INTERNATIONAL JOURNAL FOR NUMERICAL METHODS IN ENGINEERING, VOL.

8, 821-845 (1974)

VISCO-PLASTICITY-PLASTICITY AND CREEP IN

ELASTIC SOLIDS-A UNIFIED NUMERICAL
SOLUTION APPROACH

0. c. ZIENKIEWICZ*AND I. c. C O R M E A U ~
University o j Wales, Swansea, Wales

SUMMARY
The visco-plastic model of material behaviour is of much practical interest in its own right and initial strain
techniques for its solution have been developed and proved efficient. More important however is the fact that
the visco-plastic model can be used to generate plasticity solutions in a simple manner when stationary con-
ditions are reached and, at the other extreme, can reproduce standard creep phenomena. Used in this sense it
allows the treatment of non-associated plasticity and strain softening situations which present difficulties in
conventional plasticity approaches. A standard programme thus allows the treatment of a wide range of
materially non-linear problems.
The paper discusses various applications of the new general formulation and introduces certain numerical
information on solution stability.

INTRODUCTION
The visco-plastic model
In solid mechanics it has become customary to separate the two important groups of phenomena
described respectively by ‘creep’ and ‘plasticity’. The first includes all time effects and results in
the creep strains, developed at a finite rate. The second group apparently develops permanent
(plastic) strains instantaneously and time does not enter directly into consideration. Physically
it is apparent that
(a) Phenomena of creep and plasticity cannot be treated separately as only the combined
effect is measurable.
(b) That the concept of instantaneous development of permanent, irreversible strain is at
variance with experiment and that plasticity is, at best, only a convenient mathematical fiction.
To deal with observed facts a fairly general model of elasto-visco-plasticity has been intro-
duced, and is capable of reproducing some observed effects of creep and plasticity. In such a model
we have, in addition to reversible elastic strains, E ~ an
, additional set of strains, E”P, which are
characterized by a strain rate which is zero when stresses are below a certain threshold (or yield)
value and exhibit a finite strain rate only when this threshold i s exceeded.

* Professor of Civil Engineering.

7 Aspirant F.N.R.S.,Universite Libre de Bruxelles (formerly at Swansea).
Received 5 March 1974
0 1974 by John Wiley & Sons, Ltd.

821
822 0. C. ZIENKIEWICZ AND I. C. CORMEAU

In a uniaxial context such a model is shown in Figure l(a) where the visco-plastic element
remains inactive when la1 < ay.

4 nkP
plastic element
inactive i f IcrI<Qy

E =E)E"+E' pure
$4u creep
f%(plca-cl,))

Figure 1. Usiaxial visco-plastic rheological models

By placing more than one visco-plastic model in series a very general uniaxial behaviour can
obviously be introduced.
The concept of visco-plasticity is by no means new ;the so-called Bingham material is one of its
earliest versions introduced in 1922.' Much development has taken place since and PerzynaZ
presents an excellent survey of the present stage of evolution of this general model. In the present
paper we use his formulation with some modification ; quite generally we thus divide the total
strain (in a multiaxial context) into three parts
E = E~+E"P+E' (1)
where
L~ stands for elastic strain
E~~ stands for visco-plastic strain
and
co stands for 'initial', autogenous, strain-such as that due to temperature changes
In above all strains are represented by vectors of six quantities in a three-dimensional case and
have the following components

ET = r&x,&y.&~~YyzrYrx.Yxyl (2)
Defining the corresponding stress vector as <r we can immediately relate, for a linear elastic
material,
ze = D-'<r (3)
where D- is a symmetric elasticity matrix (compliance matrix).
 VISCO-PLASTICITY 823

The visco-plastic strains have now to be defined. If a yield condition is given as

F(G,K ) G F(G,K ) - - K ) =0 (4)
with F < 0 denoting the purely elastic region, we note that the rate of strain will be a function
of positive values of F. In above y is a yield stress and K is a history dependent hardening (or
softening) parameter. To define the relation between the various strain components we borrow
from classical plasticity theory the idea of a plasticity potential defined as
Q(4 (5)
and write

in which y is a fluidity parameter which can be dependent on some state variables-such as time,
total strain invariants, etc. To ensure no visco-plastic flow below the yield limit we write

F,,denotes here any convenient reference value of F to render the expressions non-dimensional.
If
Q-F (8)
wespeak of an associative law and if
Q Z F (9)
we have a non-associative situation.
A sufficientlygeneral expression for visco-plastic strain rate is given by a power law for 4,
or

and we shall often use this expression simply with an exponent n equal to unity.
Various yield criteria and plastic potentials can be introduced into the formulation depending
on the nature of the materials ~ s e d .In~ what
, ~ follows we shall concentrate our attention on
isotropic materials for which both F and Q can be defined in terms of stress invariants. We shall
thus define

and similarly Q, in which om,J , ,J 3 are the three stress invariants. Using for the moment tensor
notation we define the invariants as
om= 1 s.. = o..-d..o
IJ IJ v m
J 2 = $sijsij
J -1
2 - 3SijSjkSki
824 0. C. ZIENKIEWICZ AND I. C. CORMEAU

A convenient alternative to the third invariant is defined by the Lode angle4

A very general set of expressions can be written for materials possessing a cohesion c and an
angle of internal friction $ in a single expression.

DRUCKER-PRAGER YIELD SURFACE

The first form
6 sin $ 6c cos $
F =
3-sin $ o m + J 3 J J 2 - 3-sin t+b
is an approximation given by Drucker and Prager5 for description of soil-like materials.
If )I = 0 this reduces simply to the well known Von Mises criterion
F = J3 JJ2-2c J3 JJ2- y ( 14b)
(y = uniaxial yield stress).

