Computers and Geotechnics 30 (2003) 419–430

www.elsevier.com/locate/compgeo

Stability of an undrained plane strain heading revisited

Charles E. Augardea,*, Andrei V. Lyaminb, Scott W. Sloanb
a
School of Engineering, University of Durham, UK
b
Department of Civil, Surveying and Environmental Engineering, University of Newcastle, NSW 2308, Australia

Received 19 July 2002; received in revised form 25 October 2002; accepted 9 January 2003

Abstract
The stability of an idealised heading in undrained soil conditions is investigated in this paper. The heading is rigidly supported
along its length, while the face, which may be pressurised, is free to move. The problem approximates any flat wall in an under-
ground excavation. Failure of the heading is initiated by a surface surcharge, acting with the self-weight of the soil. Finite element
limit analysis methods, based on classical plasticity theory, are used to derive rigorous bounds on load parameters, for a wide range
of heading configurations and ground conditions. Solutions for undrained soils with constant strength, and increasing strength with
depth are presented. Recent improvements to finite element limit analysis methods, developed at the University of Newcastle, have
allowed close bounds to be drawn in most cases. Previous research in this area has often been presented in terms of a stability ratio,
N that combines load and self-weight into a single parameter. The use of a stability ratio for this problem is shown not to be rig-
orous, a finding that may be applicable to other stability problems in underground geomechanics.
# 2003 Elsevier Science Ltd. All rights reserved.
Keywords: Stability; Heading; Limit analysis; Tunnel; Plasticity

1. Introduction open face, behind which the ground is supported by a

rigid, and infinitely strong, lining. In practice, mines are
The assessment of the safety of shallow underground supported by jacks or props spaced at close intervals to
excavations in soft ground, for tunnel construction and prevent cave-in. While these systems are not totally
mining, usually requires solutions to two separate pre- rigid, it seems reasonable to assume so here as this study
dictive problems. Firstly, it is necessary to determine the is concerned with stability of the open face, rather than
stability of the excavation, for the safety of those at the the determination of deformations. (Similar impractical
surface and underground. Secondly, to prevent damage assumptions are accepted for many problems in geo-
to surface or subsurface structures, it is also necessary mechanics, for instance ‘‘rigid’’ footings). An alternative
to determine the pattern of ground deformations that plane strain model of an underground excavation is
will result from the construction works. obtained by an orthogonal view to that shown in Fig. 1,
This paper presents the results of an investigation into although in that case the excavation must be unlined.
the first of these problems as it affects what we will term Neither plane strain models capture the three-dimen-
a ‘‘plane strain heading’’ as shown in Fig. 1. This rela- sional nature of an underground excavation but are still
tively simple configuration is applicable to a long-wall useful tools given the difficulties in using fully three-
mining operation or to any flat wall in an underground dimensional models.
excavation. While the problem of underground excava- In the work presented here, numerical procedures are
tion stability is inherently three-dimensional, much can used to find bounds on parameters describing the stabi-
be learned from the behaviour of simpler two-dimen- lity of a range of heading sizes in undrained soil condi-
sional models such as the one used in this paper. It tions. The procedures used here are based on the limit
consists of a plane strain idealisation of a long vertical theorems of classical plasticity, using discretisations
similar to the displacement finite element method. Put
* Corresponding author. Fax: +44-191-374-2550. briefly, if a statically admissible stress field can be found
E-mail address: charles.augarde@dur.ac.uk (C.E. Augarde). such that yield is exceeded nowhere in the problem
0266-352X/03/$ - see front matter # 2003 Elsevier Science Ltd. All rights reserved.
doi:10.1016/S0266-352X(03)00009-0
420 C.E. Augarde et al. / Computers and Geotechnics 30 (2003) 419–430

