3

Centrifuge modelling: practical

considerations
R.PHILLIPS

3.1
Introduction
As centrifuge modelling is an experimental science, there are practical
considerations to be made in the design and conduct of any centrifuge model
test. The primary objective of the model test is to obtain high-quality and
reliable data of a physical event. The model test should therefore be based on
experience using proven apparatus and techniques. For centrifuge modelling to
advance, gradual development of these apparatus and techniques is required.
The model test is a multi-disciplinary activity, generally involving
mechanical, hydraulic, electronic and control engineering as well as geotechnics.
The test objectives are constrained by these activities. Some of these constraints
are becoming less severe with the implementation of new technology. The
centrifuge model test is also constrained by more mundane restrictions such as
resource, payload capacity, increased self-weight and communication to the
remote centrifuge environment.
The art of centrifuge modelling is to minimise the effects of these constraints
while maximising the quality of the geotechnical data obtained. This chapter
will assist the reader in understanding some of the practical considerations and
review some of the apparatus and techniques currently available to the
centrifuge modeller.
Centrifuge modelling has been used extensively for geotechnical studies, and
is also being applied to more general civil engineering studies including rock
mechanics, hydraulics, structures and cold regions. Most of the considerations
presented in this chapter are also applicable to these research areas.

3.2
Geotechnical centrifuges
There are many different types of centrifuges, used for example in material
processing, aeronautics and motion simulation. The geotechnical centrifuge is
characterised by its rugged nature, large payload capability and low-speed
Figure 3.1 Types of beam centrifuge platform: left, fixed; centre, restrained; right,
swinging.

operation. Geotechnical centrifuges can be sub-divided into two main classes:

beam centrifuges and drum centrifuges.
The beam centrifuge generally comprises a central spindle supporting a pair
of parallel arms which hold the platform on which the test package is placed.
The radial force acting on the central spindle should be minimised as shown in
equation (3.1):
(3.1)
where m is the centre of mass of a component at a vector r from the rotor axis
and is the rotational speed of the centrifuge. This equation can be simplified
to:
(3.2)
From equation (3.2) the centrifuge can be viewed as a simple mass balance with
the central rotor replaced by a central point support. Balance must be considered
in two orthogonal directions. Traditionally balance has been achieved by
symmetry; such beam centrifuges are called balanced-beam centrifuges. In
recent years, beam centrifuge design has changed and masses at large radii, such
as the platform and test package, are counterbalanced by a larger mass at a
smaller radius. This new generation of beam centrifuges, such as the one at Delft
Geotechnics and those manufactured by Acutronic, are more efficient as the
aerodynamic drag on the rotor assembly is reduced.
Beam centrifuges traditionally rotate in a horizontal plane. The acceleration
field acting on the model is the resultant of the centrifuge acceleration field and
the Earth’s gravitational field. The behaviour of the model will depend on the
orientation of the model on the centrifuge platform to this resultant acceleration
field. Beam centrifuges can be subdivided into three platform types (fixed,
restrained and swinging) as depicted in Figure 3.1.
On the fixed platform the test package is attached to the vertical face plate. On
the centre-line, the resultant acceleration is always effectively inclined to the
platform at n: 1, where n is the centrifuge acceleration at the platform. When the
centrifuge is started or stopped, restraints are necessary to retain the soil and low
shear modulus materials, such as fluids, in the test package. Restraint techniques
have been successfully developed at the University of Manchester and elsewhere
using flexible membranes in conjunction with a vaccum. These restraints need to
be relaxed during the centrifuge test to minimise their effect on the model
behaviour.
Restrained platforms are used, for example, on the Cambridge University and
ISMES beam centrifuges. This platform type eliminates the need for a low g
retention system. At rest, the swinging platform surface is slightly inclined to the
horizontal. As the centrifuge speed increases the platform swings up into the
vertical plane at about 10 g where it is restrained by a kick bar. Further speed
increases cause the increased weight of platform to twist the support linkage so
that the platform seats against, and is carried by, the strong cradle. The
minimum inclination of the resultant acceleration to the platform surface is
about 10:1 (84°) during the swing up process. Above 10 g, the resultant
inclination increases to about n: 1 to the platform surface.
The effect of this inclination at the test acceleration level on the soil model
can be eliminated by either sloping the soil strata within the test package or
more simply by placing a wedge between the test package and the platform
surface. These measures are only fully effective at the test acceleration level.
Swinging platforms were used in the original Russian centrifuges (Pokrovskii
and Fiodorov, 1953). Swinging platforms have been adopted in the new
generation of Acutronic civil engineering centrifuges and elsewhere. The
platform surface will always be normal to the resultant acceleration, provided
the hinge is frictionless and the platform symmetrically loaded. The effects of
hinge friction have been discussed by Xuedoon (1988). There is a slight
variation of vertical stress across a horizontal plane as the model surface is not
parallel to the axis of rotation. A major benefit of these platforms is the ease of
access to the user.
Drum centrifuges are being increasingly used for geotechnical research. Some
of the first drum centrifuge tests are described by Schofield (1978). The drum
centrifuge can be considered as having a fixed platform and it is not easy to
retain models in the drum. Models can be constructed in the drum during
rotation due to the continuous circumferential extent of the soil sample. Pore
suction can be used to hold these models in place while the centrifuge is stopped
to instrument the model and place any actuators. Some drum centrifuges rotate
in a vertical plane, and the radial acceleration component on the model then has
a cyclic component due to the Earth’s gravity. This cyclic component can be
used to create water movement in model tests. The Earth’s gravity is beneficial
during model making when operating in this mode. The new mini-drum
centrifuge under development at Cambridge University has the ability to rotate
from a horizontal to vertical plane during centrifuge rotation.
Figure 3.2 Movement seen around tunnel heading. Left, full model; right, half model.

3.3
Containers
In the drum centrifuge the model is normally contained within the drum itself. In
beam centrifuges, the model must be safely held within a container. The
geotechnical centrifuge model test is normally a simulation of the behaviour of
an infinite half space with a localised perturbation. The container boundaries
should replicate the behaviour of the far field half-space.
In models of static events, this generally requires creating one-dimensional
consolidation boundary conditions using ideally frictionless vertical walls of
high lateral stiffness to prevent significant lateral soil displacement. For
modelling dynamic events, particularly earthquakes, the boundary conditions are
more complex with prevention of energy reflections while maintaining the
correct dynamic shear stiffness and permitting the evolution of complementary
shear stresses (Schofield and Zeng, 1992). Other boundary condition
requirements may include thermal control for cold-regions’ problems and
hydraulic control for environmental problems. Boundary conditions at the top
and bottom of the model must also be considered.
The half space of interest may include vertical planes of symmetry. These
planes can be replaced by low-friction stiff surfaces, such as the side-walls of
the container walls to reduce the size of the scale model. Sub-surface
deformation patterns along these planes can be visualised if these frictionless
surfaces are transparent, as depicted in Figure 3.2.
Practically, the container walls will be frictional. The coefficient of friction
can be estimated from direct shearbox tests of the boundary conditions. For clay
tests, side-wall friction has been reduced by lubricating the smoothly coated stiff
walls with a water-resistant grease. For sand testing, a hard glass sheet can be
placed between the sand mass and the wall. For small strain problems in sand,
the frictional resistance can be further decreased using lubricated latex sheets at
the boundaries. However, the frictional resistance may increase since membrane
tensions can become appreciable as boundary shear displacements increase.
For plane strain models, the model could be very narrow. Since side-wall
friction is always present to some extent, plane strain models should be
sufficiently wide so that side-wall friction is not a significant proportion of the
resisting forces. The effect of side-wall friction can be further reduced by taking
measurements along the centre-line of the model. Similar considerations are
required in one-dimensional experiments and large strain events, such as
consolidation.
Containers for two- and three-dimensional studies should be about twice as
long as the soil depth they contain to minimise boundary effects. The tops of the
container walls are useful for providing support for actuators and interfaces, and
as a datum surface during model making. For static testing, the lateral
displacement of the container wall should be less than about 0.1% of the
retained height of soil to have minimal effect on lateral earth pressures.
In the centrifuge model simulation vertical planes are mapped into the
centrifuge cylindrical coordinate system (r, ,z) as radial planes. During large
strain events, such as consolidation, the arc length between these radial planes
lengthens with increasing radius which would cause lateral straining of the soil
sample. To maintain an overall condition of zero lateral strain, the arc length (r
) should be kept constant. This condition is reasonably approximated by
having the container side-walls parallel.
For general use, circular containers, or tubs, are very versatile due to their
inherent lateral stiffness and consequent light mass. The tubs retain the
maximum soil plan area with the minimum boundary material and are relatively
inexpensive to construct. Also, tubs can be easily sealed to retain pressure.
Rectangular containers, or boxes, are more massive and more expensive to
construct than a tub of the same carrying capacity and stiffness. However, since
centrifuge platforms are usually rectangular, boxes permit the available soil plan
area to be maximised. Transparent side-walls can be used in the box to permit
visualisation of sub-surface events. Sometimes, these boxes are not stiff enough
to retain the soil pressures during the preparation of overconsolidated samples.
This difficulty can be circumvented by consolidating the soil into a liner and
then transferring the soil and the liner into the box. Care must be taken to ensure
that the internal dimensions of the consolidometer and the box are identical,
otherwise lateral straining of the soil sample will occur and the lateral earth
pressures will change. Similar effects will occur when using bulkheads within
the box to retain the soil mass.
The containers should not leak. The container boundaries may require glanded
ports for instrumentation and service channels. The base of the container should
be stiff enough to prevent significant disturbance of the sample when the base is
unsupported during model preparation and mechan ical handling, and the base
should be flat to be evenly supported by the centrifuge platform. The base of the
container should also accommodate selective base drainage of the soil sample as
required.

3.4
Test design
A geotechnical centrifuge test is normally designed to model a generic prototype
situation. As with many other reduced-scale modelling techniques, such as
hydraulic modelling, not every aspect of the prototype behaviour can be
correctly modelled. Attention must be made to model directly those factors
which are expected to dictate the prototype behaviour, such as the effective
stress conditions in the soil. For those factors which cannot be directly modelled,
the modeller must still ensure that the correct class of behaviour is simulated if
possible. A commonly occurring example is modelling the flow of pore fluid
through the soil skeleton. If the same pore fluid and soil are used in the model
and the prototype then the Reynolds number is larger by the scaling factor, n in
the model. Laminar flow conditions in the model soil skeleton can still be
maintained by ensuring that the Reynolds number in the model is less than unity
(Bear, 1972).
Although the model may not be an exact replica of the prototype, it is still a
unique physical event. In prototype terms, the engineer can view the model as
‘the site next door’ where conditions are very similar if not identical to their
own.
In the centrifuge model only those processes which are dominated by
gravitational effects will be automatically enhanced. To verify the effects that
these processes have on the prototype behaviour, the technique of ‘modelling of
models’ can be employed as described by Schofield (1980). Similarities between
the different models can be attributed to these processes. Differences in
behaviour can assist in separating the effects of different processes: Miyake et
al. (1988) modelled the process of soft-clay sedimentation and consolidation and
separated the two effects by modelling of models techniques.
The principle of ‘modelling of models’ was discussed by (for example) Ko
(1988) and is demonstrated in Figure 3.3: the same 10m high prototype could be
modelled at full scale, at 1/10th scale, or at 1/100th scale at points A1, A2 and
A3, respectively. Normally the range of scales used for modelling of models is
narrower than indicated and does not include full-scale tests; more care is then
required to extrapolate the results to prototype scale. The effects of stress and
size must also be considered when comparing tests (Ko, 1988).
For small-size models the effect of particle size is also important. Soil is a
particulate medium. Modellers, for example Fugslang and Ovesen (1988), have
found that at least 30 particles must be in contact with each linear dimension of
the model structure for the observed behaviour to be representative of the
prototype behaviour. Care must be taken before scaling down the particle size in
Figure 3.3 Modelling of models principle. (After Ko, 1988).

