Dokumen - Tips Modelling in Geotechnics

Modelling in Geotechnics - Script
Part 1
Prof. Sarah Springman

In cooperation with
Dr. Jan Laue & Dr. Jitendra Sharma
http://igtcal.ethz.ch/mig
ETH Zürich Modelling in Geotechnics
Institute of Geotechnical Engineering
1 Design ...................................................................................................... 1
1.1 Modelling ............................................................................................................. 4
1.1.1 Numerical modelling ....................................................................................... 5
1.1.2 Physical modelling.......................................................................................... 6
1.1.3 Validation and calibration of models ............................................................. 12
1.2 Full scale (FS) 1g testing................................................................................... 13
1.3 Geotechnical centrifuge modelling .................................................................... 15
1.3.1 History of geotechnical centrifuges............................................................... 15
1.3.2 Types of centrifuge ....................................................................................... 15
1.3.3 Principles of modelling in the centrifuge ....................................................... 16
1.3.4 Scaling effects .............................................................................................. 17
1.3.5 Verification of models ................................................................................... 18
1.4 References ........................................................................................................ 19
2 Principles of numerical modelling ............................................................. 1
2.1 Why model numerically? ..................................................................................... 1
2.2 Validation of the finite element analysis (bench marking).................................... 3
2.2.1 El-Hamalawi (1997): mesh for a strip footing on clay ..................................... 4
2.2.2 Bransby (1995): mesh for lateral pressure on pile in clay .............................. 5
2.3 Prediction............................................................................................................. 7
2.4 Styles of numerical analysis using a computer.................................................... 7
2.5 Idealisation for numerical modelling as before for physical modelling................. 8
2.5.1 Geometry........................................................................................................ 8
2.5.2 Mesh design ................................................................................................... 9
2.5.3 Structure ......................................................................................................... 9
2.5.4 Loading and construction effects.................................................................. 10
2.5.5 Ellis (1997): Piled full-height abutment: 3D problem as 2D.... ...................... 14
2.5.6 Soil ............................................................................................................... 15
2.6 References ........................................................................................................ 17
3 Finite Element Method (FEM) in Geotechnical Engineering .................... 1
3.1 Introduction.......................................................................................................... 1
3.2 Numerical methods used in geotechnical engineering ........................................ 1
3.3 What is FEM? ...................................................................................................... 2
3.3.1 Historical Background..................................................................................... 3
3.3.2 The fundamental steps of the FEM ................................................................ 3
3.3.3 Approximation of the Circumference of a Circle ............................................. 3
3.4 Basic formulation of the FEM .............................................................................. 5
3.4.1 Interconnected elastic springs ........................................................................ 6
3.4.2 A plane truss element..................................................................................... 8
3.4.3 A constant strain triangular finite element .................................................... 10
3.5 Approximations, accuracy and convergence in the FEM .................................. 13
3.6 Geotechnical finite element analysis ................................................................. 15
3.6.1 Plane strain and axisymmetric problems...................................................... 16
3.6.2 Different types of finite elements .................................................................. 17
3.7 Techniques for modelling non-linear stress-strain response ............................. 20
3.7.1 Tangential stiffness approach with carry over of unbalanced load ............... 21
3.7.2 Modified Newton-Raphson method .............................................................. 21
3.8 Techniques for modelling excavation and construction ..................................... 22
3.8.1 Excavation .................................................................................................... 22
3.8.2 Construction ................................................................................................. 24
3.9 Advantages and drawbacks of the FEM............................................................ 25
3.9.1 Advantages................................................................................................... 25
3.9.2 Drawbacks.................................................................................................... 25
3.10 Some popular commercial FEM programs ........................................................ 25
3.10.1ABAQUS ...................................................................................................... 25
Page - 1
3.10.2SAGE CRISP ............................................................................................... 26

3.10.3PLAXIS......................................................................................................... 27
3.10.4ZSOIL ........................................................................................................... 27
3.11 Guidelines for the use of FEM in geotechnical engineering .............................. 28
3.12 Concluding remarks........................................................................................... 30
3.13 References ........................................................................................................ 30
4 Scaling laws and applications for centrifuge modelling ............................ 1