The bound theorems have also been used to develop

solutions for the stability of tunnels (but not plane
strain headings) in drained conditions [7–9]. Very few
researchers have attempted to use methods other than
the bound theorems to study the stability of under-
ground openings. Eisenstein and Samarasekara [10]
examine the stability of tunnels in clay using a novel mix
of limit equilibrium and displacement finite element
results. Anagnostou and Kovari [11] also employ limit
equilibrium methods to study pressures required for
earth pressure balance tunnels in drained conditions.
Neither of these approaches, however, provides the type
of results furnished by bound approaches, which give
safe and unsafe limits on stability parameters that
Fig. 1. Layout of plane strain heading problem. bracket the true solution.
Where the bound theorems of plasticity have been
used, in the works cited above, conventional analytical
region, then it is a safe solution (a lower bound) to the approaches (such as the Method of Characteristics)
problem. Alternatively, an unsafe solution (an upper have been used. Since the late 1980’s, Sloan and co-
bound) can be found from a kinematically admissible workers have developed finite element bound methods
collapse mechanism by equating internal power dis- that are more versatile than conventional use of the
sipated with external power expended. The bound the- bound theorems. These techniques, which are described
orems apply only to rigid-plastic materials. Were the in more detail below, permit inhomogeneous soil pro-
object of this research the determination of deforma- files and soil self-weight, two features that often prove
tions, rather than collapse conditions, then the dis- difficult to incorporate in a conventional bound analy-
placement finite element method, using an elasto-plastic sis. These methods have been used to study the
material model, would be appropriate. It is important to undrained stability of square and circular unlined tun-
note, however, that the elastic parameters of any elasto- nels [12,13] as well as the plane strain heading [4]. They
rigid plastic material model would not affect the solu- have also been used to predict the stability of a circular
tion for the collapse load [1]. unlined tunnel in cohesive-frictional soil [14].
Finite element implementations of the bound theo-
rems were first proposed by Lysmer [2], for the lower
bound case, and Anderheggen and Knopfel [3], for the 2. Problem definition
upper bound case. Sloan and co-workers have since
developed these methods and have used then to study a The layout of the plane strain heading stability prob-
range of stability problems in geomechanics. Indeed, lem is shown in Fig. 1. The heading has a height, D and
part of this paper updates an earlier study described in cover C. The heading is lined with a smooth, rigid liner
Sloan and Assadi [4] using recent developments in the along its length. (It is important to stress again that this
solution algorithms of the finite element limit proce- problem is not the same as a plane strain circular tunnel
dures to draw more accurate bounds on stability para- which represents an infinitely long unlined circular tun-
meters. This paper also extends previous work into nel. The problem studied here is one of an infinitely long
inhomogeneous soils. flat wall, of height D.) The presence of the liner implies
The stability of underground openings in soft ground the presence of reaction forces. These are however of no
has been of interest to researchers since the late 1960’s, consequence in the numerical formulation adopted here
although the literature on the subject is surprisingly as the liner is modelled as infinitely strong. The face of
sparse. A programme of (largely) experimental research the heading is free to move and is subject to a normal
at Cambridge was undertaken in the 1970’s, culminating stress representing an internal pressure
T. This pressure
in the work of Mair [5] who used centrifuge testing to could be provided by compressed air. The ground sur-
model two and three-dimensional models of tunnels in face is horizontal and subject to a vertical surcharge
s.
clay. Some of the results of Mair’s study, and of other For all analyses presented here, these stresses are taken
work from Cambridge at the time, are included in the as positive when directed into the heading face (
T) or
paper by Davis et al. [6]. The latter also presents a range vertically downwards (
s).
of solutions based on bound theorems, for unlined The ground around the excavation is modelled as a
plane strain tunnels, circular headings as well as the rigid plastic Tresca material with constant unit weight .
plane strain heading. Some of these solutions are used This material has a single strength property, undrained
below to check the numerical results. shear strength, and is the same as the Mohr–Coulomb
 C.E. Augarde et al. / Computers and Geotechnics 30 (2003) 419–430 421

criterion for the case of zero friction angle. The lem). Therefore, only the difference between
S and
T
undrained shear strength of the soil in the analyses needs to be considered. In the case of the upper bound
described here is permitted to vary linearly with depth, theorem, the external power expended by the loads and
according to the self-weight of the deforming soil mass is given by
ð ð ð
cu ðzÞ ¼ cu0 þ z ð1Þ S T
Pext ¼
S vn dA 
T vn dA þ  vdV ð5Þ
AS AT V
where cu0 is the undrained shear strength at the surface
and  is the rate of change of shear strength with depth where vSn is the downwards normal velocity at the sur-
z from the surface (as indicated in Fig. 1). (Obviously, a face, vTn is the outward normal velocity on the tunnel
homogeneous soil strength profile is modelled with face and v is the vertical velocity of points within the soil
=0). A linear variation of undrained shear strength mass. AT and AS are the deforming areas on the tunnel
with depth in normally consolidated (nc) clays has been face and at the surface respectively. The last integral on
observed empirically by Skempton [15] and is predicted the right hand side of Eq. (5) is taken over the soil
by Critical State Soil Mechanics [16]. Skempton [15] volume V. Since undrained behaviour is assumed, the
proposes the following relation between undrained soil deforms at constant volume and
shear strength and plasticity index Ip ð ð
S
vn dA ¼ vTn dA ð6Þ
cu ð zÞ AS AT
¼ 0:11 þ 0:0037Ip ð2Þ

v0 ðzÞ
It is then possible to rewrite Eq. (5) in terms of the
where
v0 is the effective vertical stress. Ladd et al. [17] dimensionless groups given above as
give the following expression for the undrained strength  ð  ð

S 
T T D cu0
profile in an overconsolidated (oc) deposit Pext ¼ cu0 vn dA þ vdV ð7Þ

 cu0 AT cu0 D V
cu ðzÞ=
v0 ðzÞ OC

 ¼ ðOCRÞ0:8 ð3Þ Engineers faced with a stability problem of this type
cu ðzÞ=
v0 ðzÞ NC are usually working with a given heading configuration
and a soil profile determined from site investigation. In
where OCR is the overconsolidation ratio. These rela- terms of the dimensionless groups given above, this can
tions are used to determine the range of values of the be restated as the determination of values of the 
para-
parameter  used in the parametric study presented later D C D
meter ð
S 
T Þ=cu0 given values of ; ; . In
in the paper. The use of a simple rigid–plastic material cu0 D cu0
model for the soil in this problem is necessary as the most practical situations the heading will be unpres-
bound theorems of plasticity underlie the numerical surised (
T=0), in which case the results from the sta-
procedures used in this paper. bility analysis will be the surcharge load parameter

S =cu0 .
An alternative approach, adopted by many previous
3. Problem variables and dimensional analysis researchers, is to assess stability in terms of a ‘‘stability
ratio’’ or overload factor (usually denoted N), by adding
Seven variables model instances of the plane strain a term that represents an initial overburden stress to the
heading
 problem outlined
 above, namely the group load parameter ð
S 
T Þ=cu0 . Broms and Bennermark

T ;
s ; C; D; cu0 ; ;  . A convenient set of dimension- [19] introduced this approach, giving the following defi-
less groups that follows the requirements for dimen- nition for a homogeneous soil
sional analysis (usefully described in [18]) is the
s 
T þ  ðC þ D=2Þ
collection N¼ ð8Þ
  cu0