the model to ensure that the mechanical properties of the particles are not
changed, including their angularity and crushing strength, as demonstrated by
Bolton and Lau (1988).
The geometric scale factor for the model is selected to fit the prototype
situation under study into the model container with minimal boundary effects.
The choice of scale factor will be constrained by the maximum model size,
which is related to the payload capacity of the centrifuge, and the operational
domain of the centrifuge. In general, the scale factor should be as small as
possible to maximise the size of the model: small models are more difficult to
instrument and more sensitive to the presence of the instrumentation and the
model making procedure. Small models of simple boundary value problems are,
however, valuable for performing parametric studies with multiple models in
one soil sample.
Some prototype situations may be too large for direct centrifuge modelling.
For deep problems, the effective stress levels in the soil can be increased by
downward seepage (Zelikson, 1969). This technique was used by Nunez and
Randolph (1984) in centrifuge model tests of long piles.
The appropriate centrifuge acceleration level is normally identical to the
geometric scaling factor, but may be different when equivalent materials and
partial similarity are required as described by Craig (1993). The centrifuge
acceleration level is not constant but increases linearly with centrifuge radius.
Schofield (1980) showed that the appropriate centrifuge acceleration level
should be selected at one-third the depth of interest in the soil model, and that
provided the overall soil depth did not exceed 10% of the effective centrifuge
radius the error in assuming that this acceleration level is constant with depth is
tolerable.
Ideally soil strata within the soil model should be formed at the same
curvature as that of the centrifuge. The majority of centrifuge model tests are
conducted with level strata. If the package width is about 20% of the centrifuge
radius in the circumferential direction, then a lateral acceleration of 10% of the
centrifuge acceleration will be induced at the outside edge of the level strata.
This effect can be minimised by placing the area of interest along the centre-line
of the strongbox.
The time scaling factors are determined from the appropriate centrifuge
acceleration level, N, using the scaling laws. From these time factors the
actuation frequency of the soil model can be determined. The method of
actuation and required actuation power can then be selected.
In many cases, the actuation frequency required to directly model diffusion
events is too fast for the available means of actuation or may require too much
power. The higher actuation frequency may also induce inertial effects that are
not present in the prototype. The actuation frequency must then be reduced to
more manageable levels. This reduced actuation frequency must not cause a
significant change in the dissipation of pore pressure occurring during the
actuation.
Where significant pore pressure changes would occur in granular materials,
the modeller can choose to increase the viscosity of the pore fluid to retard pore
pressure dissipation. Increasing pore fluid viscosity by the scaling factor is
common when modelling dynamic events in sand to match the time scaling
factors for inertia and diffusion. The modeller must also ensure that changing the
pore fluid does not significantly affect the mechanical behaviour of the soil
medium. Wilson (1988) has shown that damping within the soil medium is
increased close to resonance when viscous fluids are used.
The instrumentation for monitoring the model test can be selected knowing
the type and expected range of measurands and the required monitoring
frequency. The selection of modelling materials, actuators and instrumentation
are discussed in sections 3.5 to 3.8.
The model structure is normally scaled to have the same external geometry as
the prototype. Other scaling requirements might include the bearing stress, the
stiffness and the strength of the structure relative to the soil medium. Frequently,
these model structures are manufactured from different materials than the
prototype to satisfy the selected scaling criteria.
The model test must be integrated with the centrifuge. The design should
include how communication is established with the test package and how the
model will be affected by the centrifuge environment. As the centrifuge rotates,
most of the power required to rotate the centrifuge is dissipated in aerodynamic
drag creating heat and a potential increase of temperature within the centrifuge
chamber. For tight temperature control within the model test, the test package
may have to be enclosed within thermal control barriers. The heat may be partly
dissipated by ventilation of the rotating mass of air within the chamber. Air
movements may cause undesirable effects to the exposed model such as
buffeting and high evaporation rates which should be controlled by protecting
the exposed model.
 3.5
Model preparation
The most important aspect of a geotechnical soil model is the effective stress
profile. The effective stress history, the current effective stress state and the
effective stress path followed during the test will dictate the behaviour of the
model.
Centifuge model tests can be performed on undisturbed soil samples, if the
effective stress conditions in the sample are representative of the prototype.
Macro-fabric present in the undisturbed model sample, such as structure,
fissures, inclusions and potential drainage paths, may not scale to be
representative of the conditions in the prototype.
Macro-fabric present in undisturbed soil samples can be eliminated by
remoulding of the natural soils. The centrifuge model is then constructed from
these remoulded materials. A site investigation of the prototype situation is
required to reconstruct a representative centrifuge model. The use of remoulded
soil distorts the materials history including the effect of ageing. Techniques for
artificially ageing remoulded small soil samples are being developed (Tsuchida
et al., 1991). Without such techniques the strength of the undisturbed material
may not be modelled correctly in the remoulded sample. The remoulded sample
should then be treated as an equivalent material and allowance made in the test
design for the change in soil strength and behaviour.
For more generic conditions, reconstituted laboratory soils are normally used
with well-defined soil properties. The behaviour of these laboratory soils can be
altered to produce the required material behaviour by the use of mixtures (for
example, Kimura et al., 1991). The soil conditions in such models are well
controlled, permitting numerical analyses to be validated against such physical
model test data.
Remoulded granular soil models can be prepared by tamping and pluviation
techniques. The soil models are generally too large to be compacted on vibrating
tables. Tamped samples can be prepared moist or dry for most grain size
distributions. The sample is placed in layers which are then compacted by
tamping to achieve the required overall density. There may be a variation of
density within the tamped layers.
Dry pluviation techniques can be used for uniformly graded dry sands. Finer
silt material become air borne and will not pluviate. Air borne dust is an
important consideration using the pluviation technique. The dust should be
contained in the model preparation area and the modeller should wear a face
mask to prevent silicosis.
The density of pluviated samples can be accurately controlled by the energy
imparted to the sand particles: dense samples are created by pouring the sand
slowly from a height whereas loose samples are created by slumping the sand
quickly into the model container. These and other factors controlling pluviated
sand sample densities are presented by Eid (1988). Samples pluviated into a
rotating-drum centrifuge may have a lateral velocity relative to the soil surface;
this will cause densification of the sample. The pluviation technique should be
carefully chosen to reduce cross-anisotropy within the pluviated samples.
Single- point hoppers are particularly useful when creating highly-instrumented
samples of complex geometry. The dry sand surface can be accurately shaped
using a vacuum system. Mechanical shaping of the sand surface will disrupt the
surface density of the sample.
Saturated samples can be created by pluviating the sand through the pore
fluid. This technique can only be used to create lightly instrumented, relatively
loose samples of simple geometry. These samples may not have an acceptable
degree of saturation. Pluviated samples are better saturated after construction.
The simplest technique is to introduce the pore fluid from a header tank through
a base drainage layer into the dry model. The driving head should be kept below
the hydrostatic head necessary to fluidise the sand sample. Movement of the
saturation front should be controlled to prevent air pockets becoming trapped
within the saturated material.
Higher degrees of saturation can be achieved by evacuating air from the sand
sample before the pore fluid is introduced. The degree of saturation for water-
saturated samples can be further increased by replacing the air within the sample
by carbon dioxide. The carbon dioxide can be introduced after evacuating the air
from the sample. The inlet pressure of the gas must not be sufficient to cause
fluidisation of the sand sample.
It is advisable when saturating samples under vacuum to have the pore fluid in
the header tank under the same vacuum as the sand sample. This has the
advantages of de-airing the pore fluid before it is introduced into the sample and
of minimising the pressure differential between the header tank and the soil
sample. Boiling of the pore fluid under vacuum should be avoided.
Some pore fluids may be too viscous to penetrate the soil matrix. The
viscosity of some of these fluids may reduce sufficiently at elevated
temperatures to enter the soil matrix. Saturation using such fluids will require
the whole soil sample and header tank to be heated using a water bath or similar.
Remoulded clay and silt samples can be created by tamping. For better-
defined stress histories, clay and silt samples should be reconstituted from a
slurry. The slurry should be mixed at about twice the liquid limit of the material.
De-ionised water can be used as the pore fluid to minimise chemical effects and
bacterial growth within the sample. The slurry should be mixed under vacuum
for about two hours to de-air the slurry and create a smooth slurry. The resulting
slurry is then placed into a consolidometer. Care must be taken not to trap air
pockets within the slurry mass and drainage layers during placement which
would decrease the high degree of saturation of the slurry mass. Layered
samples can be created if slurry ingress is prevented into the underlying layers.
Silt and clay slurries can be consolidated in the centrifuge. Caution is needed
to prevent the generation of high pore pressures within the slurry mass which
may cause piping through the clay mass and preferential drainage paths. The
high pore pressures can be avoided by accelerating the centrifuge in stages to the
required test speed with delays at each stage to allow dissipation of excess pore
pressures. As the shear strength of the sample increases during consolidation, the
surface of the sample will hold a particular inclination. Therefore, for
centrifuges with restrained platforms, the resultant surface may not be correctly
inclined at the required test speed. Consolidation of deep clay layers in the
centrifuge is a lengthy procedure.
Silt and clay samples are more often formed in a consolidometer, which may
also be the test container. The initial consolidation increment should be about 5–
10 kPa, unless measures have been taken to prevent extrusion of the slurry from
the consolidometer. After this first increment, the consolidation pressure can be
increased after 80% of primary consolidation is achieved. The consolidation is
easily monitored by measuring the vertical settlement of the consolidometer
piston or by measuring the amount of pore fluid expelled from the sample. The
consolidation pressure can be successively doubled until the required maximum
consolidation pressure is achieved.
This consolidation technique is normally used to create a uniform
consolidation pressure with depth. Trapezoidal variations of consolidation
pressure can be created by inducing seepage within the consolidometer.
Normally the effective consolidation pressure at the surface of the sample is
required to be lower than that at the base. The consolidation pressure required at
the base of the sample is applied at the surface of the sample and the base of the
sample drained to atmosphere. The applied consolidation pressure can be
imposed by two methods. A sealed impermeable piston can be used to apply the
full base consolidation pressure which is reduced effectively at the clay surface
by the pore pressure acting in the drainage layer between the sample and the
piston. The second method is to use an unsealed piston which applies the
effective surface consolidation pressure. The base consolidation pressure is
applied by pressurising the pore fluid around the piston. This consolidation
technique has been called downward hydraulic gradient consolidation (Zelikson,
1969). Using this technique a number of different consolidation profiles can be
successively applied to the clay sample to reproduce most pre-consolidation
profiles, including normally consolidated clay. By reversing the technique and
applying fluid pressure at the base of the sample, upward hydraulic gradient
consolidation can also be undertaken, to create, for example, over-consolidated
surface crusts.
Consolidometers should be designed so that lateral displacement of the
consolidometer wall should be less than 0.1% of the retained height of soil to
maintain earth pressures in their at rest condition as indicated by Yamaguchi et
al. (1976).
After primary consolidation is complete at the maximum pre-consolidation
profile, the sample can be unloaded in the consolidometer. After each stage of
unloading, the pore pressures within the soil sample are in suction. If these
suctions are too high, cavitation with a consequential loss of strength may occur
within the soil mass. In the presence of excess pore fluid, these suctions will
dissipate in a controlled manner.
The height of the sample should be measured immediately before and after it
is unloaded in the consolidometer. The elastic heave of the sample can then be
assessed and accommodated during construction of the model. After the sample
is unloaded, the effective stresses within the sample are retained by pore
suctions. These suctions can be maintained by removing excess pore fluid from
around the sample and preventing air entry into the sample. The sample can be
sealed with plastic cling film or similar to reduce pore pressure dissipation.
For thick clay samples, the time required to establish full-equilibrium
effective-stress conditions during the centrifuge test may be excessive. This
reconsolidation time can be reduced by decreasing the drainage path lengths
within the soil model. These shorter paths can be accomplished by creating thin
granular drainage layers within the soil sample when it is first constructed. After
consolidation, these drainage layers must be connected to an external source of
pore fluid at the correct potential. After unloading from the consolidometer,
these layers will be in suction. Care must be taken when connecting into these
layers to prevent air entry or the layers may become air-locked and ineffective.
Radial drainage can also be created using vertical wick-wells formed from
washed wool or string.
Another option is to create a quasi-drainage boundary in the soil sample: the
clay sample is unloaded in the consolidometer to a uniform consolidation
pressure which is the average effective stress applied to the clay sample during
the centrifuge test. During the centrifuge test the upper part of the sample will
swell and absorb water and the lower part of the sample will consolidate and
expel water, thus creating a quasi-drainage boundary within the clay sample.
The preferred option to minimise consolidation time in the centrifuge is to
finally consolidate the clay in the consolidometer close to the effective stress
profile it will experience during the centrifuge test using downward hydraulic
gradient consolidation. The surface over-consolidation ratio must not exceed 10
or cracking of the clay surface may occur disrupting the seepage flow
(Kusakabe, 1982). The minimum amount of time should be taken between
unloading the consolidometer to starting the centrifuge test. The effective stress
within the clay sample can be monitored using pore-pressure transducers and the
model sample actuated when an acceptable effective stress profile is restored.
Reconstituted kaolin powder has been used extensively to create clay
centrifuge models. Kaolin is a coarse grained clay with a relatively high
permeability which consolidates rapidly minimising model preparation and
centrifuge test durations. Mixtures of clay, silt and sand offer the centrifuge
modeller a wider range of material behaviour and properties than those available
from a single material type.
Other centrifuge modelling materials have included equivalent materials,
particularly for rock mechanics studies, and photoelastic media (Clark, 1988).
Commercially available bulk materials, such as sand, silt and clay powder, are
frequently used as modelling materials. The mechanical properties of these
materials may change with time. Index properties of these modelling materials
should be routinely measured to ensure consistency of the material. These
materials are sometimes recycled after a centrifuge test and re-used. These
recycled materials should be tested to ensure there has been no significant
degradation of the material.

3.6
Fluid control
Fluid control in the model test is important for maintaining the correct drainage
and effective stress conditions. Water is most commonly used as the test fluid.
As the centrifuge speed changes, free fluid surfaces will flow to become
normal to the resultant acceleration field. This movement of fluid across the
sample surface may cause erosion or over-topping in the sample. These effects
can be reduced by either submerging the whole soil surface or by limiting the
amount of fluid on the surface during speed changes. When testing cohesive
samples, the presence of free fluid during speed changes will allow dissipation
of pore suctions within the sample. This dissipation can be reduced by adding
free fluid to the test sample after the centrifuge has started, and removing the
free fluid before the centrifuge is stopped. These changes in fluid mass within
the test package must be controlled to prevent excessive out-of-balance forces
developing on the centrifuge rotor.
If the soil surface is not submerged, excessive evaporation may lead to drying
and desiccation of the soil surface. This evaporation can be controlled by
covering the surface with a protective coating such as liquid paraffin, which will
minimise surface evaporation without significantly affecting the behaviour of
the soil. Near-surface pore suctions, from evaporation and capillary action, will
increase the effective stress in the soil skeleton and should be monitored and
controlled to acceptable levels. Surface pore suctions can be used to create
surface crusts.
The fluid level within the sample should be monitored and controlled. Fluid
may be lost from the sample due to evaporation or leakage from the
package.The fluid level can be controlled using standpipes. The standpipes are
external to the test package and connected with pipework to the drainage layers
within the soil sample. The pressure heads in the pipework are increased under
the centrifuge acceleration. Care is required to prevent cavitation within the
pipework from flow under these increased heads.
The required equipotential level in the standpipe and elsewhere in the test
package can be calculated allowing for centrifuge curvature and the Earth’s
gravity effects. This equipotential level can either be controlled using a fixed
overflow or a control system. The fixed overflow is simple to implement but
will not maintain a fixed level of fluid relative to the sample during
consolidation of the sample. A simple control system may comprise of a fluid-
level indicator and a dosing pump: as the fluid level in the package decreases the
dosing pump restores the fluid level to the required level.
The fixed overflow system can be used as a constant loss system, where fluid
is fed continuously into the test package and overflowed to waste. Generally,
fluid overflows such as water do not need to be retained in the package but can
be dumped into the centrifuge chamber. Care should be taken to pipe these
overflows away from any electrical devices. Other fluids, such as oil and
chemicals, must be safely retained within the test package. Fluids can be fed into
the standpipe through hydraulic slip-rings. The passage of fluid down the
centrifuge rotor to the test package is analogous to fluid flowing down into a
steep-sided valley. The standpipe then serves as a stilling tank to dissipate the
energy in the fluid and provide the fluid at the correct potential for the model.
Fluid feeds piped directly from the slip-rings into the model package may cause
significant erosion of the soil sample.
The standpipes can be used to change the fluid control level during the course
of the test. The elevation of pipework from the standpipe or the discharge point
into the test package must not exceed the standpipe control level or there will be
no fluid flow. The fluid control level can be changed by selection of a different
overflow level using valves, or by changing the control level.

