INTERNATIONAL JOURNAL FOR NUMERICAL AND ANALYTICAL METHODS IN GEOMECHANICS, VOL.

18, 689-708 (1994)

ZERO THICKNESS INTERFACE ELEMENTS-NUMERICAL

STABILITY AND APPLICATION
R. A. DAY
PO Box 604,lndooroopilly, Brisbane 4068,Australia
D.M.POTTS
Department of Civil Engineering, Imperial College of Science Technology and Medicine, Imperial College road, London
SW7 2BU. U.K.

SUMMARY
Many methods have been proposed to model joints in rocks or the interface between soil and a structure.
Many analysts have reported numerical problems when using zero thickness interface elements while others
have presented satisfactory results without comment of such difficulties. The numerical behaviour of zero
thickness interface elements is further investigated in this paper. Some simple examples illustrate the
application of interface elements to practical situations and highlight the numerical difficulties that may be
encountered. Both ill-conditioning of the stiffness matrix and high stress gradients were found to cause
numerical instability. Ill-conditioning can be reduced by careful selection of the size of the 2D elements
adjacent to the interface. The problem of steep stress gradients is entirely one of inadequate mesh design.
Contrary to other reports, this paper shows that the Newton-Cotes integration scheme has no benefit over
Gaussian integration.
Analyses of a retaining wall using interface elements confirm the analytical values of active and passive
earth pressure coefficients which are commonly used in analysis and design of retaining walls.

1. INTRODUCTION
In any soil-structure interaction situation relative movement of the structure with respect to the
soil can occur. The use of continuum 2D elements with compatibility in a finite element analysis
of these situations prohibits relative movement at the soil-structure interface. Nodal compatibil-
ity of the finite element method constrains the adjacent structural and soil elements to move
together. Interface or joint elements can be used to model the soil-structure boundary such as the
sides of a wall or pile, or the underside of a footing. Particular advantages being the ability to vary
the constitutive behaviour of the soil-structure interface (e.g. the maximum wall friction angle)
and to allow differential movement of the soil and the structure, i.e. slip and separation.

I . l . Interface models
Many methods have been proposed to model discontinuous behaviour at the soil-structure
interface. These include:
(1) Use of thin 2D finite elements with standard constitutive laws'*2
(2) Linkage elements in which only the connections between opposite nodes are c o n ~ i d e r e d ~ . ~
Usually opposite nodes are connected by discrete springs.

CCC 0363-906 1/94/ 100689-20 Received 22 September 1993

0 1994 by John Wiley & Sons, Ltd Revised 29 January 1994
690 R. A. DAY AND D. M. POTTS

(3) Special interface or joint elements of either zero or finite thi~kness.~-"

(4) Hybrid methods where the soil and structure are modelled separately and linked through
constraint equations to maintain compatibility of force and displacement at the inter-
face."-14
Wilsonlo has demonstrated that ill-conditioning of the interface:element stiffness matrix due to
large off-diagonal terms causes loss of accuracy. Ghaboussi et al.' and Wilson" therefore
advocate the use of relative displacement as independent degrees of freedom for interface
elements. The use of this relative degree-of-freedom formulation requires modification of the
adjacent 2D elements only on one side of the interface, so that they use the same relative degrees-
of-freedom.Thus, incorporation into a finite element program is complex. This formulation also
causes problems at joint intersections.'
The benefit of the relative degree-of-freedom formulation is uncertain. Pande and Sharma'
compared a series of analyses using the relative degree-of-freedom formulation with similar
analyses using the absolute degree-of-freedomformulation. They found that little ill-conditioning
is experienced with the use of very thin 2D elements. The additional effort involved in the use of
the relative displacement element may not be worthwhile.
Griffiths' recommends the use of special interface elements instead of thin standard 2D
elements for more accurate modelling of the interface when slippage is of prime concern.
Desai et ~ 1 propose
. ~ a thin 2D joint/interface element cidled a thin-layer element for
soil-structure interaction. A special constitutive law is used to model the principle deformation
modes of shear and opening. They report that with the use of zero thickness elements it is often
difficult to obtain constant and stable stress in the interface elements themselves, and therefore the
stress in the adjacent 2D elements is often adopted rather than the interface element stress itself.
It is argued that the thin-layer element is computationally more reliable than the zero thickness
element. The choice of element thickness is however important, and can effect the behaviour of
the interface given the same constitutive parameters.6 An extensive parametric study has recently
resulted in guidelines and empirical criteria for the determination of element thickness for the
thin-layer element.'
Zero thickness elements however have been used by many other^^,^*'^-'^ without reporting
similar problems. The element used has been based on that proposed by Goodman et aL9 The
essential modifications to the original element of GoQdman et al. are the extension to four and six
node isoparametric formulations to make the interface element compatible with isoparametric
quadrilateral 2D elements.