T
S D C D
; ; ; ; ð4Þ which is also adopted by Davis et al. [6] and Atkinson
cu0 cu0 cu0 D cu0
and Mair [20], although the latter rename the para-
It is possible to replace the first two by a single group meter, Tc. This approach appears to be a way of redu-
ð
S 
T Þ=cu0 because all calculations presented here cing the complexity of the final results. It does, however,
assume undrained behaviour. To justify this reduction, lead to problems as ð
S 
T Þ=cu0 itself depends on the
it is necessary to consider the bound theorems indivi- parameter ðDÞ=cu0 . (This will be demonstrated from
dually. A statically admissible stress field satisfying the first principles for the case of an upper bound solution
requirements of the lower bound theorem is also later in the paper). Additionally, it is not clear what is
admissible for any addition of isotropic stress, since an appropriate choice for the shear strength denomi-
undrained strength is independent of total mean normal nator in Eq. (8) for an inhomogeneous soil. Given these
stress [6], (although such a combined stress field would difficulties, the results from the analyses described in
not satisfy the stress boundary conditions in this prob- this paper are presented in terms of the load parameter
422 C.E. Augarde et al. / Computers and Geotechnics 30 (2003) 419–430

ð
S 
T Þ=cu0 . This approach follows both the require- linearising the yield surface for very large (i.e. finely
ments of dimensional analysis and of rigorous plasticity discretised) two-dimensional problems leads, however,
theory. to an excessive number of linear inequalities. The linear
programming problem that is produced is consequently
slow to solve using traditional methods (such as the
4. Finite element formulation of the bound theorems simplex method).
An alternative approach has been developed recently
The finite element formulations of the bound theo- where the yield function is left non-linear and the prob-
rems, as developed by Sloan for use in geomechanics lem is recast as a non-linear programming problem
problems, are described in detail in a number of refer- [22,23] to
ences (e.g. [12,21]). Only a brief description of the pro-
Maximise cT 
cedures is therefore given here, and the reader is referred
to the original publications for full details of the for- Subject to A ¼ b
mulations. It is important to note that the formulations fi ðÞ 4 0 i ¼ f1; . . . ; Ng ð9Þ
are not variations of the displacement finite element
method, but employ the same idea of discretisations of a where  is the vector of nodal stresses, c is a vector of
domain to obtain solutions. objective function coefficients to transform the stresses
The lower bound theorem requires a statically to the optimised load, A and b are a matrix and vector
admissible stress field that obeys the yield criterion respectively, of equality constraint coefficients derived
throughout the problem domain. Conventional analy- from equilibrium and the stress boundary conditions, fi
tical approaches seek to divide the domain into regions is the yield function for node i and N is the number of
in which statically admissible stress fields are defined. nodes. The algorithm used to solve this system is
Between regions, discontinuities in the normal stresses described in detail in Lyamin and Sloan [23] and will
in the direction of the discontinuity are permitted. not be repeated here. Recasting the problem in non-lin-
These allow the stress boundary conditions for the ear form, as described above, leads to a much faster
problem to be satisfied (for applied loads and pre- solution. Early use of the formulation has indicated a
scribed displacements). The finite element formulation 50-fold reduction in CPU time, as compared to the lin-
of the lower bound theorem has the same goal but the ear programming approach [22]. This gain in efficiency
regions are replaced by three-noded triangular finite allows much larger two-dimensional problems to be
elements within which stress fields can vary linearly. solved and is the technique used to obtain the lower
The nodal variables associated with each element are bound results presented later in this paper.
the stresses at the nodes. While elements are notionally A similar finite element approach can be taken with
connected at nodes, the nodal stresses are associated the upper bound theorem. A conventional upper bound
with that element only, unlike displacement finite ele- analysis proceeds by seeking a collapse mechanism that
ments. Stress discontinuities are permitted between each is kinematically admissible. This is a velocity field cov-
element in the domain. Special ‘‘extension’’ elements ering the domain that satisfies the velocity boundary
are also placed on the boundary to model an unboun- conditions, and the plasticity flow rule (which is asso-
ded domain, and hence produce a rigorous lower ciated in the case of the Tresca criterion). For problems
bound. The calculation then proceeds as an optimisa- such as the plane strain heading, a mechanism may
tion of the domain stress field, where the constraints are consist of rigid blocks of soil, moving at differing velo-
those imposed by equilibrium, the stress boundary cities. Internally, power is dissipated in the dis-
conditions and the yield criterion. The objective func- continuities between the blocks. This power is equated
tion, to be maximized, is the integral of the normal to the external power expended by the external loads
stresses over some part of the domain. (For the case of (
T and
S) and the self-weight of the soil g, to obtain
the plane strain heading, this is difference between the an upper bound solution.
surcharge
S and the tunnel pressure
T). The Tresca The original formulation of the upper bound theorem
yield criterion leads to a set of non-linear constraints on [24] proceeds similarly to the lower bound formulation,
the nodal stresses. To form a linear programming with rigid regions replaced by three-noded triangular
problem, this non-linearity is dealt with by replacement elements over which the velocity is allowed to vary lin-
of the non-linear constraints with linear inequalities early. Each node has two unknown velocities specific to
that maintain a rigorous lower bound solution [4,21]. that element. Velocity discontinuities are permitted at
This can be visualised as replacing the circular Tresca specified locations in the finite element mesh, for which
surface (in Cartesian stress space) with an n-sided the sign of shearing must be specified by the analyst.
internal prism. Unlike a rigid-block mechanism approach, internal
The procedure outlined above has proved successful power may also be dissipated in plastic deformation of
for many two-dimensional problems. The process of the continuum. To ensure kinematic admissibility, each
 C.E. Augarde et al. / Computers and Geotechnics 30 (2003) 419–430 423