3.7
Actuation
An important consideration in a centrifuge model test design is how to actuate
the model to simulate the prototype perturbation. Typically, the model actuation
frequency should be at least the scaling factor, N, times faster than the prototype
to simulate similar degrees of pore-pressure dissipation. The actuator’s power
density can be defined as the power per unit volume of actuator. Dimensional
analysis shows that the actuator power density required for the model should
then be at least N times greater than the prototype power density for correct
scaling.
This increase in power density is not normally achievable, and the model
actuator is proportionally either larger than the prototype or not as powerful.
Similarly, instrumentation, interfaces and other devices are not scaled. The
modeller would prefer the majority of the payload to be the soil model, but the
actuator and other devices must be accommodated.
The payload capacity occupied by the actuator and other devices can be
reduced by placing sub-systems away from the test package on either the
centrifuge rotor or off the centrifuge. Such sub-systems might include power
supplies, power amplifiers, controllers and conditioning modules. This
separation can be advantageous as the sub-systems are subjected to a lower
acceleration field and can occupy a larger volume than can be accommodated at
the test package. Communication, however, needs to be established between the
sub- systems and the test package. The bandwidth and number of
communication channels required may sometimes limit the performance of the
actuator. For example, long hydraulic hoses between a hydraulic power pack and
a hydraulic cylinder will limit the dynamic response of the system. There are
many electrical lines between a brushless servo-motor and controller which may
make this connection inefficient through slip-rings.
The mode of actuation should be kept simple and resource effective.
Frequently, the centrifuge model test objectives will evolve as a test programme
proceeds requiring changes to the actuator. Simple actuators are likely to be
more compact and more reliable than complex actuators. In recent years,
actuators have become simpler as motion controllers have become more
sophisticated. This development has been very beneficial to centrifuge
modelling permitting a range of complex tasks to be accomplished using a
combination of simple actuators, as demonstrated by McVay et al. (1994).
The actuator should not restrict the behaviour of the model. For example, if
model tests are being conducted on vertical bearing capacity, the footing should
be free to rotate and translate under the vertical load. If the footing is rigidly
attached to the actuator then these motions will be prevented. Allowance should
be made for flexure of actuator and support systems under load. Such flexure
may cause undesirable loading of the soil model and may inhibit stiffness
measurements of the model.
Actuators can be developed from commercially available products, or
specially constructed. There is a wide range of standard commercial products
suitable for use in centrifuge modelling. The modeller should consider using
such products before developing their own: the centrifuge model test research
objective is, after all, in geotechnics not electrical, control or mechanical
engineering!
When selecting a commercial product, the principle of operation and the
physical construction of the product must be considered to determine how the
product will perform on the centrifuge. Most manufacturers and suppliers do not
warrant their products for operation under high-acceleration fields, but are
usually interested to provide technical support for this unusual application. Most
commercial products are sized for continuous operation over a number of years.
Such products may be over-designed for the small number of duty cycles
required in the centrifuge model test, permitting smaller-sized products to be
used.
Some customised actuator parts, especially interfaces between standard
components and the model, for a centrifuge model test may be required. Designs
for these parts can be sought from other experienced centrifuge modellers, from
researchers in other fields with analogous requirements, or from specialist
design companies. The experience of other centrifuge modellers is available
through texts (such as this one), the proceedings of speciality conferences, the
theses and reports published by the many geotechnical centrifuge centres
worldwide and by discussion with the experienced modellers at these centres.
The modeller is encouraged, where possible, to standardise their modelling
techniques and equipment with those already developed.
Actuators can be powered from different sources including electricity,
hydraulics, pneumatics and latent energy, as described in the following
paragraphs. It is impractical to store sufficient electrical, hydraulic and
pneumatic power on the centrifuge to meet the total demand of the model test.
Pneumatic and hydraulic power can be generated on the centrifuge using
electrical power. Some electrical, hydraulic and pneumatic slip-rings are
therefore required on a centrifuge, even though more data acquisition and
control communications are being established using non-contact techniques such
as optical slip-rings and radio local area networks (LANs), rather than
mechanical slip-rings. Limited quantities of electrical, hydraulic and pneumatic
power can be stored on the centrifuge, using for example batteries, accumulators
and gas cylinders to meet peak demands, such as earthquake actuation. Latent
energy is easily stored on the centrifuge as potential energy in the actuator,
kinetic energy in the centrifuge rotor or in explosives.
Electrical power is normally transmitted as an alternating current which may
radiate electrical noise. This noise pick-up can be reduced by the use of shielded
twisted pair cables for the electrical power, proper attention to earthing and
electrical connections, and physical separation of the power and data cables and
slip-rings.
Hydraulic fluid, such as water or oil, will increase in weight under the
centrifuge acceleration. Hydraulic power systems must accommodate these
hydrostatic pressure increases. Water can be used for low-pressure applications
mixed with a rust-inhibitor and lubricant. Water is denser and less viscous than
hydraulic oil, and therefore can exert much higher hydrostatic pressures with
lower pressure losses. Hydraulic oil will be required for high-pressure
applications. Most hydraulic systems are designed to leak. This leakage must be
collected against the hydrostatic back-pressure on the return line. Hydraulic slip-
rings are also required for fluid feeds to the centrifuge model, such as surface
water control as described above, and for other fluids such as refrigerant.
Low-pressure pneumatic systems to about 15 bar can be driven through
pneumatic slip-rings. These slip-rings should be lubricated and cooled to extend
the slip-ring seal life. High pneumatic pressures to about 200 bar can be stored
in compressed gas cylinders on the rotor. Pneumatic pressure is excellent, as the
increased gas weight is negligible, for force-controlled loading. Displacement-
controlled monotonic loading can be achieved using a pneumatic cylinder: one
chamber of the pneumatic piston is connected to the pneumatic pressure supply
and the other chamber is filled with hydraulic fluid, which is vented through a
control orifice. When pneumatic pressure is applied, movement of the piston is
restricted by the flow rate through the orifice.
The centrifuge acceleration field can usefully be exploited as a power source.
Component self-weight has been used for the installation and compressive
loading of piles, suction caissons, penetrometers and footings. Tensile loading
can be imparted from the buoyancy of a float in water. The self-weight of water
can also be used as a loading source. Excavation events have been simulated by
the removal of dense fluid from a retaining pressure bag. The inorganic salt zinc
chloride is particularly effective because of its high specific gravity and
exceptional solubility in water. Zinc chloride is inexpensive, but also very
corrosive (see section 4.1.3).
The ‘bumpy road’ earthquake actuator utilises a fraction of the kinetic energy
stored in the Cambridge University beam centrifuge as its power source. A
medium-sized centrifuge will store about 10 MJ of energy in its rotor.
Controlled explosions are also a very useful energy source. Access to these two
energy sources may impart excessive loads to the centrifuge.
The weight of all components within the test package will be carried in
compression into the platform, probably through the container. In general,
support systems are simpler if the components are carried in compression. All
components should be firmly attached to the test package to prevent disturbance
to the model and to permit mechanical handling of a complete package. The
orientation of actuators with respect to the acceleration field is important to
ensure correct operation of the actuator under its increased self-weight.
The increased weight of pistons in hydraulic cylinders and pneumatic
cylinders may need to be supported by pressure if the cylinder is mounted
vertically, or may cause the piston to rack or leak in the cylinder if the cylinder
is mounted horizontally. If the required support pressure is too high in the
former case, it can be reduced by decreasing the weight of the piston and piston
rod: these items could be re-made in aluminium alloy. The piston rod could be
replaced by a smaller rod or hollow tube depending on the axial load
requirements. Cylinders can be acquired with a double piston rod, extending
from both ends of the cylinder. The piston rod can then be fitted with guides to
prevent racking of the piston within the cylinder. The additional piston rod is
also useful as a reference from which to measure displacement of the piston.
The orientation of shuttle valves also needs to be considered. Normally, the
shuttles should be mounted vertically, with the weight of the shuttle keeping the
valve down in its normal operating position. The weight of the spindle needs to
be considered when selecting the valve actuator. For example, for solenoid or
pneumatically actuated valves, the spindle weight can be calculated in terms of
an actuation pressure knowing the geometry of the valve. The specified pressure
rating for the actuator is then the required pressure for the test plus the actuation
pressure.
Electric motors are normally best orientated with their main rotor axis in line
with the centrifuge acceleration field. The increased rotor weight can then be
carried through thrust bearings attached to the motor output shaft. The rotor will
flex when placed across the acceleration field. This flexure and the radial play in
the rotor supports may permit the rotor and stator to short. The type of electric
motor and controller to use is dependent on the power, speed and control
requirements of the model test. Generally, three-phase electric motors are
preferable to single-phase motors as they have a higher power-to-frame size
ratio and radiate less electrical noise. Permanent magnet motors are beneficial
due to their simple construction and reduced number of electrical connections.
Brushless servo-motors have an excellent power-to-frame size ratio and a simple
construction, but require a sophisticated electronic controller and power
amplifier.
Solid state electronics, such as printed circuit boards, can be used successfully
in the centrifuge. In general, printed circuit boards should either be aligned
radially with the acceleration field, or supported on foam rubber or similar to
prevent flexure of the board and breakage of solder tracks and components. The
behaviour of large or delicate components should be considered under increased
self-weight. Large components such as transformers and heat sinks may require
additional mechanical support. Delicate components may need to be potted for
support. Electrolytic capacitors have been found to distort and change
capacitance under their increased weight. Electrical connections should be
aligned to mate better in the centrifuge and not be distressed.

3.8
Instrumentation
Centrifuge model test behaviour can be monitored by a variety of
instrumentation. Available instrumentation includes not only a wide range of
transducers but also visual techniques as described below. New instrumentation
is being developed which may be applicable to centrifuge model testing. As this
instrumentation becomes available, its suitability can be assessed using the
guidelines presented below. Particularly useful instrumentation is anticipated
from the areas of remote sensing and fibre optics.
Transducers in contact with the centrifuge model should be small and rugged
enough to resist not only their increased self-weight but also mechanical
handling during test preparation and disassembly. Solid-state transducers are
particularly suitable. The operating principle of the transducer must be
considered. Normally, the transducer is required to be capable of continuous
monitoring throughout the centrifuge test, such as pressure transducers. More
infrequent monitoring may be acceptable such as deformations before and after
an event. For continuous monitoring, the transducer should have an adequate
frequency response, which is normally one or two orders of magnitude higher
than that required in the prototype.
The transducer output may require conditioning to be interfaced to the data
acquisition system. The transducers and conditioning modules will be limited by
the space associated with the test package. The transducer should be reliable.
Transducers embedded within the soil model should be miniature with
dimensions of about 10mm. These buried transducers may act as ground
anchors. The model test must be designed such that it does not become a test of
reinforced earth. The transducers should be orientated to minimise
reinforcement effects. The transducer leads should be flexible and run
orthogonal to the direction of principal movement. Transducer lead runs should
also minimise potential drainage path effects. Buried transducers and their leads
must withstand the high ambient pressure levels within the soil mass and the
high ambient water pressures, when used in saturated media.
Pore-pressure transducers are fitted with a porous element to isolate the fluid
pressure for measurement. The movement of fluid through this porous element
will mechanically filter the frequency response of the transducer (Lee, 1991).
The response is dependent on the degree of saturation and porosity of this
element. Push-fit elements are recommended for these transducers. These
elements can then be de-aired in pore fluid and fitted to the transducer under
pore fluid to ensure a high degree of saturation of the transducer. The elements
should be replaced if they become blocked. Ceramic elements are recommended
for use in clay and coarser elements, such as sintered bronze, for use in granular
soils. These transducers can be used without porous elements for use in
standpipes and free pore fluid.
Commercially available Druck PDCR81 transducers are commonly used for
pore-pressure measurement. There are other commercially available pore-
pressure transducers, but these are generally larger than those supplied by
Druck. Pressure transducers smaller than those supplied by Druck are
commercially available but these transducers are not suitable for burial in a soil
model. The PDCR81 transducer is a differential pressure transducer. The
reference pressure is provided through the hollow electrical lead. The integrity
of this air passage must be ensured. Calibration and installation procedures for
these transducers are described by König et al. (1994).
Total stress transducers are required to define more completely the state of
stress at the boundaries or within the soil model. The stiffness of the transducer
is very important. For boundary-stress measurements, the stiffness of the
transducer should be similar to the boundary stiffness for a representative stress
measurement. Fluid-filled diaphragm transducers are suitable, but they only
measure normal force. Normal and shear forces can be quantified using the
Stroud cell, but this cell is not as stiff as those above. Soil arching also affects
pore pressure measurement (Kutter et al., 1988).
Displacements can be measured with potentiometers or linearly variable
differential transformers. Both these transducers require contact with the model.
For vertical measurements, the spindle weight may need to be carried on a pad
to prevent indentation of the spindle into the model. For horizontal
measurements, the spindle may require mechanical assistance, such as a spring
or glue, to maintain the spindle in contact with the model. Potentiometers should
be orientated to keep the wipers in contact with the resistive elements.
Temperature measurements can be made using thermocouples or thermistors.
Boundary temperature measurements can be made cheaply using a digital
thermometer in the view of a CCD camera.
Some instrumentation is test specific and requires development. Such
instrumentation may be load cells, which are also the linkage between the
actuator and the model structure, or the structures themselves such as
instrumented piles, retaining walls or geotextiles. This instrumentation is
normally designed to measure strain using foil strain gauges. If possible, the
strain gauges should be configured in a complete Wheatstone bridge to reduce
thermal effects within the instrument. These thermal effects can be minimised
by correct selection of the strain gauges and by restricting the power dissipation
within each gauge.
Strain gauges for instruments required for only two or three of months can be
bonded with super glue, as long-term instrument stability is not required. In
these model structures, the lead wires will be proportionally large. The support
and routing of these lead wires is an important consideration in the design of the
instrument. The strain gauges and lead wires will need to be protected and
sealed when used on embedded instrumentation. The completed instrument
should be exercised and calibrated over its working load range before the
centrifuge test. The instrument should be load cycled about 20 times to reduce
hysteresis within the instrument.
Surface cracking has been sensed using conductive paint or thin foil. The
crack is sensed from the break in electrical continuity. Other customised
instrumentation has included miniature resistivity probes at Cambridge
University and University of Western Australia (Hensley and Savvidou, 1993),
wave height gauges at Cambridge University (Phillips and Sekiguchi, 1991) and
radiation sensors (Zimmie et al., 1993).
Many of the electronic instruments mentioned only provide detailed point
measurements within the model. Visual measurements provide an overview of
the model behaviour. Movements of the soil model can be observed by placing
markers within the soil mass. In sands these markers can be thin coloured sand
bands. In clay these markers can be noodles or lead threads. These markers are
accurately placed during model preparation. Exposure of these markers after the
centrifuge test reveals the plastic deformation of the soil model. The sand bands
can be exposed in vertical sections: in fine sand, pore suctions are sufficient to
hold a vertical face created by dissecting the sample with a vacuum cleaner. In
coarser sands, the pore fluid can be replaced by sugar water, which under heat
will cement the sand structure together providing stability during dissection.
Lead threads are formed by injecting a suspension of lead powder in water-
soluble cutting oil into the soil sample to leave a lead-coated shaft. (These shafts
are potential drainage paths in the model.) Radiographic examination of these
lead threads reveal distortion within the model such as rupture band formation.
Sufficient time must be left for the oil to diffuse into the soil sample before the
centrifuge test, otherwise the lead thread shafts may hydrofracture under the
pressure of the heavy lead suspension. Lead threads can be injected into pre-
drilled holes in granular models. Lead-impregnated, home-made pasta noodles
can also be used.
Vectors of face movements can be determined by indenting markers into the
soil face. Some of these markers may be lead shot to define the model
boundaries during radiographic examination. Generally, these markers should be
small and about the same density as the soil they displace to minimise their
influence on the model behaviour. Markers used behind a transparent window
should have a low frictional resistance against the window in order to track the
soil movements.
Marker positions can be tracked from successive photographic negatives of
the model in rotation or television cameras. Stereoscopic cameras mounted very
close to the central axis of the centrifuge have been used at the University of
Manchester to map surface topography of centrifuge models. Photographic
cameras can also be mounted in the centrifuge containment structure. High-
resolution photographs are taken of the model in rotation using a short-duration
high-intensity flash system synchronised to the rotation of the centrifuge. Flash
durations are typically 5 s. Measurements of successive polyester-based
photographic negatives permit the marker positions to be tracked to an accuracy
of about 0.1 mm in a 500mm wide field of view. Strains within the plane of
movement can be assessed by finite differentiation of these movements.
Markers can also be tracked using image analysis systems as described by
Garnier et al. (1991) and Allersma (1991). For accurate measurements, these
images should be stored digitally rather than in analogue form to minimise
distortion of the image on the storage medium. For small centrifuges,
stroboscopes or CCD cameras can be synchronised to the centrifuge rotation to
view the centrifuge model. Small CCD cameras are now frequently used on the
test package to monitor various aspects of the centrifuge model test. Inaccessible
locations for the CCD camera can be visualised using mirrors or endoscopes
fixed to the CCD camera. Topographic mapping of soil surfaces after the
centrifuge test also provide useful information.

3.9
Data acquisition
A typical outline of a data-acquisition system is shown in Figure 3.4. Most of
the modules required for such a system, including the data acquisition and
control software, are commercially available. The suppliers handbooks for such
systems provide a free and very reasonable introduction to data acquisition.
The requirements of centrifuge data-acquisition systems are unusual. The
system is required to record multi-channel data about two orders of magnitude
faster than in the prototype. The system is also required to be flexible to
accommodate the acquisition requirements of many different types of centrifuge
test. Such a system may typically be required to acquire data from 16
transducers at 10 kHz per channel for a second during an earthquake simulation
and from 50 transducers at 0.01 Hz per channel for two days during a pollution-
migration experiment.

Figure 3.4 Typical data-acquisition system.