MOHR-COULOMB YIELD SURFACE

The surfaces described by the equations (14) are smooth and d o not possess corners as shown
in Figure 2(a). An alternative corresponding t o the Mohr-Coulomb yield ~ r i t e r i o ncan
~ . ~be
given as

F =
(
1 .
2sint+bam+ 2cos8--sin$sin8
J 3
JJ2-2ccos$
1 (154

Again if $ = 0 the above expression simply yields the well known Tresca (maximum shear stress)
criterion
F = 2 cos 8JJ, - 2 ~= 2 cos 8JJ, - y (15b)
These surfaces are angular and are shown in the principal stress space in Figure 2(b).

Figure 2. C-I/J yield surfaces in stress space

 VISCO-PLASTICITY 825

Any of the above expressions for F can be used to define suitable non-associated flow potentials
and we shall find that we can always write the general flow equation (6) in the form
p = To (16)
where r is a symmetric matrix dependent on the current stress state. Detailed description of this
matrix is relegated to Appendix I.
The general visco-plastic model described above (or indeed a series of such models assembled
now for a full multiaxial stress condition, in a manner indicated in Figure 1)can describe realistic-
ally material behaviour for a wide class of situations. Further, the numerical solution techniques
of an incremental, time stepping, type can readily be adopted for obtaining the detailed time
history of stress and displacement development in the context of a finite element discretization.6
In the next section we shall discuss and develop such techniques in more detail but it can now be
asserted that they can be applied with no serious difficulty for a variety of forms of equation (6)
which states the visco-plastic behaviour. Numerical solution of some visco-plastic problem using
finite element procedures has been formulated generally by Zienkiewicz and Cormeau’.’ and by
Zarka.9*’oIn a more limited application to rock mechanics problems, Van Winkel and Gerstle”
and Stagg et al” discuss its possibilities. We should not however overlook the possibility of
extending the use of the visco-plastic model to ‘pure’ elasto-plastic behaviour or to standard
creep problems. Both these extremes are well modelled by the visco-plastic formulation.
Elasto-plastic problems can be solved by the elasto-visco-plastic algorithm if stationary con-
ditions are reached at which no further change of displacements (and therefore strains) occurs.
Clearly, if after the application of loads on a structure such a state is obtained, the stresses must
be such that :
F<O
Further, if the stresses are such that equilibrium is satisfied the solution represents an admis-
sible plastic state and if ideal, associated, plasticity is considered such a solution represents a
lower bound of collapse. We shall show later that elasto-plastic solution can be followed by
incrementing the loading appropriately and, at each increment, obtaining after a series of time
steps a stationary solution.
As the elasto-visco-plastic algorithm is applicable to strain softening as well as to strain
hardening situations-and the use of non-associated laws does not introduce formulation diffi-
culties-the process offers many advantages over more conventional approaches to plasticity
solution.
Creep problems represent simply situations which can be modelled by reduction of the yield
stress y to zero in equation (4). Finite values of creep strain rates occur now at all levels of stress.
The reader can verify that the standard formulae of metallic creep result from the application of
the von Mises yield criterion to the expression given for r of equation (16) using the appropriate
formulae of Appendix I and using the expression (10)with an exponent n which for most metals
lies in the range of 3-7. We now can write explicitly
ivp =peep - -
- - y(lc)J”,’a

in which 7 is a fluidity parameter. This formula is readily recognized as the well established
metallic creep law frequently used in (Norton’s law)
Again, as is readily seen, another phenomenon which has usually been dealt with as distinct
has been brought within the scope of the general elasto-visco plastic formulation.
The remarks given above show that the elasto-visco-plastic approach is not only physically
realistic but incorporates the possibility of dealing in a unified way with the extremes represented
826 0. C. ZIENKIEWICZ AND I. C. CORMEAU

by pure plasticity or creep. Is such a unification efficient from the computational standpoint?
The object of the subsequent description is to show that it is and illustrative examples will be
quoted to show the range of application.

THE NUMERICAL ALGORITHM FOR ELASTO-VISCO-PLASTIC

FORMULATION
Initial strain approach
As is common in all finite element analysis we shall start from the virtual work equilibrium
statement. Thus if the vector u denotes the displacement field, which satisfies prescribed dis-
placements on boundary ru,E the strains, and ts the stresses, we have for a virtual and compatible
displacement-strain changes Su and SE the condition that the sum of virtual work performed
must be zero. Thus

Jb &‘a dR - Jn GuTbdR - Jru SuTfdT = 0

In above b denotes prescribed body forces, f prescribed boundary tractions on boundary roand
(18)

it is implicitly assumed that 6u = 0 on ru.For compatibility strains and displacements are

related by a linear differential operator as

a a a
E =Lu LT = 0, -’ 0, - 0, -
aY aZ’ ax

0, 0,
a ay’
-&’
a &’
a O
- -

s, (L~5u)~ts
dR - Jn SuTbdR - 1. SuTf dT = 0

We assume a trial expansion of u in terms of known ‘shape’ functcons N ias

u = CNiai = Na (21)
where a is a series of parameters determining the approximate solution and substituting into
equation (20) we have
[
6aT Jn (LN)TtsdR - In NTbdR - NTfd r ] = 0 (22)

From equation (22) we have the standard equilibrium equation (8) by equating the expression
inside the brackets to zero, i.e.6

BTadR - NTbdR - Jru NTfdT = 0

or
.fnBTtsdR-F = 0
with
 VISCO-PLASTICITY 827

To complete the formulation the constitutive relations for the material have to be introduced.
From equation (1) and (3) we have
cr =D (&-~"p-~o)+~o = DBa-D(&"P+Eo)+a0 (24)
where cr,, is the initial stress present at the start of the problem. On substitution of above into (23)
we have thus

