COMMUNICATIONS IN NUMERICAL METHODS IN ENGINEERING, Vol.

13, 963±976 (1997)

A FINITE ELEMENT METHOD FOR PLANE STRESS

PROBLEMS WITH LARGE ELASTIC AND
PLASTIC DEFORMATIONS

E. KIRCHNER,1 ST. REESE 2 AND P. WRIGGERS 2*

1
Fachgebiet fuÈr Maschinenelemente und Maschinenakustik, Technische Hochschule Darmstadt, Magdalenenstrasse 4,
64289 Darmstadt, Germany
2Institut fuÈr Mechanik IV, Technische Hochschule Darmstadt, Hochschulstrasse 1, 64289 Darmstadt, Germany

SUMMARY
A two-dimensional ®nite element method is developed for large deformation plasticity. Principal axes
are used for the description of the material behaviour, and the use of principal logarithmic stretches leads
to exact formulae for ®nite deformation problems with large elastic and plastic strains. An ecient
return mapping algorithm and the corresponding consistent tangent are derived and applied to plane
stress problems. Two examples show the performance of the proposed formulation. # 1997 John Wiley &
Sons, Ltd.
Commun. Numer. Meth. Engng, 13, 963±976 (1997)
No. of Figures: 5 No. of Tables: 1 No. of References: 12

KEY WORDS plane stress; ®nite stress; ®nite strain plasticity

1. INTRODUCTION
Most engineering materials show inelastic behaviour when being submitted to large stresses.
The corresponding deformations often become very large, so that geometrically non-linear
analysis is required.
Strength computations for such devices are mostly based on simplifying assumptions to reduce
the complexity of the problem; the most common approach to large deformation is a non-linear
®nite element computation. Algorithms for the solution of large strain plasticity problems have
been discussed extensively in recent literature, e.g. Simo,1 Hopperstad2 and Wriggers et al.3
The assumptions made in this paper concern the problem dimension and the constitutive
model. Firstly, the discussion is restricted to the plane stress case, reducing the dimension of the
problem from 3 to 2. A similar approach is due to Simo and Taylor,4 which is restricted to the
geometrically linear regime extending the algorithms by Simo and Taylor5 for elastoplasticity to
the plane stress case.
Secondly, constitutive modelling is performed in the principal axis, as done by Reese,6 Simo
and Taylor7 and Wriggers et al.,3 leading to exact expressions for the stress update algorithm and
corresponding tangent operators for non-linear plane stress applications.

* Correspondence to: P. Wriggers, Institut fuÈr Mechanik IV, Technische Hochschule Darmstadt, Hochschulstrasse 1,
64289 Darmstadt, Germany

CCC 1069±8299/97/120963±14$17.50 Received 30 September 1996

# 1997 John Wiley & Sons, Ltd.
964 E. KIRCHNER, ST. REESE AND P. WRIGGERS

This paper is organized as follows. In Section 2 the constitutive equations are summarized
brie¯y. Section 3 covers the description of the developed algorithm for plane stress elasto-
plasticity being restricted to the stress update algorithm and its consistent tangent; the ®nite
element implementation is omitted since it is standard. Finally, in Section 4 two examples
illustrate the abilities of the presented approach, followed by a short conclusion.

2. CONTINUUM MECHANICAL BACKGROUND

In this Chapter the plane stress assumption and the necessary implications are discussed.
Principal axes are used for the description of the material behaviour, and the constitutive laws for
the elastic and plastic deformations are reviewed brie¯y.

2.1. Plane stress assumption

Within the framework of this paper, the discussion is focused on the treatment of the classical
plane stress assumption

t13 ˆ t23 ˆ t31 ˆ t32 ˆ t33  0 …1†

formulated in terms of the Kirchho€ stress tensor t ˆ PFT. P is the ®rst Piola±Kirchho€ stress
and F is the deformation gradient. According to the constraint t33 ˆ 0, the work conjugate left
Cauchy±Green strain b33 does not vanish; it has to be determined in such a way that (1) is
satis®ed. For elastoplasticity the total strain b33 remains undetermined; the elastic part be33 follows
then from the plane stress condition. The shear stresses t13 and t23 vanish by the assumption of a
two-dimensional problem with @/@z0.
The following form of the deformation gradient F is appropriate here:
2 3
@x @x
2 3 6 @X 0 7
6 @Y 7
F 11 F 12 0 6 @y 7
@y
F ˆ 4 F 21 F 22 0 5ˆ6 6 @X 0 77 …2†
0 0 F 33 6 @Y 7
4 @z 5
0 0
@Z