Centrifuge model tests can be relatively expensive one-off tests, so redundancy
in the data acquisition system to ensure data have been captured is advisable.
In the centrifuge environment, failures of the data-acquisition system do
occasionally occur. The system architecture should be modular to permit faults
to be easily traced and rectified. This modularity also aids the evolution,
expansion and upgrading of the system.
The possible positions of the slip-rings in the data-acquisitions system (Figure
3.4) has changed dramatically over the last 20 years. In many of the original
centrifuges, the transducers were connected directly to the slip-rings and the rest
of an analogue data-acquisition system was located off the centrifuge. With the
progress of micro-electronics and digital technology, it is possible to have the
complete data-aquisition system on the test package with remote digital
communication to the centrifuge modeller.
The slip-rings do provide a bottleneck on the system, due to the number,
bandwidth and capacity of the slip-rings available. Some of these constraints are
being bypassed with the use of optical slip-rings and radio-LAN communication,
or by multiplexing.
In general, all the transducer signals should be conditioned close to source to
provide high-level signals in a common format for onward transmission. The
signal conditioning may include decade amplification and filtering of the
transducer signals. The signal conditioning and power selection for the
transducers can be placed in an interface box attached to the test package. This
interface box can provide a very tidy connection between the multiplicity of
transducers and the centrifuge rotor wiring. Connectors on the transducer leads
and rotor wiring should be strain-relieved so that the cable weights are not
carried by the electrical connections. These interface boxes permit pre-test
calibration and verification of the transducer performance, and maximum
machine utilisation. Transducers monitoring the speed of the centrifuge or the
acceleration level at a known radius linked to the data-acquisition system are
strongly recommended.
All transducer cables and transmission cables must be adequately supported
along their length to carry their increased weight. Electrical-power and signal-
transmission cables should be physically separated and properly shielded to
reduce electrical pick-up. The slip-rings should be correctly wired for the types
of data transmission required, such as IEEE, serial, digital or analogue.
Manual logging is an important feature of the data-acquisition system.
Transducer reliability is not high in centrifuge model tests. Manual logging
allows the individual transducer outputs to be verified, independent of the
computerised data-logging system, against the predicted response from the
transducer’s calibration. The manual logging also provides a continuous
monitoring capability of selected channels. This capability could be used by the
centrifuge operator to ensure the integrity of the test package, such as leakage, or
provide feedback to a closed-loop control system without compromising the
main data-logging system performance.
The main data-logging system will typically include an analogue-to-digital
convertor (ADC) interfaced to a computing system. The front-end signal
conditioning may be required to optimise the inputs to the ADC; this
conditioning may include binary gain amplification, offsetting and filtering. The
ADC may include an input multiplexor and a single convertor. The effect of
skew between successive inputs must then be considered.
Many complete proven data logging systems are commercially available. The
modeller is advised to use one of these systems, rather than develop their own.
Modellers should also beware of becoming a test site for the most up-to-date
commercial data-acquisition products. Customised systems will be required for
specific tests but should only be developed if essential: hardware and software
development is time-consuming.
Centrifuge model tests are resource intensive. Each model test is unique,
yielding a wealth of data, which if successful is unlikely to be repeated.
Modellers should consider backing up their main data-acquisition system with a
mass data storage system, such as magnetic tape or optical disc. In the event of
problems with the main system or unexpected events, the mass data storage
system can be used to recover essential data. The mass data storage system
should sample at higher data rates than the main system to ensure that the details
of unexpected events can be defined. In an analogue recording system, a
magnetic tape recorder provides an excellent mass data-storage system. Digtal
mass storage systems are becoming more readily available.
The data obtained from the acquisition software should be suitable for input to
the data processing, analysis and reporting software to streamline the procedure
of model test reporting. Centrifuge model tests are probably better defined and
as highly instrumented as any field trial. Data reduction rates are, however, not
accelerated because the data were acquired from a centrifuge test! The modeller
should allow sufficient time to fully interrogate their data set and develop a
better understanding of their problem.

3.10
Test conduct
The previous sections have considered mainly the design and construction of
centrifuge models. Conduct of centrifuge model tests is important. The majority
of effort is required in preparation for a centrifuge model test. To benefit from
this effort, the modeller is encouraged to prepare a full checklist of all activities
required to successfully complete the test. This checklist is best prepared by the
modeller carefully thinking through every step of the test: from the design,
through equipment construction, technique development, model-making,
package assembly, integration with the centrifuge, package verification, conduct
of the test, post-test investigation through to data processing and reporting. The
test is a multi-disciplinary project with many different factors to consider. As
many test sub-systems as possible should be verified before the centrifuge is
started.
The actual centrifuge test is a culmination of effort which may prove
exhausting to the modeller especially during extended centrifuge runs. The
checklist prompts the user, for example, to turn on the data-acquisition system
before actuating the model! The modeller is in control of the model test, and
should take time to cross-check their actions to ensure the success of their test.
Safe conduct of every centrifuge model test is the concern of every centrifuge
modeller. The safety of the centrifuge must always have a higher priority than
the successful completion of any model test. The cost of the centrifuge alone is
more expensive than the centrifuge model tests. Centrifuges are very powerful
and should be treated with respect. Personnel training in the safe operation of the
centrifuge is important.
Every centrifuge model test should be discussed and planned with an
experienced centrifuge modeller. Calculations of the stress conditions in the test
package and the balance of the centrifuge are strongly advised. Test package
configurations change frequently. Centrifuge tests to prove the integrity of the
package beyond its normal working condition in the presence of an experienced
modeller are advisable.
Centrifuge models are normally constructed away from the centrifuge.
Mechanical handling of the test package onto the centrifuge should not cause
undue disturbance to the soil model. Sensitive soil models can be transported
under vacuum to increase the effective stresses within the model. Very sensitive
models can be formed in the centrifuge chamber, but care should be taken not to
contaminate the chamber with soil, which may shot blast the centrifuge and
associated systems during rotation. Every centrifuge centre should develop a
code of practice for centrifuge operations; an example of that used at Cambridge
University was presented by Schofield (1980).
When starting and stopping the centrifuge, the angular acceleration should be
kept low to prevent the inducement of significant lateral force on the model test.
The action is required to prevent swirl of free surface water in drum centrifuges.
Consolidation of the test sample to the required effective stress condition is well
monitored by the integrated effect of surface settlement rather than individual
pore-pressure measurements.
Shortly after the centrifuge run and investigation of the model behaviour, the
centrifuge test package should be dismantled and cleaned and the test
components safely stored. Most centrifuge facilities are shared by a number of
different centrifuge modellers and are in a continual state of flux. Each modeller
must be responsible for their own test.

3.11
Conclusion
Centrifuge model testing is a challenging and exciting experimental science. It is
a tool for the geotechnical engineer. Like any tool, the success or otherwise of
the model test will reflect the effort and aptitude of the modeller. The practical
considerations presented in this chapter are intended to assist the modeller to
select the correct tools for the job.

References

Allersma, H.G.B. (1991) Using image processing in centrifuge research. Centrifuge ’91
(eds H.Y. Ko and F.G.McLean), pp. 551–558. Balkema, Rotterdam.
Bear, J. (1972) Dynamics of Fluids in Porous Media. American Elsevier, New York.
Bolton, M.D. and Lau, C.K. (1988) Scale effects arising from particle size. Centrifuge ’88
(ed. J.F. Corté), pp. 127–134. Balkema, Rotterdam.
Clark, G.B. (1988) Centrifugal testing in rock mechanics. In Centrifuges in Soil
Mechanics (eds W.H.Craig, R.G.James and A.N.Schofield), pp. 187–198. Balkema,
 Rotterdam.
Craig, W.H. (1993) Partial similarity in centrifuge models of offshore platforms. Proc.
4th Canadian Conf. Marine Geotechnical Engineering, St Johns, Newfoundland,
Vol. 3, pp. 1044–1061. C-CORE, Memorial University of Newfoundland.
Eid, W.K. (1988) Scaling Effect in Cone Penetration Testing in Sand. Doctoral thesis,
Faculty of Engineering, Virginia Polytechnic Institute and State University.
Fuglsang, L.D. and Ovesen, N.K. (1988) The application of the theory of modelling to
centrifuge studies. In Centrifuges in Soil Mechanics (eds W.H.Craig, R.G.James and
A.N.Schofield), pp. 119–138. Balkema, Rotterdam.
Garnier, J., Chambon, P., Ranaivoson, D., Charrier, J. and Mathurin, R. (1991) Computer
image processing for displacement measurement. Centrifuge ’91 (eds H.Y.Ko and
F.G.McLean), pp. 543–550. Balkema, Rotterdam.
Hensley, P.J. and Savvidou, C. (1993) Modelling coupled heat and contaminant transport
in groundwater. Int. J.Numerical Anal. Methods Geomech., 17, 493–527.
Kimura, T., Takemura, J., Suemasa, N. and Hiro-oka, A. (1991) Failure of fills due to
rain fall. Centrifuge ’91 (eds H.Y.Ko and F.G.McLean), pp. 509–516. Balkema,
Rotterdam.
Ko, H.Y. (1988) Summary of the state-of-the-art in centrifuge model testing. In
Centrifuges in Soil Mechanics (eds W.H.Craig, R.G.James and A.N.Schofield), pp.
11–18. Balkema, Rotterdam.
König, D., Jessberger, H.L., Bolton, M.D., Phillips, R., Bagge, G., Renzi, R. and Garnier,
J. (1994) Pore pressure measurements during centrifuge model tests—experience of
five laboratories. Centrifuge ’94, pp. 101–108. Balkema, Rotterdam.
Kusakabe, O. (1982) Stability of Excavations in Soft Clay. Doctoral thesis, Cambridge
University.
Kutter, B.L., Sathialingam, N. and Herrman, L.R. (1988) The effects of local arching and
consolidation on pore pressure measurements in clay. Centrifuge ’88 (ed. J.F.Corté),
pp. 115–118. Balkema, Rotterdam.
Lee, F.H. (1991) Frequency response of diaphragm pore pressure transducers in dynamic
centrifuge model tests. AST M Geotech. Test. J., 13(3), 201–207.
McVay, M., Bloomquist, D., Vanderlinde, D. and Clausen, J. (1994) Centrifuge
modelling of laterally loaded pile groups in sand. ASTM Geotech. Test. J., 17(2),
129–137.
Miyake, M., Akamoto, H. and Aboshi, H. (1988) Filling and quiescent consolidation
including sedimentation of dredged marine clays. Centrifuge ’88 (ed. J.F.Corté), pp.
163–170. Balkema, Rotterdam.
Nunez, I.L. and Randolph, M.F. (1984) Tension pile behaviour in clay—centrifuge
modelling techniques. In The Application of Centrifuge Modelling to Geotechnical
Design (ed. W.H. Craig), pp. 87–102. Balkema, Rotterdam.
Phillips, R. and Sekiguchi, H. (1991) Water Wave Trains in Drum Centrifuge. Cambridge
University Engineering Department Technical Report CUED/D-SOILS/TR249.
Pokrovskii, G.I. and Fiodorov, I.S. (1953) Centrifugal Modelling in Structures Designing.
Gosstroyizdat Publishers, Moscow.
Schofield, A.N. (1978) Use of centrifuge model testing to assess slope stability. Canad.
Geotech. J., 15, 14–31.
Schofield, A.N. (1980) Cambridge geotechnical centrifuge operations. Géotechnique, 20,
227–268.
Schofield, A.N. and Zeng, X. (1992) Design and performance of an Equivalent-Shear-
Beam Container for Earthquake Centrifuge Modelling. Cambridge University
Engineering Department Technical Report CUED/D-SOILS/TR245.
Tsuchida, T., Kobayashi, M. and Mizukami, J. (1991) Effect of ageing of marine clay and
its duplication by high temperature consolidation. Soils and Found., 31(4), 133–147.
Wilson, J.M.R. (1988) A Theoretical and Experimental Investigation into the Dynamic
Behaviour of Soils. Doctoral thesis, Cambridge University.
Xuedoon, W. (1988) Studies of the design of large scale centrifuge for geotechnical and
structural tests. In Centrifuges in Soil Mechanics (eds W.H.Craig, R.G.James and
A.N.Schofield), pp. 81–92. Balkema, Rotterdam.
Yamaguchi, H., Kimura, T. and Fujii, N. (1976) On the influence of progressive failure
on the bearing capacity of shallow foundations in dense sand. Soils and Found.,
16(4), 11–22.
Zelikson, A. (1969) Geotechnical models using the hydraulic gradient similarity method.
Géotechnique, 19, 495–508.
Zimmie, T.F., Mahmud, M.B. and De, A. (1993) Application of centrifuge modelling to
contaminant migration in seabed waste disposal. Proc. 4th Canadian Conf. Marine
Geotechnical Engineering, St Johns, Newfoundland, Vol. 2, pp. 611–624, C-CORE,
Memorial University of Newfoundland.
 4
Retaining walls and soil-structure
interaction
W.POWRIE

4.1
Embedded retaining walls

4.1.1
General principles
An embedded wall uses the passive resistance of the soil in front of the wall
below formation level to counter the overturning effect of the lateral stresses in
the retained ground ( Figure 4.1). The provision of props in front of the wall will
reduce the depth of embedment required for stability, but if there are props at
more than one level, the problem becomes statically indeterminate even if the
wall is analysed when it is on the verge of collapse.

Figure 4.1 Idealized effective stress distributions at collapse for (a) an unpropped
embedded wall and (b) an embedded wall propped at the crest. Reproduced from Bolton
and Powrie (1987). Institution of Civil Engineers, with permission.
Until the 1960s, embedded retaining walls were typically formed of steel sheet
piles, and were used almost exclusively in granular deposits. In these soils, it is
generally accepted that the movement of an embedded wall required to mobilize
the full passive pressure of the soil in front is very much greater than that required
to achieve active conditions in the retained ground. Under working conditions,
therefore, it might reasonably be assumed that the lateral effective stresses on the
back of the wall had fallen to the active limit, but that the lateral effective stresses
in front of the wall should be calculated with the fully-passive earth pressure
coefficient Kp divided by a factor Fp in the range 1.5 to 2. Equilibrium
calculations carried out using a factored (reduced) value of Kp have traditionally
formed the basis of design of embedded retaining walls in granular materials, the
aim being that the wall should neither collapse outright nor deform excessively in
service.

