0
Rk = i I G, X ik = U

(36)
(37)
(38)

The numerical calculation error exists in the practical calculation. When the value of X ik is too small and tends to zero,
let X ik equal to a positive decimal which is far less than the
integer 1, that means:


Rk = i I G, X ik = , << 1

(39)

The partition of rigid and plastic

 zones is realized based on
k
Tk1 K i U
k1 . The positive
the inspection of value X i = U
8 ) X
in the actual
decimal is usually set as (104 10
IG
1

numerical computation, where X = I G i=1 X i1 and X i1 =


T0 K i U
0 . If the value of X k for integral point i at iteration
U
i
step k(k 1) is greater than , this means that integral point
i is in the plastic
 state and locates in the plastic zone, and
Tk1 K i U
k1 in this case. If the value of X k
we set X ik = U
i
for integral point i at iteration step k(k 1) is smaller than
or equal to , this means that integral point i is in the rigid
state and locates in the rigid zone, and we let X ik = in
this situation. The division of rigid zones and plastic zones
can distinguish the rigid integral points from total integral
points and can guarantee the goal function to be meaningful
and logical at each iterative step. For purposes of brevity in
the next discussions, we define an intermediate matrix H ik as
follows:

H ik

2
i |J| i K i
MP
3
X ik

(40)

123

Comput Mech

In this way, the numerical iterative formula of Eqs. 3435

can shortly be rewritten as the following form:

k = k F
H ik U
(41)
iI G
T

F Uk = 1

(42)

Accordingly, the minimization problem corresponding to the

solution of Eqs. 4142 can be represented as:
T
Hk U

= min :
U
(43)
i
iI G

=1
s.t. FT U

(44)

Equations 4344 is a quadratic minimization problem with

is the unknown optimization
an equality constraint, where U
variable and H ik is the known intermediate matrix at iteration
step k. Using the Lagrange multiplier method to introduce the
normalization condition to the goal function, we have:


T


=
Hk U
2 FT U
1
U
(45)
L k U,
i
iI G

where denotes the Lagrange multiplier. Assembling Gk =


k
iI G H i and introducing the displacement velocity boundary conditions Eq. 4, we can obtain:


) = U
T Gk U
2 FT U
1
(46)
L k (U,
)
L k (U,

= 0 and
According to the minimization conditions
)
L k (U,
= 0, we can see that the solution of Eqs. 4344

is equivalent to solving the following linear algebraic equations:

= F
Gk U
T
F U=1

(47)
(48)

Setting:

iI G

s.t.

=1
FT U

(52)

On the basis of the mathematical treatments in Eqs. 4548,

Eqs. 51 and 52 are equivalent to solving a set of linear algebraic equations. In this way, the initial nodal displacement
0 can be calculated out and the initial limit
velocity vector U
load multiplier can be obtained as follows:



2
T0 K i U
0
i |J|i U
(53)
0 =
MP
3
iI G

Iterative Step k(k = 1, 2, ): According to the computa k1 at iterative step (k 1), we firstly detertional results U
mine the rigid zone subset Rk and the plastic zone subset
Pk by using the partition criterion given in Eqs. 3638, and
accordingly obtain the values of X ik and H ik . Then, solving
the minimization mathematical programming problem as follows:
T
HkU

U
(54)
= min :
i
iI G

=1
s.t. FT U

(55)

Similarly, the above programming formulation is equivalent

to solving Eqs. 4748. Thus, the nodal displacement veloc k can be got and the limit load multiplier at this
ity vector U
iterative step can be written as:



2
Tk K i U
k
MP
i |J|i U
(56)
k =
3
iI G

=
U

(49)

Respectively substituting Eq. 49 into Eqs. 47 and 48, we can

1
accordingly gain = G1
k F and = FT . Substituting and
back into Eq. 49, we can easily calculate the nodal displace Finally, the limit load multiplier can
ment velocity vector U.
be written as:



2
TKi U

MP
i |J|i U
(50)
=
3
iI G

4.3 The implementation of numerical iterative algorithm

In accordance with the aforementioned numerical procedure,
we construct the following direct iterative algorithm.
Iterative Step 0: At the beginning of iteration, suppose the
whole plate is in perfectly plastic state. Let X i0 = 1.0 and

123


H i0 = 23 MP i |J|i K i . Solving the following initial minimization mathematical programming problem:
T
H i0 U

U
(51)
= min :

The above-mentioned iterative process will be stopped when

the following two convergence conditions are fulfilled:


U
k1 
k U
|k k1 |


vol1,
vol2
(57)
U
k1 
k1
where vol1 and vol2 are the error tolerances and are decided
by the anticipated computation precision.
In the above iterative procedure, the plastic incompressibility condition is introduced into the goal function. By
distinguishing the rigid zones from the plastic zones and
modifying accordingly the goal function at each iterative
step, the difficulties triggered by the nondifferentiable and
nonsmooth characteristic of goal function are overcome. Each
iteration is equivalent to solving a relevant linear elastic problem. It is clearly seen that this iterative algorithm has the
advantages of simple solution formulation and easy procedure implementation.

