INTERNATIONAL JOURNAL FOR NUMERICAL AND ANALYTICAL METHODS IN GEOMECHANICS VOL.

14, 35 1-366 (1990)

FINITE ELEMENT ANALYSIS O F TWO CYLINDRICAL

EXPANSION PROBLEMS INVOLVING NEARLY
INCOMPRESSIBLE MATERIAL BEHAVIOUR

H. J. BURD A N D G. T. HOULSBY
Department of Engineering Science. Oxford University. Parks Road, Oxford, OX1 3PJ. U . K .

SUMMARY
The displacement formulation of the finite element method is well suited to the analysis of elasto-plasticity
problems involving compressible material behaviour, but it is well known that numerical difficulties occur
when the material is incompressible or nearly incompressible. The effect of these additional constraints
depends on both element formulation and mesh topology. A two-dimensional plane strain finite element
formulation suitable for the solution of problems involving large strains and displacements (but small
rotations) based on the isoparametric approach is described. The kinematics of deformation are defined in
terms of the Eulerian strain rates that are invariably used in small strain analysis; the formulation therefore
retains some of the character of small strain theory but includes additional geometrically non-linear terms.
The results of a series of plane strain finite element analyses of two cylindrical expansion problems are
presented. These results confirm the previously observed trend that as Poisson’s ratio approaches 0 5 then
the quality of the calculated stress deteriorates. The study also indicates that the solution quality depends
increasingly on mesh topology as perfect incompressibility is reached.

1. INTRODUCTION
It is well known that severe numerical difficulties arise when using the conventional displacement
method to analyse incompressible, or nearly incompressible, materials. These difficulties are
caused by the additional kinematic constraints imposed on the nodal velocities by the incom-
pressibility condition which reduces the number of free degrees-of-freedom in the finite element
mesh. The effect of incompressibility constraints on the accuracy of the solution depends on the
type of analysis being performed. In a collapse load calculation the effect of these additional
kinematic constraints is to produce an over-stiff response; in these circumstances the finite
element solution often overestimates the exact collapse load by an unacceptably large amount.’ -’
An alternative type of problem involves the calculation of the stress fields set up within a structure
when it is loaded. In finite element computations of this sort, the effect of incompressibility
constraints is to produce spurious variations of stress across each element. As perfect incom-
pressibility is reached then these variations can become significant and lead to considerable
scatter in the results. This effect has been illustrated by Naylor4 who performed a series of
numerical experiments involving small plane strain analyses of a thick cylinder under internal
pressure. The results obtained from this study also indicate that as perfect incompressibility is
approached then the errors in the stresses are mainly confined to the mean normal stresses, with
the deviatoric stresses remaining relatively unaffected.
The study of problems involving incompressible material behaviour is of considerable import-
ance in geomechanics where several of the currrently used constitutive models of behaviour
0363-9061/90/05035 1-16SO8.00 Received 9 August 1985
0 1990 by John Wiley & Sons, Ltd. Revised 7 April 1989
352 H. J. BURD AND G . T. HOULSBY

