COMPUTER METHODS IN APPLIED MECHANICS AND ENGINEERING 28 (1981) 265-274

NORTH-HOLLAND PUBLISHING COMPANY

MINIMIZATION APPROACH TO LIMIT SOLUTIONS OF PLATES

Wei H. YANG
The University of Michigan, Ann Arbor, MI 48109, USA

Received 24 October 1980

Revised manuscript received 8 January 1981

Limit analysis which predicts the ultimate load carrying capacity of a structure or machinery
provides very useful information especially for designs that must survive accidents and abnormal
conditions. In this paper, maximization of lower bound formulation and minimization of upper bound
formulation are presented as primal and dual problems respectively. The dual problem is solved by a
minimization procedure for circular, triangular and square plates with three types of boundary
conditions and two types of loadings. A detailed parametric analysis for a cracked plate is also given to
demonstrate the effectiveness of this approach.

1. Introduction

Plastic design of structures [l, 121 has generally been considered superior to the con-
ventional elastic design. It gives a more uniform strength distribution and therefore reduces
the weight of the structures. It is less sensitive to imperfections and misalignments of structural
components during construction. It provides more realistic prediction of structural behavior
under severe and unexpected conditions such as earthquakes and accidents.
Analysis of plastic design is more difficult. Even with the powerful existing elastoplastic
computer programs [2, 31, plastic analysis of large and complex structures remains a formid-
able task. Costs of analytic man power and computer time often offset the gain in material
savings for ordinary structures. This is the reason why most of the construction codes are still
based on elastic analysis.
With light-weight design now made necessary for vehicular structures, the plastic range of
the materials should be considered in designs. With higher demand on reliability of earth-
quake proof structures and accident safe nuclear installations, more thorough elastic-plastic
analysis and prediction of ultimate strength must be understood.
Limit analysis has a simple mathematical structure yet describes a realistic nonlinear
behavior. When an exact solution is obtainable by the analysis, it gives the ultimate load limit
of a structure. Approximate solutions provide bounds on the limit loads.
For convenience of presenting the ideas in this paper, the classical definitions of statical and
kinematical admissibilities [4] are slightly modified and augmented with the additional con-
stitutive admissibility. A solution of any mechanics problem satisfies simultaneously the
equilibrium equation, the kinematical compatibility condition and a mathematical description
of material behavior (constitutive relation). Consider a function space of mechanics solutions.
The solutions satisfying the equilibrium equation and its boundary conditions, if present,

0045-7825/81/0000-0000/$02.75 @ 1981 North-Holland

266 W.H. Yang, Minimization approach to limit solutions of plates

belong to a point set S in the space, named statically admissible. The solutions derived from a
single-valued displacement field and its boundary conditions form another set K named
kinematically admissible. Solutions which do not violate the constitutive equation (or relation)
establish the constitutively admissible set C. The exact solution, or solutions if not unique, lies
in the intersection of these three sets.
Limit analysis studies the solutions in the intersections of two of the three sets. The lower
bound solutions lie in S f~ C. The upper bound solutions lie in either K rl C or K f~S. These
statements agree with the well known lower and upper bound theorems [5] although they may
not be in the same wording. A limit solution is an extreme point of one of these convex sets
defined by either the lower or the upper bound solutions.
The plates under consideration are assumed to be stiff and have well defined limit values of
bending moment. The modern structural plates such as the sandwich plates [6] and composite
plates [7] are designed to have these properties. High stiffness implies infinitesimal elastic
deformation and enables the equilibrium equation to be written in the undeformed coor-
dinates and therefore to be linear. A yield criterion without hardening models a class of plate
behavior in which infinite curvature can occur under a constant yield moment. An initially
smooth flat plate may develop a displacement field containing ridge lines which allow
discontinuous gradients. These ridge lines deform like piano hinges.
Limit analysis predicts the initial collapse condition under which the plate deforms con-
tinuously with non increasing load. The initial collapse may soon stabilize when large
deformation of the plate causes part of the load to be carried by the membrane stresses. The
large deformation is outside the scope of limit analysis.
The formulation starts with the lower bound theorem from the set point of view. Although
the resulting maximization problem is difficult to solve, it gives a clear concept of uniqueness
of optimality and possible nonuniqueness of optimal solutions. An upper bound to the
optimality is obtained by the virtual work principle and by forcing the constitutive inequality
to the bounds everywhere. This leads to a minimization problem.
Several simple solutions for circular, triangular and square plates with various boundary
conditions and loads and a detailed analysis for a cracked plate with various crack angles and
lengths are obtained by minimization.