4.1.2
Embedded walls in dry granular material
Seminal research into the behaviour of embedded sheet pile retaining walls in
sand was carried out during the 1950s by Rowe, who used both large scale
laboratory tests at normal gravity (1 g) and analytical methods. Rowe’s 1 g tests
on unpropped walls (Rowe, 1951) showed that for this type of structure,
conditions at collapse were reasonably well represented by the idealized
theoretical distribution of effective stresses shown in Figure 4.1(a), with Kp
based on the peak angle of shearing resistance and wall friction
At factors of safety greater than unity, however, a triangular
pressure distribution in front of the wall with the full passive pressure coefficient
Kp reduced by a factor Fp was found to overestimate the lateral stresses near the
toe of the wall, and hence bending moments. This is probably because with the
wall rotating about a point close to the toe, there is only limited movement of the
wall into the soil at this level, restricting the development of lateral stress.
Rowe’s model tests on walls propped or anchored at the crest (Rowe, 1952)
demonstrated that only a small movement of the prop or anchor was required for
the earth pressures behind the wall to fall to the active limit. The effective stress
distribution in front of a stiff wall was found to increase linearly with depth
below formation level, consistent with the application of a reduction factor Fp to
the full passive pressure coefficient Kp (Figure 4.1(b)). A wall whose deflexion
at formation level is less than or equal to its deflexion at the toe may be defined
for this purpose as stiff.
If the wall is more flexible, so that the deflection at formation level is greater
than that at the toe, the centroid of the lateral pressure distribution in the soil in
front of the wall is raised. This leads to a reduction in bending moments and
prop load to below the values obtained using the conventional limit equilibrium-
based calculation. Rowe (1955) presented a design chart which can be used to
apply a reduction factor to the calculated bending moments to allow for the
redistribution of lateral stresses in the soil in front of a comparatively flexible
retaining structure, such as a sheet pile wall. The main limitations of Rowe’s
moment reduction chart are that the reference calculation, in which it is assumed
that fully active conditions are achieved in the soil behind the retaining wall, is
applicable only to granular soils with initial in situ earth-pressure coefficients Ko
close to the active limit, and that the relative wall flexibility is quantified in
terms of an unusual and non-fundamental soil-stiffness parameter which is
difficult to measure or even estimate reliably. It should also be noted that
Rowe’s experiments were carried out on dry sand: in a real situation with non-
zero pore- water pressures, the moment reduction factor should only be applied
to the proportion of the overall bending moment which is due to effective
stresses.
Lyndon and Pearson (1984) reported the results of two centrifuge tests on
rigid unpropped retaining structures, embedded in ballotini (small glass balls) of
150 – 250 m diameter, with =19.5° and =38° (both measured in
plane strain) at a density of 1600 kg/m3. Their retaining wall was 185 mm high
(representing a real wall of overall length 11.1 m at a scale of 1:60) and 32mm
(1.92m at field scale) thick. Both faces of the wall incorporated slots 12mm
wide×57 mm long×12mm deep which, when filled with ballotini, would
encourage the development of full friction over most of the soil-wall interface.
Some of the slots housed boundary pressure cells comprising simply-supported
strain-gauged beams, which were used to obtain stress distributions both behind
and in front of the wall. Excavation in front of the wall was carried out in
increments, stopping and restarting the centrifuge at each stage, until collapse
occurred.
For problems of this type (i.e. retaining walls in dry sand), both the self-
weight stresses which drive failure and the ability of the soil to resist shear
increase with the applied g level. The results of the stress analyses shown in
Figure 4.1 may be presented in terms of the embedment ratio d/h at failure as a
function of the angle of shearing resistance of the soil. The unit weight of the
soil does not affect the embedment ratio at collapse, so that a wall which does
not fail at 1 g should in theory be stable at any gravity level to which it is
subjected in the centrifuge—provided that the angle of shearing resistance of the
soil does not change. In reality, the peak angle of shearing resistance of a soil of
a given void ratio (density) will decrease as the applied effective stress
increases, due to the suppression of dilation. Thus a small-scale model may be
stable at 1 g, but fail at a higher g level because the peak strength is reduced. If
this happens, the implication is that a large retaining wall of a given embedment
ratio will be less stable than a smaller wall of the same embedment ratio in the
same material, because the peak angle of shearing resistance which maintains
the stability of the smaller wall cannot be mobilized at the higher stresses which
exist in the soil around the larger wall.
The implication of the foregoing is that considerable caution must be
exercised in the selection of values for the extrapolation of 1 g model test
results to field scale structures. Rowe (1951, 1952) seems to have based his
back-analyses on :
In the application of his results to a larger structure, should be measured at
the highest applicable stress level, in which case it may be similar to the critical
state angle of shearing resistance, . A further point is that in small-scale
laboratory tests at 1 g, the range of stress is not great and the use of a uniform
value of in back-analysis is probably justifiable. At field scale, or in a
centrifuge model, the range of stress is much larger, and (quite apart from the
possibility of progressive failure) the existence of a uniform value of is
extremely unlikely. In these circumstances, the use of rather thanin back-
analysis and design would seem to be appropriate.
Thus one reason for testing models of retaining walls in dry sand in a
centrifuge rather than at 1 g is that the stress state of the soil (and hence its
stress- strain behaviour, including the potential or lack thereof for the
development of peak strengths) in the centrifuge is similar to that for a large
structure in the field. A second reason is that stresses, bending moments and
prop forces are increased in proportion to the g level at which the model is
tested, making them easier to measure reliably.
The two walls tested by Lyndon and Pearson (1984) would have been
expected to behave identically, but they did not. The first wall deformed
primarily by forward rotation about the toe, whereas the mode of displacement
of the second wall was predominantly translation. Lyndon and Pearson suggest
that this may have been due to the observed slight backward inclination of the
second wall at the start of the test, which perhaps allowed the enhanced self-
weight of this unusually thick retaining structure to act eccentrically, developing
a restoring moment in the opposite sense to the overturning moment exerted by
the retained soil. The effective stress distributions measured by Lyndon and
Pearson are in both cases broadly consistent with those of Rowe (1951).
The first wall (which deformed by rigid body rotation about a point near the
toe) failed catastrophically at an embedment ratio d/h of 0.414, but whether this
was at the test acceleration of 60 g or while the centrifuge acceleration was still
increasing towards this value following the removal of the last layer of soil in
front of the wall is not stated. The mobilized soil strength for an
embedment ratio of 0.414 is approximately 36° according to the stress analysis
shown in Figure 4.1 (a), using the earth-pressure coefficients given by Caquot
and Kerisel (1948) with wall friction Failure of the second wall was
less readily identifiable, although at an embedment ratio of 0.512 its movement
was quite large. The mobilized soil strength for an embedment ratio of 0.512 is
just less than 32° (again assuming ). These values of at failure lie
between the peak and critical state values quoted by Lyndon and Pearson, which
does not seem unreasonable for a material (ballotini) with a rather stronger
potential for dilation than most real soils. However, the difference between the
embedment ratios at failure suggests that it would be unwise in a calculation
forming the basis of a design to rely on the uniform mobilization of peak
strengths at collapse. A further point which emerges from the second of the tests
reported by Lyndon and Pearson is that the identification of failure is often not
obvious: the comparison of data from different research workers may therefore
be complicated by the various different definitions of failure they use.
King and McLoughlin (1993) summarized a series of tests on more flexible
unpropped embedded cantilever walls, retaining dry sand of typical grain size 0.
16 mm. The model wall was fabricated from stainless-steel sheet 2 mm thick,
representing a Frodingham No. 5 section sheet pile wall of bending stiffness 10.
2×104kNm2/m and total length 11m, at a scale of 1:92. Tests were carried out on
walls with three different surface finishes, giving angles of soil/wall friction
and intermediate values of to
retaining either dense sand (density ) or loose
sand (density ). Bending moments were measured
directly by strain gauges glued to the wall, because it is not usually possible to
incorporate boundary pressure cells into model embedded walls of realistic scale
thickness. As with the tests reported by Lyndon and Pearson (1984), the tests of
King and McLoughlin were carried out on the Liverpool University centrifuge,
and the soil in front of the wall was excavated in stages by stopping and
restarting the machine. This procedure is considered by King and McLoughlin
not to have affected the behaviour of the model in any significant way: given the
stabilizing effect of the likely increase in at centrifuge accelerations below
the test value, this conclusion is probably not unreasonable.
Table 4.1 compares the embedment ratios (d/h) at collapse observed by King
and McLoughlin with those predicted using the stress field calculation shown in
Figure 4.1(a) with Caquot and Kerisel’s earth-pressure coefficients for the
values of and stated. For the centrifuge tests, it is not possible to identify the
embedment ratio at failure precisely: a range is given, corresponding to the last
stable excavation depth, and the removal of a further 0.5 m of soil (at field scale)
from in front of the wall, which resulted in collapse. It may be seen that the
calculations using the peak angles of shearing resistance quoted by King and
McLoughlin consistently overestimate the embedment ratio at failure,
particularly in the case of the tests on dense sand. Unfortunately, the critical
state angle of shearing resistance is not given by King and McLoughlin. Since,
however, the predicted embedment ratios for the loose sand models ( for
which is probably only slightly in excess of ) are close to or just outside the
upper limits observed in the centrifuge tests, it seems likely that the use of earth
pressure coefficients based on in the stress field calculation shown in
Figure 4.1 (a) would lead to a generally correct or only slightly conservative
prediction of the collapse limit state.
The use of a limit-type stress distribution with earth pressure coefficients
based on peak soil strengths and a passive pressure reduction factor Fp was
found by King and McLoughlin generally to overestimate maximum bending
moments under conditions where Fp=1.5. This is consistent with the earlier work
of Rowe (1951), and arises because the centroid of the pressure distribution in
front of the wall is higher than in the idealized distribution shown in Figure-
4.1(a).
For King and McLoughlin’s rough walls with = , however, the
maximum observed bending moments were very close to those calculated using
the limit-type stress distribution. The measured bending moments in the rough
walls were up to 80% higher than those in the smooth walls, and up to 50%
higher than those in the walls of intermediate roughness.
Table 4.1 Comparison of actual and theoretical retained heights (in metres at field scale)
at collapse for centrifuge model tests of unpropped walls in dry sand by King and
McLoughlin (1992)
Wall roughness was not investigated by Rowe. Although there are insufficient
data to be certain, and one possibility is that their calculations were based on
unrealistically high values of and therefore overestimated the real factors of
safety quite significantly, King and McLoughlin’s results would tend to militate
against the use of moment reduction factors in design where the wall is very
rough.
King and McLoughlin also observed that the deformations of all walls at
Fp=1. 5 (based on peak strengths) were much larger than would be tolerated in
reality. This demonstrates one of the major disadvantages of using a factor on
passive pressure coefficient Fp in design, which is that the additional embedment
required to increase the numerical value of Fp from 1.0 to 1.5 or even 2.0 may be
very small, especially when the angle of shearing resistance of the soil is high
(Simpson, 1992). The problem is exacerbated by the inappropriate use of
( rather than ) as a design parameter.

4.1.3
Embedded walls in clay
Since the advent in the 1960s of in situ methods of wall installation, such as
diaphragm walling and secant and contiguous piling, embedded retaining walls
have become increasingly constructed from reinforced concrete to retain clay
soils. It gradually became apparent that the use of factored limit-based methods
of analysis in the conventional way led to the calculation of large depths of
embedment (Hubbard et al., 1984; Garrett and Barnes, 1984). This was
primarily because the factor Fp was applied to the passive pressure coefficient
rather than to the soil strength directly, and the numerical values of Fp which had
been found satisfactory for use with granular soils having a critical state angle of
shearing resistance of 30° or more could lead to apparently uneconomic
depths of embedment in clay soils with values of only 20° or so. A further
concern arose from the fact that the stress history of a typical overconsolidated
clay deposit is such that the in situ lateral stresses are comparatively high,
perhaps approaching the passive limit. Under these conditions, the conventional
assumption that more movement is required to reach the active state than the
passive is clearly suspect, and there is a possibility that a wall which moves
sufficiently for the lateral stresses in the retained soil to fall to the active limit
will have become unserviceable. Alternatively, if the embedment of the wall is
sufficient to limit movement, the lateral effective stresses exerted by the retained
soil might be considerably in excess of those calculated assuming fully active
conditions. There was also (and there still is) some uncertainty concerning the
length of time it would take for pore water pressures in clay soils to reach
equilibrium following the construction and excavation in front of the wall, and
— particularly for temporary works—the need to consider both short- and long-
term conditions in design.
Bolton and Powrie (1987, 1988) carried out a series of centrifuge model tests
using the Cambridge University centrifuge to investigate both the collapse and
serviceability behaviour of diaphragm walls in clay. Figure 4.2 shows a cross-
sectional view of a typical model, which represents a wall with a retained height
of 10m at a scale of 1:125, deforming in plane strain. Walls were made from
either 9.5mm or 4.7mm aluminium plate, corresponding to full-scale flexural
rigidities of approximately 107kNm2/m and 1.2×106kNm2/m, respectively.
The clay used in the model tests was speswhite kaolin, which was chosen
primarily because of its relatively high permeability, k=0.8×10−9m/s. The kaolin
sample was consolidated from a slurry to a vertical effective stress of 1250 kPa
before being unloaded to a vertical effective stress of 80kPa. The clay was
removed from the consolidation press, and the model was prepared.

Figure 4.2 Cross-section through a model diaphragm wall (dimensions in millimetres at

model scale). Reproduced from Bolton and Powrie (1987). Institution of Civil Engineers,
with permission.