4.1 Introduction.......................................................................................................... 1
4.1.1 Scaling laws ................................................................................................... 1
4.1.2 Scaling of time ................................................................................................ 1
4.2 Scale effects ........................................................................................................ 2
4.2.1 Stress distribution in centrifuge model: Depth ................................................ 3
4.2.2 Stress distribution in a centrifuge model......................................................... 5
4.2.3 Particle size effects........................................................................................ 6
4.2.4 Coriolis acceleration ....................................................................................... 6
4.2.5 Boundary effects............................................................................................. 7
4.3 Scaling under earthquake conditions ................................................................ 10
5 Practical considerations: mechanical ....................................................... 1

5.1 Beam Centrifuges................................................................................................ 1
5.1.1 Capacity.......................................................................................................... 1
5.1.2 Swing platform, package and liner ................................................................. 2
5.2 Drum Centrifuges ................................................................................................ 4
5.2.1 Capacity.......................................................................................................... 4
5.3 Site investigation devices (penetrometers, vane)................................................ 8
5.3.1 Vane: .............................................................................................................. 8
5.3.2 Penetrometer:................................................................................................. 8
5.3.3 Cylindrical T-Bar: ............................................................................................ 8
5.4 Post-test investigation devices ............................................................................ 9
5.4.1 Photographic: camera & flash ........................................................................ 9
5.4.2 Digital Images and PIV analysis ..................................................................... 9
5.4.3 X-ray .............................................................................................................. 9
6 Practical considerations: geotechnical ..................................................... 1
6.1 Introduction.......................................................................................................... 1
6.2 Design of soil model: real or laboratory ............................................................... 2
6.3 Kaolin as a model soil.......................................................................................... 7
6.4 Preparation of soil samples in the DRUM centrifuge......................................... 21
7 In-situ testing, instrumentation, data acquisition....................................... 1
7.1 Measurement of soil properties ........................................................................... 1
7.1.1 Vane shear testing to determine su at discrete locations .............................. 1
7.1.2 Cone penetration testing (CPT): to determine a profile of su .......................... 3
7.1.3 T-Bar penetration testing: to determine su .................................................... 8
7.2 Measurement of displacement .......................................................................... 10
7.2.1 Spotchasing.................................................................................................. 10
7.2.2 Digital Images and PIV analysis ................................................................... 10
7.2.3 Displacement measurements ....................................................................... 11
7.2.4 Radiography ................................................................................................. 11
7.2.5 Excavation .................................................................................................... 12
7.3 Electronic/electrical instrumentation .................................................................. 12
7.3.1 Other considerations: ................................................................................... 17
Page - 2
7.4 Summary ........................................................................................................... 19

7.5 References ........................................................................................................ 20
8 Finite Difference Analysis using FLAC ..................................................... 1
8.1 Basics .................................................................................................................. 1
8.1.1 Specific to Geotechnics via FLAC .................................................................. 1
8.2 Finite Difference .................................................................................................. 3
8.3 Details of FLAC Program................................................................................... 11
8.3.1 Null model group .......................................................................................... 16
8.3.2 Elastic model group ...................................................................................... 16
8.3.3 Plastic model group ...................................................................................... 17
8.4 Example analyses ............................................................................................. 18
8.5 References ........................................................................................................ 26
Page - 3
Modelling in Geotechnics
Introduction to Modelling

1 Design
• The main focus of modelling is to achieve optimal design, which is cost effective, safe
and also aesthetic.
Codes
Structure Ground
Interaction
Load model
Rock/soil layering/properties
Idealisation
Physical model
Calculation model
Failure
Serviceability &
(ULS)
deformations (SLS)
Dimensioning
Design
Figure 1.1: Design Process
• frequent modelling is necessary in geotechnics to 'engineer' solutions (design)
• modelling implies idealisation of the real life 'prototype'
• understanding 'system' behaviour in response to perturbations (various loads) is crucial:
→ e.g. fundamental understanding of soil behaviour is required and should be
modelled effectively,
→ i.e. directly in physical models or through constitutive modelling and numerical
analysis.
We need to decide the WORST CASE SCENARIO in terms of the worst possible combina-
tions of loads, soil properties, geometry, local environmental and construction effects and to
design for them as indicated above by using Limit States to examine the potential structural
damage as well as any unserviceable deformations.
So what are the relevant Limit States?