require that pointwise incompressibility is enforced in the plastic regime. A considerable amount
of attention has therefore been paid in the past to developing special finite element formulations
in which the detrimental effect of the pointwise incompressibility constraint is reduced. One
possible approach is to retain the basic features of the displacement method but to lessen the
constraining effect of the incompressibility condition by reducing the order of quadrature used to
evaluate the element stiffness matrices; this is the so-called 'reduced-integration' technique.' The
effectiveness of using the eight-noded quadrilateral element with a four Gauss point integration
rule is often cited as an example of this advantages of this A number of researchers
have investigated a variant of this approach known as 'selective-integration' in which different
integration rules are used for the volumetric and deviatoric strain terms in the element stiffness
matrix. 1 * 7 * 8 The procedure generally adopted in these cases is to use reduced-integration for the
volumetric terms but retain a high order of integration for the deviatoric components. Since the
detrimental effects of the incompressibility constraints are confined mainly to the volumetric
terms, this approach allows the incompressibility condition to be relaxed without sacrificing too
much accuracy in the calculation of the deviatoric strains and stresses. Marti and Cundall'
suggest that the performance of displacement elements may also be improved by using a 'mixed
discretization procedure', in which different geometrical discretization methods are used to
approximate the deviatoric and volumetric strains.
An alternative modification to the standard displacement formulation which has been shown
to reduce the effect of incompressibility constraints is to use a formulation in which the primary
variables are not all of the same type. These formulations are generally known as 'mixed
methods'. An early example of this type of approach is described by Herrmann" who describes a
variational formulation in which mean pressure is taken as an additional independent variable;
this formulation is discussed further by Nagtegaal et al.' The hybrid method is an alternative
form of mixed method in which both stresses and displacements are used as primary
It has been shown that variants of the displacement method based on reduced-
integration or mixed methods are essentially e q ~ i v a l e n t . ' ~ -A' ~good review of the application of
mixed methods to incompressible analysis is given by Zienkiewicz el al.'
An alternative philosophy for the application of the displacement method to problems
involving material incompressibility is described by Sloan and Randolph' who show that as the
order of polynomial used to describe the strains within an element is increased then the rate of
increase of the number of incompressibility constraints is less than the increase in the degrees-of-
freedom introduced by the additional nodes. If the total number of degrees-of-freedom in a finite
element mesh is held constant, therefore, then the total number of constraints may be reduced by
using higher order elements. A satisfactory finite element solution should, therefore, be possible
without having to resort to the use of reduced-integration. This hypothesis was confirmed by
Sloan3 who performed a series of small strain collapse load calculations using 15-noded
triangular elements.
It has been shown that for the case of the constant strain triangle the incompressibility
condition imposes a single constraint on the nodal velocities of each element.' If four constant
strain triangles arc coalesced to form a quadrilateral, however, then provided that the interior
node is situated at the intersection of the two diagonals then the four constraint equations are
linearly dependent and the total number of independent constraints is three.'"' In a finite
element calculation involving incompressible material behaviour, therefore, it would be advant-
ageous to use these quadrilateral 'super-elements' if the discretization is based on the use of
constant strain triangles. Mercier' and Argyris et a1.20 suggest that a similar geometric
arrangement of linear strain triangles also results in linear dependencies between the constraint
equations of the individual elements, but that the improvement is not so marked as with constant
strain elements.
 INCOMPRESSIBLE MATERIAL 353

In order to investigate the effect of variations in mesh topology on the accuracy of stresses
calculated using the displacement finite element method for the case of nearly incompressible
material behaviour, two sets of numerical experiments have been performed and the results
presented in this paper. These experiments are limited to the case of plane strain problems
involving radial displacements only; the first set relates to small displacement elastic analysis of a
thick cylinder, and the second set consists of large displacement elasto-plastic cavity expansion
analysis. The finite element formulation used in this study is described in a form suitable for the
analysis of problems in which displacements and strains are large but rotations are small; an
extended form of the formulation which is suitable for the analysis of finite material rotation is
given by Burd.21The formulation described in this paper is based on the isoparametric approach
and is presented in a general form. The calculations described, however, were all performed using
six-noded triangular elements with three Gauss points (except where noted othewise), and are
based on either elastic or elastic perfectly-plastic material behaviour.

2. FINITE ELEMENT EQUATIONS

2.1. Strain and stress definitions

In the infinitesimal theory of elastic-plastic deformations it is possible to define strain in a
unique and unambiguous way. This is not the case in large displacement theory, however, since
different co-ordinate systems may be used to describe the kinematics of deformation and these
give rise to different strain measure^.^^"^ The most usual approaches are either to use a
formulation based on Green-Lagrange strain, with an appropriate stress measure, or to use an
Eulerian approach in which the constitutive equation relates Cauchy stress rate to Eulerian strain
rate. In this formulation, the latter approach is adopted and the Cauchy stresses at each element
Gauss point are accumulated as the calculation proceeds. In an Eulerian approach, the kin-
ematics of deformation may be described by strain rates which are defined in the same way as for
infinitesimal theory; this type of formulation, therefore, has the desirable feature that some aspects
of the character of the small strain theory are preserved for the case where the total deformation is
finite.
Three separate strain rates, familiar from small strain theory, are defined:
au
Ex, =-
ax
av
E,, = -
aY
. aU av
y =-+- (3)
xy ay ax
where x and y indicate the Cartesian co-ordinates of a material point and velocities in the x and y
directions are given by u, v. The superior dot indicates the time derivative.