Comput Mech

at the limit states are displayed by employing the associated

drawing software.
5.1 Circular plate

q
h

x
Z

Fig. 4 Geometric parameters of clamped circular plate

5 Numerical examples
In this section, some numerical examples of plastic limit
analysis for thin plate problems are considered, which are
employed to investigate the performances of the present solution procedure. The trial function of the C1 NEM is constructed on the basis of a transformation of the C1 natural
neighbour interpolant in the framework of the Galerkin
method, and the numerical integration is carried out by adopting 3 point quadrature rule in each Delaunay triangle unless
otherwise indicated. All the thin plates considered in the following are governed by the von Mises criterion and made
up of perfectly rigid-plastic material. The relevant material
parameters are assumed as: the yield stress s = 200MPa,
the Youngs modulus E = 2.1 105 MPa, the Poisson ratio
v = 0.3, basis load pressure q = 1.0 N/m2 . The error tolerances are chosen as vol1 = vol2 = 1.0 104 .
Because the shape functions of the Sibson interpolant are
nondifferentiable at the nodal locations, and the C1 natural
neighbour interpolant is constructed by embedding the Sibsons natural neighbour coordinates in the BernsteinBzier
surface representation of a cubic simplex, the obtained C1
shape functions are also nondifferentiable at the nodal locations. Therefore, the nodal strains and stresses of plates in the
C1 NEM cannot be gained directly as in the Sibson-based
NEM. Tabbara et al. [39] presented a simple and accurate
method to recover the nodal strain by adopting the MLS technique. Owing to the motivation presented by Tabbara et al.,
this article firstly obtains the nodal displacement velocities
and the strain rates at the Gaussian points of plates by using
the C1 NEM, and then approximates the nodal strain rates
in view of the strain rates at the Gaussian points by using
the MLS technique. By this means, the smooth strain rate
fields of the whole plates are fitted, and the smooth plastic
dissipation work rates can be easily calculated in the light of
Eq. 5. The visualizations of the plastic dissipation work rates

Initially, a clamped circular plate subjected to uniform transverse pressure q is studied, as shown in Fig. 4. The radius
and thickness of the circular plate are respectively assigned as
R = 1.0m and h = 0.01m. Six different nodal distributions
represented in Fig. 5 are used to implement the upper bound
limit analysis of this circular plate by using 3 point quadrature rule in each Delaunay triangle, and the corresponding
numerical limit load multipliers are listed in Table 1. The
analytical solution 12.5MP /q R 2 provided by Hopkins and
Wang [40] for the clamped circular plate with uniform load
is also given in Table 1. It is obvious that the present upper
bound limit load multipliers gradually tend to a stable value as
the increase of the node number, and the numerical solution
with 801 nodes is 12.486MP /q R 2 and just 0.112% slightly
smaller than the analytical solution. Figure 6 indicates the
plastic dissipation work rate distribution of clamped circular plate at the limit state when the upper bound limit load
multiplier is 12.486MP /q R 2 . Figure 7 illustrates the iterative convergence processes of the limit load multipliers for
clamped circular plate with six different nodal distributions,
where we can see that the limit load multipliers decrease
monotonically to the anticipated computation precision just
after 69 iterative steps.
As pointed out by Sukumar and Moran [31], the numerical integration errors in the C1 NEM originate from the
non-polynomial form of integrand shape functions and the
supports of the shape functions are not coincident with
the Delaunay triangles, and the selection of an appropriate numerical quadrature scheme in the C1 NEM is still an
open issue. In order to compare the computational accuracies of using different quadrature rules, we have respectively
adopted 3, 6, 13, 25 integral points to carry out the numerical
integrations when the circular plate is respectively discretized by 457 and 801 nodes. The relevant numerical results are
illustrated in Table 2. The numerical results clearly indicate
that different quadrature schemes lead to different numerical computational errors, all these errors between numerical
results and analytical solution are very small, and the errors
of using 3 point quadrature scheme are smaller than those
of using other quadrature schemes. We think that adopting 3 point quadrature scheme to implement the numerical integration in the C1 NEM for the upper bound limit
analysis seems to be a sufficient, reasonable and effective
choice. We derive this conclusion based on the following
factors: (1) the numerical results of Sukumar and Moran
[31] and our elastic analysis indicate that the elastic displacement fields obtained by using 3 point quadrature scheme are
less accurate than those obtained by using 6, 13, 25 point