Ka- JnBTDsuPdR-F = Ka-V =0

in which F = F +I,BTDz0 dR-J,BTcro dR represents prescribed 'force' values and K is the

standard 'stiffness matrix't

K = 1,B'DBdR

The visco-plastic strain is obtained from the relation (16)

p = ra
Equations (25), (24)and (16) present a discretized system of ordinary differential equations from
which solution can be obtained in a time stepping manner.
The procedure is identical to the so-called 'initial strain' method described in Reference 6
and in the simplest form proceeds as follows :
(a) Starting from known values of a",a", ( E " ~ ) " and F" at a time instant t" compute the rate of
visco-plastic straining (i"")"
by equation (16).
(b) Determine approximately the change in cupas
(Ac'P)" N (B"P)". At (274
If any change of 'forces' occur determine and V"+' as

vn+l = p.1 + 1, BTD(~"p)"+dn

where
(&"p)"+ = (cup)" +(A&"")"
(c) Compute the value of a"+ using equation (25)
a f l + l = K-lVfl+l
(274
(d) Determine d+'using equation (24) and thus arrive at all the starting values of the next
step for which the same process can be started in the next time interval.
The accuracy of this time stepping algorithm is dependent only on the approximate integration
in equation (27a)which is equivalent to the Euler extrapolation procedure and in order to improve
this an inner iterative loop can be imposed on the sequence (aHd). In this we replace step (b)
by calculating
(Ad"')" 2: (i"+ i"+ ')At/2 (28a)

t It is necessary to mention here that if non-zero displacements are prescribed some of the parameters a are given and
the matrix K is not square. As is usual in the finite element analysis the appropriate columns of K multiplied by the
known parameters are then incorporated into the 'force' vector F.
828 0. C. ZIENKIEWICZ AND I. C. CORMEAU

Taking initially

and repeating the process to (d) for a certain number of cycles improved estimates of visco-
plastic strain and hence improved accuracy can be obtained.
In the practical implementation of the program the reader will observe that the imposition
of this iterative loop involves a substantial increase in the storage requirements as an additional
vector of Sup has to be stored. As the magnitude of the time step which can be taken is severely
limited by stability and iterption convergence requirements which are not relaxed by the above
process, it is found in practice that it is convenient to proceed simply with the scheme outlined
in equations (27) for most practical problems.
In the constitutive law of equation (24) we have deliberately introduced an initial stress q,
from which the computations are started. The importance of this cannot be over-emphasized as
in many practical problems residual stresses are present at the starting stage from which the
displacements are measured. Further, however, it allows us to tackle problems in which a change
of boundary conditions or indeed of configuration is introduced during the time sequence as at
such a change the stresses are noted and simply used as a start of the next sequence.
In the discretizations considered in this paper we use throughout a two-dimensional iso-
parametrid element of a parabolic type (with eight nodes)6 and use for numerical integration a
mesh of 2 x 2 Gauss points which has been proved very effective in numerous application^.^,^

MAGNITUDE OF TIME STEPS-STABILITY

In the early work on visco-plasticity (and indeed on creep problems) it was observed that accuracy
can deteriorate violently with the increase of the time step length. Further, for time steps above a
certain magnitude, oscillating results are obtained indicating instability of the process. In the
consideration of accuracy several empirical rules have been tried in which the time step magnitude
was related to the anticipated changes of visco-plastic strain. A convenient scalar measure of the
visco-plastic strain is obtained by the second strain invariant defined as

In many of the examples described in this and other papers we limit the maximum increment
of this scalar visco-plastic strain to a certain fraction of total accumulated strain at all integrating
points of the system. Thus

7 in the range 0.1, to 0.15 was found to be effective for 'contained' problems but near collapse
lower values in the range 0.01 to 0.05 were found necessary.
A further limit is generally imposed relating the change of time step between successive inter-
vals. In general we limit

At"' d 1.5At" (31)

None of these empirical criteria is entirely satisfactory and none prevents some form of
instability occurring. In considering stability it is worth while to rewrite the governing equations
 VISCO-PLASTICITY 829

(16)(24)and (25)in an alternative form. Differentiating (25) and (24)with respect to time we note
that

a = K-'(JaBTDiuPdQ+F)

and

Eliminating a, using equation (16)and noting that our interest focusses in numerically integrated
elements on a finite set of values of t i we have a set of equations (C is a set of weighting parameters)
6 = DBK- '(CCB'DTo + F)- D r o - D&O (33)
or a standard form of coupled non-linear ordinary differential equations
do
-= f(o,t) (34)
dt
For equations of the above form various limits of stability have been discuss d in lite ature
and in a separate paper CormeauI5 investigates in detail the limits which have to be imposed on
At. The algebra and principles of this investigation will not be discussed here but as the results
are of primary importance, these are summarized below. We shall present these only for associ-
ated visco-plasticity of the kind in which Q = F. For general non-associated visco-plasticity
the stability bounds have not yet been established although experience to date seems to indicate
that stable solutions can be obtained using the same limits as those used in associated behaviour.

DRUCKER-PRAGER ASSOCIATED VISCO-PLASTICITY

Using expression (14a) for the yield function and plastic potential, the stability limit is given by
the smallest of two expressions :

(354
(1+~)(1-2~)Fo (3- sin +)2
At G
yE4' 31-2v)(3-sin +)'+6(1 +v)sin2 +
where

and E and v are the elastic constants of the material.