2. Lower bound formulation

The equilibrium equation for plates with a frequently used sign convention is given in [S]

d2MXX
_ a2Mx,+ a2W, _
ax2
2
ayZ -
axay -q(‘7 Y) inI3 7

where M,,, MxY, MY, are moment components and q(x, y) is the load distribution function in
the domain D.
Let (a/ax, alay) be a two dimensional vector operator and let the moment components be
arranged in a symmetric 2 x 2 matrix denoted by M. The above equation can be written in the
compact form,
 W.H. Yang, Minimization approach to limit solutions ofplates 267

M
V*@+M)=q inD, M= -En _~n], (2)
I XY YY

where - is the inner product of vectors or a vector and a matrix.

Eq. (2) defines a set in the space Rzx2. The set is convex since for any two points M, and MZ
in the set, aMZ + (1 - (Y)M; (0 I CI!5 1) is also in the set, a consequence of the linear operator.
Consider a yield criterion that models the limiting behavior of the plastic plate by the
inequality,

f(M)~O, (3)

where f is a map from Wzx2to R. We need not specify f now except to require that f is convex.
Therefore, the relation (3) also defines a convex set in R2x2. The points M lying in the
intersection of the sets defined by (2) and (3) are called lower bound solutions. Lower bound
solutions form a convex set since the intersection of convex sets is convex.
If q is given in the form of propo~ion~ loading such that q = q&(x, y) where q. > 0 and
4(x, y) is a given integrable function, the constrained maximization problem,

max q.
s.t. V~(V~M)=qo#, fWW (4
has a solution (M, qo) E W2x2 x W. The boundary conditions on moment and shear if present
should also be included in the constraint set. Some problems may not have these static
boundary conditions. q. can be uniquely determined, but A4 may not be unique. These are the
properties of the convex programming problem given in (4). A proportional loading is also
called a one parameter dist~bution.
If q has a finite N-parameter distribution such that q = Xz’=, qi#g(X, y), qi > 0, we have
another maximization problem,

s.t. V(VM)=q, fWk% (5)

which has a unique maximum for the integral I SD q dA.
The problems (4) and (5) are in principle solvable by a finite dimensional space ap-
proximation but the numerical tasks are formidable. Nevertheless, we know the solution exists
and the optimality is unique for each problem. These assertions provide the basis for the
method of solution to the upper bound formulation derived in the following section from (4)
by the virtual work principle.

3. Upper bound fo~u~tion

We shall now satisfy the equilibrium equation (2) in a weak sense (virtual work) such that
268 W.H. Yang, Minimization approach to limit solutions of plates

wV(VM)dA= wqdA forall wE W,

f fD

where W is the space of real, kinematically admissible functions which are continuous in D
and satisfy certain boundary conditions.
The left-hand side of (6) can be manipulated into another form by the use of Green’s
theorem

f I D
wV(VM)dA= V(wVM)dA- (VM)VwdA

= wit (VM)dS- V(MVw)dA

f ar,

+
II D
M:VVwdA

= [wn(VM)-n.MVw]dS+ M:VVwdA, (7)

f aD

where ft is the unit normal vector along the boundary 3L) and : denotes the inner product of
two matrices such that