123

Comput Mech
Fig. 5 Six different nodal
distributions for circular plate

(a) 121 nodes

(b) 209 nodes

(d) 457 nodes

Table 1 Limit load multipliers
of circular plate forsix different

nodal distributions MP /q R 2

(e) 617 nodes

(f) 801 nodes

Nodes

121

209

321

457

617

801

Limit load multipliers

12.814

12.696

12.609

12.541

12.488

12.486

Analytical solution [40]

12.5

quadrature schemes, but the differences between them are

very small. It seems that using 3 point quadrature scheme
can also obtain relatively satisfied and accurate displacement
velocity fields. Thus, the limit load multipliers calculated by
using the displacement velocity fields and 3 point quadrature
scheme in our upper bound limit analysis can be considered
as the relatively satisfied and accurate ones; (2) the investigations on the quadrature of the clamped support circular
plate and the underneath-mentioned simply supported and
clamped support square plate indicate that the upper bound
limit load multipliers respectively obtained by using 3, 6, 13,
25 point quadrature schemes are close to each other under the
same condition, though sometimes the differences between
some of them are slightly distinct; (3) according to the theorem of upper bound limit analysis, the smallest upper bound
limit load in all the upper bound limit loads is the real limit
load. So we prefer to choose the smallest one from the abovementioned upper bound limit load multipliers respectively
obtained by using 3, 6, 13 and 25 point quadrature schemes
when they are all relatively accurate and satisfied; (4) the last
factor is on the computational efficiency. Using 3 point quadrature scheme costs less computational time than adopting 6,
13, 25 point quadrature schemes.

123

(c) 321 nodes

Fig. 6 The plastic dissipation work rate distribution

 of clamped
 circular plate discretized by 801 nodes at the limit state 106 N m

Comput Mech
14.6

limit load multiplier MP / qR

14.4
14.2
14.0

121 nodes
209 nodes
321 nodes
457 nodes
617 nodes
801 nodes

13.8
13.6
13.4
13.2

IV

III

II

13.0
12.8
12.6
12.4
0

10

iterative step k
Fig. 7 The iterative convergence processes of the limit load multipliers
for clamped circular plate
Table 2 The error comparison for limit load multipliers
of clamped


circular plate by using different quadrature schemes MP /q R 2
Quadrature

Analytical
results [40]

12.5

457 nodes

801 nodes

Numerical
results

Errors
(%)

Numerical
results

Errors
(%)
0.112

12.541

0.328

12.486

12.981

3.848

12.935

3.480

13

12.806

2.448

12.714

1.712

25

13.025

4.200

12.884

3.072

5.2 Rectangular plate

A rectangular plate with uniform transverse pressure q and
different displacement velocity boundary conditions is investigated herein. Figure 8 depicts the geometric parameters of
this rectangular plate with clamped displacement velocity
boundary conditions, where the width a = 1m and the thickness h = 0.01m. In this example, four different length-width
ratios of rectangular plate are used and the corresponding
nodal layouts are respectively 29 15, 33 15, 37 15 and
41 15. Table 3 summarizes the computational limit load
multipliers, which reveals that the limit load multipliers of
rectangular plate increase as the growth of the length-width
ratio under clamped, simply supported, 3 clamped and 1 short
side supported, and 2 clamped and 2 short sides supported
displacement velocity boundary conditions, and the opposite conclusion can be drawn under 3 clamped and 1 long
side free, and 2 clamped and 2 long sides free displacement
velocity boundary conditions.
Additionally, as summarized by Fischer et al. [32] and
Rajagopal et al. [33], the C1 NEM has some advantages
over other traditional numerical methods. For example, it is
not limited to regular non-deforming grids, demands fewer

Fig. 8 Geometric parameters of clamped rectangular plate

DOF than some classical C1 FEMs, and can easily impose

the essential boundary conditions, etc. This paper chooses a
famous rectangular non-deforming plate element named as
the ACM [41] to compare with the C1 NEM. Meanwhile, in
order to further investigate the influence of irregular nodal
arrangement on the numerical results of the C1 NEM, this
paper adopts one regular nodal distribution and three irregular
nodal distributions when the length-width ratio of rectangular plate is b/a = 2, as shown in Fig. 9. Le et al. [23] studied
this example by adopting the EFG method and second-order
cone programming. The numerical results obtained by this
article are compared with the results of Le et al. in Table 4.
The comparisons and our computational experiences reveal
that: (1) the limit load multipliers obtained in this paper
agree well with the available ones; (2) different nodal distributions influence the numerical results obtained in the C1
NEM, though this influence is very little and not obvious;
(3) the C1 NEM is more flexible than the ACM, which can
be used to calculate the examples not only with the regular
nodal distributions but also with the irregular ones. Meanwhile, the numerical examples investigated in this paper also
indicate that the C1 NEM can be used to analyse not only
the rectangular plates but also other shapes of plates; (4) in
contrast to the EFG method used by Le et al., the C1 NEM
can more easily impose the essential boundary conditions
as in the ACM and other FEMs; (5) the drawback of the
C1 NEM is that it costs more computational time than the
ACM.
When the length-width ratio of the rectangular plate is
b/a = 2, the plastic dissipation work rate distributions of
29 15 regular nodes at the limit states under four different
displacement velocity boundary conditions are respectively
displayed in Fig. 10. For the same length-width ratio and
the same nodal arrangement, the iterative convergence processes of the limit load multipliers are shown in Fig. 11.
It is obvious that the solution algorithm has the fast convergence because the calculated limit load multipliers monotonically decrease to the steady minimums only after few itera-