2.2. Element formulation

If X is a vector of nodal velocities for a single element, then the strain rate vector .is
iobtained
from an equation of the form
L. = [ E ] X (4)
where the [ E l matrix is usually written in terms of reference co-ordinates, which are related to the
global co-ordinates (i.e. the co-ordinates used to describe the geometry of the problem) by an
354 H.J. BURD AND G.T.HOULSBY

isoparametric mapping. The principle of virtual work applied to a single element may be used to
derive an equation for the nodal forces P in equilibrium with the stresses n:
r r
P = J J E [B]' n det [J] da d j

where a, /?are the reference co-ordinates, the Jacobian matrix of the transformation from global
to reference co-ordinates is [J] and the integration is performed over the area of the element.
Equation (5) may be differentiated to give the rate equation

P= [IE ([B]' U det [J] + [ [BIT'det [J]]u)da dj (6)

Note that in this equation the superior dot associated with the second term in the integrand
denotes the time derivative of the complete contents of the bracket. Equation (6)may be rewritten

P= IjE( [BIT [D] [B] det [J] + [C] )X do! d j (7)

where the material stiffness matrix [D] relates stress rate to strain rate

and the matrix [ C ] is defined by

[C]X = [ [B]" det [J]]u (9)
The incremental element stiffness matrix is thus

The stiffness matrix, therefore, contains the conventional small displacement terms but
additional terms are present, in the form of the [C] (distortion rate) matrix, that account for the
rate of change of element geometry. The distortion rate matrix contains terms which are of the
order of the stresses and these are insignificant when the magnitudes of the stresses are much less
than the material moduli (i.e. when strains are small). As the strains increase, these terms become
increasingly important and need to be included in the evaluation of the element stiffness matrix.
The distortion rate matrix must be determined by expanding equation (9) on a term-by-term
basis; the resulting matrix for the case of the six-noded triangle is given by Burd." The distortion
rate matrix is generally non-symmetrical and introduces geometric non-linearities into the
equations; these factors need to be considered in the selection of a suitable algorithm to invert the
global stiffness equations. The material stiffness matrix is derived from the constitutive model
used to describe the soil; a derivation of this matrix for the case of an elastic perfectly-plastic
material where plastic behaviour is defined by a von Mises model is given in Appendix I.
In these calculations, the finite element equations are solved using a modified Euler procedure3
which is illustrated in Figure 1 for the case of a single degree-of-freedom system. At the start of
the ith load step, the individual element stiffness matrices are assembled and inverted to give a set
of incremental nodal displacements Aai corresponding to an increment of load AFi. The next
stage is to calculate a set of stress increments Au at each Gauss point, corresponding to the set of
incremental nodal displacements ASi. The stress increments are obtained by performing an
integration of the form
An = [+M [DlEdt (1 1)
 INCOMPRESSIBLE MATERIAL 355

Force A

Figure 1. Modified Euler solution scheme

where 1. is a strain rate vector which, when integrated over the time step, is compatible with the
incremental nodal displacements Adi. The increment begins at time t , and ends at time t + At. In
order to proceed with this calculation it is necessary to find a set of strain rates L. which, when
integrated over the time step, are compatible with the incremental nodal displacements. Since the
basis of the finite element method is to calculate values of the primary variables at discrete times
during the loading, no information is available regarding the precise way that the nodal
displacements vary during the calculation step. In order to obtain a set of consistent strain rates,
therefore, it is necessary to make an arbitrary assumption about the nature of the variation of
either the velocities or the strain rates occurring during the calculation step. In this formulation
the assumption is made that the strain rates remain constant. A closed-form solution may be
obtained relating the (constant) strain rates to the incremental nodal displacements; these strain
rates are then used as the basis of a calculation in which the Gauss point stresses are updated.
The assumption of constant strain rates clearly also implies that the velocity gradients (&/ax
etc.) remain constant during the time step at each Gauss point. In the region of the Gauss point,
the strain rates may be assumed to be spatially constant, in which case equations (1) and (2) are be
integrated to give

u = uo +( g ) x +G>Y

u = 00 + ( $ ) Y + (:)x

where uo and uo are constants, and the higher order terms due to the spatial variation of velocity
gradients are neglected.
Given that
u = i and u = j (14)
then equations (12) and (13) may be rearranged to give a differential equation in x:

(- -)i (--
x - au + av
ax ay
ayax
au - au - --
au
a x a yu ) x - (5). + (
$)u. =0
356 H. J. BURD A N D G.T.HOULSBY

A similar differential equation may be written in y. These equations may be solved to give a set of
constant strain rates which are consistent with the incremental nodal displacements; these
solutions are given in Appendix 11. When the compatible strain rates have been determined, the
Gauss point stresses may be updated by integrating equation (8). A closed-form solution exists for
the integral in some simple cases (i.e. elastic constitutive behaviour*', or von Mises pla~ticity,'~
but otherwise a numerical method must be used. The final step in the solution increment is to
calculate the nodal forces for each element, P,in equilibrium with the updated stresses at the new
geometry from equation ( 5 ) (these forces are represented by Pi on Figure 1). A set of unbalanced
nodal forces (Rion Figure 1) are then obtained which are negated by applying then in the
opposite sense during the subsequent load step.

3. A SMALL STRAIN THICK CYLINDER PROBLEM

In order to investigate the effect of mesh topology and Poisson's ratio on the quality of Gauss
point stresses calculated using this displacement formulation, a series of small strain analyses
have been performed on a thick, plane strain cylinder. This problem is radially symmetric but it is
solved here as a plane strain problem in order to give insight into the behaviour of two-
dimensional problems involving nearly incompressible material behaviour.
Four meshes have been used in this study and these are shown in Figure 2. In each case the
number of mesh degrees-of-freedom (defined as twice the number of nodes minus the number of
externally applied degree-of-freedom restraints necessary to define the boundary conditions) is

M-h 1 : 48 Mcsh 0.0.F. Mesh 2: %Mesh D0.F.

Mesh 3: 66 Mesh n0.E Mesh 4: 500 Mesh D.0.E

Figure 2. Thick cylinder finite element meshes

 INCOMPRESSIBLE MATERIAL 357

indicated. In the finite element calculations a radial displacement of 0.03 per cent is prescribed to
the inner boundary; this is sufficiently small for geometrical non-linearities to be absent. The
material is taken to be linearly elastic. In order to investigate the effect of incompressibility
constraints on the accuracy of the stresses obtained by the finite element calculations, a series of
analyses were performed for values of Poisson’s ratio of up to 0.4999; it was found that values of
Poisson’s ratio closer to 0.5 than this gave rise to substantial round-off errors in the numerical
computation. The quality of the principal stresses obtained from the finite element calculations is
characterized in terms of the r.m.s. error:

r.m.s. error = -
n
where a’@) is a Gauss point stress from the finite element calculation, r is the radius of the Gauss
point, dr)is the exact value of the stress at radius r calculated from elasticity theory based on the
value of Poisson’s ratio used in the numerical computation, and the total number of Gauss points
is n. The errors in the radial stresses are plotted in Figure 3; the errors in the other two principal
stresses are not plotted since they show identical trends. The results of this numerical experiment
have several interesting features. Firstly, in all cases the r.m.s. error increases as Poisson’s ratio
approaches 0.5 (i.e. 6 K I E approaches infinity where K and E are the bulk and Young’s moduli
respectively). These errors arise because the incompressibility condition imposes additional
constraints on the nodal velocities, thus reducing the number of ‘free’ degrees-of-freedom in the
mesh. A further feature of the results is that Mesh 1 performs rather better than Mesh 2 and
Mesh 3 in spite of the fact that it has a smaller number of mesh degrees-of-freedom. It
would appear, therefore, that the special arrangement of elements used in Mesh 1 results in a
considerable improvement in the performance of the mesh. The r.m.s. errors for the deviatoric
radial stresses are plotted in Figure 4 and the errors for the mean normal stresses plotted in
Figure 5. It is clear that the errors are mainly associated with the mean normal stresses, the

rms
error 1
+-+ Mesh 1
+-+ Mesh 2
+----+ M-h 3
+----+ Mesh4

/
t /

t
0.3

0.1
Oe2

Figure 3. Errors in calculated radial stresses

358 H. 1. BURD A N D G . T. HOULSBY

r ms
error
0 3-
+....... + Mesh 1
+-+ Mesh 2
+--t Mesh 3

002 - +--4 Mesh C

- /.”-+ .-.---+

0.01 -

$..- ---.....-.c---.
............ -* ................ +
I
&

+ -
_... ...-..+ -...- ._
__._
.+ --.. -.-. -.-+I
I

1 2 3 C l%ln(y)