1.2. Constitutive laws

The development of interface or joint elements has been motivated by the desire to model the
behaviour of joints in fractured rock. The proposed constitutive laws are therefore strongly
oriented towards describing this behaviour.
Non-linear elastic hyperbolic relationships are generally used to describe the shear
stress-strain b e h a ~ i o u r . ~ * ~ "Similar
~.' relationships have been used to describe the normal
''
stress-strain behaviour in rock joint^.^.^' For soil-structure i.nteraction,linear elastic normal
stress-strain behaviour has often been considered sufficient7*8. '
The yield behaviour of interface elements has not generally been formulated in terms of
plasticity theory. In these simple formulations the Mohr-Coulomb strength criterion is used to
define the maximum permissible shear stress. When the shear stress reaches this limit, the value of
the shear stiffness is set to zero. The normal stiffness remains unchanged. Tensile stress in the
 ZERO THICKNESS INTERFACE ELEMENTS-NUMERICAL STABILITY AND APPLICATION 691

interface element is generally not permitted. If tensile stress occurs both the normal and shear
stiffness are set to zero and the tensile stress is redistributed by the solution algorithm.
In more complex models dilation has been included. These models are based on the sawtooth
model for rock joint behaviour. The models include the effect of dilation during the pre-failure
shearing phase of the Joint.5920
Models based on the theory of plasticity in which the dilation is controlled by a yield criterion
and a plastic potential function have been proposed by Ghaboussi et al.,’ Desai et Gens
et al.” A more complex plasticity model incorporating the disturbed-state concept for more
realistic modelling of joints and interfaces is introduced by Desai and Ma.” Their model allows
for the elastic-plastic, dilation, roughness, and hardening and softening response of joints.
Navayogarajah et al.” present a model based on the Hierarchical Single-Surface (HISS) ap-
proach to describe the behaviour of interfaces under static and cyclic loadings, including strain
softening.

I .3. Soil-structure and soil-reinforcement analysis

The zero thickness element formulation using a simple linear elastic perfectly plastic constitu-
tive model appears to have been successfully used for retaining wall-soil problems and for
soil-reinforcement interaction.7*17,1 This paper highlights and examines some of the numerical
problems the authors have encountered when using zero thickness interface elements in finite
element analysis. The linear elastic perfectly plastic model was adopted for the analyses presented
in this paper.

2. INTERFACE ELEMENT FORMULATION USED IN THIS STUDY

The isoparametric interface element is described by Beer7 and Carol and A l o n ~ oThe . ~ element
(Figure 1) with four or six nodes is fully compatible with four and eight-node isoparametric 2D
elements. The interface stress is characterized by the normal and shear stresses. The normal stress,
0,and the shear stress, T, are related by the constitutive law to the normal and tangential interface
element ‘strains’, E and y.

The interface element ‘strain’is defined as the relative displacement of the top and bottom of the
interface element, i.e.
y = Au, = up”’ - upo’ (2)
E = - vbol - ybo‘ (3)
1- I I

where
u, = u cos a + v sin a (4)
uI = - u sin a + v cos a
and, u and u are the global displacements in the x and y directions, respectively.
Hence,
y = (ut0p- ubot)cos a + - vbol)sin a
(vtOp

E= -(u lop - ubot)sin a + ( d 0 p - vbot) cos a

692 R. A. DAY AND D. M. POTTS

U,

XU

Figure 1. Isoparametric interface element

2.1. Constitutive laws

A linear elastic perfectly plastic model using a Mohr-Coulomb failure criterion as the yield
surface is used in the analyses presented here (Figure 2). The formulation of the constitutive
behaviour was based on plasticity theory. The elastic constitutive [D] matrix is
KllO
(7)
= [0 K.]
where K , and K , are the elastic shear stiffness and normal stiffness respectively (in units of F/L3).
The failure criteria are formulated in terms of effective stress. Pore fluid may exist in the
interface. Undrained behaviour is modelled by including the effective bulk modulus of the pore
fluid in the stiffness matrix.

t
Figure 2. Mohr-Coulomb yield function
 ZERO THICKNESS INTERFACE ELEMENTS-NUMERICAL STABILITY AND APPLICATION 693

The Mohr-Coulomb failure criterion defines the yield function, F , and the gradient of the
plastic potential function, G.
F = JzJ+ ortan@ - c’ (8)
aG 8G
= tanu; - -- + 1 (9)
aa aY
where Q is the effective normal stress, 4r is the maximum angle of shearing resistance, c’ is the
cohesion (Figure 2) and u is the dilation angle. Dilation of the interface is represented by a value of
u < 0.

Opening and closing

If the interface moves such that the maximum normal tensile strength is exceeded (c’/tan $f) the
interface is allowed to subsequently open and close and the residual tensile stress is redistributed.
When the interface is open the normal stress remains equal to c’/tan # and shear stress remains
equal to zero. This is essentially the same as setting both normal and shear stiffnesses to zero. The
amount of opening of the interface is recorded. When the interface recloses and reforms contact,
the constitutive model again defines the interface behaviour.

3. NUMERICAL BEHAVIOUR OF INTERFACE ELEMENT

Numerical problems such as ill-conditioning, poor convergence of solution and unstable
intergration point stresses have been experienced by the aurhors when using the zero thickness
interface element. To investigate these problems three simple examples, crudely representing
practical situations, have been analyzed to highlight and illustrate the behaviour of the interface
element.
In the analyses results using full (three point) and reduced (2 point) Gaussian integration were
compared. Results using the Newton-Cotes (3 point) integration scheme are also discussed. The
interface is assumed fully drained with pore pressure equal to zero.

Example I (simply supported split beam). This example demonstrates the ability of an interface
element to model frictionless slip between 2D elements (e.g. rock joints). The beam consists of two
rows of ten equal length 8-node 2D elements, separated by ten, 6-node interface elements
(Figure 3). Only half of the beam was modelled because of symmetry. The problem is specified in
dimensionless units. The Young’s modulus and Poisson’s ratio of the beam are, E = 5 . 0 x lo4
and p = 0. A uniform load of lO/unit length was applied to the top surface of the upper row of
elements.