The entry F33 is determined iteratively from t33 ˆ 0 according to the elastic model; see Section 2.3.
For the quadratic stored energy function, F33 can be calculated explicitly. By F from (2) the left
Cauchy±Green tensor b ˆ FFT and its spectral decomposition (6) with principal values l2i ,
i ˆ 1, 2, 3, follow immediately:
v"
#
u 
b11 ‡ b22 u b ÿ b 2
l21;2 ˆ + t 11 22
‡b12 b21 ; l23 ˆ b33 ˆ …F 33 †2 …3†
2 2

The eigenvectors associated with the principal values l2i are then
T T T
n1 ˆ …cos W sin W 0† ; n2 ˆ …ÿsin W cos W 0† ; n3 ˆ …0 0 1† …4†

COMMUN. NUMER. METH. ENGNG, VOL. 13, 963±976 (1997) # 1997 John Wiley & Sons, Ltd.
 FINITE ELEMENT METHOD FOR PLANE STRESS PROBLEMS 965

referred to a Cartesian frame in the instantaneous con®guration. The angle W denoting the
orientation of the principal system is W ˆ 12 tanÿ1 ‰2b12 =…b11 ÿ b22 †Š. It is obvious by (3) that l21 is
not necessarily the largest eigenvalue of b.

2.2. Description in the principal axis

For simplicity the constitutive modelling is performed in the principal axis.1,6,7 A Lagrangian
rank-2 tensor S, e.g. the second Piola±Kirchho€ stress, has a polar representation using the
principal values si , i ˆ 1, 2, 3, and principal axis Ni ,

X
3
Sˆ s i Ni
Ni
iˆ1

If hyperelasticity is assumed for the elastic deformations, then the principal values SI of the
second Piola±Kirchho€ tensor S follow from a polyconvex strain energy function W as

@W 1 @W
SI ˆ ˆ …5†
@E I lI @lI

The lI are the square roots of the principal values of the Cauchy±Green tensors C ˆ FTF and b;
the EI are eigenvalues of the Green±Lagrange tensor E ˆ 12…C ÿ 1†. The principal Kirchho€
stresses bi follow by push-forward of the principal values SI of S, formulated in terms of the
eigenvalues li of the deformation gradient F by bi ˆ l2i Si ˆ li @W=@li .
Concerning the orientation of the di€erent principal systems, it follows from the isotropy of the
problem that the left Cauchy±Green tensor C and the Kirchho€ stresses t possess the same
spatial principal axis ni , i.e.

X
3 X
3
bˆ l2i ni
ni ; tˆ bi ni
ni …6†
iˆ1 iˆ1

Since logarithmic stretches are introduced in Section 2.4 to describe the plastically deformed
con®gurations, it is helpful to have the following results available. With ei ˆ ln li it follows by the
chain rule that

@W…lj † ^
@W…e k†
b i ˆ li ˆ
@li @ei

holds. The argument of the strain energy function W changes in accordance with the form of the
stretches adopted; see Section 2.3.2.
The stress vector t^ ˆ ft11 t22 t33 t12 gT in Cartesian co-ordinates follows by a relation used by
Gruttmann and Taylor8 and Reese:6

t^ ˆ Rb
b …7†

# 1997 John Wiley & Sons, Ltd. COMMUN. NUMER. METH. ENGNG, VOL. 13, 963±976 (1997)
966 E. KIRCHNER, ST. REESE AND P. WRIGGERS

A hat ^ symbolizes that the quantity  was computed in the principal axis and transformed
back to the physical co-ordinate system. b ˆ (b1 b2 b3 0)T is the vector of principal Kirchho€
stresses and
2 3
…n11 †2 …n12 †2 0 2n11 n12
6 22 7
6 …n1 † …n22 †2 0 2n21 n22 7
6
Rij ˆ 6 7 …8†
7
4 0 0 1 0 5
n11 n21 n21 n22 0 n11 n22 ‡ n21 n12

the respective transformation matrix. The fourth entry of b is dropped in the sequel when relating
to principal axis. The nji are the components of the eigenvectors ni of the left Cauchy±Green
tensor b; see (3), (4) and (6).