With tests on clay soils, pore-water pressures (and the control thereof) cannot
realistically be neglected. The final stage in the preparation of a clay model
before the test proper commences is therefore usually a period of reconsolidation
in the centrifuge, during which the clay sample is brought into equilibrium under
its enhanced self-weight at the appropriate g level. After this time, the clay is in
a state which corresponds to idealized field conditions, with effective stresses
increasing with depth and hydrostatic pore-water pressures below the
groundwater level set by the modeller. The one possible exception to this is a
mechanistic study, in which the g level might be increased steadily to initiate the
rapid collapse of a model in a clay sample having a constant profile of undrained
shear strength tu with depth. In such a case, a low-permeability clay should be
used to ensure that any changes in volume—and hence in undrained shear
strength—which occur during the gradual increase in centrifuge acceleration are
insignificant.
At the end of the reconsolidation period, it is necessary to simulate excavation
in front of the wall without stopping the centrifuge. The most common
technique, as used by Bolton and Powrie (1987, 1988), is to form the excavation
before mounting the model on the centrifuge. The soil removed is replaced by a
rubber bag filled with zinc chloride solution 1 mixed to the same unit weight as
the clay. A valve-controlled waste-pipe is used to drain the zinc chloride
solution from the rubber bag, simulating the excavation of soil from in front of
the wall, at an appropriate stage following reconsolidation of the clay sample in
the centrifuge.
The vertical stress history imposed on their clay samples by Bolton and
Powrie corresponded to the removal by erosion of about 150m of overlying soil,
which is reasonably representative of a typical overconsolidated clay deposit.
Although the in situ lateral stresses in such a soil are likely to be high, the slurry
trench phase of diaphragm wall construction is certain to result in a significant
alteration to this initial condition. The exact effect of the installation of the wall
will depend on a number of factors, but an approximate analysis can be used to
estimate the likely range of the pre-excavation lateral earth pressure coefficient
(Powrie, 1985). In London clay, for example, the slurry trench phase might
reduce an initial effective stress earth-pressure coefficient of 2.0 to between 1.0
and 1.2. A pre-excavation lateral earth- pressure coefficient Ko of unity might
therefore be considered appropriate for centrifuge tests on model diaphragm
walls, which start with the wall already in place. As the zinc chloride solution in
the tests reported by Bolton and Powrie was mixed to the same unit weight as
the soil it replaced, the lateral stresses in front of the wall above formation level
were consistent with the condition Ko=1 after reconsolidation in the centrifuge.
The bending moments measured during reconsolidation were generally very
small, indicating that Ko=1 was quite closely achieved in the retained soil as
well.
Although it might at some time in the future be possible to replicate exactly in
a centrifuge the processes of diaphragm wall installation and excavation in front,
this is at present an unattainable ideal. In investigating the long-term (post-
excavation) behaviour of diaphragm walls in clay, it is therefore necessary to
start the centrifuge test (like Bolton and Powrie) with the wall already in place.
Even if it is attempted to replicate the likely pre-excavation pore pressures and
effective stresses rather than their in situ values, the problem remains that in the
centrifuge model the changes in stress between the in situ and pre-excavation
conditions will probably have to be applied across the entire sample, whereas in
reality they would be confined to the vicinity of the wall. It seems reasonable to
argue, however, that the stress state of the soil remote from the wall should not
influence the behaviour of the wall to any great extent, and that it is more
important to model correctly the stress state of the soil adjacent to the wall.
A second concern relates to the mobilization of soil-wall friction during
reconsolidation. The stress history of the clay samples used by Bolton and
Powrie was such that overall soil surface settlements were observed during
reconsolidation in the centrifuge. If there were any relative soil-wall movement
during reconsolidation, the soil would have tended to move downward relative
to the wall. On excavation, the relative soil-wall movement would have
continued in the same sense as far as the retained soil was concerned, but its
direction would have been reversed for the soil remaining in front of the wall.
This might have increased the rate of mobilization of soil-wall friction on the
active side of the wall (where it makes comparatively little difference to the
earth pressure coefficients), but delayed the mobilization of soil-wall friction on
the passive side (where its effect is considerable). Overall, therefore, the models
would be expected to err on the conservative side in that they would tend to
overestimate the displacements of a corresponding full-scale construction, in
which the mobilized soil-wall friction at the start of the excavation process was
zero.
The stress history of an overconsolidated clay deposit immediately prior to
diaphragm wall installation might have been one-dimensional swelling (on
geological unloading or an increase in groundwater level) or compression
(following reloading or underdrainage). It might, in principle, be attempted to
replicate any of these in carrying out a centrifuge model test. It should, however,
be noted that the argument for conservatism from the point of view of the rates
of mobilization of soil-wall friction would not necessarily apply if the stress
history to which the sample had been subjected during preparation had been
different.
The layout of the instrumentation installed in a typical model diaphragm wall
as tested by Bolton and Powrie is shown in Figure 4.2. Bending moments were
measured using strain gauges glued to the wall, which were water-proofed and
protected by a coating of resin 2mm thick. 2 The resin coating also served to
increase the thickness of the model wall so that both the thickness and the
flexural rigidity of a typical reinforced concrete field structure were modelled
correctly. Pore-water pressures were measured using Druck PDCR81 miniature
transducers, and soil-surface settlements were measured using Sangamo
(Schlumberger) LVDTs. Black markers embedded into the front face of the
model enabled vectors of soil movement to be measured from photographs taken
at various stages during a test. The density of the kaolin used in the centrifuge
tests was approximately 1768 kg/m3, the critical state angle of shearing
resistance =22° and the peak angle of shearing resistance = 26°. The
angle of friction between the soil and the resin used to coat the model wall was
found (in shear box tests carried out at appropriate normal stresses) to be the
same as the critical state angle of shearing resistance of the soil.
In most tests, a full-height groundwater level on the retained side of the wall
was modelled and purpose-made silicone rubber wiper seals were used to
prevent water from leaking between the edges of the wall and the sides of the
strongbox. During the initial reconsolidation period, water was supplied at the
elevation of the retained ground to the soil surfaces behind and in front of the
wall and to an internal drain formed by a layer of porous plastic overlying
grooves machined into the strongbox at the base of the soil sample. 3 After
excavation, solenoid valves were used to switch drainage lines to isolate the base
drain and to keep the water level within the excavation drawn down to the
excavated soil surface. It can be shown that the presence of the isolated drainage
sheet at the base of the model results in a steady state seepage regime similar to
that for a somewhat deeper clay stratum with an impermeable boundary at the
base.
In addition to the considerations of the stress state and the stress-strain
behaviour of the soil, one of the main reasons for carrying out centrifuge model
tests on in situ retaining walls in clays is that the time taken for the pore-water
pressures to move towards their long-term equilibrium state following
excavation in front of the wall are reduced by a factor of N2 (in a 1/N-scale
model) compared with a full-size structure. Unless the wall is well supported by
props during excavation, it is usually the long-term conditions which are the
more critical. This is because excavation processes tend to cause transient pore
pressures which are below the long-term steady-state values, aiding stability in
the short-term. At a scale of 1:100, the changes in pore pressure which would
occur over a period of approximately 14 months at field scale may be observed
in one hour in a centrifuge model made from the same soil. A further reason for
carrying out centrifuge rather than 1 g tests is that in a 1 g test, the relationship
between the undrained shear strength and the self-weight stresses, which
governs the short- term behaviour, is likely to be so unrepresentative of any real
situation as to be effectively meaningless.
The centrifuge tests on model diaphragm walls carried out by Bolton and
Powrie demonstrated quite graphically the calamitous effect that groundwater
can have on these structures. With an excavated depth of 10 m at prototype scale
and a full-height groundwater level in the retained soil, unpropped walls of 5 m
and 10 m embedment failed almost immediately on excavation. The initial soil
deformations were so large that a tension crack opened between the wall and the
soil. Surface-water ponding in the settlement trough behind the wall filled the
crack, pushing the wall over almost instantaneously. The retained ground was
left standing in a cliff as shown in Figure 4.3. Although in a field situation
surface water is unlikely to be as readily available as it was in the model tests,
either a burst water main or a thin gravel aquifer could have a similar effect.
An analysis based on the limiting lateral stresses associated with the estimated
undrained shear strength profile of the clay sample used in the centrifuge model
tests indicates that the theoretical maximum depth of a tension crack is 5.7 m dry
and 31 m flooded. In the absence of water to fill cracks, an unpropped wall
retaining 10m of clay would require only a small embedment of about 2 m to
maintain short-term equilibrium. If a crack between the wall and the soil should
flood, however, it could remain open to a considerable depth transferring
hydraulic thrust to the wall.

Figure 4.3 Flooded tension crack failure of unpropped model diaphragm wall in clay
with h=d=10m at field scale. Reproduced from Bolton and Powrie (1987). Institution of
Civil Engineers, with permission.

The wall would then be forced outwards as the crack widened, provided that
the rate of inflow was sufficient to maintain the hydraulic head. According to a
limit equlibrium stress analysis, the embedment at which a water-filled crack
 could no longer cause complete failure of the wall would be 14 m: this is
consistent with the observed short-term behaviour of the model walls. With a
nominally full-height groundwater level on the retained side, a wall of 15 m
embedment suffered large movements on excavation but remained in contact
with the soil. Wall movements continued after excavation with no sign of
abatement. The deformations were clearly influenced by the development, some
time after excavation, of one or more slip surfaces: Figure 4.4 illustrates the final
pattern of rupture lines. The deepest unpropped wall tested had an embedment of
20 m which, with a full-height groundwater level in the retained soil, was still
insufficient to prevent unacceptably large ground movements.

Figure 4.4 Rupture pattern in the clay retained by an unpropped model diaphragm wall
with h = 10m and d = 15m at field scale. Reproduced from Bolton and Powrie (1987).
Institution of Civil Engineers, with permission.

The long-term behaviour of walls having sufficient embedment to prevent

short-term collapse must be analysed in terms of effective stresses and pore
water pressures. The equilibrium of the unpropped walls of 15m and 20m
embedment at an instant near the end of each test was investigated using the
assumed distribution of effective stresses shown in Figure 4.1 (a), together with
the measured pore water pressures. The soil strength which satisfied the
equations of horizontal and moment equilibrium for the wall was found by
iteration, assuming that the same soil strength applied on both sides of the wall
and that the angle of soil-wall friction = . Earth-pressure coefficients were
taken from Caquot and Kerisel (1948).
For the wall of 15 m embedment some time after excavation (corresponding
to about six years at field scale), the back-analysis indicated a mobilized angle
of shearing of 21.7°, almost identical to the critical state value. The
analysis also predicted a pivot point at zp=14m below formation level, which
compares well with the measured soil displacements which indicated a pivot at
13.5m depth. The main rupture surface shown in Figure 4.4 intersected the wall
at about 10m below formation level. For the wall of 20m embedment, the
mobilized angle of shearing required for equilibrium in the long-term (i.e. after
about 14 years at field scale) was 19.7°, slightly smaller than the critical state
angle. This is consistent with the less-damaging deformation and the absence of
any clear rupture surface, but the wall would in practice be deemed to have
suffered a serviceability failure.
In the case of a wall of 5 m embedment propped at the crest, significant
rupture surfaces developed in the soil at a time corresponding to approximately
two years at field scale after excavation (Figure 4.5). The steepness of the main
rupture surface near the top of the wall is due to the kinematic restraint imposed
by the prop: pure sliding would be impossible at this level along a non-vertical
slip surface. With the pore water pressures measured at the end of the test (7.38
years at field scale following excavation), the static equilibrium of the zones of
soil bounded by the rupture surfaces requires the mobilization of soil strengths
of between 22° and 26°. This is reasonable, since these values represent,
respectively, the estimated critical state and peak strengths of the kaolin used in
the centrifuge tests. However, the equivalent earth pressure coefficients deduced
from the thrusts arising from the back-analysis of the collapse mechanism are
0.23 on the retained side and 4.48 in front of the wall. According to Caquot and
Kerisel, these earth-pressure coefficients correspond to =35° and = =22°
behind the wall, and =26° and =22° in front. Strengths as large as these on the
retained side would not generally be invoked on the basis of the data from
laboratory tests on soil elements, and the discrepancy results from the use of an
earth-pressure coefficient which is conservative in that it takes no account of the
kinematic restraint imposed by the prop. Tests were also carried out on walls of
10m and 15m embedment propped at the crest. In these cases the embedment
was sufficient to prevent failure, and soil movements were correspondingly
smaller.

Figure 4.5 Rupture pattern in the vicinity of a model diaphragm wall propped at the crest
with h = 10m and d = 5m at field scale. Reproduced from Bolton and Powrie (1987).
Institution of Civil Engineers, with permission.

For unpropped walls, it was observed that significant soil movements

occurred mainly in the zones defined approximately by lines drawn at 45°
extending upward from the toe of the wall on both sides. This pattern of
deformations might be represented for an unpropped embedded wall VW
 rotating about a point O near its toe by the idealized strain field shown in Figure
4.6. For stiff walls in which bending effects are small, the shear strain increment
in each of the six deforming triangles is uniform and equal to twice the
incremental wall rotation . The stress analysis shown in Figure 4.1(a) can be
used for an unpropped wall of any geometry to calculate the uniform mobilized
soil strength required for equilibrium . The remoteness of the wall from
failure could then be quantified by its factor of safety based on soil strength, tan
/tan . It is more rational to apply a factor of safety to soil strength than
to a parameter such as the depth of embedment or the passive pressure
coefficient, which (as the centrifuge tests reported by King and McLoughlin
show) can result in a wall which is actually very close to failure.
The shear strain corresponding to a given mobilized soil strength can be
determined from an appropriate laboratory test on a representative soil element.
The idealized strain field shown in Figure 4.6 can then be used to estimate the
magnitude of the wall rotation and soil movements.

Figure 4.6 Idealized strain field for an unpropped model diaphragm wall rotating about a
point near its toe. Reproduced from Bolton and Powrie (1988). Institution of Civil
Engineers, with permission.

For example, a calculation following Figure 4.1 (a) based on the undeformed
geometry and the pore-water pressures measured immediately after excavation
in the centrifuge test on the unpropped wall of 20 m embedment indicates that a
mobilized soil strength of 17.5° is required for equilibrium with = .
The corresponding shear strain according to plane strain laboratory test data is
1.1%, which with the calculated pivot depth zp of 18.8 m leads to a crest
deflection of 158 mm at field scale.
Figure 4.7 Comparison of (a) measured and (b) calculated soil movements during
excavation in front of an unpropped model diaphragm wall with h = 10m and d = 20m.
Reproduced from Bolton and Powrie (1988). Institution of Civil Engineers, with
permission.

Figure 4.7 shows that the measured and calculated soil movements are very
similar. A similar calculation for walls propped at the crest is described by
Bolton and Powrie (1988), and a worked example is given by Bolton et al.
(1989, 1990a). The allowable mobilized soil strength for design purposes will
depend on a number of factors, including the initial stress state of the soil and its
stiffness measured following appropriate stress and strain paths.
It would seem that the limit-based stress distribution shown in Figure 4.1(a)
may be more readily adapted (by using factored soil strengths) to the
serviceability analysis of cantilever retaining walls in clays than in granular
materials (cf. Rowe, 1951; Lyndon and Pearson, 1984; King and McLoughlin,
1993). If this is so, it is probably because the clay remaining in front of the wall
below formation level is brought almost to passive failure by the removal of the
overlying soil, so that very little wall movement is required to mobilize high
horizontal stresses, even near the pivot.

4.1.4
Other construction and excavation techniques
The draining of fluid from a rubber bag to simulate the stress changes which
result from excavation is sometimes criticized on the grounds that it is only
possible to model a pre-excavation earth pressure coefficient of unity. This is not
so: Lade et al. (1981) used paraffin oil with a density of 7.65 kg/m 3 to give a
pre-excavation earth pressure coefficient of approximately ( ) in a
granular material. At the other extreme, Powrie and Kantartzi (1993) describe
the use of sodium chloride solution of density 1162 kg/m3 , contained in a rubber
 bag filled to above the level of the soil surface, to impose a profile of Ko with
depth similar to the in situ conditions in an overconsolidated clay deposit in the
field (Figure 4.8). Following reconsolidation, the salt solution is drained to the
level of the soil surface to simulate the stress changes caused by excavation of a
diaphragm wall trench under bentonite slurry. This technique is being used in an
investigation into the effects of diaphragm wall installation in overconsolidated
clays, currently in progress at Queen Mary and Westfield College (University of
London) using the geotechnical centrifuge operated jointly with City University.

Figure 4.8 Pre-excavation effective earth pressure coefficients in centrifuge model tests
to investigate installation effects of diaphragm walls in clay. Tedd et al. (1984), cited in
Powrie and Kantartzi (1992); Symons and Carder (1990)—centrifuge model. Reproduced
from Powrie and Kantartzi (1993). Institution of Civil Engineers, with permission.

A second, more valid, criticism of the ‘draining fluid from a rubber bag’ method
of simulating excavation in the centrifuge is that the rate of reduction of lateral
stress at any depth is approximately constant until the fluid level reaches that
depth, which would not necessarily be the case in a real excavation. Other
methods of simulating excavation in the centrifuge include the removal of rigid
supports (Craig and Yildirim, 1976). Alternatively, the soil within the
excavation could be contained within a flexible porous fabric bag, which could
be winched clear at the appropriate stage of the centrifuge test using an electric
motor. The latter method is proposed by Ko et al. (1982), but it is not clear
whether it has actually been successful in practice. Further more, unless the
excavation were to proceed in several stages using a number of fabric bags, the
problem concerning
the rate of lateral unloading would not be overcome.
The evolution of centrifuge modelling techniques to investigate the behaviour
of embedded retaining walls has reflected the development of methods of
construction of these structures in geotechnical engineering practice. Centrifuge
model tests carried out by Powrie (1986) demonstrated the efficiency of props at
formation level in terms of minimizing wall movements for a given embedment
ratio. The main shortcoming of these tests (of which some of the results were
summarized by Powrie and Li, 1991) was that the propping system had to be
installed prior to reconsolidation and excavation in the centrifuge. New tests
currently in progress at Queen Mary and Westfield College, in which the
behaviour of in situ walls propped at both crest and formation level is under
investigation, incorporate props with locking devices so that the props will not
begin to take load or resist movement until they are required to do so by the
modeller. Walls of this type are being increasingly used for underpasses at major
interchanges on arterial roads in urban areas.

4.1.5
Centrifuge tests for specific structures
The model tests described so far were undertaken primarily to identify and
investigate the fundamental mechanisms of deformation or collapse associated
with a particular class of retaining structure. Centrifuge modelling is equally
suited for use in connection with individual projects, especially where the effect
of ground conditions is uncertain or some aspect of the proposed design is
untried. An example of this is given by Rigden and Rowe (1975), who carried
out centrifuge model tests to investigate the feasibility of using an unreinforced
concrete diaphragm wall, circular on plan, to retain the ground around a
proposed underground multi-storey car park in Amsterdam. One of the main
issues was whether the hoop compression in the circular diaphragm wall would
be sufficient to prevent tensile cracking due to longitudinal bending. The
centrifuge model tests indicated that tensile stresses in the circular diaphragm
wall could be substantially eliminated, provided that the depth of embedment
below formation level was not excessive.

4.2
Gravity and L-shaped retaining walls

4.2.1
Background
The scientific study of gravity retaining walls can be traced back at least as far
as the start of the 18th century, and was one of the subjects of Coulomb’s
celebrated memoir on statics (Heyman, 1972). Perhaps for this reason, the basic
problem of the stability of gravity retaining structures does not seem to have
been investigated very extensively using centrifuge modelling techniques.
Centrifuge model studies on gravity-type retaining walls (including L-walls and
bridge abutments) have tended to focus on the additional analytical difficulties
 which result from (for example) the application of a line load to the retained soil
surface (Bolton and Mak, 1984), cyclic surcharge and lateral loading (Hird and
Djerbib, 1993) and the effects of construction on a clay foundation (Bolton and
Sun, 1991).