Example 2 (simple pull-out test). In view of potential applications to soil reinforcement and
pile analysis this example demonstrates the interface element in a problem controlled by the
sliding mode of deformation, This problem consists of a thin strip of 2D elements, representing a
reinforcing membrane, on a rigid base (Figure 4). The contact between the membrane and the
base is modelled by interface elements. Five, 8-node 2D elements and five 6-node interface
elements were used to model the test. Initially a uniformly distributed load of 10 kPa was applied
to the top of the membrane. End A of the membrane was then moved horizontally by the amount
6, in the direction shown under displacement control. This problem involves a sliding front that
moves progressively from the point of load application towards the other end of the membrane.
This problem is a simplified version of the pull-out test studied by Gens et a l l 7
694 R. A. DAY AND D.M. POTTS

Figure 3. Split beam-finite element mesh

Figure 4. Pull-out test-finite element mesh

I
C
c

Figure 5. Overturning of elastic block

 ZERO THICKNESS INTERFACE ELEMENTS-NUMERICAL STABILITY AND APPLICATION 695

Wing
Unloading

Figure 6. Displacement of point D

Example 3. (overturning of an elastic block). This problem is crudely representative of a

gravity retaining wall. It consists of an elastic block, Young's modulus E = lo6 kN/m2 and
Poisson's ratio p = 0, separated from a rigid footing along the base A-B by interface elements
(Figure 5). For clarity, the interface elements are shown artificially expanded in Figure 5. Ten, 6-
node interface elements and fifty, 8-node 2D elements were used to model the problem. A
downwards vertical stress (200 kN/m2) and a shear stress in the positive x direction were applied
to the top boundary C-D. The interface was assumed to be non-dilatant (u = 0) with strength
parameters 4' = 20" and c' = 0.
Since tension in the interface is not permitted separation first occurs in the interface at point A
when the applied horizontal stress is 16.66 kN/m2. As the shear stress is increased the zone of
separation spreads towards B. Overturning instability of the block occurs when the applied shear
stress is 50 kN/m2. Figure 6 illustrates the load-displacement behaviour of the point D at the top
of the block. When all horizontal load is removed there is a net horizontal displacement of the
block due to the plastic shear strain that occurs during loading and unloading.

3.1. IIZ-conditioning
The analysis of the overturning block (Example 3) was used to investigate the reported
problems of ill-conditioning." A large number of elastic analyses were undertaken and in these
the interface element stiffness was varied from K , and K, = lo6 to K , and K , = 1OO ' kN/m3, with
Young's modulus of the 2D elements, E = lo6 /m2 and from K , and K, = lo3 to K , and K,
= lo' kN/m3 with E = lo3 kN/m2. Note that in general K , was not equal K,. The shear stress
applied to the top of the block was 50 kPa.
Ill-conditioning of the stiffness matrix caused fluctuating stress in the interface element when
the interface element stiffness was large with respect to the stiffness of the 2D elements.The results
of the analysis with K , = K, = lo9 kN/m3 and E = lo6 kN/m2 using 3 point Gaussian integra-
tion are shown in Figure 7 in which the integration point stresses are plotted. The effect of ill-
conditioning was noticeable if either K , or K, was greater than about 100E.(This is dependent on
696 R. A. DAY AND D.M.POTTS

the units used. The units used here are kN and m.) The use of reduced integration (2 point) does
not prevent the fluctuation in interface stress which is observed in Figure 7 for full (3 point)
integration.
The linear, normal and shear stress distributions shown in Figure 7 are the theoretical
distributions for a rigid block on elastic springs. Since the vertical faces of the block are stress free,
the shear stress within the base of the block must reduce to zero at the corners (A and B), however
the shear stress applied by the interface element to the base does not. To satisfy continuum theory
a singularity therefore occurs in the shear stress and vertical stress distributions at the corners
within the block. The theoretical shear stress distribution is parabolic for a long flexible block on
a rigid f~undation.’~ The shear stress distribution in the interface elements lies between the two
extremes.
An analysis with a thin layer of 2D elements in place of the interface elements was also
performed. The elastic properties of the thin layer are Ei and pi and the thickness is t. The

Q ’.a
\
a\*
, Linear distribution

0 I J

h
I .*‘-.
2
25 ‘4.
8 -
i-
z
0 Interface elements
\
-7
m 2-D elements
A B
-1ooO-

-----
Figure 7. Effect of ill-conditioning on interface element stn:ss (full integration)
 ZERO THICKNESS INTERFACE ELEMENTS-NUMERICAL STABILITY AND APPLICATION 697

equivalent normal and shear stiffnesses of the thin layer are

It is noted that when Kin = Kis, (1 - p!) = 2(1 + pi). The solution of this equation gives
p i = - 1, in which case the stiffnesses are infinite. To avoid this paradox and to be consistent
with the overlying elements the Poisson's ratio was chosen to be zero (pi = 0). With the thickness
chosen to be equal to 0.02m and Ei= 2.0x lo7 the normal stiffness of the thin elements is
equivalent to that of the interface element and the shear stiffness is half that of the interface
elements. Fluctuations also occurred in the 2D element stresses. The integration point stress
along the centre line of the 2D elements is also shown in Figure 7.
Reducing the size of the 2D elements adjacent to the interface elements reduced the effect of ill-
conditioning even though the total number of elements remained the same. Figure 8 shows the
results of the analysis when different size 2D elements are used adjacent to the interface element

Figure 8. Reduced ill-conditioning with smaller elements (full integration)

698 R. A. DAY AND D. M.POTTS

(see insert on Figures 7 and 8). In this case the contributions to the global stiffness matrix from the
2D elements is increased in the locations concerning the interface element degrees of freedom. The
contributions from the interface elements remains unchanged. Thus there is less difference
between the stiffness of the interface elements and the stiffness of the 2D elements and hence less
ill-conditioning.