2.3. Elastic deformations

2.3.1. Material tangent matrices in principal axis. For the tangent matrix D used in the
incremental stress±strain relations
dtt ˆ D dee …9†

we have, as described by Reese,6 the following expressions:

 
1 @ 1 @W 2 2
Diijj ˆ l l …10†
lj @lj li @li j i

with indices i, j covering values from 1 to 3. For the shear components Dijij the following
equations6 are used:
bi l2j ÿ bj l2i
Dijij ˆ ; i; j ˆ 1; 2; 3; i 6ˆ j …11†
l2i ÿ l2j

All other members of D which are not de®ned by (10) or (11) vanish due to the isotropy of the
problem.
If, as in Section 2.2, logarithmic stretches are used in the description of the constitutive
behaviour, then it is possible to put the spatial tangent matrix D in a special form. This is
achieved by introducing a (spatial) material tangent C related to logarithmic stretches

@ti @2 W ^ @tt @2 W ^
C ij ˆ ˆ ,Cˆ ˆ …12†
@ej @ei @ej @ee @ee @ee

Remark 1. It is clear by the indices in (12) that the tangent modulus C is a 3  3 matrix de®ned in
the principal axis of the left Cauchy±Green tensor b.

Application of the chain rule and use of logarithmic stretches in the stored energy function WÃ
yields, with @ei/@lj ˆ dij(lj)71, the desired form of D,
Diijj ˆ C ij ÿ 2ti dij …13†

COMMUN. NUMER. METH. ENGNG, VOL. 13, 963±976 (1997) # 1997 John Wiley & Sons, Ltd.
 FINITE ELEMENT METHOD FOR PLANE STRESS PROBLEMS 967

where dij is the Kronecker symbol. Note that D is a fourth-order tensor, as predicted by (9). One
should note that W depends on the principal values li , while WÃ is assumed to depend on the
logarithmic stretches ej .
For the transformation of the computational tangent modulus D back to Cartesian axis D, ^ the
6
transformation matrix R from equation (8) is employed. It can be shown that the relations from
tensor calculus for the transformation of fourth-order tensors reduce to
^ ˆ RT DR
D …14†
^ and D are put in a 4  4 matrix as predicted by t^ ; see (7).
The components of D

2.3.2. Stored energy functions. The simplest strain energy function is a quadratic form of the
logarithmic stretches e and leads to the same structure of the constitutive equations as in the
in®nitesimal theory. With the bulk modulus L and shear modulus m the stored energy function
reads

^ ˆ m‰…e †2 ‡ …e †2 ‡ …e †2 Š ‡ L …e ‡ e ‡ e †2
W …15†
1 2 3
2 1 2 3

Kirchho€ stresses and the elastic tangent modulus C related to logarithmic stretches are

@W ^
tˆ ˆ L tr…ee†1 ‡ 2mee …16†
@ee
C ˆ L1
1 ‡ 2mI …17†

As pointed out by Simo,1 the function (15) is not, strictly speaking, polyconvex, but has the
correct limits when approaching extreme strains, ei ! + 1.
For a material that obeys the neo-Hookean strain energy function
m 2 2 2 L 2
W ˆ …l ‡ l2 ‡ l3 ÿ 3† ÿ m ln J ‡ …J ÿ 1 ÿ 2 ln J† …18†
2 1 4
the stresses and tangent follow again by straightforward di€erentiation with respect to the
principal stretches li .

2.4. Finite strain plasticity

In this paper the multiplicative split of the deformation gradient F into elastic Fe and plastic
contributions Fp is used:
e p
FˆFF …19†

This leads to an intermediate con®guration, whose orientation `is in principle arbitrary'

(Lubliner,9 p. 455). It is convenient to assume that all mechanical equations depend on the elastic
deformation Fe only through its principal values lei .
The consequence of the multiplicative split (19) is an additive decomposition of the
incremental logarithmic stretches; see Simo1 or Wriggers et al.:3
e p
dee ˆ dee ‡ dee …20†

# 1997 John Wiley & Sons, Ltd. COMMUN. NUMER. METH. ENGNG, VOL. 13, 963±976 (1997)
968 E. KIRCHNER, ST. REESE AND P. WRIGGERS