Figure 4. Errors in calculated deviatoric radial stresses

rms
error
A
06 -
+....-4 Mcsh 1
0.5 - +-+ Mesh 2
+--+ Mcsh 3

01 -
0.3 -
0.2 -
0.1 -

Figure 5. Errors in calculated mean normal stresses

deviatoric components remaining relatively unaffected. This feature of the results of incom-
pressible analysis was previously reported by Naylor4 who performed a similar study of this
problem based on the use of eight-noded quadrilateral elements.
In these computations, the stresses are calculated at each of the three Gauss points in each
element. The quality of the mean normal stress results may be improved in some cases if the
computed values are averaged over the element and the resulting value taken to be the mean
normal stress acting at the element centroid. The r.m.s errors in the mean normal stresses
 INCOMPRESSIBLE MATERIAL 359

ura 1
a28 -
0.26 -
0.24 - +....--+ Mesh 1

0.22 - +-+ Mesh 2

0.20 - +-.- + Mesh 3

0.18 - +--- + Mesh 4 / #

Figure 6. Errors in calculated averaged mean normal stresses

calculated in this way are plotted in Figure 6 and it is clear that this procedure results in a
substantial reduction in the magnitudes of the errors in the range 6 K / E < 10oO. This improved
accuracy is at the expense, of course, of reducing the number of material points at which values of
stress are obtained. This data smoothing procedure is equivalent to the use of an integration
method in which a single Gauss point is used to sample the stresses and calculate the element
stiffness matrices. The errors in the averaged mean normal stresses increase rapidly as 6 K / E
exceeds loo0 and for meshes 1 and 4 the r.m.s. errors for 6 K / E = lo4 exceed the errors at the same
value of Poisson’s ratio for the mean normal stresses at each individual Gauss point. It is thought
that this sharp increase in r.m.s. error is caused by round-off errors in the computational
procedure rather than the effect of incompressibility constraints.

4. A LARGE STRAIN CAVITY EXPANSION PROBLEM

A large strain cavity expansion problem has been studied as a further investigation into the use of
the displacement method in the analysis of incompressible materials. In this calculation the
material is assumed elastic perfectly-plastic; plastic behaviour is defined by a fully associated von
Mises model which implies that plastic volumetric strains are zero.
The analysis of plane strain cavity expansion is a rare example of a large strain elasto-plastic
problem with a known closed-form solution for the stresses. This problem has been studied
theoretically by Gibson and Anderson” for an elastic-perfectly plastic Tresca material (with
small strain in the elastic region) and by SagasetaZ6for an elastic-perfectly plastic von Mises
material using large strain analysis throughout. This latter solution, which is summarized in
Appendix 111, is used here to check the numerical results. The Sagaseta solution is for a material
that is incompressible in the elastic regime, and this feature is approximated in the finite element
calculations by using a Poisson’s ratio of 0.49.
360 H.I. BURD AND G. T.HOULSBY