Ultimate Limit State Serviceability Limit State
ULS controlled by SLS controlled by
Figure 1.2: Ultimate Limit State (ULS) and Serviceability Limit State (SLS)
Design Page 1 - 1
Failure…or ULS
• avoiding 'failure' of engineering structures is central to the design process
• includes social responsibility
• together with a legal requirement
• manifest in design codes via extremes of simple through to complex methods of analysis
(dependent on relative risk)
• necessitating engineering judgement
e.g. Heathrow Airport: tunnel collapse
• Failure October 1994
• half face, 3 stage excavation
• sprayed concrete lining
• construction method not properly applied
• 80m of tunnel and concourse collapsed (fortu-
nately without loss of life)
• subsequently stabilised by structural and light-
weight foamed concrete
• creating major delay and huge costs (tunnel
came into operation in 1998, nearly 3 years
late)
Figure 1.3: Heathrow Airport: tunnel collapse
The parties found to have been responsible (after lengthy court proceedings) have been
fined recently, in the case of the main contractor, 1.2 million pounds or 3 million SFr.
• Human error is at the core of most engineering failures, due to
→ conceptual/modelling errors,
→ inadequate components,
→ poorly considered design/construction changes.
• often these combine to form a critical chain of events leading to failure.
Figure 1.4: Failure of the Teton Dam, USA (1976)
Design Page 1 - 2
Figure 1.5: Teton Dam during failure
Figure 1.6: Teton Dam after failure
• e.g. failure of 90m high Teton Dam (Fig. 1.4 - 1.6): (1976) poor design of central core in
terms of material (silt), shape of core trench, and an irregular stepped longitudinal
abutment cross section which promoted hydraulic fracture in the core.
But, we learn more from our failures than from our successes…
Factors of safety/reliability against failure may be defined as
→ available strength/required strength or
→ available resistance/required resistance
whereas at the moment of failure, we KNOW with certainty that
→ the available strength/resistance = the required strength/ resistance
and we can often establish the failure mechanism... which is extremely helpful for subse-
quent back analysis.
Design Page 1 - 3
Development of engineering judgement is crucial and often underestimated and observing

failures of models provides a fast, comprehensive track to developing engineering
judgement and experience,
... as long as the model is appropriate!
1.1 Modelling
So what models can we use?
numerical
low
• simple, theoretical or empirical
• complex, iterative/computational
Low-medium (relatively low cost)
risk, quick and
“accuracy”
cheap
physical
Medium-high
risk, more time • full scale (high cost)
• small scale, 1 gravity
• small scale, enhanced gravity high
(medium cost)
Figure 1.7: Types of models
e.g. simple small scale 1g physical model….
• Domenico Fontana (in 1585)

• 300m move: 330 ton, 30m Vatican obelisk
• 1/50th model to demonstrate procedure
• but geotechnical modelling - i.e. with soil - is more complex
How do we model?
primarily through….
1. Numerical modelling
2. Physical modelling
Stochastic or statistical methods are also valid forms of modelling. These are, at present,
less often employed in geotechnics, other than for hazard assessment of earthquake
engineering, and are not considered further in this course. Simple forms of these methods
will, however, be used in the future Eurocodes, in order to be able to reduce the relevant
partial factors.
Design Page 1 - 4
1.1.1 Numerical modelling

• specific fundamental (classical plasticity) or empirical models
• finite element
• finite difference
• boundary element etc... (will be ignored here)
Type Sketch of Key Advantages Disadvantages

Model/Prototype assumptions
specific Soil • exact solution from • no strain before
fundamental continuous classical plasticity yielding
(classical homogeneous, (when upper and • significant
plasticity) rigid perfectly lower bound agree) idealisation required
plastic • fundamental basis • uniform strength in
in (soil) mechanics failure zone
empirical Calculation • quick and cheap • past data may not

model method based (back of envelope) suit current design
on past tests - • field validation of conditions
small/full scale frequently used • basic assumptions
measurement - construction may differ
y and/or lab tests methods • usually does not
in and account for
approximate fundamental
constitutive behaviour
x models
out?
finite Soil continua • general analytic tool • approx. solution+

element with partial • divide geometry into engineering
(FEM) differential elements judgement versus
equations to • adaptive methods apparently complex
describe can refine mesh and analysis
physical reduce errors • strain must vary
phenomena • spatial variation of according to type of
and extensive material properties element selected
integration • more descriptive • element
method with constitutive models concentration
solution of • computing power up required for area of
stiffness matrix • ideal for high strain variation
serviceabilty • numerical instability
analysis at large strain
finite Soil continua • competitive with • not ideal for linear
difference iterative finite FEM when highly problems
(FDM) difference non linear (large • strict limitations on
formulation i.e. strain) mesh patterns unless
similar to FEM • range of constitutive at expense of
with reduced models/applications calculation efficiency
integration inc. user-specified • rel. stiffness can
cause instability?
Tab. 1.1: Numerical modelling
Design Page 1 - 5
1.1.2 Physical modelling