The additive split (20) leads to the proper description of a plastic deformation even if the
orientation of the principal systems changes remarkably during the deformation.
This paper is restricted to classical, rate-independent, plasticity models. By the restriction to
isotropy the yield criterion depends on t only through its invariants; when restricting to classical
J2-models, the yield function F is only a function of J2 ˆ 2kdev t k2.
A well-known example for the yield function F is the pressure-insensitive Mises yield function
with isotropic exponential work hardening k(x),
p q p q
1 ÿdx
F…J 2 ; x† ˆ J 2 ÿ 23k…x† ˆ J 2 ÿ 23‰sY ‡ Hx ‡ …sY ÿ sY †…1 ÿ e †Š …21†

Remark 2. Despite the restriction to the Mises yield function, the proposed algorithm is
applicable to any single-surface yield function.
Next, the evolution equation for the plastic strain is given. The increments in the plastic
logarithmic stretches dee p follow from the yield function F by partial di€erentiation:

p @F
dee ˆ Dg
@b
b

It is not necessary to specify the elastic constitutive model at this stage of the discussion. Seen
from a mechanical point of view all elastic models are treated in the same way. The equation of
evolution for the work hardening parameter x follows also from the yield function F by partial
di€erentiation:
q
dx ˆ Dg 23 …22†

A quite comprehensive discussion of the derivation of (22) is given by Simo1 and Wriggers et al.3
To motivate (22) physically one should interpret x as an equivalent plastic strain e p being work
conjugate to the equivalent stress 2kdev t k.

3. ALGORITHMIC TREATMENT OF FINITE PLASTICITY

A product algorithm is used for inelastic problems leading to an overall structure identical to the
in®nitesimal regime. The classical corrector step is extended to large strain plasticity with
isotropic hardening, and its linearization is provided. A modi®cation of the algorithmic quasi-
elastic tangent is appropriate to ful®l the plane stress condition also in the elastoplastic regime.

3.1. Operator split

Despite the non-linearities of the problem the FEM formulation remains comparatively simple.
This is achieved by the use of the principal axis; the computation time for the transformation
back to physical co-ordinates is comparatively low. A consequence of the description in
the principal axis is that the algorithmic modulus Dep
n‡1 needs only two indices with values from 1
to 3 relating Kirchho€ stresses bi and stretches ej to each other. The entry Dep n‡144 for the
shear sti€ness is given in (35).
The algorithmic treatment of ®nite strain plasticity relies on the usual operator split; see Simo.1
The asterisk * denotes trial quantities from an elastic predictor step. As usual, it is assumed that

COMMUN. NUMER. METH. ENGNG, VOL. 13, 963±976 (1997) # 1997 John Wiley & Sons, Ltd.
 FINITE ELEMENT METHOD FOR PLANE STRESS PROBLEMS 969

all necessary data for the pseudo-time instance n are known. The elastic left Cauchy±Green
tensor ben ˆ Fen …Fen †T and the hardening parameter xn were determined during the return mapping.
The determination of the trial state is based on the relative deformation gradient1,3 fn‡1 mapping
the deformation Fn on time instance n ‡ 1 by Fn‡1(X) ˆ fn‡1(X) Fn(X). The trial stresses t * follow

with the elastic left Cauchy±Green tensor ben‡1 ˆ f n‡1 ben f Tn‡1 for the trial state.
The yield (21) rule for the computation of the principal Kirchho€ stresses bi,n‡1 appears in the
usual Kuhn±Tucker form with a consistency parameter Dgn‡1 ,

Dgn‡1 5 0; F…bn‡1 ; xn‡1 † 4 0; Dgn‡1 F…bn‡1 ; xn‡1 † ˆ 0 …23†

The physical notion associated with (23) is closely connected to the small strain regime: an
incremental loading, say DPext , leads either to a purely elastic deformation increment Duh with
Fn‡1 4 0 or an elastoplastic deformation, Dgn‡1 > 0.