3.2. Convergence of the non-linear solution algorithm

For analyses in which the interface was elasto-plastic, a modified Newton-Raphson algorithm
incorporating a sub-stepping stress point algorithm with error contr01,’~has been used to solve
the finite element equations. It is not uncommon to use a constant elastic stiffness matrix for all
iterations with this approach. The benefit being that the global (elasto-plastic) stiffness matrix
does not have to be recalculated and inverted at every stage of the analysis. This method works
well in cases where the changes in the elasto-plastic stiffness are not large. Where the behaviour of
a problem is highly non-linear the use of the constant elastic stiffness matrix can result in large
numbers of iterations and convergence problems.
During analyses of the overturning block (Example 3), convergence of the solution become
slower on successive increments. In this problem the global stiffness is rapidly changing as
successive interface elements open (Figure 6).In an attempt to increase the speed of convergence,
the elastic global stiffness matrix used in the solution algorithm was replaced by the tangent
elasto-plastic stiffness matrix (D,,)calculated the beginning of each increment. This dramatically
increased the speed of convergence up to a shear stress of 35 kPa., but caused divergence at later
stages. The solutions for Example 3 were obtained by using the Depmatrix calculated at a load
level of 30 kPa for the remaining loading increments (35,40 and 45 kPa) and first two unloading
increments (40 and 35 kPa). For the subsequent unloading increments the elastic stiffness matrix
was used. This technique was used siqce the ‘stiffness’ of the problem remains very low in early
stages of unloading until reclosure of the interface occurs over a sngnificant portion of the base of
the block.
The solution algorithm used also assumed the global stiffness matrix is symmetric and
therefore stores only the upper triangular part. (The benefit being a large saving in storage space.)
If a non-symmetric D,, matrix is encountered, an ‘averaged’ global stiffness matrix is used in the
solution algorithm. The full non-symmetric D,, is used when updating the iterative out-of-
balance load for successive iterations so that the final solution fully satisfies the non-symmetric
constitutive equations. The D,, matrix for a non-associated interface element is non-symmetric.
In the analysis of the overturning block, it is because the incorrect ‘averaged’ stiffness matrix was
being used at later stages of the analysis that the solution diverged. When the interface element
properties were changed so that the plastic flow was associated (the global stiffness was
symmetric), the solution converged very rapidly when the D,, matrix calculated at the beginning
of each increment was used.
Poor convergence also occurred when analysing the split beam (Example 1). In this problem
the two beams separated over a short distance near the simply supported end. This caused large
changes in the stiffness matrix. In this case the use of the D,, matrix, updated at the beginning of
each increment, allowed rapid convergence of the solution.

3.3. Steep stress gradients

Gens et a1.” found the behaviour of the interface element in the pull-out test was unsatisfactory
when Gaussian integration was used. Large stress oscillations occurred near the sliding front.
Gens et a1.17 reported that if the Newton-Cotes integration scheme was adopted the results were
 ZERO THICKNESS INTERFACE ELEMENTS-NUMERICAL STABILITY AND APPLICATION 699

quite satisfactory.The pull-out test in Example 2 was therefore analysed with a range of interface
and membrane stiffness to investigate whether similar problems would be encountered. The
analyses were carried out using 3 point Gaussian and Newton-Cotes integration.
The difference between Gaussian and Newton-Cotes integration is the location of the inte-
gration points and their corresponding weights." The integration points for the 3 point
Newton-Cotes method correspond to the positions of the three nodal pairs (i.e. at each end and
the midpoint). The stress in the element-at each integration point-is then given by the relative
displacement between the pair of nodes at the integration point location. The displacement of the
other nodes in the element has no influence. The element is therefore essentially a linkage
element.3.5Three point Newton-Cotes integration is equivalent to Simpson's rule for integration.
For Gaussian integration the integration points are located between the end and the midpoint of
the element, and at the midpoint of the element. The relative displacement of all nodes affects the
stress at each integration point.
In the results presented below the Poisson's ratio of the elastic membrane is zero and the
+'
interface element parameters are K , = K , = lo6 kN/m3, = 30°, v = 0" and c' = 0. Analyses
were also performed with K , = lo6 and K , = lo9. The results of these analyses are the same as
those with K , = K , and therefore are not presented. This is a problem dominated by sliding
behaviour. The membrane is free to move up and down as the interface dilates or contracts. The
normal stress in the interface element is controlled by the boundary condition of 10 kPa applied
to the top of the membrane. In this example the value of the normal stiffnessis not important.
The integration point stress was found to oscillate widely near the sliding front when the
membrane stiffness,E, was reduced with respect to the interface stiffness. Figure 9 shows the shear
stress distribution in the interface elements when the displacement, 6, of point A (Figure 4) is
0.12 mm. The Newton-Cotes integration scheme appears to greatly improve the behaviour of the
interface element. However further investigation of the pull-out test with E = lo4 kN/mZ shows
that the stress distribution given by the Newton-Cotes integration is actually a poor approxima-
tion to the correct solution.