The stability for the von Mises material case is simply obtained by taking + = 0 and gives an
extremely simple expression
4(1+ V) Fo
At<--
3YE 4'
It will be noted that the stability limit in the form given above is independent ofthe mesh sub-
dioision, a feature of considerable computational importance.
8 30 0. C. ZIENKIEWICZ AND 1. C. CORMEAU

Mohr-Coulomb type materials yield similar formulas for associated behaviour. Defining F
by equation (15a) we have

At G
+
( 1 v ) ( 1 - 2v)Fo
+
y ~ [ 1 -2v sin’ +]4’
For the Tresca case the above expression is reduced by inserting + = 0 and we obtain

It will be observed that the Mohr-Coulomb-Tresca potential gives critical time step value
in the range of 0.5 to 0.75 of those obtained by the Drucker-Prager-von Mises potential.
The dangers of exceeding stability criteria are illustrated on a simple example. This concerns
the ‘autofrettage’ of a cylinder of visco-plastic material initially inserted inside an elastic tube
(liner) clearing it by a specified-amount. A pressure is applied inside the visco-plastic cylinder
causing it to flow (Stages 1-111 of Figure 3) until the clearance is closed. In stages IV-V the load
is transferred to the outer cylinder until static elasto-plastic solution is reached. In stages VI-VIII
the pressure is removed and a re-adjustment recurs leaving a set of residual stresses. The dis-
cretization uses six isoparametric elements as shown on Figure 3 and the problem illustrates
well the need for the introduction of the initial stress concepts previously discussed. The stress
history development is shown in Figure 4 while Figure 5(a), 5(b) show respectively the stationary
effective stress and hoop stress in the inner cylinder with and without limiting the increments
to the stability limit. The general oscillations in stresses when stability limits are exceeded and
the reduction of the plastic stresses below the plastic yield should be noted.

SOME SPECIFIC DETAILS O F THE COMPUTATION PROCEDURE

Asymptotic approach to stationary condition
As we have mentioned already in the introduction, the general algorithm of elasto-visco-
plasticity can be used for solution of pure plasticity problems continuing the solution until
‘stationary’ conditions are reached. In such calculations the exact form of the visco-plastic rate
function 4 of equation (6) or indeed of the fluidity y is immaterial and time plays the r61e of a
fictitious variable. As the approach to stationary conditions is asymptotic complete stationariness
is seldom reached and a simple test is included in the program to stop time stepping when F
approaches some prescribed proportion of F o at all points of the domain. This has been fixed
in most problems discussed later as 0.5 per c e n t 4 1 per cent and within this limit the lower
bound solution is always obtained. In practice cruder limits often suffice.

Relation of the visco-plastic solution to other solution procedures of plasticity problems

In conventional solution of plasticity problems two general techniques are available. First the
incremental approach in which at each stage the tangential stiffness matrix is updated, second
the initial stress methodology in which a constant elastic stiffness is used iteratively correcting
the ‘residual forces’ obtained from the difference of elastic and plastic stresses calculated. These
techniques and many variants thereof are described in Reference 3 and all introduce complexities
due to inexact satisfaction of yield conditions at various iteration stages. All these complexities
and artifices disappear in the visco-plastic formulation. However it is of interest to compare
the procedures and their efficiency. While exact relationship cannot in general be established it
is easy to show (Appendix II)I6 that for a specific case of uniaxial elements the initial stress
 LOAO HISTORY
9-
I -PRESSURE STEP & d;mcnsionless mean hoop
EXCEEDING CORE P s t r e s s in l i n e r
BEAR IN G CAPAC I T Y
- E L A S T I C RESPONSE P
6-
II FLOW
- VISCOPLASTIC pressur
IN CORE drop
-STRESS REDISTRIBUTION

m -STATIONARY STRESSES
-STEADY VELOCITIES
-GAP CLOSES

IY - L I N E R PROGRESSIVELY p=o
LOADED
-RATE OF CREEP SLOWS
DOWN s
NOTE: i n
- t h e absence o f viscoplast ;c. , x0
P -STATIONARY STATE
UNDER PRESSURE c o r e the m e a n h o o p s t r e s s
i n t h e L i n e r i s dgl= 9.992 p
(LAME formulr) ’
l!I - INSTANTANEOUS
PRESSURE REMOVAL
- E L A S T I C RESPONSE

YII -REVERSED VISCOPLASTIC V I SC 0 PL AS T IC

FLOW IN CORE DUE TO
HOOP TENSION I N LINER

X D I - F I N A L STATIONARY
STATE AND RESIDUAL m d;mens;onlesS time
STRESSES I I
- AUTOFRETTAGE COMPLETED 0 0.5 1 1.5
CORE:RADII
- land 2 cm L I N E R RADII 2 a n d 2 . 2 cm
E. l o 7 bars
E : 2x107bars L I N E R STRESS H I S T O R Y
4:0 . 3 3 4 : 0.33
Y : 104 bars pure\ y e l a s t ic
quas ilinear von MISES

Figure 3. Autofrettage in a composite thick walled cylinder. Quasilinear von Mises flow function @(FIFO)= F/F, with
F, = us
832 0. C. ZIENKIEWICZ AND I. C. CORMEAU

I
w . 1-1-1
I LH -I - -
 VISCO-PLASTICITY 833

-
Y
e f f e c t i v e ' stress a t s t a t i o n a r y s t a t e VIII

0.90 -
-maximum t i m e step e n c o u n t e r e d
I ' equals'3.3 t i m e s t h e s t a b i l i t y l i m i t
I I - z-0.05
I/
Y