• 1g, full scale testing
• (+ field monitoring)
• 1g, small scale laboratory testing
• ng, small scale centrifuge modelling
Type Sketch of Model/ σv at A Advantages Disadvantages

prototype
[kPa]
1g, 100 • stress correct • time to construct and
to • can control soil for diffusion processes
full scale 140 conditions • boundary effects
7m • cost
Field 100 • ‘the real thing’ • time (for diffusion)

to • stress correct • cost
monitoring 140 • soil, geometry, • failure not OK
boundaries realistic • boundary/soil
7m conditions often not
A clear
1g, 1 • time (very quick) • stresses incorrect

• cost (very cheap) • potential for suction
small scale • good preliminary test to and dilatancy to affect
check equipment and results
7cm
testing principles • boundary effects
A
100g, 100 • stress correct • radial ‘g’ field (in a

to • idealise to reveal key beam centrifuge)
1/100th 140 mechanism of • ng varies with depth
ω behaviour • coriolis effect
scale in
• select soil and soil • size of particles,
centrifuge parameters instrumentation, site
1m • design stress history investigation devices
• control loading system • stress path may be
• time different
• cost • construction method
100g • allowed to fail: observer different?
witnesses deformation • boundary effects
and failure mechanism -> so take care over
idealisation
7cm
A
Tab. 1.2: Physical modelling
• 1g small scale testing is not useful for exacting work, because the stresses are not
correct and since soil behaviour is nonlinear, the modelling is unsatisfactory.
• all other methods should have correct order of magnitude of stresses but need to take
care over stress paths.
Design Page 1 - 6
• time, cost and of course technical factors (e.g. scaling laws, model preparation) are
important.
• parametric studies can be extremely useful in exposing mechanisms of behaviour
relevant first at SLS and later at ULS.
What concerns do we have about modelling 'effectively'?

i.e. As designers:
• is it comparable or relevant to the design?
• will it help the design process?
• will it reveal the key mechanisms of behaviour?
• will it reveal the secondary, more complex, mechanisms of behaviour (sometimes
important)?
i.e. As researchers (in addition):
• can we produce a robust design method which reproduces all the important character-
istics?
• is this method potentially usable by engineers working in industry?
Idealisation: i.e. model an 'ideal' prototype - not necessarily the exact field condition.
• range of critical modes of behaviour identified
• factors affecting these studied in detail
• intelligent simplifications to replicate key features which control the prototype behaviour
pattern.
Will the idealisation reproduce the full scale 'prototype' behaviour?
e.g. for the Severn bridge approach embankments
Design Page 1 - 7
Idealise any combination of:

• geometry
• soil
• structure
• loading
• construction effects.
Figure 1.8 contains an idealisation

exercise to emphasise these points in
terms of centrifuge modelling on a bridge
abutment and approach embankment Figure 1.8: Severn Bridge
(e.g. Severn Bridge).
Design Page 1 - 8
The prototype problem

• embankment on soft clay Piled full-height bridge abutment
• clay moves laterally further than the piles δu
• therefore “passive” lateral thrust on piles
Embankment
→ ?magnitude and ?distribution
Soft
• increased pile bending moments (BM) and
lateral deformation at deck (δu < 25 mm) Stiff
→ ?magnitude and ?distribution
1
1st idealisation
INPUT
Single free-headed pile
• vertical load (pressurised air bag)
• row of single free-headed piles
• soft clay (su ≅ 15 kPa) over stiff sand Vertical load, q
δup
OUTPUT
Soft p?
• lateral “passive” pressure p as f(q)
• BM & δup Stiff
3rd idealisation
INPUT Pile group
• pile group with cap
• stiffer clay (su ≅ 40 kPa) over stiff sand Vertical load, q
δup
OUTPUT for both rows of piles
Soft p?
• lateral “passive” pressure p as f(q)
• BM & δup Stiff
also
Bransby PhD, 1995
• drag under pile cap
Inflight construction of embankment