3.2. Return mapping algorithm in principal axis

The construction of a locally quadratically convergent return mapping is supported by the
introduction of a residual vector1

e e @F…b
bn‡1 ; xn‡1 †
r ˆ e n‡1 ÿ e n‡1 ÿ Dgn‡1 …24†
@b
b

The norm of r vanishes if the algorithmic yield rule (23) is satis®ed. It is important to keep the

trial state ben‡1 ®xed in the return map to achieve quadratic convergence. The key equation in the
return mapping is the relation for dDgn‡1,( j) . It is obtained by linearizing the yield function and
the equation for the algorithmic residual around the actual point in the current iteration ( j). Since
Fn‡1,( j) 6ˆ 0 and jr( j)j 6ˆ 0 holds, the goal is to ®nd a state with Fn‡1,( j‡1) ˆ 0 and jr( j‡1)j ˆ 0 in the
neighbourhood of state ( j). Cast into a more mathematical form, this appears as

@F @F
Fn‡1;… j ‡1† ˆ Fn‡1;… j† ‡ db
b ‡ dDgn‡1;… j† ˆ 0 …25†
b n‡1;… j† @Dg
@b
e @F @2 F
r… j ‡1† ˆ r… j† ÿ dee n‡1;… j† ÿ dDgn‡1;… j† ÿ Dgn‡1;… j† db
b ˆ0 …26†
@b
b @b
b@b b n‡1;… j†

b n‡1,( j) is obtained
b is the vector of principal Kirchho€ stresses, as before. An equation for db
from the hyperelastic constitutive law,

@2 W ^ e e
b n‡1;… j† ˆ
db dee ˆ C dee n‡1;… j† …27†
@ee @ee n‡1;… j†

Observe that the matrix C is related to the logarithmic stretches; this must be taken into account
for the neo-Hookean model discussed in Section 2.3.2, where the strain energy function was
written in terms of the (elastic) principal stretches lei .

Remark 3. In contrast to Section 3.3 the trial state given by e en‡1 is held ®xed. This is a
consequence of the assumption of a frozen plastic state when determining the trial state at the
beginning of the algorithm.

# 1997 John Wiley & Sons, Ltd. COMMUN. NUMER. METH. ENGNG, VOL. 13, 963±976 (1997)
970 E. KIRCHNER, ST. REESE AND P. WRIGGERS

When inserting (26) and (27) in (25) and resolving for dDgn‡1,( j), it proves to be reasonable to
introduce a modi®ed quasi-elastic modulus by
( )ÿ1
^ ee
@2 W…e
ÿ1 ÿ1 @2 F n‡1;… j† † @2 F
fhn‡1;… j† g ˆC ‡ Dgn‡1;… j† ˆ ‡Dgn‡1;… j† …28†
@b
b @bb @ee e @ee e @b
b@b b

It is worth mentioning that until now neither a speci®cation of the hyperelastic constitutive model
nor of the yield function has been necessary.
The last point in the derivation of the return mapping scheme is to resolve (25)±(27) for
dDgn‡1,( j) . Using the modi®ed modulus hn‡1,( j) from (28), a lengthy manipulation yields
(  T )"   #ÿ1
@F @F T @F 2 @k
dDgn‡1;… j† ˆ Fn‡1;… j† ‡ hn‡1;… j† r… j† hn‡1;… j† ‡
@b
b @b
b @b
b 3 @x

It is clear that the derivatives of the yield function F must be evaluated for the actual iteration
state b n‡1,( j) , xn‡1,( j) to obtain quadratical convergence of the return map. The update formula
for the logarithmic strains is
 
e e ÿ1 @F
e n‡1;… j ‡1† ˆ e n‡1;… j† ‡ C hn‡1;… j† r… j† ÿ dDgn‡1;… j† …29†
@b
b

The Kirchho€ stresses follow next by evaluation of the stored energy function for e en‡1;…j ‡1† . Since
hyperelasticity is assumed, the calculation of the principal Kirchho€ stresses directly from Wà is
appropriate in the return map rather than using explicit increments db b n‡1,(j) .

3.3. Consistent elastoplastic material tangent

The relation de®ning the consistent material tangent reads1

ep @b
b n‡1
Cn‡1 ˆ  …30†
@ee en‡1

This modulus is related to logarithmic stretches used in the return map from Section 3.2. The
connection to the spatial tangent used to calculate the element tangent matrices is provided by
equation (13), the derivatives in Dep
n‡1 interpreted as being taken with respect to trial increments
of the principal values of ben‡1 .
The main di€erences to Section 3.2 when linearizing and resolving for Cep n‡1 follow clearly by
inspection of (25) and (26) on the background of the product formula algorithm described in
Section 3.1. Firstly, the linearization of the yield function F leads to the usual consistency
condition dF ˆ 0 with a form as in the small strain regime,

@F @F
dF ˆ db
b ‡ dDgn‡1 ˆ 0 …31†
b n‡1 @Dg
@b

The iteration index ( j) is left aside since db

b n‡1 follows as the solution of a system of linearized
equations rather than from an iteration scheme as in Section 3.2.