-1

9e
Y

-
-2

t i m r interval kept within s t a b i l i t y l i m i t

-+- t i m e i n t e r v a l a l l o w e d t a e x c e e d t h e
stability limit

Figure 5. Composite thick cylinder. Effect of numerical instability on the stress distribution in the visco-plastic
region
approach and the visco-plastic time stepping are identical for a certain value of At. Indeed it is
simple to show that At = Atcri,,corresponding to the stability limit for such cases is equivalent
to the initial stress procedure in which the residuals are exactly doubled in the iteration process-
and thus corresponds to an over-relaxation process. In a general continuum situation this is
true in specific circumstances only but the general principle is still valid making the uisco-plastic
approach equivalent to an accelerated initial stress procedure and thus being basically more
efficient.
Application of fictitious time and 'viscosity' constants to the solution of the problem could be
accomplished as a pure mathematical fiction. (Indeed this process is used in the so-called dynamic
relaxation method"^ ' where inertia forces are also perhaps unnecessarily introduced.) Consider
a non-linear set of equations such as may be derived by the finite element process for some non-
linear elastic situation, e.g.
f(a) = 0 (37)
834 0. C. ZIENKIEWICZ AND I. C. CORMEAU

it is always possible to amend this to a time variable problem of the kind

da
a- + f(a) =‘O
dt
and seek the solution to the non-linear system of equation (37) in an asymptotic manner as the
final steady state solution of equation (38). Clearly much depends on the structure of the matrix a
and vector f to ensure existence of a steady state and a stable and efficient computation algorithm.
In the visco-plasticapproach not only is such an algorithm supplied on the basis of sound physical
reasoning but, as usually follows this course, an efficient process is discovered.

Treatment of ‘corners’ in plastic-potential surfaces

As in plasticity-a difficulty arises when ‘corners’ on the plastic-potential surface, Q , exist-as
they do in case of Mohr-Coulomb or Tresca materials. On such corners the direction of straining
is indeterminate and in the procedures used here a simple device of ‘average directions’ is used.
This is identical to the procedure used in Reference 3 and when a ‘corner’ is approached within a
certain tolerance (e.g. when 8 in equation (13) approaches f71/6 by 3 a degree) the direction of
straining of either side of the ‘corner’ is computed and the average taken. This corresponds to
‘rounding’ the corner and involves no computational problems.

APPLICATIONS : CREEP PROBLEMS

As already mentioned, if the yield stress in the von Mises expression for visco-plastic material
is taken as zero, pure creep results. To test this feature of the algorithm a problem solved by
~ recomputed and results are shown in Figure 6. Excellent
Greenbaum and R ~ b i n s t e i n ’was
comparison between the two results is shown despite a coarse mesh of isoparametric elements.

APPLICATIONS : ASSOCIATED PLASTICITY

Perforated strip-strain hardening
In this case again a von Mises criterion is used-now in a plane stress situation of a perforated
strip on which numerous measurements were taken by Theocaris and Marketos. l 9 This problem
’
has been used frequently for assessment of plasticity algorithms20*2and in Figure 7 and 8 we
show some of the results obtained in the present investigation together with those of Reference 20.
The results compare well with alternative plasticity computations although the strain dis-
tributions are somewhat different from measured ones.
This example serves to illustrate a further point on which some controversy exists. It has
always been asserted that it is essential to proceed in pure plasticity in small load increments to
reproduce the incremental nature of plasticity laws. In the results shown this process of increment-
ing loads was pursued with 6 steps only. At each stage the visco-plastic time stepping process
was used to obtain stationary results. The number of time steps needed at each stage with a limit
on z = 0.05 and further limited by stability criteria imposed is recorded in Figure 8. To test the
need for load incrementation the full load of stage V was imposed instantly (one shot solution).
With identical stability and step limits this required about 46 per cent less time steps of which
however most were at the stability limit. Stresses obtained in general differed by less than 1 per
cent (maximum difference 1.5 per cent) while displacement differed by a maximum of 0.17 per
cent. Clearly in many cases considerable economy may be achieved without following thefull load
incremental process providing the loading is monotonically increasing. A further test was carried
out to study the need for imposing any other time stepping criteria beyond the stability limit.
 VISCO-PLASTICITY 835

With imposing the latter only the number of steps was reduced from 45 to 34 for complete
stationariness in ‘one shot’ loading. The maximum error in stresses was now however increased
to 2.8 per cent.

t 0.121
0445

800

900

0 r e s u l t s of GREENBAUM and
R U B INST E I N (1968 J
a l t e r 3 hours c r e e p

300
___--_---------- - l-
Figure 6. Creep in a pressure vessel. Power flow function @(F/F,,) = (F/F0)3.61 with F, = 1 psi. Contours of ,/3J, =
‘effective’ stress: (a) Upon instantaneous pressurization at t = 0 (b) At t = 3 hours. 86 times steps. 9.8s < At < 346s,
T = 0.03
136 M. GOTOH

where = c”
2,k, b: = 2~;.,, C
: = c:,,, U: =~:,k,
bk, = c;&, ck, = 2c;,,, ,k 3 - c‘;,,+c;,,, bkj = c;,,+~c;,,,

ckj = ~C;,,+C’;~,, p: = --2cy3,,, q: = 0, r: = -2cY6,,,

s t = -12cy7,,9 t: = 0, pi = -2c74+k9 qk, = -2c76,k3
rk, = si = 0, tk, = -12c;8,k, pi = 2c;5,,, qkj = 4c76,k,

ri = 4c;6,&, skj = ti = 0, and x i , x i . . . are nodal co-ordinates.