In order to optimize the accuracy of the finite element calculations it is desirable to avoid as far
as possible the detrimental effects of the additional kinematic constraints imposed by the
incompressibility condition. With this in mind, the mesh is based on the special arrangement of
elements that was found to be successful for the small strain radial expansion problem described
in Section 3; the mesh is plotted in Figure 7. Since the Sagaseta cavity expansion analysis is based
on a continuum of infinite extent it is necessary to add a correcting layer to the perimeter of the
mesh to simulate an infinite boundary. The outer edge of this boundary is fixed and the layer has
Poisson's ratio, v', and Young's modulus, E', given by
5E
v'=O.25 and E l = -
12
~-
(17)
where E is the Young's modulus of the inner mesh and the outer radius of the layer is arbitrarily
chosen to be twice the inner radius. Details of the calculation of the material properties for the
correcting layer are given in Appendix IV.
In the finite element calculation an expansion ratio of four is applied to the radius of the cavity
and the calculation performed in 400 displacement increments. The ratio (G/c),where c is the
shear strength in triaxial compression, is taken to be 112. The numerical results are plotted in
Figure 8, where r is the radius of a material point and a is the final radius of the cavity, and these
show good agreement with the Sagaseta solution. Some scatter of the data is apparent near the
inner boundary (a/r = 1). This may be due to the fact that the displacements of all of the nodes on
the inner boundary are prescribed with the result that this portion of the mesh is heavily
constrained.
A repeat of this finite element calculation in which the mesh was based on an arbitrary
arrangement of triangles resulted in a large scatter in the results. This scatter could be consider-
ably reduced by using a selective integration procedure in which three Gauss points are used to

layer

Figure 7. Cavity expansion finite element mesh

 INCOMPRESSIBLE MATERIAL 361

Figure 8. Results of cavity expansion analysis

integrate the deviatoric components and a single Gauss point for the volumetric terms in the
element stiffness matrices.

5. CONCLUSIONS
Two cylindrical expansion problems have been studied in order to investigate the effect of mesh
topology and material incompressibility on the quality of stresses calculated using the displace-
ment finite element method. The results of the small strain thick cylinder calculations suggest that
for low values of Poisson’s ratio, variations in mesh topology are relative unimportant, whereas
as perfect incompressibility is reached a small change in mesh topology can have a significantly
detrimental effect on the quality of the results. This feature of the finite element displacement
method suggests that great care should be exercised in the choice of finite element mesh when the
material behaviour is incompressible or nearly incompressible.
It has been shown by other workers that if four constant strain triangles are coalesced to form a
quadrilateral ‘super-element’ then the total number of independent constraints is one less than the
sum of the number of constraint equations for each triangular element. If this special arrangement
of elements is used in an analysis involving incompressible material behaviour then it would
generally be expected that the quality of the calculated stresses would be greater than if a mesh
with the same number of elements arranged in an arbitrary fashion were used. It is usual,
however, to use elements with higher orders of polynomial strain expansions and it is interesting
to speculate whether similar special arrangements of elements exist for these higher order
elements. The results of the thick cylinder calculations suggest that, for problems involving radial
expansion at least, the accuracy of the calculated stresses is greater when the mesh consists of
quadrilateral ‘super-elements’ than if an arbitrary arrangement of elements is used. The possible
advantages of coalescing linear strain triangles in this way has been mentioned briefly by other
authors but a full investigation into the possible advantages of using special arrangements of
elements in this way has yet to be carried out.
362 H. J. B U R D A N D G. T. HOULSBY

APPENDIX I: MATERIAL STIFFNESS MATRIX FOR AN

ELASTIC PERFECTLY- PLASTIC VON MISES MATERIAL
The Eulerian strain rate vector E and the Cauchy stress rate vector b are defined

The total strain rate may be decomposed into elastic and plastic components:
E = '.2 + L'P (19)
where E' is the elastic component, and EP the plastic component. The stress rate is related to the
elastic strain rate by
b = [D]'E' (20)
where [D]' is the elastic material stiffness matrix
2G
K-- K-- 0
3 3
4G 2G
K-- K+- K-- 0
[D]' = 3 3
L K i F K - - 2G
3
0 O G
The plastic strain rates are derived from the plastic potential g(a):

where iis a scalar multiplier.