element therefore also has a specified number of plastic parameters in the group [Eq. (4)]. For both upper and
multiplier rates to ensure plastic deformation obeys the lower bound analyses, the optimum value of the load
flow rule for the yield criterion used. The formulation parameter is found by setting
S to zero and optimising
also ensures that zero deformation occurs in regions for
T alone. For the upper bound a uniform tunnel
where computed stresses lie within the yield surface. An pressure is optimised by imposing the loading condition
optimisation problem then evolves with the objective at the tunnel face equivalent to
function (to be minimized) being the internal power ð
dissipation. The constraints arise from the need to vTn dA ¼ 1 ð10Þ
AT
satisfy kinematic admissibility (for continuous velocity
fields between discontinuities, the plastic flow rule and This does not restrict the upper bound solution to
velocity boundary conditions). In a similar fashion to have a constant velocity profile over the heading height,
the lower bound formulation, the problem can be solved an aspect of the finite element upper bound solutions
by linear programming methods, by making the yield which will be shown to have important consequences
surface linear in Cartesian stress space. for the use of stability ratios, such as the one expressed
The original formulation has a number of short- in Eq. (8).
comings that have been addressed in recent years. Results are presented for values of the weight para-
Firstly, the need to define the location and nature of meter ðDÞ=cu0 of 0, 1, 2 and 3 and for five values of the
velocity discontinuities a priori has been removed by strength inhomogeneity parameter ðDÞ=cu0 between
techniques described in Sloan and Kleeman [25] with a zero and unity. These values are chosen to cover a wide
novel formulation that permits velocity discontinuities range of normally consolidated and overconsolidated
along every element edge in the mesh. This has also soil conditions, following the relations given in Eqs. (2)
removed the need to adopt fixed element patterns to and (3).
ensure incompressible material behaviour. A typical mesh used to analyse a plane strain heading
More recently, Lyamin and Sloan [26] have developed with C=D ¼ 2 is shown in Fig. 7. Similar meshes are
an upper bound finite element formulation based on used for upper and lower bound analyses as the new
non-linear programming. The new finite element for- formulation, based on non-linear programming, allows
mulation of the upper bound theorem uses the same the use of a very large number of two-dimensional ele-
linear velocity elements as in the original formulation. ments. It is therefore not necessary to experiment with a
Unlike the original formulation, however, each element range of meshes to arrive at narrow bounds. The mesh
is associated with a constant stress field and a single shown in Fig. 7 has 27 936 nodes and 9312 triangular
plastic multiplier rate, as the yield surface is not line- elements. The stress boundary conditions (for the lower
arised in this formulation. The optimisation problem bound analyses) ensure zero shear and normal stress at
can then be cast in terms of the nodal velocities and the the surface, zero shear stress on the heading face and
element stresses. Once again, the solution algorithm along the lined section of the heading. Extension ele-
used for this non-linear programming problem is fully ments are also included along the soil domain bound-
described elsewhere [26] and will not be repeated here. aries for the lower bound analyses. An upper bound
The new procedure is much quicker than the original analysis using the mesh in Fig. 7 includes velocity
linear programming formulation, permitting very fine boundary conditions to ensure zero velocity at the
discretisation for two-dimensional problems. boundaries, and zero vertical velocity along the lined
A previous study of the stability of a plane strain section of the heading. As indicated in the description of
heading [4] was restricted to the linear programming the upper bound formulation, velocity discontinuities
approach of both lower and upper bound finite element are present between each element in the mesh.
formulations. The new techniques employed here have All results presented here were obtained using an
allowed much finer two-dimensional finite element AMD Athlon processor running at 1200 MHz and
meshes to be used, thus improving the quality of bound generally required between 40 and 480 CPU seconds per
solutions. analysis. These timings are impressive given the com-
plexity of the finite element meshes used in the analyses,
and demonstrate the efficiency of the non-linear pro-
5. Numerical results gramming approach.

The results of over four hundred finite element bound

analyses of the plane strain heading problem, using the 6. Analytical comparisons
methods outlined above, are presented in Table 1. The
results are also presented as dimensionless stability It is important to check the results of any numerical
charts in Figs. 2–6. In all analyses, the load parameter model using existing analytical solutions where these are
ð
S 
T Þ=cu0 is optimised for set values of the other available. In this case, the main source of useful analy-
424 C.E. Augarde et al. / Computers and Geotechnics 30 (2003) 419–430

Table 1
Bounds on the load parameter for stability of an undrained plane strain heading