When the shell material has anisotropy, change of direction of its principal axes must be taken
into account. Then let the principal axes (x,,y , , zl) at one stage before move to (x’, y’, z‘) at
current stage. The direction cosines [A] of (x’, y’, z’) with reference to the fixed co-ordinates
(x, y, z ) are obtained as the following :

PI = [Aisjl (23)
and
I.,,, + ajp3 + J[(a;P2 + ajp3)2 -(I +/I; + b:)(a;’ + a;’
= [“;fl2 - 1)]]/(1 + +
A,.,, = a; -P2A,,,, 3 -p’ 3 A
AXfZ= arl X’X)

lyx= [a’;& + a;& + J[(a;p; + a’;&)’ - (1 + + p?)(a;2 +

p;2 .‘;2 - I)]]/( 1 +a;’ + g),
/Iv‘, = c Y ; - p l ; / I y , x , i.,.
=,.’j-p;i,.,,
a; = e ; i z ’ z / ( u 2 i z ’-a3Azfy),
z

p2 = .( lJbz’z - u 3 i b z ’ x ) / ( u 2 A z ’ z -u3Az‘y),

a; = ~ ; ~ z , y / ( ~ 3 A-2 a, y2 M,

8 3 = (‘l~z’y-u2/2z’x)/(a3~z’y -a2i2’2)?

a’; = e;a;/e;, /3; = p2, a’; = e;uh/ei, 8; = P 3 ,

e; = uIl’[+u2l‘’+u3l‘i, e; = a,m‘;+u,m‘;+a,m‘;,
1’; = ,,?L, cos 68,, + A y I x sin 68,,,
1’; = A,,, cos do,, +Ayly sin dd,,,
1; = A,x,zcos 60,, + Aylz sin 68,,,
my = -A,,,sin6%,,+A,,,cos 68,,,
m; = -,IxI, sin dd,, +Ayly cos 68,,,
m’; = --Axlz sin 68,, +Aylz cos 68,,,
and Ailj’sare the known direction cosines of (xl,y,, zl),6d,, is the incremental rotation around
the z1 axis at one step before which can be estimated from equation (2),and
a, = (Azly~~z-
’z ~ ~ , ~ ~ ~ , y ) / [ ( ~ ”-2 ~, z,l y~ J2- z, ’yx ) 2+ ( ~ z I y ~ z-, z~ ~ z l z ~ . z ’ y+) 2( ~ z 1 2 ~ z-, x~ z l x ~ z ‘ z ) 2 1 * ?
a2 = a1 .( J - z , z ~ z-
~ x~ 2 1 x ~ 2 ~ z ~ / -
~ ~~” z
2 11z ~yz ~
, y z) 9 ~ 2

a3 = a1 . (AzIxJLz.y- ~ ~ l y ~ z ~ x ~ / - ~ ~ z l y ~ 2 ~ z AZ,Z~Z’Y)>

and Az,.r, AZPr and are easily estimated from the current position of three nodal points.
 GENERAL DEFORMATION OF SHEET METALS 737

For large deformation, the stiffness equation (15) must be modified to express geometrical
non-linearity. Then we obtain the next total stiffness equation in reference to the fixed co-
ordinates.
+
{6F} = [K] (6) [T*IT{SL)+ [6T*IT[T*]{F} (24)
where matrix [K] is the total stiffness matrix for small deformation, [T*] is the translation matrix
for whole structure, and {S;} is related to geometrical or initial stress stiffness and is given as :
{s;)= h0,~{6C’;,,+26C~,, . I1 +6C’;,, . ~ , } + h r T - , ~ { 6 ~ ~ , , + .~I~1 ’+2&:,,
;,, . Iz}
+ h T X . , ~ { 6 C ‘ ; , ,+ dc‘;, + (6c’;,, + 264,) .11 + (264, + CSC’;,) . 12),
where m means row number (1,2,..., 18), and I, = A . d x & , I , = A . 6 y & , and (6x&,6y&) is
position shift of centre of gravity. The last term in (24) corresponds to so-called initial load stiff-
ness. We must note that (6F) includes incremental quantities besides external forces’ increments,
if {F} contains variable terms besides external forces. The solution for one stage of calculation
of equation (24) is obtained only by repeated process.
Finally, boundary conditions or loading conditions must be formulated for any given prob-
lems. For example, we treat hydraulic bulging of sheet metals. Then deformation is constrained
at the clamped end and the loading condition is expressed in terms of nodal force increments as
follows :

where
6Fij = j-components of nodal force increments of node i,
j = either x, y or z,
mi= number of elements to which node i belongs,
Ai = area of element k ,
(&‘j)k = direction cosine of z‘-axis of the element k with respect toj-axis.
All increments of external couples (6M,, 6 M , and 6M,) are put equal to zero for nodal points
not located on the clamped end.

NUMERICAL EXAMPLES
A few examples of hydraulic bulging of circular sheet metals are examined numerically. Calcula-
tion is performed by the computer FACOM 230-60 at Nagoya University Computation Center.
Firstly, only elastic deformation is considered for three types of element division. Those are
shown in Figure 2, and have 42, 82 and 148 elements in a quarter part of the circular sheet,
respectively. Material properties used in the calculation are: ho = 0.8 mm, a = 25 mm, E =
21,000 kg/mm2, v (Poisson’s ratio) = 0.29, and isotropy is assumed.
Figure 3 shows calculated deflection curves along a radial line for pressure p = 1 kg/cm2.
In this figure, we can see the evident effect of clamping at the end.
Figure 4 shows w,,, (maximum deflection) against number of elements used in calculation.
From this diagram, we can confirm considerable rapidity of convergence of w,,, when the number
of elements is increased.
 838 0. C. ZIENKIEWICZ AND I. C. CORMEAU