The line integral in (7) vanishes for all types of natural boundary conditions [lo]. These
include the simply supported, the clamped and the free edge conditions as well as the
combinations of these boundary conditions.
Consider the case where q has a one parameter distribution such that q = qoq(x, y)_ We can
write (6) in the form

qo(W= j I, M :VVwdA
/If D
Q’+‘u

provided that s SD QW dA# 0. There is no loss of generality if we Set ,f _fD cpw dA >O. The
problem (4) becomes

max q&f),
s.t. f(M) s 0 , (10)
The solution M which maximizes q&f) gives the limit load qoL. This maximization problem is
by no means easy to solve and there exists no satisfactory algorithm for its solution. We shall
construct the dual of (10) which leads to a minimization problem. Approximate solutions of
the minimization problem give upper bounds to the limit load qoL.
Note that M in (10) is constitutively admissible therefore it lies inside or on the yield
 W.H. Yang, Minimizaticm approach to limit solutions of plates 269

surface. Consider a state M* which is on the yield surface such that f(M*) = 0 and is
associated with a w leading to the yield state. The inequality

(M*-M):vvw~o (11)
states the well-known condition of Drucker [15] on stable materials. Using (ll), we have an
upper bound to q*(M) such that

M*:VVwdA qowdA = q”(w) f (12)

where the form of M*(w) in q*(w) depends on the specific f(M). The least upper bound q2
can be obtained by minimizing q*(w) over all kinematically admissible w(x, y) such that

M*:VVwdA cpwdA. (13)

mathematically, (13) is called the dual problem of (10) which is called the primal problem. For
some f(M), M* can be made explicit and the dual problem involves only the kinematical
variable w(x, y) which is easier to visualize than M(x, y). For simple domains, w(x, y) can be
parameterized and the minimization can be carried out in a small parameter space. For
complex domains, the correct w may be iteratively approached by a sequence of trial functions
in a finite dimensional space. Even if w is not optimal, qt so obtained by (13) always bounds
the exact limit load from above such that

q2rq0,. (14)

The examples in the following sections belong to the class of simple domains. A specific
yield function [ll]

f(M) = llMlli4
- A& = Ait,,, - M; (15)

is used where A2,,,(M) is the largest eigenvalue of M* and MO is the constant yield moment of
the plate. The function is convex and A,,, occurs in the major principal direction of M. It is
physically reasonable to assume that along the lines of maximum moment ridges will form.
For a given 4, we may choose a w(x, y) in the following way. Let the domain D be divided
into finite number of subdomains Di such that

L)= 64. (I@

r=l
Within each Di, there exists a local curvilinear orthogonal coordinate system (5,~) such that
M& = 0. If w is linear along one coordinate say 6, this direction corresponds to the minor
principal direction along which the moment is less than MO. Along the major principal
direction 7, jM&T,I= M,, and a2w/@* # 0. If w is linear in both directions, the subdomain
deforms like a rigid body.
270 W.H. Yang, Minimization approach to limit solutions of plates

The function w(x, y) must be continuous and satisfies the kinematical boundary conditions
given in (7). The intersections of all Di (i = 1,2, . . . , N) are the ridge lines. Let (n, s) denote
the coordinates along each ridge line with s being the line. Then awl&r has a discontinuity
across the ridge line and IMznl = MO.
Since M* : VVw is invariant to coordinate rotation, we may write

M* : VVw = -M,<
a*w_
2
M* a*w
(17)
at ?v aq* .

For each choice of w, the quotient in equation (13) is an upper bound. We choose M$ and
Mz,, to be both on the bounds (&MO) to give

where fl is the union of all subdomains, where VVwf 0, S is the union of all ridge lines and
[awl&t] is the jump of dw/an. J is the Jacobian of the transformation from (x, y) to (6,~).
We have established an upper bound to qoL for a kinematically admissible function w such
that

The least upper bound 4: may be obtained by miAmizing the function q*(w) such that

Eq. (20) is a special case of (13) and is the one used in the examples.

4. Simple examples

In order to present a large number of solutions in the limited pages allowed, the detailed
derivation for each problem is omitted. They are interesting exercises of interaction between
intuition and mathematical optimization process. The choices of functions w(x, y) are guided
by intuition to establish a few variable parameters. The computation then proceeds to obtain
the values of the parameters that minimize q*. The choices of w(x, y) in fl are linear and
conical which intersect at linear and circular ridge lines. The parameters in w(x, y) which
minimize q* characterizes the configurations of collapse mode.
Another interpretation of the above algorithm is the finite element concept. The special
finite element method employed here does not fix its mesh system. But the basis function in
each element is predetermined. The mesh system is parameterized so that q* is minimized in
the parameter space.
 W.H. Yang, Minimization approach- to limit solutions of plates 271