4th idealisation Supply hopper
INPUT
• construct embankment in stages inflight
(slow or fast build)
• soft clay (su ≅ 20 kPa) over stiff sand
(with or without wick drains in clay)
Ellis PhD, 1997
Design Page 1 - 9
4th idealisation
INPUT Embankment loading on clay
• construct embankment in stages inflight Supply hopper
• soft clay (su ≅ 20 kPa) over stiff sand
also NOTE
• drag under embankment
• drag under pile cap
Ellis PhD, 1997
4th idealisation
Piled full-height bridge abutment
OUTPUT load on abutment
• active lateral pressure on wall
• arching of embankment loading onto wall arching δup
base
• lateral “passive” pressure p as f(q) p?
• BM & δups
(Springman, 2001)
Ellis PhD, 1997
It is often advisable to omit some detail to focus on the key conditions which will affect the
behaviour.
Typically this will be a function of the:
geometry / soil / structure / loading / construction effects
Geometry
• simplify
→ from three dimensions to two dimensions
→ soil strata & structure
→ location for load application.
Soil
• constitutive model (numerical)
• laboratory soil or real soil (physical)
→ create or design soil stress history
→ specify soil strength (for clays) or relative density (for silts and sands)
→ ensure homogeneity of the prepared soil model (when this is required).
Design Page 1 - 10
Structure
• model key aspects which will affect soil-structure interaction
→ use hollow aluminium piles / retaining wall to have same flexural stiffness (E I)
(Young's modulus in MPa x 2nd Moment of Area in m4) as equivalent concrete
piles / wall
→ e.g. 1st idealisation - single row of free-headed piles
→ 2nd idealisation - pile group (2 vertical piles x 3) with pile cap raised above
ground surface
→ 3rd idealisation - as for 2nd, but pile cap rests on ground surface
→ 4th idealisation - as for 3rd, but with retaining wall (and embankment built in-
flight).
Loadings
• point loading or line loading
• normal, uniformly distributed or average pressure i.e. modelling embankment by a
surcharge load, ignoring tractions acting at surface of soft layer - idealisations 1-3 above
(and these tractions really do have an effect).
Construction effects
it is quite difficult to build/excavate in-flight - soil is heavy under ng and equipment must be
small, light but strong and manoeuvrable!
• pile installation
→ lateral loading - 1g installation is OK: stress in upper layers of soil (which
control the lateral response) is not too badly affected
→ axial loading - installation must be at correct stress levels (i.e. in-flight) since
pile response / load capacity is affected significantly by lateral stresses
generated during this process
→ bored piles are not easily modelled in the centrifuge but better modelled
numerically.
• tunnel construction
→ excavate clay at 1g and replace with airtight rubber membrane which can be
pressurised later
→ pour sand around a polystyrene tunnel (with or without a liner) at 1g and
dissolve the polystyrene in-flight
→ jack Tunnel Boring Machine from side of box! This is complex in a mechanical
sense, but can be and has been done.
• building an embankment
→ at 1g and embankment 'grows' as 'gravity' force increases (not same stress
path)
→ pour embankment in-flight (better stress path) but no allowance for
compacting the embankment material
→ NOTE: sand is dry & is not the same material as the 'real' embankment.
Design Page 1 - 11
1.1.3 Validation and calibration of models

A model is of no use if it fails to represent the prototype behaviour. The modelling laws
should therefore be appropriate. Selection of relevant non-dimensional groups, which
control model behaviour, is often necessary. These should remain constant if the basic
modelling premises are to be correct.
An example concerning stability of a slope of height H, soil density ρ and undrained shear
strength su and a key non-dimensional group Hρg/su is given in table 1.3
Prototype Model 1 Model 2 Model 3 Model 4

Geom. scale 1:1 1:10 1:10 1:10 1:10
Acceleration 1g 1g 1g 1g 10g
H [m] 5 0.5 0.5 0.5 0.5
su [kPa] 15 15 1.5 15 15
ρ [t/m3] 1.65 1.65 1.65 16.5 1.65
g [m/s2] 10 10 10 10 100
Hρg/su 5.5 0.55 5.5 5.5 5.5
Tab. 1.3: Instabile Hänge und andere risikorelevante natürliche Prozesse, Monte Verità (after Bucher, 1996)
Similitude of models, when defined by Hρg/su, is guaranteed for Models 2-4 but not for
Model 1, which is therefore not appropriate. Since it is very difficult to adjust either the
density (model 3) or the shear strength (model 2) without affecting the response of the
model in other ways (e.g. modelling deformations), the recommended method is to
increase the gravity field (model 4).
Even if the agreement between non-dimensional groups is satisfied, it is often difficult to
judge whether the model is valid or not. Ideally it would be possible to compare results
against a full scale prototype however it is extremely rare to obtain sufficiently good data to
be able to do this.
Normally the physical or numerical data is checked against a known numerical or funda-
mental solution for validation. Provided this test is passed then there is confidence in the
modelling method, which may in turn be used to calibrate either soil properties or additional
modelling methods (e.g. new numerical algorithms).
For example, a strip footing on a homogeneous clay layer with uniform undrained shear
strength with depth, su, can be shown using plasticity theory to fail at a load of (2 + π)su. A
physical test set up or a numerical calculation (or computer algorithm or run) may be
checked against this to ensure that the model system is valid before further models are
used to calibrate/back-calculate for other prototypes (El-Hamalawi, 1997).
Design Page 1 - 12
1.2 Full scale (FS) 1g testing