40 b a r s

I 1 upper b o u n d IILIOUCHINE[2II
__.L___.-.-. .-.-.
lower bound ILlOUCHlNE I

-
+--+: POPOV e t a1.[22]

viscop\astic r e axat i o n

-.+ continuing viscoplastic flow

0.1 0.2 0.3 0

Figure 9(b). Clamped circular plate. Stress-strain statical curves (exponential fitting). Load-central displacement curves
 RIGID BOUNDARY
A S S O C I A T E D M A T -E- R I A L I NON A S S O C I A T E D M A T E R I A L
Figure 10. Lined tunnel in a pre-stressed visco-plastic soil. Quasilinear Mohr-Coulomb flow function @((FIFO)= F / F o
with F , = 1 psi. Yield tolerance 0.1 per cent. Time step control parameter T = 0.03. Associated material: 96 time steps
8.3 x lo-" iyAt i2 x lo-'. Non-associated material: 84 time steps 8.3 x lo-" iyAt i2 x lo-'
840 0.C . ZIENKIEWICZ AND I. C. CORMEAU

Stationary vtscoplastic stresses Elastic stress diagram

associated MOHR-C0UU)MB

0 1 2 3Xlo-barS 2
.
all stresses are - w i v e
1p ft

P 1 2 gm x

Figure 11. Lined tunnel: lining extreme fibre stresses

The number of time steps in this example is fairly large due to a requirement of 0.1 per cent
approach to the yield surfaces which for practical purposes is too stringent.
Circular plate-strain hardening-strain softening
This problem is brought in to illustrate a case in which strain softening characteristics can be
included in the solution and is illustrated in Figure 9(a) and (b). A very fine mesh is used and for
comparison results obtained for a strain hardening material by Popov et ~ 1 are. given. ~ With
~
strain softening of the form given no comparisons are available here but bounds on the perfectly
plastic solution given by I l i ~ u c h i n eare
~ ~indicated and the present solution is shown to lie
between these.

APPLICATION-ASSOCIATED VERSUS NON-ASSOCIATED MATERIALS OF

MOHR-COULOMB TYPE
In this last example we apply the algorithm to a soil mechanics problem of a tunnel inserted into
a sandy material (Figure 10). It is assumed that the soil is supported until the lining is placed-
and that the initial stresses are thus instantly released. While much controversy exists with regard
to the uniqueness and existence of solution for non-associated plasticity it is well known that the
volumetric changes occurring in soil-like materials do not correspond to associated assumptions.
We thus solve the above problem using both associated and non-associated rules. In the latter
case a set of Tresca surfaces is simply taken as the plastic potential. Both solutions show complete
 VISCO-PLASTICITY 84 1

convergence and indeed the spread of the plastic zones (Figure 10) is not much afrected by the
flow rule adopted. Figures 11 and 12 show the development of stresses and deformation in the
lining and here considerable differences between the two rules are detected at the stationary
solution though the average stress level in the lining is broadly maintained.

70 80 90 100 i t
I
Figure 12. Lined tunnel : lining deformation (magnification 200)
842 0. C. ZIENKIEWICZ AND I. C. CORMEAU

CONCLUSIONS
The theory outlined and the examples given here show that the elasto-visco-plastic algorithm
presents a powerful, efficient and unifying process of approach to a wide range of materially
non-linear problems. While only four kinds of isotropic yield surfaces were considered it is clear
that most other material non-linearities can be inserted. Indeed the solution technique is adapt-
able to visco-elastic behaviour and this can be readily coupled with visco-plastic phenomenon in
series. At the present time such widely different behaviours as those of laminated rock and con-
crete with limited tension are being investigated and the common framework allows a simple
programming approach.
Two questions which as yet are not completely resolved are those of the most efficient way of
dealing with large deformations and with dynamic effects, and research is continuing in these
directions.

ACKNOWLEDGEMENT

The authors are grateful to the Belgian Fund for Scientific Research for providing support to
I. C. Cormeau during his work at University College of Swansea.

APPENDIX I
The visco-plastic strain rate law
In equation ( 6 ) a very general visco-plastic rate law is presented.

Various forms of the yield function F are given in equations (14a) and (b) and (15a) and (b) and
the calculation of the scalar quantity in the square brackets presents no difficulty. The calculation
of the vector aQ/an is more complex and we give here a form suitable for expressing (6) as given in
equation (16).

We can write quite generally, in the isotropic case,

with Mo, M' and MI' being symmetric matrices summarized in Table I.
 VISCO-PLASTICITY 843

Table I
-
0 0
0 0
0 0
0 0
2 0
2-
-4 0 0 0-
- + o o o
- 4 0 0 0
1 0 0
1 0
1-

All that remains therefore to establish the matrix r is to compute the scalar derivatives aQ/acr,
etc. In Table I1 we give the simple expressions derived from equations (14a)and (b) and (15a)and
(b). It is readily seen that the change from one form of plastic potential to another is established
with very minor computation changes.