COMMUN. NUMER. METH. ENGNG, VOL. 13, 963±976 (1997) # 1997 John Wiley & Sons, Ltd.
 FINITE ELEMENT METHOD FOR PLANE STRESS PROBLEMS 971

Another di€erence to the derivation of the return mapping in Section 3.2 is the incremental
formulation of the algorithmic yield rule. Assuming r0 when adding a logarithmic stretch
increment caused by an incremental deformation Duh leads to

e e @F @2 F
dee n‡1 ˆ dee n‡1 ÿ dDgn‡1 ÿ Dgn‡1 db
b …32†
@b
b @b
b @bb n‡1

Due to a change of the displacement during the global iteration, the trial state is not held ®xed.
Additional plastic deformation is permitted in the global Newton step.
b n‡1 , (27), remains unchanged. Inserting (27) and (32) in (31) yields dDgn‡1 ,
The equation for db
a function of the incremental elastic trial stretches:

 T " T #ÿ1
@F e @F @F 2 @k
dDgn‡1 ˆ hn‡1 dee n‡1 hn‡1 ‡ …33†
@b
b @b
b @b
b 3 @x

The corresponding elastoplastic tangent modulus is attained by inserting (33) into (27):
"  #ÿ1
ep @F @F @F T @F 2 @k
Cn‡1 ˆ hn‡1 ÿ hn‡1
hn‡1 hn‡1 ‡ …34†
@b
b @b
b @b
b @b
b 3 @x

The shear component Dep ep

n‡1 of the computational modulus Dn‡1 is
3
44

 

ep bi …lej †2 ÿ bj …lei †2
Dn‡1 ˆ   …35†
44
…lei †2 ÿ …lej †2

As pointed out by Simo,1 it is possible to derive a closed expression for the consistent elastoplastic
modulus Cep n‡1 with a form as in the small strain regime for plane strain and axisymmetry when
using the stored energy function from (15). This is a consequence of the constitutive model:
linearization of the constitutive equations (16) and (17) at the stress free point b ˆ 1 leads to the
classical set of small strain equations. However, for plane stress an explicit expression is not
available. For perfect plasticity (H ˆ 0 and s1Y ˆ sY † one recovers the same form of the modulus
Cep
n‡1 as derived by Simo.
1

The last step is the transformation by (14) back to the Cartesian co-ordinates of the
instantaneous con®guration; see Section 2.2.

3.4. Algorithmic plane stress conditions

For both purely elastic and elastoplastic loading the plane stress condition and its algorithmic
realization are obtained by inspection of the simple small strain regime. Following the arguments
of Gross et al.10 it is possible to treat both plane stress and plane strain with the same set of
formulas with tangent moduli appropriate to each problem. This idea is now extended to the
®nite strain elastoplastic regime, the linear elastic tangent modulus being replaced by hn‡1 .
It is clear by hn‡1 from (28) that the modi®ed material tangent used in the corrector algorithm
equals the purely elastic tangent matrix C for an incremental elastic loading. During the

# 1997 John Wiley & Sons, Ltd. COMMUN. NUMER. METH. ENGNG, VOL. 13, 963±976 (1997)
972 E. KIRCHNER, ST. REESE AND P. WRIGGERS

projection algorithm (see Section 3.2), the change in principal Kirchho€ stresses can be
computed by
 
@F
b n‡1;… j† ˆ b n‡1;… j ‡1† ÿ b n‡1;… j† ˆ hn‡1;… j† r… j† ÿ dDg
db
@tt

The incremental algorithmic expression for the stress perpendicular to the X±Y-plane is

X  
3
@F def: X3
db3;n‡1;… j† ˆ h3i;n‡1;… j† ri;… j† ÿ dDgn‡1;… j† ˆ h3i;n‡1;… j† dZi …36†
iˆ1
@ti iˆ1

and holds in this form due to the above discussion for both the purely elastic and the elastoplastic
regime. Applying the arguments of Gross et al.,10 the third member of the square brackets dZ3 in
(36) is eliminated analytically by the incremental plane stress condition db3,n‡1,( j) ˆ 0, yielding
ÿ1
dZ3 ˆ ÿ‰h33;n‡1;… j† Š ‰h13;n‡1;… j† dZ1 ‡ h23;n‡1;… j† dZ2 Š …37†