NewtmCotes integration
Gauss integration

Figure 9. Shear stress in interface elements

700 R. A. DAY AND D. M. POTTS

NewtonCOtes integration
-Gaussintegration
_.-

-8L
a) Shear stress on interface

b) Axial force in membrane

c) Displacement of membrane

Figure 10. Effect of element size on pull-out results

 ZERO THICKNESS INTERFACE ELEMENTS-NUMERICAL STABILITY AND APPLICATION 701

Figure 9 indicates that as the stiffness of the membrane reduces, the gradient of the shear stress
distribution on the interface becomes steeper. When E = lo6 kN/m2 the stress reduces from the
maximum value, 5 . 8 kPa, to nearly zero over a distance of 0.4 m. This reduction therefore occurs
over 2 elements or 6 integration points. When E = lo5 kN/m2 the stress reduces over a distance
of 0.2 m, which is the size of a single element. When E = lo4 kN/mZ the same stress change
probably occurs over a distance less than the represented by one element. This is likely to cause
the stress oscillation seen in Figure 9. Smaller elements (less than 0.2 m) are clearly necessary for
-=
the analysis of this problem if E los kN/mz.
Further analyses were performed with E = lo4 kN/mZ but with smaller elements at end A of
the membrane and interface. Elements with lengths of 0-05 m and 0.025 m were placed at the end
of the membrane and interface. Figure 10 shows the results of these analyses when 6 = 0.12 mm
and also the results taken from Figure 9 where the element size is 0.2 m.
At end A of the interface the shear stress is greater than the expected value of 5.77 kPa
(10tan 30”) and the distribution of shear stress in the first 0.05 m is rather surprising. The
membrane is experiencing a shear stress acting on its underside only. Being thin, the bending
stiffness of the membrane is very small. The shear stress acting on the side of the membrane causes
the end A of the membrane to bend downwards increasing the normal stress in the interface. Thus
the maximum permissible shear stress in increased. At a short distance from the end of the
membrane the curvature of the membrane causes a reduction in normal stress in the interface
element resulting in a complex distribution of shear stress. When the membrane was constrained
so that it could not bend, the maximum normal stress in the interface element remained at 10 kPa
and the maximum shear stress at 5.77 kPa. This confirms that bending of the membrane causes
the unusual distribution of shear stress, shown in Figure 10.
The problem of oscillating stress is due to the use of elements too large to model adequately the
steep stress gradient that occurs in this problem. Stress oscillation does not occur when small
elements are used. The numerical problems encountered here are not due to poor performance of
the interface element but are due to inadequate modelling of the problem at hand. Only high-
order elements that allow for a complex distribution of stress across the element will be able to
accurately model the sliding front with the use of larger elements. Newton-Cotes integration is
unnecessary and also undesirable as it has the effect of ‘glossing over’ or ‘smoothing out’ the steep
stress gradient, thus hiding the real solution. When sufficiently small elements are used to describe
adequately the stress gradient, Newton-Cotes and Gaussian integration give similar results.
Steep stress gradients across the interface element were also noted to cause poor convergence in
the split beam analysis (Example 1). In this problem the two halves of the beam separate near the
simply supported end. There is a large stress gradient in the normal stress distribution at the end
of the beam where the two halves make contact. The stress gradient becomes greater as the
normal stiffness of the interface element is increased. Oscillation in the stress occurs and
convergence is inhibited. Reducing the size of the elements at the end of the beam allowed more
accurate representation of the steep stress gradient in this region and improved the speed of
convergence.

4. USING INTERFACE ELEMENTS WITH RETAINING WALLS

To assess the performance of interface elements in retaining wall analysis a series of analyses of
the passive and active failure modes of deformation were carried out. A rigid wall embedded 5 m
into the soil was translated uniformly in the horizontal direction. Interface elements placed on
both sides of the wall allowed the magnitude of the maximum angle of wall friction, 6, to be
702 R. A. D A Y A N D D. M. POTTS

varied. The minimum active force, the maximum passive force and the failure surfaces are
compared. Similar analyses were carried out by Potts and Fourie.26 Interface elements were not
available to them and consequently they only studied the two extremes of a fully rough wall, 6
= 4' and a smooth wall, 6 = 0.

4.1. Details of analyses

The finite element mesh used in the analysis and the applied boundary constraints are shown in
Figure 11. The rigid wall was modelled as part of the boundary. Details of the mesh around the

t ZOm - I- 70 m 1

I
Figure 11. Retaining wall analysis-finite element mesh

Zero thickness
interface elements
(artificially expanded)

,-
Figure 12. Interface element details
 ZERO THICKNESS INTERFACE ELEMENTS-NUMERICAL STABILITY AND APPLICATION 703

Table I. Interface element properties

Elastic stiffness Angle of friction Angle of dilation

Analysis Kw K" 6 0
No. (kN/m3) (degrees) (degrees)
1 E* 25.0 250
2 lOOE 25.0 25.0
3 E 25.0 00
4 E 17.5 17.5
5 E 12.5 125
6 E 5.0 5.0
l E 5.0 00

* E is the Young's modulus of the soil

bottom of the wall is shown in Figure 12. The nodes on the boundaries A-B and A - B and nodes
C,D, and E were moved horizontally to the right at a rate of 1 mm per increment. A total of
120 mm displacement was applied. Vertical displacement of these nodes was not permitted.
The Mohr-Coulomb failure criterion was adopted as the yield function for the soil. The
material properties of the soil are the same as those used by Potts and Fourie.26 These are: E
= 60,000 kN/m2, p = 0.2, c' = 0, 4' = 25". Fully associated plastic flow was assumed. The bulk
unit weight y = 20 kN/m3. The initial horizontal stress was assumed to be equal to the vertical
stress ( K O= 1.0). Fully drained plane strain analyses were performed,
The zero thickness interface element with elastic perfectly behaviour described in Section 2 was
used to model the interface between the wall and the soil. Table I gives the interface element
properties used in each analysis. In all analyses the Mohr-Coulomb failure model was used and
the effective cohesion. c' = 0.
An analysis in which there were no interface elements was also performed.