C D D D D D
¼0 ¼1 ¼2 ¼3
D cu0 cu0 cu0 cu0 cu0

Lower Upper Lower Upper Lower Upper Lower Upper

bound bound bound bound bound bound bound bound

1 0.00 4.00 4.39 2.46 2.89 0.85 1.39 0.74 0.11

1 0.25 5.32 5.59 3.82 4.09 2.29 2.59 0.68 1.09
1 0.50 6.47 6.77 4.98 5.27 3.46 3.77 1.87 2.27
1 0.75 7.60 7.93 6.12 6.43 4.62 4.93 3.04 3.43
1 1.00 8.73 9.09 7.25 7.59 5.76 6.09 4.19 4.59
2 0.00 5.05 5.68 2.40 3.18 0.20 0.68 2.84 1.82
2 0.25 7.93 8.10 5.43 5.60 2.91 3.08 0.37 0.54
2 0.50 10.42 10.66 7.93 8.17 5.43 5.66 2.90 3.14
2 0.75 12.89 13.19 10.41 10.71 7.91 8.21 5.40 5.70
2 1.00 15.35 15.72 12.87 13.24 10.38 10.74 7.87 8.24
3 0.00 5.75 6.50 2.20 3.00 1.40 0.50 5.03 4.00
3 0.25 10.20 10.50 6.70 7.00 3.19 3.48 0.35 0.05
3 0.50 14.23 14.66 10.74 11.16 7.24 7.66 3.72 4.14
3 0.75 18.23 18.79 14.75 15.30 11.25 11.80 7.75 8.30
3 1.00 22.22 22.91 18.74 19.42 15.25 15.93 11.75 12.43
4 0.00 6.25 7.21 1.71 2.71 2.86 1.79 7.49 6.29
4 0.25 12.52 12.82 8.03 8.32 3.52 3.81 1.01 0.72
4 0.50 18.28 18.72 13.79 14.23 9.29 9.72 4.78 5.21
4 0.75 24.01 24.59 19.52 20.10 15.03 15.60 10.53 11.09
4 1.00 29.72 30.45 25.24 26.00 20.75 21.46 16.25 16.96
5 0.00 6.70 7.70 1.15 2.20 4.48 3.30 10.11 8.80
5 0.25 14.81 15.19 9.32 9.69 3.81 4.18 1.71 1.35
5 0.50 22.40 22.98 16.92 17.48 11.42 11.98 5.91 6.47
5 0.75 29.96 30.72 24.48 25.23 18.99 19.73 13.49 14.23
5 1.00 37.51 38.46 32.03 32.97 26.54 27.47 21.05 21.97
6 0.00 7.02 8.12 0.41 1.62 6.10 4.88 12.90 11.38
6 0.25 17.14 17.58 10.65 11.08 4.14 4.57 2.39 1.95
6 0.50 26.67 27.36 20.18 20.87 13.69 14.36 7.18 7.86
6 0.75 36.17 37.11 26.69 30.61 23.20 24.11 16.70 17.61
6 1.00 45.66 46.84 39.17 40.34 32.69 33.85 26.19 27.35
7 0.00 7.33 8.49 0.32 0.99 7.90 6.51 15.53 14.01
7 0.25 19.47 20.02 11.98 12.50 4.47 5.00 3.05 2.52
7 0.50 31.01 31.87 23.53 24.37 16.03 16.87 8.53 9.36
7 0.75 42.53 43.70 35.04 36.20 27.55 28.71 20.06 21.21
7 1.00 54.02 55.51 46.54 48.02 39.06 40.53 31.56 33.03
8 0.00 7.46 8.83 1.21 0.33 9.80 8.17 18.43 16.67
8 0.25 21.82 22.46 13.33 13.96 4.82 5.45 3.70 3.06
8 0.50 35.44 36.49 26.95 27.99 18.46 19.49 9.96 10.99
8 0.75 49.02 50.48 40.55 41.99 32.06 33.49 23.56 24.99
8 1.00 62.60 64.46 54.12 55.97 45.64 47.47 37.14 38.97
9 0.00 7.57 9.08 2.16 0.42 11.75 9.92 21.39 19.42
9 0.25 24.19 24.95 14.70 15.46 5.19 5.95 4.33 3.57
9 0.50 39.94 41.21 30.46 31.72 20.96 22.22 11.46 12.71
9 0.75 55.66 57.43 46.17 47.94 36.68 38.44 27.19 28.94
9 1.00 71.37 73.64 61.88 64.15 52.40 54.65 42.90 45.16
10 0.00 7.70 9.32 3.16 1.18 13.75 11.68 24.39 22.18
10 0.25 29.58 27.48 16.09 16.98 5.58 6.48 4.94 4.03
10 0.50 44.51 46.03 34.02 35.54 23.52 25.04 13.02 14.53
10 0.75 62.40 64.54 51.92 54.05 41.43 43.55 30.93 33.06
10 1.00 80.29 83.06 69.80 72.56 59.32 62.06 48.82 51.57

tical checks on the numerical results comes from Ref. can be obtained by adapting the solution for a v-not-
[6]. They present analytical solutions for lower and ched bar under tension, given in [27]
upper bounds of the load parameter ð
S 
T Þ=cu0 for 

S 
T C
undrained soil with constant strength with depth. For 5 2 þ 2log þ1 ð11Þ
cu0 D
the case of a weightless soil ðDÞ=cu0 ¼ 0, a lower bound
 C.E. Augarde et al. / Computers and Geotechnics 30 (2003) 419–430 425

Fig. 2. Bounds on load parameter for ðDÞ=cu0 ¼ 0. Fig. 4. Bounds on load parameter for ðDÞ=cu0 ¼ 0:5.

Fig. 5. Bounds on load parameter for ðDÞ=cu0 ¼ 0:75.

Fig. 3. Bounds on load parameter for ðDÞ=cu0 ¼ 0:25.