Using (37) explicitly leads to a plane stress version of the modi®ed tangent hpl:str:
n‡1;… j† ,
2 3
h11 ÿ h13 …h33 †ÿ1 h31 h12 ÿ h13 …h33 †ÿ1 h32 0
pl:str: 4
hn‡1;… j† ˆ h21 ÿ h23 …h33 †ÿ1 h31 h22 ÿ h23 …h33 †ÿ1 h32 05 …38†
0 0 0

For purely elastic behaviour governed by the quadratic stored energy function (15) the matrix
hpl:str:
n‡1;… j† from (38) is identical to the Hookean tangent modulus for plane stress.
1,10

4. NUMERICAL EXAMPLES AND CONCLUSIONS

The algorithm presented in this paper is implemented in the fully non-linear FE program FEAP
by Taylor and Zienkiewicz.11

4.1. Notched bar under uniaxial tension

This example was chosen to demonstrate the in¯uence of the hyperelastic constitutive model
on the large strain behaviour of a specimen. The magnitude of t33 was less than 10712 in all
computations. The global iteration achieved a quadratic rate of convergence. For this example an
enhanced strain formulation for the isoparametric four-node element is used; see Kirchner.12
To investigate the in¯uence of the hyperelastic law the two models from Section 2.3.2 are used.
Both reduce to the small strain Hookean material law when linearizing around b ˆ 1. The elastic
material parameters L ˆ 1731 N mm72 and m ˆ 1154 N mm72 in the strain energy functions are
taken from Simo1 coupled with a linear isotropic hardening, i.e. s1 Y ˆ sY . The specimen
thickness is 1; the geometry is shown in Figure 1.
To discuss the in¯uence of the elastic law the plasticity parameters are varied; see Table I. For
the elastic simulation I, sY and H are taken from Wriggers et al.;3 to change the order of
magnitude of the elastic deformations the magnitude of sY and H is changed. The in¯uence on
the overall behaviour is shown in Figures 2 and 3.

COMMUN. NUMER. METH. ENGNG, VOL. 13, 963±976 (1997) # 1997 John Wiley & Sons, Ltd.
 FINITE ELEMENT METHOD FOR PLANE STRESS PROBLEMS 973

Figure 1. Geometry of the notched bar

By Figures 2 and 3 it is clear that the hyperelastic constitutive model has only a small in¯uence
on the solution of this problem, both for large (simulation II) and for moderate plastic deforma-
tions (simulation III). To obtain considerable di€erences the elastic deformations must be one
order of magnitude larger than the inelastic deformations. For large plastic deformation, necking
occurs; the magnitude of the displacements di€ers by more than a factor of 2 between simulation
I (elastic) and simulations II and III (both elastoplastic). No di€erence between the necking
displacement curves of the di€erent elastic models can be observed for large deformations.
One conclusion can be drawn from this example: for metals, the use of the quadratic stored
energy function (15) is appropriate since the elastic deformations remain small. Nevertheless,
good results are obtained. The advantange of this model for 3D-, plane strain or axisymmetric
computations is that the consistent tangent is known explicitly.

4.2. Uniaxial stretching of a perforated plate

This example is chosen to show a comparison with elastoplastic plane stress computations from
the literature. Hopperstad2 applied a material model with non-linear isotropic and kinematic
hardening within a plane stress shell formulation to predict the behaviour of a perforated plate;
see Figure 4. His material data for the aluminium alloy 6060-T4 were ®tted to experiments; a
comparison between simulation and experiment proved that this rather complex model was
restricted to small strains.

Table I. Plasticity parameters

Simul. Yield stress Hardening
I sY ˆ 1 Ð
II sY ˆ 24.3 N mm72 H ˆ 300
III sY ˆ 243 N mm72 H ˆ 1500

# 1997 John Wiley & Sons, Ltd. COMMUN. NUMER. METH. ENGNG, VOL. 13, 963±976 (1997)
974 E. KIRCHNER, ST. REESE AND P. WRIGGERS