Three shapes of plates, circle, equilateral triangle and square, are included. Each plate is
imposed with three cases of boundary conditions: resting, simply supported and clamped
edges. The resting edge condition restricts downward displacement but allows the plate to
leave the support at places when compressive reaction from the support is absent. fn each
case, two distribution functions are chosen: &(x, y) = 1 and #& y) = 6(x, y) where 8(x, y) is
the delta function with origin at the center of each plate. The optimal qE corresponds to $1 is
the uniform collapse load intensity. The optimal qt corresponds to & is the magnitude of the
concentrated collapse load denoted by P.
The integrals in (20) can be easily calculated in the conical subdomains and along the ridge
lines since c?*w/c~~~and [awl&z] have simple expressions and can be integrated exactly. There
is no contribution by the integration in the remaining domain since VVw = 0.
The results for all 18 combinations of plate configurations, boundary conditions and load
distributions are given in table 1 and table 2. The meanings of the angles & and f&,and lengths

Table 1
Uniformly distributed collapse load intensity

Boundary
condition Shape q~i2/6M~

% 0.92495064 35.78 18.44 0.72069224

Resting Cr l.oooooooo - - -
Tr 2.61836929 50.45 19.09 0.69913792

sq 1.~ 45.00 0.00 1.0~

Simply Cr 1.~ - - -
supported Tr
2.83179668 53.46 13.08 0.77910598

Sq 1.81098918 31.05 27.90 0.60204897

Clamped Cr 2.oOOOOOOO - -
Tr 4.95895700 42.75 35.50 0.53369682

Table 2
Concentrated collapse load at plate center

Boundary
condition Shape P/MO

sq 6.62741700 22.50 45.00 0.41421357

Resting Cr 6.28318531 - - -
Tr 6.92820323 30.00 60.00 0.33333333

Simply 2 8.~ 45.00 0.00 1.~

6.28318531 - -
supported
Tr 9.14159266 45.00 30.00 0.57735027

Sq 12.56637000 0.00 90.00 o.OOOOOOOo

Clamped Cr 12.56637000 - -
Tr 12.56637000 0.00 120.00 0.00~0000
272 W.H. Yang, Minimization approach to limit solutions of plates

RESTING SIMPLY SUPPORTED CLA?fPED

Fig. 1. Plate configurations and ridge line patterns.

1 and d are defined in fig. 1. The conical surfaces are shown as fans. The ridge lines are heavier
than other lines. Only one independent parameter & appears in minimization. Some solutions
contained in [ll] are reproduced here for verification.

5. A cracked plate

To present a more complex example, a simply supported square (21 X 21) plate with a
centered crack is considered in this section. The plate is under a uniform load and the crack
has an arbitrary angle a and length c as shown in fig. 2. The effect of the crack on the load
carrying capacity of the plate is studied by the method of limit analysis.

Fig. 2. Configuration and parameters of a cracked plate.

 W.H. Yang, Minimization approach to limit solutions of plates 273

The square plate suggests the ranges of (Y and c such that

Ola 145”. 0 I c I llcos a . (21)

The ridge lines shown in fig. 2 may or may not intersect the crack tips. The correct ridge lines
will be determined by two angle parameters I% and & which are optimized to obtain the limit
load. The function w(x, y) is assumed to be linear in each of the four subdomains separated by
the ridge lines as shown. It is an easy exercise to calculate the integrals in (19) and construct
the function

q*w = q*cA, 02, a, c/l). (22)

With two parameters (Y and c/Z assigned, q*(w) can be minimized with respect to & and e2.
The dimensionless limit load of the cracked plate, qi12/6Mo, is shown in fig. 3 for the entire
ranges of (Yand c/l.
For small crack lengths, the orientation of the crack does not play a significant role in the
limit load which decreases with increasing crack length. But the effect of the crack angle is
quite pronounced at large crack lengths. It is also interesting to note that for any crack angle
other than 45”, there is a range where crack length has no effect on the limit load. The limit
load initially decreases with increasing crack length to a point then remains constant. The
points separate the decreasing range from the constant range of the limit load are shown along
the dashed curve.
The minimization procedure used in the &-e2 space is the gradient descent method [9]
which converges in 4 to 5 iterations for each set of (a, c/l). This fast convergence is achieved
by the sequential computation such that a converged solution is used as the initial iterate for
an adjacent pair (a, c/Z).