For the case of a soil with self-weight, ðDÞ=cu0 6¼ 0, written

Davis et al. [6] suggest it is acceptable to add a hydro-  

S 
T C D C 1
static stress field to the weightless case to obtain a solu- 5 2 þ 2log þ1  þ ð12Þ
cu0 D cu0 D 2
tion. This is not a strict lower bound, as the tunnel
pressure now varies linearly with depth, rather than Davis et al. [6] suggest this is, however, a safe solution
being constant over the depth; the condition used to based on their results for the other stability problems.
derive the original weightless lower bound solution. If These lower bound solutions are by inspection, also

T is taken instead to represent the mean tunnel pres- lower bound solutions for the case of inhomogeneous
sure over the heading height then this solution can be soil, albeit very poor ones.
426 C.E. Augarde et al. / Computers and Geotechnics 30 (2003) 419–430

S 
T

D ¼
c u0
cu0 ¼0

 
2C sin2 sin4
D sin1 sin3 sin5 cosð1 þ 2 þ 3 þ 4 þ 5 Þ

þ cot1 þ 2cot2 þ 2cot3 þ 2cot4 þ cot5

 
cosð1 þ 2 þ 3 þ 4 Þ D C 1
þ  þ ð13Þ
sin5 cosð1 þ 2 þ 3 þ 4 þ 5 Þ cu0 D 2

This mechanism can also be used for the case of

varying soil strength with depth, in which case the
expression for the load parameter is

S 
T

S 
T

D

¼
þ
cu0 D
6¼0 cu0 D
¼0 cu0
cu0 cu0

0 1
 cot1 þ 2cot2 þ 2cot3 þ 2cot4 þ cot5  2
Fig. 6. Bounds on load parameter for ðDÞ=cu0 ¼ 1:0. CB C
C
@ cosð1 þ 2 þ 3 þ 4 Þ A þ
D þ D
sin5 cosð1 þ 2 þ 3 þ 4 þ 5 Þ
An analytical upper bound solution can be obtained 
from the five variable mechanism shown in Fig. 8 [4]. sin2 sin4
This is a rigid block mechanism where the internal sin1 sin3 sin5 cosð1 þ 2 þ 3 þ 4 þ 5 Þ
power dissipation takes place along the interfaces 0 1
sinð1 þ 2 Þ
between blocks only. The mechanism may appear to be 1 B  cosð1 þ 2 Þ C
þ @ sin1 A
incompatible in that the tip of the triangle at the face of 2sin2
the heading has to penetrate the soil beneath. This dif- ðcos1 þ sin1 ð2cot2 þ 2cot3 þ cot4 ÞÞ
ficulty is dealt with by noting that if the tip deforms 0 cosð þ  þ  þ  Þ 1
1 2 3 4
plastically to accommodate this movement the energy ðsin3 ðcot4 þcot5 ÞÞ
dissipated is of second order to that dissipated in the B sin4 C
B C
interfaces between blocks for a given movement, and 1 B B sin3 cosð1 þ 2 þ 3 þ 4 Þ C
C
can be neglected in the analysis [28]. Similar mechan-
þ Bþ C
2sin2 B sin5 cosð1 þ 2 þ 3 þ 4 þ 5 Þ C
B C
isms are also used by Davis et al. [6]. For the case of a @ sinð þ  Þ A
3 4
constant undrained strength ðDÞ=cu0 ¼ 0, equating þ
external and internal power dissipation and rearranging sin4
gives the following expression for the load parameter, ð14Þ

Fig. 7. Typical finite element mesh for C/D=2.

 C.E. Augarde et al. / Computers and Geotechnics 30 (2003) 419–430 427

uniform outward velocity field vTn ¼ is imposed on the

heading face then, using Eq. (15) the volume integral in
Eq. (7) is
ð 
D
v dV ¼ D C þ ð16Þ
V 2

Substituting this result into the expression for the

external power expended [Eq. (7)] and equating to the
internal power, Pint gives
  

S 
T D D
Pint ¼ cu0 D þ cu0 D C þ ð17Þ
cu0 cu0 2

Hence we can write the result for soil with self-weight

in terms of the solution for the weightless soil as
Fig. 8. 5-variable rigid block mechanism for analytical upper bound.
   

S 
T
S 
T D C 1
¼  þ ð18Þ
Upper bounds can then be found by numerically cu0 D
6¼0 cu0 D
¼0 cu0 D 2
cu0 cu0
optimising Eqs. (13) and (14). Values thus obtained are
included, where appropriate, with the plots of the finite This is similar in form to the upper bound solutions
element results, in Figs. 2–6. A limited number of cases using the rigid block mechanism in Fig. 8 [Eqs. (13) and
only, for the analytical solutions given above, are inclu- (14)]. As is now evident, however, the second term in
ded in these figures, because the finite element results Eq. (18) changes according to the chosen velocity dis-
usually provide much better bounds and also for the tribution on the heading face. If, for example, the velo-
sake of clarity. city vTn varies linearly from at the top of the heading to
It is notable that upper bound solutions for soils zero at the base, as shown in Fig. 9, the  second
 term on

where D=cu0 > 0 can be derived from the weightless D C 1
the right hand side of Eq. (18) becomes þ .
solutions by subtraction of a term including two of the cu0 D 3
dimensionless groups.
This apparently convenient result was derived inde-
pendently for any upper bound solution of this problem 7. Discussion
by Sloan and Assadi [4]. It appears to make the use of a
stability ratio N, as defined in Eq. (8), rather attractive The numerical results presented in Figs. 2–6 show
as the term deducted from the upper bound solutions very close bounds on the load parameter ð
S 
T Þ=cu0
for D=cu0 > 0 is equal to the term added to the load for a wide range of problem configurations. The close-
parameter to obtain N. One stability parameter there- ness of the bounds, and the feasibility of conducting
fore covers all cases of soil with and without self-weight, such a large study, can be attributed to the improve-
for a particular configuration of heading. This is
applicable only in certain cases, however, as will be
demonstrated below.
The external power term required for the upper
bound analysis [Eq. (7)] includes an integral of the vertical
velocity v of points in the soil mass. Due to the incom-
pressibility of the soil it is possible to replace this integral
with a simpler term that can be evaluated in terms of the
geometry of the problem and the nature of the horizontal
velocity profile applied to the tunnel face as follows:
ð ð CþD
v dV ¼ zvTn dz ð15Þ
V C

where vTn is the horizontal outward velocity at the face

of the heading (as before) and z is the vertical coordi-
nate, as shown in Fig. 1. [A detailed proof of the deri-
vation of Eq. (15) is given by Sloan and Assadi [4]]. If a Fig. 9. Linear variation of velocity over heading depth.
428 C.E. Augarde et al. / Computers and Geotechnics 30 (2003) 419–430