123

Comput Mech
Table 3 Limit load multipliers of rectangular plate for four different length-width ratios (MP /qab)
b/a

Clamped

Simply supported

3 Clamped
1 Short side free

3 Clamped
1 Long side free

2 Clamped
2 Short sides free

2 Clamped
2 Long sides free

2.0

55.09

28.90

45.18

18.82

40.39

9.28

2.5

63.47

33.48

54.59

18.31

49.94

7.63

3.0

71.93

37.57

64.00

18.15

59.61

6.33

3.5

79.27

39.81

72.47

18.01

68.63

5.36

Fig. 9 The regular and

irregular arrangements of the
435 nodes for rectangular plate
when b/a = 2

(a) node 1 (regular,2915)

(b) node 2 (irregular)

(c) node 3 (irregular)

(d) node 4 (irregular)

Table 4 Limit load multipliers of rectangular plate in comparison with the available solutions for b/a = 2 (MP /qab)
Authors

Clamped

Simply supported

3 Clamped
1 Short side free

3 Clamped
1 Long side free

2 Clamped
2 Short sides free

2 Clamped
2 Long sides free

[23]

54.61

29.88

43.86

9.49

ACM (node 1)

57.26

29.88

46.31

19.74

40.28

9.62

Node 1

55.09

28.90

45.18

18.82

40.39

9.28

Node 2

55.05

28.75

44.75

18.96

39.86

9.30

Node 3

55.08

28.76

44.72

19.02

39.86

9.33

Node 4

55.58

28.73

45.66

19.15

40.51

9.38

C1 NEM

tion implementations, and the rectangular plate subjected to

different boundary conditions has different iterative convergence steps.
5.3 Rhombic plate
A further investigation is undertaken to consider a rhombic
plate with uniformly loaded pressure q and varying skewness
angle . The geometric parameters and two nodal distributions adopted here are respectively shown in Fig. 12. Mansfield [42] described an upper bound analytical method to
study the limit loads of this rhombic plate both with clamped

123

and simply supported edges based on a square yield criterion,

and Lowe [43] provided an alternative analytical method to
estimate the lower bound limit load of the clamped edge
rhombic plate. Recently, Capsoni and Vicente da Silva [15]
developed a Mindlin plate finite element to investigate the
upper collapse loads of this rhombic plate by adopting the
von Mises yielding criterion, and provided a set of helpful
numerical solutions. In the following numerical calculation,
the radius in Fig. 12 is chosen as R = 0.5 m and the thickness
of rhombic plate is chosen as h = 0.02m. Table 5 gives
the numerical results respectively evaluated by this paper
and Capsoni and Vicente da Silva, and clearly indicates that

Comput Mech

(a) 3 clamped and 1 short side free

(b) 3 clamped and 1 long side free

(c) 2 clamped and 2 short sides free

(d) 2 clamped and 2 long sides free

limit load multiplier MP / qab

Fig. 10 Plastic dissipation

 work rate
 distributions of 29 15 regular nodes for the rectangular plate at the limit states under four different boundary
conditions when b/a = 2 106 N m
Clamped
Simply supported
3 clamped and 1 short side free
3 clamped and 1 long side free
2 clamped and 2 short sides free
2 clamped and 2 long sides free

64
60
56
52
48
44
40
36
32
28
24
20
16
12
8
0

8 10 12 14 16 18 20 22 24 26 28 30 32

iterative step k
Fig. 11 The iterative convergence processes of the limit load multipliers for the rectangular plate (b/a = 2)

the present upper bound limit load multipliers are slightly

smaller than those solutions of Capsoni and Vicente da Silva.
Fig. 13a, b, respectively, depict the plastic dissipation work
rate distributions of uniformly loaded rhombic plates with
clamped and simply supported displacement velocity bound-