Full scale 1g tests are not usually carried out on an 85m high earth dam or a 100m deep
quarry which is about to be landfilled. It is much more likely that elements or aspects of a
full scale test will be modelled. E.g. a trial embankment may be built quickly to failure in the
appropriate clay to test the construction and compaction procedures to be used for a clay
cored dam, and to measure in situ density, permeability and strength. This can be back
analysed numerically and the results applied to model the behaviour of the complete
prototype structure.
Figure 1.9: ETH Big Box for 1g models
Equally Hertweck (1998) and Brinkmann (1999, 2001) have both used the IGT 5.5m x 4m x
3m Big Box to model an aspect of the behaviour of a clay barrier which they have also
analysed numerically using finite element analysis with the same boundary conditions as in
the Big Box. Subsequently they have also modelled the full scale behaviour of the prototype
with the field boundary conditions.
Below are some figures from the doctoral work of Michael Hertweck (1998).
Design Page 1 - 13
Figure 1.10: Dissertation Hertweck (1998)
Investigation of stress-deformation behaviour for two-phase manufactured diaphragm walls

for encapsulation of waste materials has been carried out by Andreas Brinkmann (2001).
Critical external loads are related to internal stress and strains in the wall. Element tests,
finite element calculations and full scale tests have been carried out.
Figure 1.11: Dissertation Brinkmann (2001)
Design Page 1 - 14
1.3 Geotechnical centrifuge modelling

1.3.1 History of geotechnical centrifuges
Centrifuge models have been recognised for more than 100 years as being capable of
creating homologous points of stress and strain in both model and prototype. Bucky (1931)
first used a centrifuge for mine roof stability investigations in Columbia USA, but the major
thrust of development came from the Soviet Union. This was largely stimulated by the
scaling advantages applicable to the weapons industry, and large scale explosions in
particular. Pokrovsky made a major presentation concerning the use of a geotechnical
centrifuge at the first International Conference on Soil Mechanics and Foundation
Engineering at Harvard in 1936.
Since then, Professor Andrew Schofield FRS has been in the forefront of centrifuge devel-
opments both at the University of Manchester Institute of Science and Technology and later
at Cambridge University (Schofield, 1980).
Further details are available in Chapter 1: Geotechnical Centrifuge Technology, Taylor
(1995).
1.3.2 Types of centrifuge
BEAM
A beam is mounted on a central spindle and can rotate about this to allow models made in
packages located at each ends of the beam to be subjected to increased gravity. Usually
these packages are fixed to a swinging platform so that they are hanging in the vertical
plane initially and can swing up to lie in the horizontal plane as the centrifuge acceleration is
increased.
Centrifuge modelling: BEAM Centrifuge modelling: DRUM
r r
ω
ng
ω
ng
.. ..
ng = ω2r - r ng = ω2r - r
g = 9.81 m/s2 g = 9.81 m/s2
Figure 1.12: Types of centrifuge
DRUM
A different style of centrifuge, in which a drum of diameter between 800mm and >2m (e.g.
ETH Zürich) may be rotated respectively, between 90 and 650 r.p.m., to present a bed of
soil between 0.1 up to 0.5 km wide by 1-3 km long in a gravity field of up to 500 g. This
allows examination of soil behaviour for specific problems, which relate to shallow construc-
tions (shallow foundations onshore and offshore, transport processes for environmental
geotechnics, tunnelling, ice forces on structures, slope stability, etc.).
Design Page 1 - 15
1.3.3 Principles of modelling in the centrifuge