Fig. 10. Load parameters replotted for varying weight parameter.

ments in the algorithms used to solve the non-linear the load parameter for weightless soils conditions is
programming problem, resulting from the finite element only 1.6 times that at the highest weight parameter. The
bound formulation. In the majority of cases the bound opposite effect happens when the heading is located at a
solutions bracket the real solution to within
4%. As greater depth. Fig. 10 shows the variations in the load
the soil weight parameter D=cu0 is increased, the gap parameter as weight increases, from C=D ¼ 1 to
between the bounds widens slightly, as does the calcu- C=D ¼ 3. The plots for the shallow heading are closely
lation time to obtain the results. spaced, in contrast to those for the deeper heading. This
A negative value of the load parameter ð
S 
T Þ=cu0 confirms, perhaps obviously, that the determination of
indicates the heading configuration is inherently the variable  assumes greater importance as the head-
unstable; i.e. a positive heading pressure is required to ing depth increases.
prevent collapse, with zero surcharge. This situation Fig. 10 also shows a linear variation in load para-
occurs for a limited range of headings, generally for meter with weight parameter, for both upper and lower
heavy soils, with ðDÞ=cu0 5 2 where there is little or no bound results. This follows the predictions from the
increase of strength with depth. For all cases where analytical upper bound solutions discussed above,
there is appreciable inhomogeneity where solutions for soils with self-weight are found by
[withðDÞ=cu0 5 0:5], all headings are stable for zero subtracting a term dependent on D=cu0 from the
surcharge and zero heading pressure. weightless load parameter.
The analytical and rigid-block bound solutions [Eqs. Fig. 11a shows the velocity vectors at nodes in an
(11)–(14)] improve on the finite element results for one upper bound mesh for a heading where
case only (C=D ¼ 2; D=cu0 ¼ D=cu0 ¼ 0) and in C=D ¼ 1; D=cu0 ¼ 3; D=cu0 ¼ 0. Horizontal
general provide much poorer solutions. For shallow components at the tunnel face (i.e. values of vTn ) are
headings with low soil weight, however, the analytical clearly non-uniform over the tunnel height. In this case,
upper bound solutions [Eqs. (13) and (14)] are very close therefore, the stability number given in Eq. (8) would
to those given by the finite element methods. The lower not be valid. This demonstrates that the safe use of a
bound solution for soils with self-weight, suggested by stability number, N, requires prior knowledge of the
Davis et al. [6] are indeed safe, as can be seen in Fig. 2. velocity profile over the heading height. Similar conclu-
This solution also appears to follow the same trend as sions may be drawn for other underground stability
the finite element lower bound solutions. problems in geomechanics. Non-uniform horizontal
Another noticeable feature of the results is the relative velocity profiles at the heading face, obtained in the
importance of soil weight and strength profile. As the finite element solutions, are shown to produce better
latter is increased from 0.5 to unity, the importance of upper bound solutions and to invalidate the use of the
the soil weight profile decreases considerably. When stability ratio as described by Eq. (8). There are also
ðDÞ=cu0 ¼ 0:5, the load parameter is reduced three-fold interesting differences in the patterns of these velocities
as the weight parameter changes from zero to 3. For the as soil weight increases and as headings are deepened.
steepest increase in strength with depth ðDÞ=cu0 ¼ 1:0, The effect of increasing soil weight is to make the velo-
 C.E. Augarde et al. / Computers and Geotechnics 30 (2003) 419–430 429

Fig. 11. Velocity vectors for finite element upper bound analyses
C=D ¼ 1; D=cu0 ¼ 0: (a) D=cu0 ¼ 0, (b) D=cu0 ¼ 3.

city profile over the heading depth less uniform. Fig. 11b
shows the profile for C=D ¼ 1; D=cu0 ¼
0; D=cu0 ¼ 0 and indicates relatively uniform velo-
city vectors over the heading depth. Fig 10a, in contrast
shows the velocity vectors for the same analysis with
D=cu0 ¼ 3. The profile, as highlighted above, is clearly
Fig. 12. Velocity vectors for finite element upper bound analyses
less uniform with greater velocities to the base of the
C=D ¼ 5; D=cu0 ¼ 0: (a) D=cu0 ¼ 0, (b) D=cu0 ¼ 3.
heading and reduced velocities at the top. This pattern
is evident for all depths of heading studied here. Fig. 12a
and b show a similar effect for the case where C=D ¼ 5. 8. Conclusion
Clearly, increased soil weight leads to a larger variation
in vertical stress across the heading and this may lead to The stability of a plane strain heading in undrained
earlier yield and a greater contribution to continuum conditions has been investigated using finite element
plastic flow in the base of the heading. Fig. 12 also bound methods. Improvements to solution algorithms
shows the finite element upper bound to depart from the have permitted very close bounds to be drawn in most
rigid block mechanism of Fig. 8 once the heading is at cases, providing useful charts for those assessing the sta-
depth. Non-zero velocities are predicted for nodes bility of this type of underground opening. It is impor-
above the lined section of the tunnel for this case, tant to remember that, in practice, the drained stability
C=D ¼ 5, and deeper cases. For the case of C=D ¼ 1, should also be considered, an area not covered in this
velocity discontinuities appears to line up vertically paper. The use of a universal stability ratio [19] has been
from the top of the heading, similar to the rigid block shown not to be advisable, depending as it does on the
mechanism of Fig. 8. nature of collapse which cannot be determined a priori.
430 C.E. Augarde et al. / Computers and Geotechnics 30 (2003) 419–430