ary conditions when the skewness angle is = 60 , which

are in good agreement with the dissipation patterns given by
Capsoni and Vicente da Silva. Figure 14 shows the iterative
convergence processes of the limit load multipliers for rhombic plates with five different skewness angles and two kinds
of displacement velocity boundary conditions when nodal
distribution 2 is used, which indicates that the iterative curve
decreases monotonically as the increase of iterative step and
the obtained numerical limit load multiplier tends to a steady
minimum after several iterations. We can also see that for the
same skewness angles, the iteration steps of clamped plates
are usually fewer than those of simply supported plates and
the limit load multipliers of clamped plates are much greater
than those of simply supported ones.
In addition, the rhombic plate turns into a square plate
when the skewness angle is = 90 . Up to now, the limit
load capacity of square plate has been deeply investigated,
and the limit loads under different kinds of loads and different types of displacement velocity boundary conditions
have been reported in many published literatures [5,7,9,10,
15,23,44] over the past years. By operating the unit conversion of the limit load multiplier, i.e. 6.267MP /q R 2 =
25.068 25.07MP /q L 2 , Table 6 collects and summaries
some available solutions of square plates applied by uniform pressure and respectively imposed by simply supported

123

Comput Mech
Fig. 12 Geometric parameters
and nodal distributions of
rhombic plate

L = 2 R / sin

(b) Nodal distribution 1 (1717) (c) Nodal distribution 2 (2121)

(a) Rhombic plate



Table 5 Limit load multipliers of rhombic plates subjected to uniform load MP /q R 2

Simply supported

Clamped

Nodal distribution 1

Nodal distribution 2

Nodal distribution 1

Nodal distribution 2

[15]

Present

[15]

Present

[15]

[15]

30

5.228

4.808

5.140

4.808

9.852

8.979

9.575

45

5.666

5.447

5.609

5.447

10.847

9.992

10.596

9.939

60

5.997

5.925

5.966

5.916

11.641

10.885

11.394

10.781

75

6.212

6.196

6.197

6.186

12.143

11.266

11.893

11.065

90

6.289

6.274

6.278

6.267

12.314

11.438

12.062

11.296

and clamped boundary conditions. It clearly reveals that the

present numerical results are very close to those reference
solutions. The plastic dissipation work rate distributions of
square plates with clamped and simply supported displacement velocity boundary conditions at the limit states are
respectively exhibited in Fig. 13c, d when 21 21 node
arrangement is employed.
Finally, we also carry out the investigation on quadrature
when the square plate is respectively discretized by the nodal
distribution 17 17 and 21 21, and the detailed numerical
results are depicted in Table 7. The results listed in Table 7
indicate that: (1) the upper bound limit load multipliers for
square plate decrease as the numbers of nodes increase, and
this tendency is reasonable and can be well-understood; (2)
the upper bound limit load multipliers for the simply supported square plates become smaller and smaller as the numbers of quadrature points increase. However, the upper bound
limit load multipliers for the clamped support square plate
show a similar counter-intuitive tendency as listed in Table 2
for the clamped support circular plate. The upper bound limit
load multipliers for the clamped support square and circular
plates generally become greater and greater as the numbers
of quadrature points increase. Thus, the numerical results
obtained in the limit analysis by using different quadrature
schemes for the circular and square plates seem to have no
unified regularity. According to our thorough analysis based
on a large number of numerical computations and comparisons of elastic analysis and limit analysis results, we firstly
ensure the correction of the developed direct iterative algo-

123

Present

Present
8.901

rithm and relevant computer codes, and then exclude the

influences of structural discretization error, the selection of
computational data type, the selection of iterative parameters
(positive decimal , convergence error tolerances vol1 and
vol2), and the computational formula of displacement veloc = = T on the counter-intuitive tendencies of
ity U
F
numerical results listed in Tables 2 and 7. The initial iteration
of upper bound limit analysis carried out by using the direct
iterative algorithm is fully equivalent to the elastic analysis


s
of thin plate problem when we set v = 0.5 and E = 27
2 h.
However, the whole solution process of upper bound limit
analysis presented in this paper is different from the elastic analysis, and our numerical results indicate that the high
order quadrature schemes can usually but not always ensure
more accurate and satisfied upper bound limit load multipliers to be obtained in the C1 NEM. The similar numerical
results also appeared in the paper of Sukumar and Moran
[31], in which the relative error norms of the Airy stress function problem for a linear elastostatics circular plate obtained
by using the high order quadrature schemes are usually but
not always smaller than those obtained by the low order
ones. We think these numerical results may be the factor that
Sukumar and Moran [31] claimed the selection of an appropriate numerical quadrature scheme in the C1 NEM is still
an open issue. Though we mainly adopt 3 point quadrature
scheme to implement the numerical integration in this paper,
we also deem that the selection of an appropriate numerical
quadrature scheme in our upper bound limit analysis by using
the C1 NEM is still an open issue.