• Centrifuge testing accelerates the model to achieve full scale stress conditions
For non-linear materials in either small or full scale prototypes, correct stress-strain fields
must be replicated for subsequent meaningful interpretation.
ng
·
2 ⋅ ω ⋅ r· + r ⋅ ω
2
r ⋅ ω – r··
Figure 1.13: Accelerations of a rotated body
PHYSICS
If a body is rotated about a spindle in a horizontal plane, then it is subject to a combination
of radial and tangential accelerations due to r and ω and the derivatives of these: dr/dt,
d2r/dt2, dω/dt or r·, r··, ω·
For static cases, there tends to be:
no radial acceleration dω/dt so rdω/dt disappears (tangential)
no change in radius dr/dt, d2r/dt2 so 2ωdr/dt and d2r/dt2 disappear (tangential and
radial)
Example:
If the effective radius of the centrifuge is 4 m, and the gravity level is 100 g, then angular
velocity ω:
100 g = r ω2
ω = (100 g / 4)0.5
ω = 15.7 rad/sec = 15.7 * 60/(2*π) = 150 r.p.m.
A change in radius occurs during the construction of an embankment in flight: sand is
rained onto the fill, and the dr/dt term becomes important. This is described as the Coriolis
effect and the particle trajectory follows a parabolic path. Deflector plates may be used
underneath the sand pouring device to counter this effect.
Design Page 1 - 16
Optional explanation ... if a mass is rotated at a specific radius and angular velocity then
Newton's law of motion shows that the mass is pulled out of a straight line around the radial
curve causing a radial acceleration towards the centre. An associated inertial force must
apply, acting radially inwards.
Looking externally, the package must be accelerating inwards, but actually the mass
appears to be at rest in a radial sense, so relative to the package's frame of reference, the
acceleration and body force act in the opposite direction - radially outwards. This means
that the mass must be held by mechanical means strong enough to resist this outward
radial body force.
Newton first explained the concept of gravity - masses accelerated towards the centre of
the earth in terms of a gravity force on each terrestrial mass. Relativity implies that the
gravity force is identical to the inertial force - the small scale model will weigh more under
angular velocity at ω than when the centrifuge is at rest.
1g = 9.81 m/s2. In the centrifuge, radial acceleration will be a factor of g, i.e. ng. So that if
the model dimension is scaled down by a factor of n, i.e. 1/nth scale model, then the
stresses will be equivalent.
Effectively, the gravity acts on the nuclei - the centre of mass of each atom - and the net
pressure builds up with depth. This means that there is no gravity or inertial force at the
surface of a model, but this increases with depth.
1.3.4 Scaling effects
Parameter Unit Scale

(model/prototype)
Acceleration m/s2 n
Linear dimension m 1/n

Stress kPa 1
Strain - 1
Density kg/m3 1
Mass or Volume kg or m3 1/n3

Unit weight N/m3 n
Force N 1/n2
Bending moment Nm 1/n3
Bending moment / unit width Nm/m 1/n2
Flexural stiffness/ unit width (EI/m) Nm2/m 1/n3
Time: diffusion s n2
Time: dynamic s n
Frequency 1/s n
Tab. 1.4:
Design Page 1 - 17
Similarity of stress (and strain) will be achieved in both modelm and prototypep, for a
sample in soil of the same density ρ and stiffness constructed at a scale of 1/n, located at
an appropriate radius and rotated at an angular velocity to give a multiple of earth's gravity,
ng at that radius so that the vertical stress at depth z equivalent to the radius is:
σv = ρm (ng)m (z/n)m = ρp gp zp
Scaling of time
As in fluid mechanics, it is not always possible to achieve correct scaling in all dimen-
sionless groups, and so choices must be made.
In dynamics, where acceleration in m/s2 scales as n in the model, and the linear dimension
is modelled at 1/n prototype, then time is modelled n times faster in the centrifuge.
But the scaling factor for modelling time in terms of diffusion may be demonstrated to be:
n2 faster in the centrifuge.
The non-dimensional time factor, Tv = f(time/depth2) = cvt/d2, becomes independent of
gravity level for a depth of sample reduced to 1/n of the original, if the model time is also
reduced by 1/n2.
(1D Diffusion equation - saturated soil)
du/dt = cvd2u/dz2
where u is excess pore pressure and time t scales with length z2 provided cv m = cv p .
This offers a significant advantage because 27 years of prototype diffusion may be
modelled in 1 day using a centrifuge at 100 g, and is especially useful for environmental
problems or heat loss by conduction where diffusion is the main transport mechanism.
However, in offshore foundations or earthquake problems, the pore pressures are created
dynamically, with time scaling as: n times faster in the centrifuge and yet they decay in a
diffusive process where time is modelled as: n2 faster in the centrifuge.
Solution: use pore fluid in the model with a viscosity of n times that of the prototype (and
same density) or reduce the value of permeability of the soil (Attention: this will cause a
change in the properties).
1.3.5 Verification of models