Acknowledgements [13] Sloan SW, Assadi A. The stability of tunnels in soft ground. In:
Houlsby GT, Schofield AN, editors. Predictive soil mechanics.
London: Thomas Telford 1993, p. 644–63.
The work presented in this paper was carried out
[14] Lyamin AV, Sloan SW. Stability of a plane strain circular tunnel
while the first author was visiting the University of in a cohesive-frictional soil. In: Proceedings of the J.R. Booker
Newcastle, Australia. During this period he was sup- Memorial Symposium, Sydney, 2000. p. 139–53.
ported by an IREX Fellowship, awarded by the Aus- [15] Skempton AW. The planning and design of the new Hong Kong
tralian Research Council. airport. Discussion. Proc Inst Civil Eng 1957;7:305–7.
[16] Atkinson JH, Bransby PL. The mechanics of soils: an intro-
duction to critical state soil mechanics. London: McGraw-Hill;
1978.
References [17] Ladd CC, Foote R., Ishihara K, Schlosser F, Poulos HG. Stress
deformation and strength characteristics. In: Proc. 12th Int.
[1] Salencon J. An introduction to the yield design theory and its Conf. Soil Mechanics and Foundation Engineering, Vol. 2, 1977.
application to soil mechanics. European Journal of Mechanics A/ p. 421–94.
Solids 1990;9(5):477–500. [18] Butterfield R. Dimensional analysis for geotechnical engineers.
[2] Lysmer J. Limit analysis of plane problems in soil mechanics. Géotechnique 1999;49:357–66.
ASCE Journal of the Soil Mechanics and Foundations Division [19] Broms BB, Bennermark H. Stability of clay in vertical openings.
1970;96(SM4):1311–34. Journal of the Geotechnical Engineering Division, ASCE 1967;
[3] Anderheggen E, Knopfel H. Finite element limit analysis using 193:71–94.
linear programming. International Journal of Solids and Struc- [20] Atkinson JH, Mair RJ. Soil mechanics aspects of soft ground
tures 1972;8:1413–31. tunnelling. Ground Engineering 1981;14:20–4.
[4] Sloan SW, Assadi A. Undrained stability of a plane strain head- [21] Sloan SW. Lower bound limit analysis using finite elements and
ing. Canadian Geotechnical Journal 1994;31:443–50. linear programming. International Journal for Numerical and
[5] Mair, RJ. Centrifugal modelling of tunnel construction in soft Analytical Methods in Geomechanics 1988;12:61–77.
clay. PhD thesis, University of Cambridge, 1979. [22] Lyamin AV. Three-dimensional lower bound limit analysis using
[6] Davis EH, Gunn MJ, Mair RJ, Seneviratne HN. The stability of non-linear programming. PhD thesis, Department of Civil, Sur-
shallow tunnels and underground openings in cohesive material. veying and Environmental Engineering, University of Newcastle,
Géotechnique 1980;30:397–416. Australia, 1999.
[7] Mühlhaus H-B. Lower bound solutions for tunnels in two and [23] Lyamin AV, Sloan SW. Lower bound limit analysis using non-
three dimensions. Rock Mechanics and Rock Engineering 1985; linear programming. International Journal for Numerical Meth-
18:37–52. ods in Engineering 2002;55:573–611.
[8] Atkinson JH, Potts DM. Stability of a shallow circular tunnel in [24] Sloan SW. Upper bound limit analysis using finite elements and
cohesionless soil. Géotechnique 1977;27:203–15. linear programming. International Journal for Numerical and
[9] Leca E, Dormieux L. Upper and lower bound solutions for the Analytical Methods in Geomechanics 1989;13:263–82.
face stability of shallow circular tunnels in frictional material. [25] Sloan SW, Kleeman PW. Upper bound limit analysis with dis-
Géotechnique 1990;40:581–606. continuous velocity fields. Computer Methods in Applied
[10] Eisenstein Z, Samarasekara L. Stability of unsupported tunnels Mechanics and Engineering 1995;127:293–314.
in clay. Canadian Geotechnical Journal 1992;29:609–13. [26] Lyamin AV, Sloan SW. Upper bound limit analysis using non-
[11] Anagnostou G, Kovari K. Face stability conditions with earth- linear programming. International Journal for Numerical and
pressure-balanced shields. Tunnelling and Underground Space Analytical Methods in Geomechanics 2002;26:181–216.
Technology 1996;11:165–73. [27] Ewing DJF, Hill R. The plastic constraint of V-notched tension
[12] Sloan SW, Assadi A. Undrained stability of a square tunnel bars. Journal Mech Phys Solids 1967;15:115–24.
whose strength increases linearly with depth. Computers and [28] Chen WH. Limit analysis and soil plasticity. Amsterdam:
Geotechnics 1991;12:321–46. Elsevier; 1975.