Comput Mech

(a) Simply supported ( = 60 )

(c) Simply supported ( = 90 )

(b) Clamped ( = 60 )

(d) Clamped ( = 90 )



Fig. 13 The plastic dissipation work rate distributions of rhombic plates at the limit states using nodal distribution 2 106 N m

5.4 Equilateral polygon plate

The fourth set of examples considered underneath addresses
the investigation of two equilateral polygon plates, which
are known as equilateral triangle plate and equilateral hexagon plate and are depicted in Fig. 15. The radius marked
in Fig. 15 is assigned as R = 0.5m and the thickness of
plates is selected as h = 0.02m, and the discretizations of
these two equilateral polygon plates are illustrated in Fig. 16.
The uniform pressure q is applied and simply supported
and clamped boundary conditions are respectively considered here. The corresponding evaluated numerical results are
listed in Table 8.

To the authors knowledge, the limit loads of these two

plates with simply supported boundary conditions seem not
to have been reported in the available literatures. In the case
of clamped boundary conditions, Fox [45] studied the regular
polygonal shape plates under uniform pressure and obtained
a set of exact upper bound solutions by using the square yield
criterion. The solutions provided by Fox for clamped equilateral triangle plate and clamped equilateral hexagon plate
are also shown in the Table 8. It can be clearly seen that
the present results are slightly greater than those of Fox and
the errors between them are very small. The plastic dissipation work rate distributions of these two equilateral polygon
plates are displayed in Fig. 17, and obviously, all these cloud

123

Comput Mech
0

=30 simply-supported
0
=45 simply-supported

13
12

=60 simply-supported
0
=75 simply-supported
0

limit load multiplier MP / qR

11

=90 simply-supported
0
=30 clamped

10
9

=45 clamped

=60 clamped
0
=75 clamped

=90 clamped

0
0

(a) Equilateral triangle plate

(b) Equilateral hexagon plate

Fig. 15 Geometry details for equilateral polygon plates

5
0

10

12

14

16

18

20

22

24

iterative step k
Fig. 14 The iterative convergence processes of the limit load multipliers for rhombic plates using nodal distribution 2

images are symmetrical and reasonable. Finally, the iterative

convergence processes of the limit load multipliers for these
two equilateral polygon plates are indicated in Fig. 18.

(a) 397 nodes

(b) 553 nodes

Fig. 16 Nodal arrangements for equilateral triangle and hexagon plates

6 Conclusions
1.
This article presents a novel, simple and effective numerical
solution procedure for the estimation of the upper bound limit
load multipliers of thin plates by adopting the combination
of the C1 NEM and a direct iterative algorithm. The primary
remarks are concluded underneath:
Table 6 The comparison of the
limit load
of square
 multipliers

plates MP /q L 2

In the C1 NEM, the C1 natural neighbour interpolant,

which is built by embedding the Sibsons natural neighbour coordinates in the BernsteinBzier surface representation of a cubic simplex, is utilized to construct
the C1 shape function in a displacement-based Galerkin
scheme. Though the calculations of C1 shape function

Authors

Method

Simply supported

Clamped

25.00

44.13

Le et al. [23]

Upper bound

25.01

45.07

Capsoni and Corradi [9]

Upper bound

25.02

45.29

Present

Upper bound

25.07

45.18

Capsoni and Vicente da Silva [15]

Upper bound

25.11

48.25

25.38

43.93

Andersen et al. [44]

Turco and Caracciolo [10]

Table 7 The result comparisons

for the limit analysis of square
plate with uniform pressure and
different boundary conditions
by using different nodes
 and 
quadrature schemes MP /q R 2

123

Hodge and Belytschko [7]

Upper bound

26.54

49.25

Lubliner [5]

Upper bound

27.71

52.01

Quadrature

Simply supported

Clamped support

Nodal distribution 1

Nodal distribution 2

Nodal distribution 1

Nodal distribution 2

6.274

6.267

11.438

11.296

6.263

6.261

11.547

11.432

13

6.255

6.256

11.657

11.559

25

6.264

6.261

11.611

11.507

Comput Mech
Table 8 Limit load multipliers
of
 polygon plates
 equilateral
MP /q R 2

Equilateral polygon plate

Simply supported

Clamped

Present

Present

Triangle plate

5.545

10.039

9.610

4.464

Hexagon plate

6.370

11.944

11.442

4.387

Fox [45]

(a) Simply supported equilateral triangle plate

(b) Clamped equilateral triangle plate

(c) Simply supported equilateral hexagon plate

(d) Clamped equilateral hexagon plate

Error (%)

Fig. 17 The
 plasticdissipation work rate distributions of uniformly loaded equilateral polygon plates with different boundary conditions at the
limit states 106 Nm

involve some basic linear algebra matrix-vector operations, the computational costs are not observably more
than those spent in the evaluation of C0 Sibson shape

function. The acquired C1 shape functions have interpolation to the nodal function and nodal gradient values,
and the essential boundary conditions in the C1 NEM

123

Comput Mech

Simply supported equilateral triangle plate

Clamped equilateral triangle plate
Simply supported equilateral hexagon plate
Clamped equilateral hexagon plate

14

limit load multiplier MP / qR

13

the plotted plastic dissipation work rate distributions

are symmetrical and reasonable. The numerical results
indicate that using the C1 NEM and the direct iterative
algorithm to carry out the upper bound limit analysis of
plates is effective and appealing, and has the advantages
of good accuracy and fast convergence.