Table 11. Derivatives of plastic potential Q

DruckerIPrager von Mises

equation (14a) equation (14b)

6 sin
3-sinJ/
* 0

J3 J3
0 0

Mohr-Coulomb Tresca
equation (15a) equation (15b)

aQ/aam 2 sin JI 0
aQ/aJ, 2 c o s 8 [ 1 + t a n B s i n 3 0sin
+ ~ij( t a n 3 0 - t a n B ) ] 2cos8[1+tanesin30]

aQlaJ3
+
J3 sin e sin t,b cos e J3 sin e
J 2 cos 30 J , cos 30

Note: von Mises and Tresca are obtained simply by substituting + = 0 in more general ex-
pression.
844 0. C. ZIENKIEWICZ AND I. C. CORMEAU

APPENDIX I1
Equivalence of elasto-visco-plastic approach and of initial stress procedures
Consider a structure of uniaxial bars with a linear visco-plastic law of the form
E U P = (y 0 fo y ) / O , (‘44)
where

The corrective ‘force’ of equation (25) gives a change at each time step of

s, BTDEUpAtdR =
I Y
BTE-(
0,
0 f0 , ) A t

If (y . A t . E ) / a , = 1 we have precisely the initial stress a l g ~ r i t h m ~where

dR

, ~ . ~ the
~ difference
(‘45)

between the actual stress and the yield value is corrected.

For stability it is easy to show that
20
At d 2
YE
and hence the critical time step integration is precisely the initial stress procedure with an over-
relaxation factor of two.
For triaxial stress conditions the initial stress algorithm becomes more complex owing to the
need to define a tangential modulus matrix D , , depending on some stress B on the yield surface.
However, with certain assumptions regarding this procedure, a similar equivalence can be
demonstrated.“

REFERENCES
1. E. C. Bingham, Fluidity and Plasticity, McGraw-Hill, New York, 1922, Chap. VIII, pp. 215-218.
2. P. Perzyna. ‘Fundamental problems in viscoplasticity’, Advan. Appl. Mech. 9, 243-377 (1966).
3. G. C. Nayak and 0. C. Zienkiewicz, ‘Elasto-plastic stress analysis. Generalization for various constitutive relations
including strain softening’, Inr. J. Num. Meth. Engng, 5, 113-135 (1972).
4. G. C. Nayak and 0. C. Zienkiewicz, ‘Convenient forms of stress invariants for plasticity’, Proc. ASCE, 98, 949-954
(1972).
5. D. C.’Drucker and W. Prager, ‘Soil mechanics and plastic analysis on limit design’, Quart. Appl. Math. 10, 157-165
\
( I 952).
~ - - -,-
6. 0. C. Zienkiewicz, The Finite Element Method in Engineering Science, McGraw-Hill, London, 1971.
7. 0. C. Zienkiewicz and I. C. Cormeau. ‘Viscoplasticity solution by finite element process’, Arch. Mech. 24. 5-6, 8 7 s
888 (1972).
8. 0. C. Zienkiewicz and I. C. Cormeau. ‘Visco-plasticity and plasticity. An alternative for finite element solution of
material non-linearities’, Proc. Colloque Mithodes Calcul Sci. Tech. 171-199, IRIA Paris (1973).
9. J. Zarka. ‘Generalisation de la theorie du potential multiple en visco-plasticite’. J . Mech. Phys. Solids, 20. 179-195
(1972).
10. Q. S. Nguyen and J. Zarka, ‘Quelques methods de resolution numerique en elastoplasticite classique et en elasto-
viscoplasticite’, Siminaire Plasticite et Viscoplasticiti, Ecole Polytechnique, Paris (1972), also Sciences et technique
de I‘armement, 47, 407436 (1973).
1 I . B. Van Winkel and K. H. Gerstle, ‘Analysis of time dependent deformation of openings in salt media’, Int. J . Rock
Mech. Min. Sci. 9, 249-260 (1972).
12. K . G. Stagg, 0. C. Zienkiewicz, and I. C. Cormeau, ‘On the application of a numerical visco-plastic model to Rock
Mechanics problems’, Int. Symp. Untertagbau (Ed. H. Grob and K. Kovari), 327-335, Luzern (1972).
13. G. A. Greenbaum and M. F. Rubinstein, ‘Creep analysis of axisymmetric bodies using finite elements’, Nucl. Engng
Des. 7 , 378-397 (1968).
14. R. K. Penny and D. L. Marriott, Designfor Creep, McGraw-Hill, London, 1971.
 VISCO-PLASTICITY 845

15. I. C. Cormeau, ‘Numerical stability in quasi static elasto-visco-plasticity’,fnt. J. num. Meth. Engng, 9,00(M00 (1975).
16. I. C. Cormeau, ‘Viscoplasticity and plasticity in the finite element method’, Ph.D. Thesis, University College of
Swansea, 1974 (in preparation).
17. J. R. H. Otter, ‘Computation for prestressed concrete reactor pressure vessels using dynamic relaxation’, Nucl.
Struct. Engng, 1, 61-75 (1965).
18. R. D. Lynch, S. Kelsey, and H. C. Saxe, ‘The application of dynamic relaxation to the finite element method of struc-
tural analysis’, Univ. of Notre Dame Indiana, U.S.A. Techn. Report Themis-UND-68-1 (1968).
19. P. S. Theocaris and E. Marketos, ‘Elastic-plastic analysis of perforated thin strips of strain hardening materials’,
J. Mech. Phys. Solids, 12, 377-390 (1964).
20. P. V. MarGal and I. P. King, ‘Elastic-plastic analysis of two dimensional stress systems by the finite element method’,
Int. J. Mech. Sci. 9, 143-155 (1967).
21. 0. C. Zienkiewicz, S. Valliappan, and I. P. King, ‘Elasto-plastic solutions of engineering problems. Initial stress,
finite element approach’, Int. J. num. Meth. Engng, 1, 75-100 (1969).
22. E. P. Popov, M. Khojasteh-Bakht, and S. Yahmai, ‘Bending of circular plates of hardening material’, Int. J. Solids
Struct. 3, 975-987 (1967).
23. A. A. Iliouchine, ‘Plasticitt’, Diformations ilastico-plastiques, Eyrolles, Paris, 1956.