4.2.2
Line loads behind L-shaped walls
The model configuration used by Bolton and Mak (1984) to investigate the
behaviour of an L-shaped retaining wall with a line load applied to the surface of
the retained soil is shown in Figure 4.9. The dimensions shown are those of the
model, which at a scale of 1:60 represented a field structure with a retained
height of 8.82m. The base and the rear face of the wall incorporated Cambridge
contact stress transducers (Stroud, 1971), which was why the wall was so thick.

Figure 4.9 Cross-section through centrifuge model of an L-shaped retaining wall with a
strip footing at the surface of the backfill. Total wall height H=147mm. Reproduced from
Bolton and Mak (1984).
The soil used in the experiments was dry 14/25 Leigh ton Buzzard sand, with
= 33° and up to 48°, depending on the relative density. The stress
conditions under which the values were measured are not stated. At a centrifugal
acceleration of 60 g, the load applied to the footing was increased until the
retaining wall-footing system failed. A series of tests was carried out, in which
the relative density of the sand (and hence the peak angle of shearing resistance
), the length B of the wall base, and the distance DF between the footing and
the retaining wall were varied.
From the displacement patterns and contours of shear strain measured during
the centrifuge tests, an analysis based on the limiting equilibrium of a wedge of
soil below the footing and a block comprising the wall and the soil contained
Figure 4.10 Idealized mechanism of failure for model wall-footing system. Reproduced
from Bolton and Mak (1984).
within the ‘L’ was proposed (Figure 4.10). This analysis was used to relate the
mobilized angles of shearing resistance on the planes VV and II to the load
on the footing. Although values of on these planes in excess of were
inferred, the rates of mobilization of soil strength and the instants of maximum
mobilized soil strength along VV and II were not coincident: this would suggest
that peak strengths are not mobilized uniformly at collapse. The centrifuge
model tests indicated that the stability of the wall was controlled primarily by
the bearing capacity of the base, which was in turn dependent on the soil
strength mobilized along VV. The peak footing loads measured in the
centrifuge, corrected for the effects of side friction in the model (which were in
this case quite considerable), were generally overpredicted by the idealized
wedge calculation. This is only to be expected from what is essentially an upper
bound calculation, with the blocks ACEF and CDE forming a kinematically
admissible mechanism analysed by statics rather than by a work balance.

4.2.3
Cyclic loading of L-shaped walls
Hird and Djerbib (1993) reported the results of centrifuge tests on stiff and
flexible L-shaped cantilever walls, retaining dry 14/25 Leighton Buzzard sand,
subjected separately to cyclic applications of lateral load to the crest of the wall
(such as might be imposed by a fixed bridge deck due to temperature effects),
and surcharge to the backfill. A cross-section through the model is shown (with
dimensions at model scale) in Figure 4.11. At a scale of 1:60, the retained height
of the field structure modelled is approximately 9 m. The more flexible wall was
fabricated from 3.2mm thick mild steel plate, giving a bending stiffness
equivalent to a 0.4m thick reinforced concrete wall stem, and was instrumented
with strain gauges to measure bending moments. The stiff wall was made from
22 mm thick mild steel plate, representing a reinforced concrete structure
approximately 3m thick at field scale, and was instrumented with Stroud-type
Figure 4.11 Cross-section through centrifuge model of an L-shaped retaining wall
subjected to cyclic lateral loading at the crest and cyclic surcharge loading to the backfill.
All dimensions in mm. Figures in parentheses refer to stiff wall. Asterisked dimensions
increased by 19mm in lateral loading tests. Reproduced from Hird and Djerbib (1993).
Institution of Civil Engineers, with permission.

contact stress transducers. Hird and Djerbib consider that these models would
represent the extremes of wall stiffness likely to be encountered in practice. The
density of the sand as placed was 1700 kg/m3. The peak angle of shearing
resistance was estimated on the basis of research by Stroud (1971) to be 46.
5°, but the detailed reasoning behind this choice of is not given.
The cyclic surcharge was applied to the backfill by pressurizing each of four
airbags behind the retaining wall in turn, simulating a 65 mm wide load (3.9 m
at field scale) moving towards and then away from the wall. The lateral load was
applied at the top of the wall by means of a double-acting pneumatic ram. A stiff
beam was used to distribute the load evenly along the top of the wall during the
pushing stroke, but the pulling force was transmitted to the wall by means of
hooks at the edges, leading to some out-of-plane bending in the case of the more
flexible wall. Certain other technical difficulties are described by Hird and
Djerbib (1993).
The bending moments measured in the more flexible wall on reaching the
required centrifuge acceleration of 60 g were consistent with active conditions
based on the value of 46.5°. In the case of the stiff wall, the earth-
pressure coefficients at this stage were (as might be expected) rather higher,
increasing to just above Konc with = 46.5° ( ) at the base. In
both cases, significant changes in bending moment/horizontal stress and lateral
wall movement were associated only with the application or removal of the
surcharge closest to the wall. However, the residual bending moments/
horizontal stresses and wall movements at zero surcharge gradually increased as
cycling continued, with no steady-state value apparent after 15–20 load cycles.
Figure 4.12 Stress analysis for reinforced soil retaining wall.
(equation 4.3). Reproduced from Bolton and Pang (1982).
Institution of Civil Engineers, with permission.

The stiff wall showed more significant reductions in horizontal stress and
larger movements back towards its initial position following the removal of each
surcharge than the more flexible wall: in other words, the ‘locking in’ of
horizontal stresses following the application of a surcharge was more complete
for a more flexible retaining wall. This is qualitatively consistent with the
idealized theories for lateral stresses induced by compaction described by Broms
(1971) and Ingold (1979). These theories were developed for use with walls
where the backfill is compacted in comparatively thin layers, however, and so
are not strictly applicable in the present case. Some indication of the complexity
of the problem addressed by Hird and Djerbib (1993) is given by the changes in
the stress state of the soil indicated by the Stroud cells in the case of the stiff
wall, which showed a reversal in the direction of the shear stress at the soil-wall
interface during each load cycle. While this behaviour might be expected in
theory, it is nonetheless instructive to see it demonstrated in practice.
The tests in which the retaining walls were subjected to cycles of lateral load
showed similar progressive changes (‘ratcheting’) of displacements and bending
moments/lateral stresses. Although the difference between successive cycles
diminished as the number of cycles increased, a true steady state had not been
reached for either the stiff wall or the more flexible wall after 50 cycles, when
the centrifuge tests were terminated.
Where cyclic loads are expected to be applied throughout the life of a
structure, their effects must be taken into consideration by the designer. While it
is some reassurance that the rates of increase of deflection and bending moment
change are smaller at the high end of the cycle (which would normally be the
more critical condition for design) than at the low end, appropriate design
assumptions are generally far from clear. Although the failure of plane retaining
walls following a progressive increase in lateral stresses due to the cyclic effects
may be rare, there are certain classes of structure (for example reinforced
concrete ring filter beds) for which it is certainly a problem (England, 1992).
This is an area of considerable uncertainty at present, in which further research
is required.

4.3
Reinforced soil walls, anchored earth and soil nailing

4.3.1
Reinforced soil walls
In addition to the modes of collapse relevant to conventional retaining walls
(monolithic rotation or sliding, triggering of a landslide, and failure of the
materials used to construct the wall itself), the designer of a reinforced soil wall
must guard against the possibility of the failure of the reinforcement, either in
tension or by slippage. Bolton and Pang (1982) presented a simple yet rational
stress analysis (Figure 4.12), based on the limiting equilibrium of a block of soil
BCFE behind the retaining wall. This block of soil has width L (the length of the
reinforcement strips) and a general depth Z. It is assumed that the vertical
boundaries BE and CF are frictionless, and that the horizontal effective stresses
on CF increase linearly with depth Z, with and in the
absence of pore-water pressures, where is the unit weight of the retained soil,
Ka is the active earth-pressure coefficient and is
the angle of shearing resistance of the soil. The vertical stress distribution along
the base EF of the block of soil is assumed to decrease linearly from the back of
the facing panels to the end of the reinforcement, with a mean value of The
maximum value of the vertical stress on this boundary is obtained from the
condition of moment equilibrium about M, the midpoint of EF. This analysis
may be used to estimate the load on each facing panel and reinforcement strip on
the verge of failure, and hence the factors of safety against tensile failure, FT,
and frictional failure (slippage) FF, in the most critical reinforcement strip:
(4.1)
(4.2)
where P is the reinforcement load at tensile failure, Sv and Sh are the height and
width of a facing panel, respectively, B is the width of a reinforcement strip, and
is the coefficient of friction between reinforcement strip and the soil. The load
on a reinforcement strip, Tmax, at depth Z is (equation
4. 3, Figure 4.12), and its pull-out resistance is
Prior to Bolton and Pang (1982), attempts to investigate the validity of this
approach had centred on model tests carried out at 1 g, on reinforced soil
retaining walls 200–500 mm in height. Generally, the stress analysis shown in
Figure 4.12 had been found to underestimate the observed tensile strength of the
models by factors of up to 1.8. Difficulties were also encountered with local
weaknesses at the joints between the facing panels and the reinforcement strips.
The main problems with 1 g model tests on reinforced soil retaining walls in
granular soils relate to (i) the low stress levels in the model, which permit the
development of peak strengths almost uniformly down the back of the
wall that would be unlikely to occur in the field, (ii) the delicate reinforcement
needed because the soil stresses are low, which means that considerable
additional stiffness will result from the attachment of strain gauges, and (iii) that
significant local imperfections (for example at the joints with the facing panels)
are likely to occur. These problems can be overcome by testing models at an
enhanced gravity level in the geotechnical centrifuge.
Bolton and Pang (1982) report the results of a series of plane strain centrifuge
tests on model reinforced soil walls, 200mm high, retaining dry sand of density
1723 kg/m3. The peak angle of shearing resistance of the sand was 49° at a
vertical effective stress of 50kPa, and 43° at an effective stress of 400 kPa. The
facing units were flexible, and four different types of reinforcement strip were
used, with friction coefficients =tan between 0.16 and 0.75. The ratio of the
reinforcement length to the retained height L/H varied between 0.5 and 1.5. In
each case, the centrifuge acceleration was gradually increased until collapse
occurred. Tensile failure of reinforced soil retaining walls may be investigated in
this way because the factor of safety FT (equation 4.1) will reduce as the unit
weight of the soil in the centrifuge acceleration field (=png) is increased with
increasing centrifuge acceleration ng. Although the factor of safety against pull-
out (slippage) of the reinforcement FF shows no direct dependence on the unit
weight in the centrifuge acceleration field, the suppression of dilation as the self-
weight stresses of the soil are increased will in actuality have some effect
because of the resulting reduction in peak soil strength .
Collapse occurred as a result of the failure of the reinforcement in tension in
some of the models, and by slippage of the reinforcement in others. The spread
of values of FT and FF at collapse, calculated according to the stress analysis
shown in Figure 4.12 and equations (4.1) and (4.2) respectively, is indicated in
Figure 4.13. It may be seen that pull-out failure is generally well-predicted using
equation (4.2), although there is a tendency to err on the unsafe side, with
collapse in one test occurring at FF=1.33. According to Bolton and Pang (1982),
this lack of conservatism could be reduced by the use of the minimum (rather
than average) values of soil strength and soil-reinforcement coefficient of
friction obtained from laboratory tests, to give factors of safety FF at collapse
of unity ±15%, which is within the range of experimental error and repeatability.
Figure 4.13 Values of (a) FF and (b) FT at collapse of centrifuge models of reinforced soil
retaining walls, (a) Mark I tests (C): SS (stainless steel), L/H=1.00 (C31) to 0.43 (C48),
MS (mild steel), L/H=0.75; AL (aluminium), L/H=0.50. Mark II tests (P) SS, L/
H=0.80. (b) Mark I tests (C): AL, L/H=1.5 (C6, 7,8) to 0.5 (C14,18, 51, 52, 54). Mark
II tests (P): AL, L/H=0.5. Reproduced from Bolton and Pang (1982). Institution of Civil
Engineers, with permission.

The use of critical state soil strengths might well have eliminated entirely the
tendency of the calculation sometimes to err on the unsafe side, and is perhaps
therefore to be recommended in design.
In contrast, the values of FT at collapse (calculated according to equation 4.1)
vary between 0.5 and 1.06, indicating that the stress analysis shown in Figure
4.12 is generally (but not always) unduly conservative. Tensile failure was well
predicted by equation (4.1) only in tests on walls with narrow reinforced zones
(L/H=0.5), which were already close to pull-out failure. In other tests, it seemed
that some redistribution of reinforcement loads was possible after the most
critical reinforcement strip had yielded, thereby delaying the onset of collapse.
In order to clarify the apparent over-conservatism of the stress analysis shown
in Figure 4.12 in predicting tensile failure, Bolton and Pang carried out a second
series of centrifuge tests. The basic geometry of the model was the same as that
of the first series, but the reinforced soil zone was underlain by a wooden, rather
than a sand, foundation. This enabled the vertical stress distribution at the base
of the reinforced zone (AD in Figure 4.12) to be measured, without significantly
affecting the behaviour of the model (some of the series 1 tests were duplicated
with the new arrangement to verify this). Wall facings were either flexible (0.15
mm thick aluminium foil, as before) or more rigid (1 mm thick aluminium
plate), and strain gauges were fixed to certain of the reinforcement strips in
order to measure the tensile loads developed.

Figure 4.14 Peak reinforcement load as a function of depth for centrifuge models of
reinforced soil retaining walls at 63 g. For equation (4.3) see Figure 4.12. Reproduced
from Bolton and Pang (1982). Institution of Civil Engineers, with permission
 The breaking strain of the 4 mm×0.1 mm aluminium reinforcement strips was
0.35%, which was considered to be representative of a typical field situation in
which extensive ductility of the reinforcement could not be relied on due to
localized corrosion and/or construction defects.
In the second series of tests, the distribution of maximum reinforcement load
with depth was found to be similar for five nominally identical models (having
L/ H=0.8), up to a centrifuge acceleration of 63 g (Figure 4.14). It may be seen
that the stress distribution shown in Figure 4.12 leads to the overprediction of
the tension in the layer of reinforcement at the base of the wall by a factor of
about 1. 4, and that the most critical level (at which the tensions are greatest) is
at a depth (below the soil surface) of 150 mm or 0.75H. This is probably the
result of the reduction in vertical stresses in this zone (as indicated by the
pressure cells incorporated into the foundation), due to shear stresses between
the facing panel and the soil, and the frictional resistance of the foundation
which tends to limit the lateral movement of the lowest facing panel. This is one
of the reasons why equation (4.1) is apparently overconservative in its prediction
of tensile failure. However, the vertical stress reduction at the base of the wall
might not occur if the facing panel were smooth, or if the foundation layer were
comparatively compressible.
The centrifuge acceleration at collapse of Bolton and Pang’s five series 2
models with L/H=0.8 was in the range 63–92 g. In the model which collapsed at
63 g, one reinforcement strip had failed at a reduced load near to its joint with
the facing panel at 300, due to fatigue resulting from its repeated use in
successive models. This precluded the possibility of the redistribution of
reinforcement loads, leading to instant collapse following the yield of the
reinforcement at the critical location.