12
11
10

Finally, the developed C1 NEM-based numerical solution

procedure can also be extended to implement the kinematic
shakedown analysis of thin plate problems. Further efforts
will be made along this line.

9
8
7
6
5
0

10

12

14

16

18

20

22

24

iterative step k
Fig. 18 The iterative convergence processes of the limit load multipliers for equilateral polygon plates

2.

3.

can be imposed as easily as those in the FEM. Thus, it

is an attractive and interesting choice to investigate the
limit analysis of plate problems by using the C1 NEM.
The developed iterative algorithm is implemented only
by a series of direct iteration computations. The plastic incompressibility condition is realized in the goal
function by introducing a constant matrix. By distinguishing the rigid zones from the plastic zones and modifying accordingly the goal function at every iteration
step, the difficulties caused by the nondifferentiable and
nonsmooth goal function are overcome. The whole solution procedure in principle needs to solve a series of
quadratic minimization problems with an equality constraint and every quadratic minimization problem is
equivalent to solving a relevant linear elastic problem.
The algorithm has the advantages of simple formulation
and easy programming.
In the numerical examples, circular plate, rectangular
plate, rhombic plate and equilateral polygon plate are
selected to verify the performances of the proposed
numerical solution procedure. The MLS technique is
employed to approximate the nodal strain rates, and
the plastic dissipation work rate distributions of plates
are calculated accordingly and displayed visually. The
provided iterative convergence curves illustrate that the
limit load multipliers monotonically decrease to steady
minimum values after several iterations, and for the
same geometries and loadings, the iteration steps of
clamped plates are usually fewer than those of simply
supported ones and the limit load multipliers of clamped
plates are much larger than those of simply supported
plates. The obtained upper bound limit load multipliers agree well with the previously reported results and

123

Acknowledgements This study was supported by the National Natural Science Foundation of China (10872104) and by the Program for
New Century Excellent Talents in University (NCET-04-0075).

References
1. Hodge PG (1963) Limit analysis of rotationally symmetric plates
and shells. Prentice-Hall, New Jersey
2. Save MA, Massonnet CE (1972) Plastic analysis and design of
plates, shells and disks. North-Holland, Amsterdam
3. Zyczkowski M (1981) Combined loadings in the theory of plasticity. Polish Scientific, PWN and Nijhoff
4. Xu BY, Liu XS (1985) Plastic limit analysis of structures. China
architecture & building Press, Beijing
5. Lubliner J (1990) Plasticity theory. Macmillan, New York
6. Yu MH, Ma GW, Li JC (2009) Structural plasticity limit, shakedown and dynamic plastic analyses of structures. Springer,
New York
7. Hodge PG, Belytschko T (1968) Numerical methods for the limit
analysis of plates. J Appl Mech 35(4):795802
8. Nguyen HD (1976) Direct limit analysis via rigid-plastic finite elements. Comput Methods Appl Mech Eng 8(1):81116
9. Capsoni A, Corradi L (1999) Limit analysis of plates-a finite element formulation. Int J Numer Methods Biomed Eng 8(4):325341
10. Turco E, Caracciolo P (2000) Elasto-plastic analysis of Kirchhoff
plates by high simplicity finite elements. Comput Methods Appl
Mech Eng 190(57):691706
11. Corradi L, Vena P (2003) Limit analysis of orthotropic plates. Int
J Plasticity 19(10):15431566
12. Corradi L, Panzeri N (2003) Post-collapse analysis of plates and
shells based on a rigid-plastic version of the TRIC element. Comput
Methods Appl Mech Eng 192(3334):37473775
13. Tran TN, Kreissig R, Staat M (2009) Probabilistic limit and shakedown analysis of thin plates and shells. Struct Saf 31(1):118
14. Le CV, Nguyen-Xuan H, Nguyen-Dang H (2010) Upper and lower
bound limit analysis of plates using FEM and second-order cone
programming. Comput Struct 88(12):6573
15. Capsoni A, Vicenteda Silva M (2011) A finite element formulation
of Mindlin plates for limit analysis. Int J Numer Methods Biomed
Eng 27(1):143156
16. Belytschko T, Lu YY, Gu L (1996) Element free Galerkin methods. Int J Numer Methods Eng 37(2):229256
17. Liu WK, Jun S, Zhang YF (1995) Reproducing kernel particle
methods. Int J Numer Methods FL 20(89):10811106
18. Duarte CA, Oden JT (1996) An h-p adaptive method using clouds.
Comput Methods Appl Mech Eng 139(14):237262
19. Atluri SN, Zhu T (1998) A new meshless local Petrov-Galerkin
(MLPG) approach in Comput Mech. Comput Mech 22(2):117
127