Iff(a) = 0 is the yield function then

[El f i = o
T

Substituting equations (19), (20) and (22) into the above expression gives

[ ]'[ [D]' ( E - I.:)] =0

and

Equations (19), (20) and (22) may be used to derive a relationship between the stress rate and the
total strain rate:
U = [[D]' + [DlP]E
= [DIE (27)
 INCOMPRESSIBLE MATERIAL 363

where [D] is the material stiffness matrix. If the Gauss point stress lies on the yield surface, then
the plastic stiffness matrix, [DIP, is given by

[DIP =
- CDI' (2)(gy CDI'

If the stress point lies inside the yield surface then [DIP is set to zero on the basis that the plastic
strain rates are zero.
The von Mises yield function is defined as
f(~=
) (ayy- o,,)' + (oXx- o,,)' + (oXx- oYy)'+ 6 7 r y - 8 ~ ' (29)
where c is the shear strength of the material in triaxial compression. The plastic strain rates are
derived from an associated flow rule. In equation (28), therefore, g(u) =flu) and the plastic
material stiffness matrix is given by

where the primes denote deviatoric stresses.

APPENDIX 11: CALCULATION OF STRAIN RATES FOR

A FINITE DISPLACEMENT INCREMENT
Let x', y' be the Cartesian co-ordinates of a point at the start of a time step, and x' y' be the
+

co-ordinates of the point at the end. The Jacobian matrix of the transformation of the co-
ordinates during the time step is

If the length of the time step is normalized to unity, then the strain rates, which if held constant
over the time interval produce the prescribed deformation, are
Ex, - Eyy = F ( D x x - Dyy)
Exx + Eyy = log,(det[J])
Yxy = F(Dxy+4 x 1
The factor F depends on the value of the discriminant D where
D = ( D x x - DYY)'+ 4DxyDyx
364 H. 1. BURD A N D G . T. HOULSBY

If D > 0 then

(Note, no solution exists for the case D > (Dxx+ Dvv)zin which case the determinant of the
Jacobian is negative.)
If D = 0 then

If D < 0 then

APPENDIX 111: SERIES SOLUTION FOR LARGE DISPLACEMENT

ELASTO-PLASTIC PLANE STRAIN CAVITY EXPANSION
In the elastic region the principal stresses are

where
G
f, =-
C

az - a;
‘12 = ~

r2

A(q) = ‘1’
v4 + -
+- v8 + . . .
v6 + -
4 9 16
G is the shear modulus, c is the triaxial compression shear strength, a, is the initial radius of the
cavity, a is the final radius of the cavity and r is radius corresponding to the stresses a,,, uzz
and age.
The position of the elasto-plastic boundary is given by

The stresses in the plastic zone are

u,, = p + 2c
J3
azz = P
2c
beg =p --
J3
 INCOMPRESSIBLE MATERIAL 365
where

APPENDIX IV: CORRECTING LAYER FOR THE CAVITY EXPANSION PROBLEM

Consider a cavity of radius r in a continuum of infinite extent of Young’s modulus E and
Poisson’s ratio v. If an internal pressure, p, is applied, then for small elastic deformations under
plane strain conditions, the radial displacement is

6, =
pr(1 + v)
E
For an outer correcting layer of internal radius r and outer radius Kr, whose outer edge is fixed,
made from a material of Young’s modulus E’ and Poisson’s ratio v’, then the radial displacement
corresponding to an internal pressure p (again assuming small elastic plane strain deformations)
is
pr(1 - 2v’)( 1 + v’) (K’ - 1 )
6, =
E‘(1 + K ’ ( 1 - 2v’))
For the finite layer to be of equivalent stiffness to the infinite continuum,

This equation clearly does not give a unique set of values of E’ and v’. For reasons of
computational accuracy it is desirable to choose a low value of Poisson’s ratio; thus suitable
material properties for the correcting layer are
5E
E’ = - V‘ = 0-25
12