Figure 4.15 Earth pressures mobilized at 50 g in models remote from tensile failure:
frictional capacity of models P I7 and P 19 was double that of P20. Reproduced from
Bolton and Pang (1982). Institution of Civil Engineers, with permission.
In another model which failed at 70 g, two reinforcement strips had ruptured at
or soon after 63 g. In the model which collapsed at 92 g, significant plastic
redistribution of reinforcement loads away from the critical zone had occurred,
increasing the load carrying capacity of the construction as a whole by a factor
of nearly 1.5. Plastic redistribution of reinforcement loads is the second reason
for the apparent overconservatism of equation (4.1). This cannot be relied on,
however, if the reinforcement is effectively brittle (for example, due to local
corrosion or a weak joint with the facing panel), or if the factor of safety against
pull-out FF is close to unity.
Further tests showed that walls with a higher factor of safety against pull-out
FF are likely to be subjected to higher lateral earth pressures under working
conditions, with (where ) rather than
(Figure 4.15). Earth pressures at collapse, however, are slightly overpredicted
(due to the effects of friction at the base of the wall and plastic stress
redistribution) by the use of the active earth pressure. (In back-analyses, Bolton
and Pang used values measured in shear box tests at a vertical effective
stress equal to that at the base of the model wall at the centrifuge acceleration in
question.)
The comprehensive series of centrifuge model tests described by Bolton and
Pang gives a considerable insight into the behaviour of reinforced soil retaining
walls in granular materials. In particular, the collapse of these walls can be
conservatively predicted using the simple stress analysis shown in Figure 4.12,
based on the peak angle of shearing resistance of the soil, measured at the
appropriate vertical effective stress. If data on appropriate values of are not
available, the use of would be acceptable in design. The apparent over
conservatism which occurs in certain cases is due to factors which may be
difficult to quantify or even control, including the exact details of wall
construction, and cannot necessarily be relied upon. Furthermore, a degree of
conservatism in design is necessary because it was not generally possible in the
centrifuge tests to identify incipient collapse from an increase in the rate of wall
movement with increasing g level or decreasing factor of safety. Empirical
modifications to the stress analysis of Figure 4.12, involving for example the
assumption of an erroneous ‘mechanism’ of failure, might not be appropriate in
every case, and should therefore be used with considerable caution.
Mitchell et al. (1988) describe a series of centrifuge model tests on reinforced
soil retaining walls, in which the effects of reinforcement type, facing panel
stiffness, foundation stiffness and surcharge loading were investigated. The
model walls were 150mm high, and (as with Bolton and Pang’s tests) were
subjected to an increasing centrifuge acceleration until collapse occurred. The
reinforcement strips were not strain-gauged, but the values of FT at collapse, and
data from earth-pressure cells, were generally supportive of Bolton and Pang’s
observations and hypotheses. Walls with stiffer facings collapsed at generally
higher g levels than similar walls with flexible facings. Walls with aluminium
foil reinforcement strips tended to fail catastrophically, whereas walls with
plastic or geotextile reinforcement strips were more ductile.
The vertical stresses measured near the base of the wall were generally in
excess of the overburden pressure, which is consistent with the trapezoidal
distribution of base pressure indicated in Figure 4.12. With stiffer facing panels,
however, the vertical stresses in this region were reduced, which is consistent
with Bolton and Pang’s observations and the hypothesis that stiff facing panels
carry some of the vertical load in shear. Mitchell et al. tested some walls on
compressible foundations of foam rubber or compacted clay. Not surprisingly,
this did not significantly affect the factor of safety FT at outright collapse for
walls with flexible facings. Deformations prior to collapse were increased
significantly, especially where both the facing panels and the reinforcement
strips were of low stiffness. Unfortunately, no walls with stiff facings were
tested on the compressible foundation: the settlement of a compressible
foundation would be expected to destroy the capacity of friction at a stiff facing
to reduce vertical stresses near the base of the wall, so that an increase in factor
of safety (manifest by a reduction in g level) at collapse towards the value for a
wall with flexible facings should be anticipated in such a case.
Mitchell et al. also report a limited number of tests on reinforced walls with a
surcharge load at the retained soil surface. The data are too few to draw general
conclusions; however it seems that (as with the walls with no surcharge) failure
by pull-out tends to occur at FF ~ 1, while the prediction of tensile failure
remains generally somewhat conservative. Two problems with ductile
reinforcements are demonstrated. The first is that gross deformations occur
without pull-out or breakage of the reinforcement, which implies that a wall
which is to meet the usual requirements of serviceability must be designed to be
very remote from either of these collapse limit states. The second is that certain
reinforcements may creep, leading to failure after a certain time at a factor of
safety greater than that which promotes instantaneous collapse. This is
demonstrated by two tests carried out by Mitchell et al., in which identical
model walls with fabric sheet reinforcement collapsed instantaneously at 20 g,
and after 70 min at 13 g. The problem of creep of geotextiles is well known and
well researched in its own right, and must be addressed by centrifuge modellers
investigating the behaviour of structures incorporating these materials,
especially if the ‘increasing g level to failure’ method of testing is adopted.

4.3.2
Soil nailing
In recent years, soil nailing has developed as an economical method of retaining
(at least temporarily) near-vertical cuts in clays and other soils. The soil nails are
essentially reinforcing bars installed into the soil through the cut face, attached
to a wire mesh-reinforced shotcrete covering. The nails may be installed in pre-
drilled holes and grouted in place, or they may be fired into the ground using a
compressed air gun.
Shen et al. (1982) carried out five centrifuge model tests to investigate failure
mechanisms of nailed soil support systems in a silty sand. The retained height of
the model was 15cm in all cases, but the length and spacing of the grouted piano
wire reinforcing elements were varied from test to test. Deformations were
monitored using photography and video recordings, but nail loads were not
measured. The method of testing was to increase the centrifuge acceleration
until failure occurred. Failure was defined by Shen et al. as the onset of surface
cracking, because the models consistently suffered large deformations without
sudden collapse.
The results of the model tests were compared with analyses based on the
limiting equilibrium of a block of soil defined by a parabolic arc through the
base of the cut face. The measured and calculated g levels at failure were
generally close (i.e. failure tended to occur at a factor of safety of unity), and the
cracks observed in the retained soil surface were reasonably consistent with the
positions of the critical parabolae. As the parabola does not constitute a
kinematically admissible slip surface, the analysis is of the ‘limit equilibrium’
type, rather than a true upper bound. From an analytical point of view, there
would seem to be little difference between a reinforced soil retaining wall and a
nailed soil support system. The universal applicability of the method proposed
by Shen et al. is perhaps, therefore, open to question—especially in situations
(such as a compressible foundation layer) where the factors which lead to the
general over conservatism of the stress analysis shown in Figure 4.12 are no
longer present. However, it might be argued that a less conservative analysis is
justified if the wire mesh-reinforced shotcrete facing is comparatively stiff (and
therefore able to reduce vertical stresses at the base of the retained soil by
friction), and because some of the vagaries of reinforced soil systems associated
with tolerance gaps between the facings and uncertain construction practices are
eliminated.

4.4
Concluding remarks
This chapter has described some examples of the ways in which geotechnical
centrifuge models have been used to enhance our understanding of the behaviour
of earth retaining structures. In some cases, a comparatively limited number of
tests has served to corroborate previous work, and perhaps revealed some
limitations (Lyndon and Pearson, 1984, and King and McLoughlin, 1993, on
embedded cantilever walls in dry granular materials). In other cases, the
behaviour of a particular class of retaining structure has been investigated
comprehensively for the first time (Bolton and Powrie, 1987, 1988, on
diaphragm walls in clay; Bolton and Pang, 1982, on reinforced earth retaining
walls). Sometimes, new areas of uncertainty are revealed, which may have very
significant implications in design (Hird and Djerbib, 1993, on cyclic lateral
loading of L-shaped walls). In only a few reported cases has the centrifuge been
used directly as part of an individual design (Rigden and Rowe, 1975, on an
unreinforced circular diaphragm wall).
Centrifuge model tests have not only led to the clarification of mechanisms of
deformation and collapse, but have also demonstrated that the use of the critical
state soil strength will generally lead to the most reliable (if sometimes
conservative) calculation of conditions at collapse. The evidence suggests quite
strongly that it is unreasonable to assume that peak values of will be
mobilized uniformly throughout the soil mass at failure. If non-critical state
strengths are used, for example in extrapolating the behaviour of small-scale
models tested at 1 g to field structures, the values of should be selected very
carefully with reference to the relative densities and the maximum effective
stresses which apply to each case.
The constraints of centrifuge modelling can be significant, and their effects on
the behaviour of the model must be considered in the interpretation of centrifuge
model test results. A particular example of this is the way in which excavation in
front of an in situ or sheet pile retaining wall is simulated. Nonetheless, it is
doubtful that any of the work described in this chapter could have been carried
out to quite the same effect without using a geotechnical centrifuge. In granular
materials, it is the realistic variation of self-weight stresses with depth which
forces the geotechnical engineer to address the crucial question of the
appropriate value of to use in calculations. In clay soils, centrifuge modelling
also enables the effects of excess pore pressure dissipation to be observed under
a reasonable timespan. The potential of the geotechnical centrifuge has been
amply demonstrated over the last two decades by the many successful general
mechanistic studies which it has been used to carry out. It is a source of regret
that this potential has not yet been fully realized in the application of centrifuge
modelling techniques to individual problems in civil engineering design.

References

Bolton, M.D. and Mak, K. (1984) The application of centrifuge models in the study of an
interaction problem. Proc. Symp. Application of Centrifuge Modelling to
Geotechnical Design, Manchester, 16–18 April 1984 (ed. W.H. Craig), pp. 403–422.
Balkema, Rotterdam.
Bolton, M.D. and Pang, P.L.R. (1982) Collapse limit states of reinforced earth retaining
walls. Géotechnique, 32(4), 349–367.
Bolton, M.D. and Powrie, W. (1987) Collapse of diaphragm walls retaining clay.
Géotechnique, 37(3), 335–353.
Bolton, M.D. and Powrie, W. (1988) Behaviour of diaphragm walls in clay prior to
collapse. Géotechnique, 38(2), 167–189.
Bolton, M.D. Powrie, W. and Symons, I.F. (1989) The design of stiff in situ walls
retaining overconsolidated clay. Part I. Short term behaviour. Ground Eng., 22(8),
44–8.
Bolton, M.D., Powrie, W. and Symons, I.F. (1990a) The design of stiff in situ walls
retaining overconsolidated clay. Part I. Short term behaviour. Ground Eng., 22(9),
34–40.
Bolton, M.D., Powrie, W. and Symons, I.F. (1990b) The design of stiff in situ walls
retaining overconsolidated clay. Part II. Long term behaviour. Ground Eng., 23(2),
22–28.
Bolton, M.D. and Sun, H.W. (1991) The displacement of bridge abutments on clay.
Centrifuge ’91 (eds H.-Y.Ko and F.G.McLean), pp. 91–98. Balkema, Rotterdam.
Broms, B.B. (1971) Lateral earth pressures due to compaction of cohesionless soils.
Proceedings of the 4th Budapest Conference on Soil Mechanics and Foundation
Engineering (3rd Danube-European Conference) (ed. A.Kezdi), pp. 373–384.
Akademiai Kiado, Budapest.
Caquot, A. and Kerisel, J. (1948) Tables for the calculation of passive pressure, active
pressure and bearing capacity of foundations . Gauthier Villars, Paris.
Craig, W.H. and Yildirim, S. (1976) Modelling excavations and excavation processes.
Proc. 6th Eur. Conf. Soil Mech. Found. Eng., Vol. 1, pp. 33–36.
England, G.L. (1992) Ring-Tension Filter Beds Subjected to Temperature Changes.
MACE Centre Short Course Notes: The Behaviour of Granular Materials and their
Containment. Imperial College, London.
Garrett, C. and Barnes, S.J. (1984) The design and performance of the Dunton Green
retaining wall. Géotechnique, 34(4), 533–548.
Heyman, J. (1972) Coulomb’s Memoir on Statics: An Essay in the History of Civil
Engineering. Cambridge University Press.
Hird, C.C. and Djerbib, Y. (1993) Centrifugal model tests of cyclic loading on L-shaped
walls retaining sand. Retaining Structures (ed. C.R.I.Clayton), pp. 689–701. Thomas
Telford, London.
Hubbard, H.W., Potts, D.M., Miller, D. and Burland, J.B. (1984) Design of the retaining
walls for the M25 cut and cover tunnel at Bell Common. Géotechnique, 34(4), 495–
512.
Ingold, T.S. (1979) The effects of compaction on retaining walls. Géotechnique, 29(3),
265–283.
King, G.J.W. and McLoughlin, J.P. (1993) Centrifuge model studies of a cantilever
retaining wall in sand. Retaining Structures (ed. C.R.I.Clayton), pp. 711–720.
Thomas Telford, London.
Ko, H.-Y., Azevedo, R. and Sture, S. (1982) Numerical and centrifugal modelling of
excavations in sand. Deformation and Failure of Granular Materials (eds
P.A.Vermeer and H.J.Luger), pp. 609–614. Balkema, Rotterdam.
Lade, P.V, Jessberger, H.L., Makowski, E. and Jordan, P. (1981) Modelling of deep
shafts on centrifuge tests. Proc. 10th Int. Conf. Soil Mech. Found. Eng., Vol. 1, pp.
683–691.
Lyndon, A. and Pearson, R. (1984) Pressure distribution on a rigid retaining wall in
cohesionless material. Proc. Symp. Application of Centrifuge Modelling to
Geotechnical Design, Manchester, 16–18 April 1984 (ed. W.H.Craig), pp. 271–281.
Balkema, Rotterdam.
Mitchell, J.K, Jaber, M., Shen, C.K. and Hua, Z.K. (1988) Behaviour of reinforced soil
walls in centrifuge model tests. Centrifuge ’88, pp. 259–271. Balkema, Rotterdam.
Powrie, W. (1985) Discussion on 5th Géotechnique Symposium-in-Print. The
performance of propped and cantilevered rigid walls. Géotechnique, 35(4), 546–548.
Powrie, W. (1986) The Behaviour of Diaphragm Walls in Clay. PhD Thesis, University
of Cambridge.
Powrie, W. and Kantartzi, C. (1993) Installation effects of diaphragm walls in clay.
Retaining Structures (ed. C.R.I.Clayton), pp. 37–45. Thomas Telford, London.
Powrie, W. and Li, E.S.F. (1991) Finite element analysis of an in situ wall propped at
formation level. Géotechnique, 41(4), 499–514.
Rigden, W.J. and Rowe, P.W. (1975) Model performance of an unreinforced diaphragm
wall. Diaphragm Walls and Anchorages, pp. 63–67. ICE, London.
Rowe, P.W. (1951) Cantilever sheet piling in cohesionless soil. Engineering, 51,
316–319.
Rowe, P.W. (1952) Anchored sheet pile walls. Proc. ICE, 1(1), 27–70.
Rowe, P.W. (1955) A theoretical and experimental analysis of sheet pile walls. Proc.
ICE, 1(4), 32–69.
Shen, C.K., Kim, Y.S, Bang, S. and Mitchell, J.F. (1982) Centrifuge modelling of lateral
earth support. Proc. ASCE, J.Geol. Engg. Div., 108(GT9), 1150–1164.
Simpson, B. (1992) Retaining structures: displacement and design (32nd Rankine
Lecture). Géotechnique, 42(4), 541–576.
Stroud, M.A. (1971) The Behaviour of Sand at Low Stress Levels in the Simple Shear
Apparatus. PhD Thesis, University of Cambridge.
Symons, I.F. and Carder, D.R. (1990) Long term behaviour of embedded retaining walls
in overconsolidated clay. In Geotechnical Instrumentation in Practice, pp. 289–307.
Thomas Telford, London.
Tedd, P., Chard, B.M., Charles, J.A. and Symons, I.F. (1984) Behaviour of a propped
embedded retaining wall in stiff clay at Bell Common Tunnel. Géotechnique, 34(4),
513–532