4.2. Results of analyses

Earth pressure coeficients. The equivalent earth pressure coefficient, K, is calculated from K
= 2Ph/(yH2)where H is the height of the wall and P, is the sum of the horizontal nodal reactions
along the boundary representing the wall. Calculation of the reaction considered the stress only in
the elements adjacent to the wall (i.e. on boundaries AB and AB'). The limiting values of K, i.e.
K , , Kab,K, and Kph,are listed in Table 11.
The angle of wall friction given in Table I1 is calculated from arctan(P,/P,) where P, and P h
are the sum of the vertical and horizontal reactions along the boundary representing the wall. It is
therefore the angle from the horizontal of the direction of the resultant reaction. On the passive
side of the wall, friction is fully mobilized over the full length of the wall. On the active side the
average mobilized wall friction is slightly less than the maximum wall friction angle. This is likely
to be caused by fluctuations in the soil stress (see discussion below on earth pressure) due to large
stress gradients in the elements around the base of the wall.
The magnitude of the average wall friction in the analysis without interface elements is greater
than 25". This does not satisfy the soil constitutive model. The friction angle has been calculated
from the horizontal and vertical reactions on the boundary representing the face of the wall. In
finite element analysis, the reactions are indirectly determined by extrapolating the integration
point stresses to the boundary. The integration point stresses satisfy the constitutive law. The
104 R. A. DAY AND D. M. POTTS

Table 11. Calculated earth pressure coefficients

Active pressure Passive pressure

Analysis Horizontal Resultant Average wall Horizontal Resultant Average wall

No Kah Ka friction Kph KP friction

No interface
element 0.31 0.35 26.18 4.00 4.44 25.70
1 0.32 0.35 24.1 1 4.00 4.4 1 24.99
2 0.32 0-34 23.76 4.03 4.45 24.99
3 0.31 0.34 24.45 3.99 4.40 25.00
4 0.33 0.34 17.47 3.69 3.87 17.49
5 0.34 0.35 12.46 3.37 3.45 12.50
6 0.37 0.37 4.88 2.84 2.85 5.00
7 0.37 0.37 4.88 2-82 .2-83 5.00

extrapolated boundary stresses and reactions do not necessarily satisfy the constitutive law,
particularly if high stress gradients exist. The integration points for the zero thickness interface
elements are on the interface and therefore no extrapolation towards the wall is necessary. Hence
the interface stress and the average angle of wall friction satisfy the constitutive law.
The ultimate earth pressure coefficients are dependent on the maximum wall friction angle but
are independent, for all practical purposes, on the dilation and the elastic stiffness of the interface.
The calculated values of KP1 and K , , are compared with various approximate analytical
methods in Table 111. The analytical values of K,h are all very similar and have little variation due
to wall friction. The finite element values are slightly lower than all of the analytical methods for
all values of wall friction angle. The differences are however quite small. The various analytical
values of Kph are similar at low wall friction angles (for S = 0 they are all equal) but have a

Table 111. Comparison of Kph and Kahwith other methods

Wall friction angle

Methods 5" 12.5" 17.5" 25"

Passive pressure coeficients, Kp,

Finite element analysis 2.84" 3.31 3.69 4 w
Caquot and Kerisel" 2.85 3.38 3.73 3.89
ChenZ8 2.81 3.36 3.7 1 4.19
Coulomb30 2.82 3.47 4.01 5.07
Packshaw" 2.79 3.28 3.58 3-90

Active pressure coeficients, Ka,

Finite element analysis 0.37" 0.34 033 0.32b
Caquot and KeriselZ7 0.38' 0.36 0.35 0.333
ChenZs 0.386 0.360 0.346 0.327
Coulomb30 0.385 0.359 0.343 0.322
PackshawZ9 0.385 0.359 0.343 0-322

Analysis No. 1
" Analysis No. 6
Interpolated value
 ZERO THICKNESS INTERFACE ELEMENTS-NUMERICAL STABILITY AND APPLICATION 705

substantial variation for a fully rough wall. The values calculated by the finite element analyses
with the interface element are in reasonable agreement with those calculated by Caquot and
Keri~el,'~ Chen,2* and PackshawZ9at all values of wall friction. The values given by Packshaw
are slightly less than the finite element values at all wall friction angles. The values given by
Coulomb30 overestimate K,, considerably for higher wall friction angles. This is widely agreed.
The values of Caquot and Kerisel are practically the same as the finite element calculations for
low wall friction angles but is about 3 percent less than the finite element result when 6 = 6'.

Earth pressure. The ultimate stress distribution on both sides of the wall is nearly linear in each
analysis. A linear distribution is generally assumed in the analytical earth pressure theories.
Towards the base of the wall the passive pressure tends to be greater than, and the active pressure
less than, an equivalent linear distribution. The distribution of stress in the 2D elements adjacent
to the interface is essentially the same as the distributions in the interface elements. Some
oscillation of the integration point stress occurs in the bottom two elements. This is likely to be
due to very large strains and high stress gradients in the soil elements around the base of the wall.
At the base of the wall the horizontal stress in the soil changes from about 400 kPa on the passive
side to about 40 kPa on the active side over the width of the wall (0.2 m).

100 E

5 00
~ o n n astress
i (kPa)

Figure 13. Normal stress in interface elements (Analyses 1 and 2)

706 R. A. D A Y A N D D. M. POTTS

In Figure 13 the stress distribution in the interface elements for analysis No. 2 ( K , = 100E)is
compared with the distribution in the interface from analysis No. 1 ( K , = E ) . Oscillation in the
stress occurs in analysis No. 2, particularly in the lower half of the wall.