REFERENCES
I. J. C. Nagtegaal, D. M. Parks and J. R. Rice, ‘On numerically accurate finite element solutions in the fully plastic
range’. Comp. Meth. Appl. Mech. Engng, 4, 153-177 (1974).
2. S.W. Sloan and M. F. Randolph, ‘Numerical prediction of collapse loads using finite element methods’, Int. j . numer.
analy. methods geomech., 6, 47-16 (1982).
3. S. W. Sloan, ‘Numerical analysis of incompressible and plastic solids using finite elements’, Ph.D. Thesis, University of
Cambridge, U.K., 1981.
4. D. J. Naylor, ‘Stresses in nearly incompressible materials by finite elements with application to the calculation of
excess pore pressures’, Int. j . numer. methods eng., 8, 443460 (1974).
5. 0. C. Zienkiewin. The Finite Element Method, 3rd edn, McGraw-Hill, London, 1977.
6. J. H. Prevost and T. J. R. Hughes, ‘Finite-element solution of elastic-plastic boundary-value problems’, J . Appl. Mech.,
48, 69-74 (1981).
7. D. S . Malkus, ‘A finite element displacement model valid for any value of the compressibility’,Int. J. Solids. Struct., 12,
731-738 (1976).
8. E. Hinton, E. M. Salonen and N. Bicanic,‘A study of locking phenomena in isoparametricelements’, The Mathematics
of Finite Elements and Applications I I I . Ed. J. R. Whiteman, 437-447, Academic Press, New York, 1979.
9. J. Marti and P. Cundall, ‘Mixed discretization procedure for accurate modelling of plastic collapse’, Int. j . numer.
analy. methods geomech., 6, 129-139 (1982).
10. L. R. Herrmann. ‘Elasticity equation for incompressible and nearly incompressible materials by a variational
theorem’, A I A A J., 3, 18961900 (1965).
11. T. H. H. Pian, ‘Hybrid elements’ in Numerical and Computer Methods in Structural Mechanics, eds S. H . Fenves,
N. Perrone, A. R. Robinson and W. C. Schnobrich. Academic Press, New York, 1973, pp. 59-78.
366 H. J. BURD AND G. f.HOULSBY

12. T. H. H. Pian and S. W. Lee, ‘Notes on finite elements for nearly incompressible materials’, AIAA J., 14, 824-826
(1976).
13. R. L. Spilker, ‘Improved hybrid-stress axisymmetric elements including bchaviour for nearly incompressible mater-
ials’, Int. j . numer. methods eng., 17, 483-501 (1981).
14. T. J. R. Hughes, ‘Equivalence of finite elements for nearly incompressible elasticity’, J. Appl. Mech. 44, 181-183 (1977).
15. D. S. Malkus and T. J. R. Hughes,‘Mixed finite element methods, reduced and selective integration techniques: a
unification of concepts’, Comp. Merh. Appl. Mech. Engng, 15, 63-81 (1978).
16. H. Shimodaira, ‘Equivalence between mixed models and displacement models using reduced integration’, Inr.j . numer.
methods eng., 21, 89-104 (1985).
17. 0.C. Zienkiewicz, R. L. Taylor and J. M. W. Baynam,‘Mixed and irreducible formulations in finite element analysis’,
Chapter 21 in Hybrid and Mixed Finite Element Methods, eds S. N. Atluri, R. H. Gallagher and 0. C. Zienkiewicr,
Wiley, Chichester 1983.
18. J. R. Rice, R. M. McMeeking, D. M. Parks and E. P. Sorensen, ‘Recent finite element studies in plasticity and fracture
mechanics’, Comp. Meth. Appl. Mech. Engng, 17/18, 41 1-442 (1979).
19. B. Mercier. ‘A conforming finite element method for twodimensional incompressible elasticity’, Inr. j . numer. methods
eng., 14, 942-945 (1979).
20. J. H. Argyris, P. C. Dunne, T. Angelopoulos and B. Bichat, ‘Large natural strains and some special difficulties due to
non-linearity and incompressibility in finite elements’, Comp. Meth. Appl. Mech. Engng, 4, 219-278 (1974).
21. H. J. Burd, ‘A large displacement finite element analysis of a reinforced unpaved road‘, D.Phil. Thesis, University of
Oxford, U.K.,1986.
22. K. J. Bathe, E. Ramm and E.L. Wilson, ‘Finite element formulations for large deformation analysis’, Inr. j . numer.
methods eng., 9, 353-386 (1975).
23. M. S. Gadala, M.A. Dokainish and G. AE. Oravas, ‘Formulation methods of geometric and material nonlinearity
problems’, Int. j . numer. merhods eng., 20, 887-914 (1984).
24. J. R. Booker, Personal Communication, 1984.
25. R. E. Gibson and W. F. Anderson, ‘In situ measurement of soil properties with the pressuremeter’, Civil Engineering
and Public Works Review, 56, 6 1 5 4 1 8 (1961).
26. S. Sagascta, Personal Communication 1984.