Comput Mech
20. Chen JS, Wu CT, Yoon S, You Y (2001) A stabilized conforming nodal integration for Galerkin mesh-free methods. Int J Numer
Methods Eng 50(2):435466
21. Liu GR, Gu YT (2001) A point interpolation method for twodimensional solids. Int J Numer Methods Eng 50(4):937951
22. Cueto E, Sukumar N, Calvo B, Martinez MA, Cegonino J, Doblar M (2003) Overview and recent advances in natural neighbour
Galerkin methods. Arch Comput Methods Eng 10(4):307384
23. Le CV, Gilbert M, Askes H (2009) Limit analysis of plates using
the EFG method and second-order cone programming. Int J Numer
Methods Eng 78(13):15321552
24. Le CV, Gilbert M, Askes H (2010) Limit analysis of plates and
slabs using a meshless equilibrium formulation. Int J Numer Methods Eng 83(13):17391758
25. Le CV, Askes H, Gilbert M (2010) Adaptive element-free Galerkin method applied to the limit analysis of plates. Comput Methods
Appl Mech Eng 199(3740):24872496
26. Onate E (1994) A review of some finite element families for thick
and thin plate and shell analysis. In: Hughes et al (eds) Recent
Developments in finite element analysis, CIMNE, Barcelona,
pp 98111
27. Zienkiewicz OC, Taylor RL (2005) The finite element method.
Butterworth-Heinemann, Oxford
28. Barros FB, de Barcellos CS, Duarte CA (2007) p-Adaptive C-k
generalized finite element method for arbitrary polygonal clouds.
Comput Mech 41(1):175187
29. Xiang JW, Chen XF, He ZJ, Zhang YH (2008) A new waveletbased thin plate element using B-spline wavelet on the interval.
Comput Mech 41(2):243255
30. de Barcellos CS, Mendonca PDR, Duarte CA (2009) A C (k) continuous generalized finite element formulation applied to laminated
Kirchhoff plate model. Comput Mech 44(3):377393
31. Sukumar N, Moran B (1999) C-1 natural neighbor interpolant for
partial differential equations. Numer Methods Part D E 15(4):417
447
32. Fischer P, Mergheim J, Steinmann P (2010) On the C-1 continuous discretization of non-linear gradient elasticity: a comparison
of NEM and FEM based on BernsteinBezier patches. Int J Numer
Methods Eng 82(10):12821307

33. Rajagopal A, Fischer P, Kuhl E, Steinmann P (2010) Natural element analysis of the Cahn-Hilliard phase-field model. Comput
Mech 46(3):471493
34. Chen G, Liu YH (2006) Numerical theories and engineering methods for structural limit and shakedown analyses. Science press,
Beijing
35. Sukumar N, Moran B, Belytschko T (1998) The natural element
method in solid mechanics. Int J Numer Methods Eng 43(5):839
887
36. Sukumar N, Moran B, Semenov AY, Belikov VV (2001) Natural
neighbour Galerkin methods. Int J Numer Methods Eng 50(1):1
27
37. Farin G (1990) Surfaces over Dirichlet tessellations. Comput
Aided Geom D 7(14):281292
38. Zhang PX, Lu MW, Hwang KC (1991) A mathematical programming algorithm for limit analysis. Acta Mech Solida Sin
23(4):433442
39. Tabbara M, Blacker T, Belytschko T (1994) Finite-element derivative recovery by moving least-square interpolants. Comput Methods Appl Mech Eng 117(12):211223
40. Hopkins HG, Wang AJ (1954) Load-carrying capacities for circular plates of perfectly-plastic material with arbitrary yield condition. J Mech Phys Solids 3(2):117129
41. Melosh RJ (1963) Basis for derivation of matrics for the direct
stiffness method. J AIAA 1(7):16311637
42. Mansfield EH (2000) Collapse pressures for rhombic plates. Int J
Mech Sci 42(3):635643
43. Lowe PG (2002) Collapse pressures for rhombic plates: discussion. Int J Mech Sci 44(9):20252026
44. Andersen KD, Christiansen E, Overton ML (1998) Computing
limit loads by minimizing a sum of norms. SIAM J Sci Comput
19(3):10461062
45. Fox EN (1974) Limit analysis for plates: exact solution for a
clamped square plate of isotropic homogeneous material obeying
square yield criterion and loaded by uniform pressure. Philos T R
Soc A 277(1265):121155

123