Failure mechanism. The failure surfaces calculated by the analyses are close to planar in the
case of 6 = 5" and spiral shaped in the case of 6 = 4'. The results of the other analyses indicate
failure surfaces between these two extremes. These results agree with analytical limit equilib-
rium*' and limit analysis2* methods which also indicate failure on plane surfaces for smooth
walls (6 = 0) and on curved surfaces (e.g. log spiral) for rough walls (6 = #). The zone of failure
appears independent of the stiffness and angle of dilation of the interface elements.

Interface properties. In the analyses of the retaining wall the elastic shear stiffness is assumed
equal to the elastic normal stiffness. In reality this may often be unrealistic.
However, in retaining wall problems the elastic parameters of the interface are not important
except for the oscillations reported above if K , is large. The interface and surrounding soil yields
very quickly and the elastic-plastic behaviour then dominates. The use of a much smaller value of
K , will result in greater shear displacement prior to yield but will not effect the value of the
limiting earth pressure. The value of the limiting earth pressure is governed by the soil properties
and the strength parameters of the interface.
Reduction in K, will decrease the shear stiffness contribution of the interface in the global
stiffness matrix. This is similar to the effect on the elasto-plastic stiffness matrix of reducing the
friction angle. It does not result in ill-conditioning of the stiffness matrix or numerical difficulties.
The results presented here are also valid for the case when K , Z> K,.

5. CONCLUSIONS
The zero thickness interface element is useful for modelling relative slip and opening and closing
on predefined surfaces. Numerical problems can however occur through ill-conditioning of the
stiffness matrix and high stress gradients in the interface elements. In many situations stress
gradients are likely to be high and are increased with increased interface stiffness. This problem
can therefore easily be confused with ill-conditioning. Ill-conditioning was noticed in the
problems analysed here when the stiffness of the interface element was greater than 100 times the
Young's modulus of the surrounding soil. This however depends on the units used in the problem
and the size of the surrounding soil elements.
In analyses in which the interface is opening and closing, and where non-dilatant interface
properties are assumed, large changes in the stiffness matrix occur. This can cause convergence
difficulties with a modified Newton-Raphson solution algorithm in which a constant elastic
stiffness matrix is being used. The use of the tangent elasto-plastic stiffness at each stage of the
analysis will in most cases significantly accelerate convergence. When non-associated behaviour
of the interface or 2D elements is assumed, the D,,matrix is non-symmetric. For some problems it
may be necessary to store and solve the full non-symmetric global stiffness matrix in order to
achieve convergence.
Reduced (2 point) Gaussian integration does not help to solve the numerical problems
illustrated. Reduced and full Gaussian integration produce similar results.
Newton-Cotes integration tends to improve the numerical behaviour of interface elements at
the possible expense of hiding the true solution. Where sufficiently small elements are used to
model the interface behaviour, Newton-Cotes and Gaussian integration give similar results. No
advantage is therefore gained through the use of Newton-Cotes integration.
 ZERO THICKNESS INTERFACE ELEMENTS-NUMERICAL STABILITY AND APPLICATION 707

The Use of interface elements with an elasto-plastic Mohr-Coulomb failure criterion has
predicted ultimate values of earth pressure for various angles of wall friction in good agreement
with other analytical solutions. These analyses have provided numerical confirmation of the
analytical solutions of Caquot and Kerisel” which are widely accepted and commonly used in
retaining wall design. These values however apply only to the situation of uniform wall
translation. Differentmodes of wall movement, for example rotation, will affect the ultimate value
and distribution of the earth pressure on the
In the analysis of the active and passive failure modes of a retaining wall, the ultimate earth
pressure is independent of the elastic stiffness and the angle of dilation of the interface elements.
The results obtained for a rough wall modelled with interface elements assuming associated
plastic flow are essentially the same as those obtained if no interface element is used to model the
soil-wall boundary.
In the retaining wall problem failure occurs through the soil mass. The ultimate failure load for
given interface behaviour is governed by the soil properties rather than the interface properties.
The interface element simply provides a stress boundary condition to the soil mass. Hence it is
only the strength parameter (i.e. interface friction angle) that is important. A different result will
be obtained when different soil properties are used with the same interface properties. Conversely,
for given soil properties and interface strength the same result is obtained with different interface
stiffness and dilation.
In problems in which the failure mode is by sliding on the soil-structure interface, this would
not be expected to be so. In analysis of piles and ground anchors for example, the ultimate load
for given interface element properties could conceivably be largely independent of the soil
properties if the strength of the interface is exceeded before the strength of the soil. The normal
stiffness and dilation properties of the interface, and the boundary conditions in this type of
problem would be expected to play a dominant role. The use of the more complex elasto-plastic
models21*22 may be necessary in these situations to obtain more realistic and accurate results.
The zero thickness interface element provides a useful means to model reduced interface
friction in finite element analysis of retaining walls. The simple elasto-plastic Mohr-Coulomb
constitutive model appears to be satisfactory for the modelling of interface behaviour in this type
of analysis.

ACKNOWLEDGEMENTS
This work was done while R. A. Day was a research student at Imperial College of Science,
Technology and Medicine and contributes to a project partly funded by British Steel plc.
Additional financial support for R. A. Day was provided by the Association of Commonwealth
Universities, the Overseas Research Student Award scheme and the Dalby Bursary.