0.2 0.4 0.6 0.8 1.0

C/1

Fig. 3. Limit load of cracked plate with angle and length variations.
274 W.H. Yang, Minimization approach to limit solutions of plates

6. Final remarks

Establishment of upper and lower bound solutions to limit analysis has been known for
many years as indicated by a number of references listed. The list could have been much
longer if a survey is intended. But a systematic search for the limit solution in a convex set
using the developments of nonlinear programming methods is relatively recent. The com-
putational methods certainly provide added power for solving more complex limit analysis
problems. The minimization approach presented here computes the least upper bound from a
given family of test functions (collapse modes). If the family contains all possible collapse
modes, the minimum solution obtained is the limit solution. Otherwise, the minimum is an
upper bound to the limit solution. Some solutions of the simple problems presented agree with
the available known results. The other new solutions give physically correct quality and trend.
The solutions of the cracked plate problem point out a new phenomenon. For some
configurations, the crack tip is not the most critical area of high stresses. This is contrary to the
common belief developed from elastic analysis. The plastic analysis should give more realistic
estimates on load capacity of ductile plates.
Limit analysis which studies the ultimate behavior rather than the elastoplastic response is a
valuable aid to structural and machine designs. Its power can be better utilized when more
convenient methods are made available. This paper and some similar developments [13, 141
offer a few effective methods. Methodology of limit analysis enjoyed much less development
than that of elasticity.

References

[l] S.R. Daniels, Inelastic Steel Structures (University of Tennessee Press, Memphis, 1966).
[2] H.A. Balmer, J.St. Doltsinis and M. Konig, Elastoplastic and creep analysis with the ASKA program system,
Comput. Meths. Appl. Mech. Engrg. 3 (1974) 87-104.
H.A. Balmer and J.St. Doltsinis, Extensions to the elastoplastic analysis with the ASKA program system,
Comput. Meths. Appl. Mech. Engrg. 13 (1978) 363-401.
[3] T.G. Butler and D. Michel, NASTRAN; a summary of the functions and capacities of the NASA structural
analysis computer system, Scientific and Technical Information Office, NASA SP-260 (1971).
[4] W. Prager and P.G. Hodge, Jr., Tlieory of Perfectly Plastic Solids (Wiley, New York, 1951).
[5] P.G. Hodge, Jr., Plastic Analysis of Structures (McGraw-Hill, New York, 1959).
[6] F.J. Plantema, Sandwich Construction, The Bending and Buckling of Sandwich Beams, Plates and Shells
(Wiley, New York 1966).
[7] R.M. Jones, Mechanics of Composite Materials (McGraw-Hill, New York, 1975).
[8] S.P. Timoshenko and S. Woinowsky-Krieger, Theory of Plates and Shells (McGraw-Hill, New York 1959).
[9] D.G. Luenberger, Introduction to Linear and Nonlinear Programming (Addison-Wesley, Reading, MA, 1973).
[lo] K. Washizu, Variational Methods in Elasticity and Plasticity (Pergamon Press, New York, 1975).
[ll] E.H. Mansfield, Studies in collapse analysis of rigid-plastic plates with a square yield diagram, Proc. Roy. Sot.
London Ser. A 241 (1957) 311-338.
[12] M.A. Save and C.E. Massonnet, Plastic Analysis and Design of Plates, Shells and Discs (North-Holland,
Amsterdam, 1972).
[13] W.H. Yang, On a class of optimization problems for framed structures, Comput. Meths. Appl. Mech. Engrg.
15 (1978) 85-98.
[14] W.H. Yang, A practical method for limit torsion problems, Comput. Meths. Appl. Mech. Engrg. 19 (1979)
151-158.
[15] D.C. Drucker, A definition of stable inelastic material, J. Appl. Mech. Trans. ASME 81 (E 101) (1959).