Figure 2. Maximum stress t22 against end displacement

Figure 3. Notch ground displacement at point A against total elongation

Hopperstad's material model covers 21 parameters to model monotonic and cyclic hardening
phenomena; this is not reviewed here. The elastic deformations are computed from the stored
energy function (15); the isotropic hardening follows (21). The actual parameters for the present
model are chosen as
ÿ2 ÿ2 ÿ2
L ˆ 33;084 N mm ; m ˆ 25;000 N mm ; sY ˆ 115 N mm
1 ÿ2 ÿ2
sY ˆ 300 N mm ; d ˆ 20; H ˆ 8000 N mm

COMMUN. NUMER. METH. ENGNG, VOL. 13, 963±976 (1997) # 1997 John Wiley & Sons, Ltd.
 FINITE ELEMENT METHOD FOR PLANE STRESS PROBLEMS 975

Figure 4. Geometry of the perforated plate problem

Figure 5. Comparison with load±displacement curves obtained by Hopperstad

To account for the ¯exibility of Hopperstad's testing machine, the tested specimen is considered
as a non-linear spring in series with a linear one representing the compliance of the testing
machine.
For the load±displacement curve (see Figure 5) remarkable agreement is achieved. The results
match reasonably well in the range of interest. When using a real parameter identi®cation it
should be possible to achieve even better results than with the experimental data.

4.3. Conclusions
Despite its simplicity the present approach to large strain elastoplasticity works well and is
capable of satisfying the plane stress condition almost exactly. It seems to be possible to extend

# 1997 John Wiley & Sons, Ltd. COMMUN. NUMER. METH. ENGNG, VOL. 13, 963±976 (1997)
976 E. KIRCHNER, ST. REESE AND P. WRIGGERS

this method to complex models suitable for modelling cyclic plasticity. Since hyperelasticity is
assumed during the derivation of the stress update algorithm and the corresponding tangent
operator, it can be applied to any hyperelastic model. Furthermore, extensions to non-associative
plasticity are possible.
We conclude that the present approach is an ecient tool for the solution of plane stress
problems for elastoplastic material behaviour. Only a few simplifying assumptions were made
during the derivation of the present algorithm; nevertheless, it is able to reproduce the behaviour
of real structures with remarkable agreement.

REFERENCES

1. J. C. Simo, `Algorithms for static and dynamic multiplicative plasticity that preserve the classical return
mapping schemes of the in®nitesimal theory', Comput. Methods Appl. Mech. Eng., 99, 61±112 (1992).
2. O. S. Hopperstad, `Modelling of cyclic plasticity with application to steel and aluminium structures',
Technical Report 1993 : 7, Department of Structural Engineering, The Norwegian Institute of
Technology, Trondheim, 1993.
3. P. Wriggers, R. Eberlein and S. Reese, `Comparison between shells and 3D-elements in plasticity',
to appear in Int. J. Solids Struct.
4. J. C. Simo and R. L. Taylor, `A return mapping algorithm for plane stress elastoplasticity', Int. j. numer.
methods eng., 22, 649±670 (1986).
5. J. C. Simo and R. L. Taylor, `Consistent tangent operators for rate-independent elasto-plasticity',
Comput. Methods Appl. Mech. Eng., 48, 101±118 (1985).
6. S. Reese, Theorie und Numerik des StabilitaÈtsverhaltens hyperelastischer FestkoÈrper, Dr.-Ing.
Dissertation, Technical University Darmstadt, 1995.
7. J. C. Simo and R. L. Taylor, `Quasi-incompressible ®nite elasticity in principal stretches. Continuum
basis and numerical algorithms', Comput. Methods Appl. Mech. Eng., 85, 273±310 (1991).
8. F. Gruttmann and R. L. Taylor, `Theory and ®nite element formulation of rubberlike membrane
shells using principal stretches', Int. j. numer. methods eng., 35, 1111±1126 (1992).
9. J. Lubliner, Plasticity Theory, Macmillan, New York, 1990.
10. D. Gross, W. Hauger, W. Schnell and P. Wriggers, Technische Mechanik 4, Springer, Berlin, 1993.
11. O. C. Zienkiewicz and R. L. Taylor, The Finite Element Method, Vol. 1, 4th Edn, McGraw Hill,
London, 1989.
12. E. Kirchner, Finite Elementformulierungen fuÈr elasto-plastische Probleme mit groûen Deformationen,
Diploma thesis, Institut fuÈr Mechanik IV, Technical University Darmstadt, 1995.

COMMUN. NUMER. METH. ENGNG, VOL. 13, 963±976 (1997) # 1997 John Wiley & Sons, Ltd.