REFERENCES
1. D. V. Griffiths, Numerical modelling of interfaces using conventional finite elements,Proc. 5th Int. Con$ on Numerical
Methods in Geomechanics, Nagoya, 1985, p p . 837-844.
2. G . N. Pande and K. G. Sharma, On joint/interface elements and associated problems of numerical ill-conditioning’,
i n t . j . number. anal. methods. geomec., 3, 293-300 (1979).
3. R. Frank, A. Guenot and P. Humbert, Numerical analysis of contacts in geomechanics’, Proc. 4th Int. Con! on
Numerical Methods in Geomechanics, Rotterdam, 1982,3742.
4. L. R. Hermann, ‘Finite element analysis of contact problems’, Proc. ASCE, 104 ( E M )1043-1057 (1978)
5. I. Carol and E. E. Alonso, ‘A new joint element for the analysis of fractures rock‘, 5th Int. Congress on rock mechanics.
vol. F, Melbourne, 1983, 147-151.
708 R. A. DAY AND D. M. POTTS

6. C. S. Desai, M. M. Zaman, J. G. Lightner and H. J. Siriwardane, ‘Thin-layer element for interfaces and joints’, Int. j .
numer. anal. methods geomech., 8, 19-43 (1984).
7. G. Beer, ‘An isoparametric joint/interface element for finite element analysis’, Int. j . numer. methods eng., 21, 585-600
(1985).
8. J. Ghaboussi, E. L. Wilson and J. Isenberg, ‘Finite element for rock joint interfaces’, J. Soil Mechs Foundations Div.
ASCE. 99 (SM10) 833-848, (1973).
9. R. E. Goodman, R.L. Taylor and T. L. Brekke, ‘A model for the mechanics ofjointed rock’, J. Soil Mechs Foundations
Div. ASCE, 94 (SM3) 637-659 (1968).
10. E. L. Wilson, ‘Finite elements for foundations, joints, and fluids’, in a G. Gudefius(ed.) Finite Elements in Geomechanics,
Chap. 10, Wiley , New York, 1977.
11. A. Francavilla and 0. C. Zeinkiewica, ‘A note on numerical computation of elastic contact problems’, Int. j . numer.
methods eng., 9, 913-924 (1975).
12. M. G . Katona, ‘A simple contact-friction interface element with applications to buried culverts’, Int. j. numer. anal.
methods geomech., 7, 371-384 (1983).
13. J. Y. Lai and J. R. Booker, ‘A residual force finite element approach to soil-structure interaction analysis’, Research
Report No. 604, University of Sydney, 1989.
14. T. D. Sachdeva and C. V. Ramakrishnan, ‘A finite element solution for the two dimensional elastic contact problem’,
1nt.j. numer. methods eng., 17, 1257-1271 (1981).
15. K. G. Sharma and C. S. Desai, ‘Analysis and implementation of Thin-Layer element for interface and joints’, J. Eng.
Mech. ASCE, 118, 2442-2462 (1992).
16. G. W. Clough and J. M. Duncan, ‘Finite element analysis of retaining wall behaviour’, J. Soil Mechs Foundations Div.
ASCE, 97 (SM12) 1657-1672 (1971).
17. A. Gens, I. Carol and E. E. Elonso, ‘An interface element formulation for the analysis of soil-reinforcement interaction’,
Comput. Geotech., 7 (1,2) 133-151 (1989).
18. S. K. Bhatia and R. M. Bakeer, ‘Use of the finite element method in modelling a static earth pressure problem’, 1nt.j.
numer. anal. methods geomech., 13, 207-213 (1989).
19. A. Varadarajan and K. G. Sharma, ‘Effect of a shear seam in the foundation of Karjan dam’, Int. j. numer. anal. methods
geomech., 13,435-442 (1989).
20. R. E. Goodman and J. Dubois, ‘Duplication of dilatancy in analysis of jointed rocks’, J. Soil Mechas Foundations Diu.
ASCE, 98 (SM4) 399-422, 1972.
21. A. Gens, I. Carol and E. E. Alonso, ‘A constitutive model for rock joints formation and numerical implementation’,
Comput. Geotech., 9, 3-20, (1990).
21. C. S. Desai and Y Ma, ‘Modelling ofjoints and interfaces using the disturbed-slate concept’, lnt.j . numer. anal. methods
geomech. 16, 623-653 (1992).
22. N. Navayogarajah, C. S. Desai and P. D. Kiousis, ‘Hierarchical single-surface model for static and cyclic behaviour of
interfaces’, J Eng. Mech. ASCE, 118,990-1011 (1992).
23. W. C. Young, Roark‘s formulas for stress and strain, 6the edn. McGraw-Hill, New york, 1989.
24. S. W. Sloan, ‘Substepping schemes for numerical integration of elasto-plastic stress-strain relations”, Int. j . numer.
methods eng., 24, 893-911 (1987).
25. 0.C. Zienkiewicz, The Finite Element Method, 3rd edn, McGraw-Hill, London, 1977.
26. D. M. Potts and A. B. Fourie, ‘A numerical study of the effects of wall deformation on earth pressures’, Int. j . numer.
anal. methods geomech., 10, 383-405 (1986).
27. A. Caquot and J. Kerisel, Tables for the Calculation of Passive Pressure, Active Pressure and Bearing Capacity of
Foundations, Gauthier-Villars, Paris, 1948.
28. W. F. Chen, Limit analysis and soil plasticity, in Developments in Geotechnical Engineering, Vol. 7, Elsevier,
Amsterdam, 1975.
29. E. Packshaw, ‘Earth pressure and earth resistance’, J . ICE, 25, 233-256 (1946).
30. C. A. Coulomb, ‘Essai sur une application des regles de maximis et minimus a quelques Problemes de statique, relatifs a
l’architecture’, Memoires de Mathematique et de Physique Presences a PAcademie Royale des Sciences, I , Paris, 1776.