If the prototype is at full scale under earth's gravity, then the model behaviour is scaled
according to the value of n. If a 1/100th scale model at 100g predicts the same prototype
behaviour as a 1/200th scale model at 200g, then verification of models (often called
‘modelling of models’) has been achieved and we can then use that predicted prototype
behaviour to check numerical analyses.
Design Page 1 - 18
1.4 References
1. El-Hamalawi, A., Adaptive refinement of finite element meshes for geotechnical

analysis. PhD thesis, University of Cambridge, 1997.
2. Brinkmann, A. and Amann, P., Small and large scale tests for the determination of the
mechanical behaviour of a clay-cement stabilised slurry wall. Bauingenieur, Band 74,
Heft 9, Sept. 1999,
3. Brinkmann, A., Untersuchungen zum mechanischen Verhalten von ton-zement-gebun-
denem Dichtwandmaterial für Zweiphasen-Verfahren. PhD Thesis, Swiss Federal
Institute of Technology, ETH Zürich, 2001.
4. Hertweck, M., Untersuchung des Tragverhaltens von Steilwandbarrieren in Deponiebau
mit grossmassstäblichen Modellversuchen. PhD Thesis, Swiss Federal Institute of
Technology, ETH Zürich, 1998.
5. Springman, S. M., Soil structure interaction: idealisation, validation and calibration of
models. 1st Albert Caquot Conference, Paris, 2001.
6. Taylor, R. N., Geotechnical centrifuge technology. Geotechnical Engineering Research
Centre, City University, London, 1995.
7. Schofield, A. N., Cambridge geotechnical centrifuge operations. 20th Rankine lecture,
Géotechnique 30 , No.3, p. 227-268, 1980.
8. Bucky, P. B., Use of models for the study of mining problems. American Institution of
Mining and Metallurgical Engineers, Tech. Pub. 425, p. 3-28, 1931.
Design Page 1 - 19
Numerical Modelling

2 Principles of numerical modelling
2.1 Why model numerically?

This is inevitably part of any calculation and Code-based design methods, in which respon-
sibility for safe design must be assured. This must be validated either by known experience
or another proven calculation or physical models.
A classic example of local experience applied in a foreign environment was the - at the time
- surprising series of embankment failures which occurred in South East Asia when
extremely experienced and well known geotechnical engineers from Scandinavia designed
these embankments based on the undrained shear strength obtained from vane shear
tests. What experience had NOT shown beforehand was that the plasticity index of the soil
affected the values of su obtained (LHA p.79, figure 6.17). The vane strength values of the
relatively low plasticity Scandinavian clays required no correction factor but the strengths of
the high plasticity S.E. Asian clays should have been reduced by 2/3rds.
The first lesson to learn about numerical modelling is that the results are only valid when
both the input data and the calculation method (algorithm) are appropriate…..GIGO or
Garbage In...Garbage Out.
Figure 2.1: Understanding computer technology
The simplest forms of numerical modelling would be the 'back of the envelope' calculations
that are carried out for a preliminary judgement on a particular engineering problem.
E.g. a relatively homogeneous clay deposit has an undrained shear strength su ~ 20 kPa
and vertical load of 200 kN/m will be applied onto a strip footing / strip pile cap. Given that
the width of the footing is limited for reasons of lack of space to 3m, will it be necessary to
use piled foundations?
Principles of numerical modelling 2-1

Considering the Ultimate Limit State (ULS) at first and knowing that the maximum vertical
load on a strip footing qmax is approximately 5su, qmax ~ 100 kPa. The load applied to the
footing = 200/3 = 67 kPa, so a global factor of safety would be 100/67 = 1.5 which is
certainly insufficient. We would also remember that if the load was inclined then this value
would be reduced, and that even if the ULS conditions had been fulfilled that the Servicea-
bility Limit State (SLS) should also be checked.
If the soil deposit is extremely variable (with uneven layering and fairly soft or sensitive
contents), and the structure to be built on it is extremely expensive (or indeed potentially
dangerous if failure occurred e.g. nuclear power station), then a far more extensive
numerical modelling process will be necessary. This may well entail more complex
analyses of continua using a computer to solve a series of equations based on a mesh with
appropriate boundary conditions, a range of loading scenarios and a suitable constitutive
model (e.g. elastic, elasto-plastic, critical state).
These are often called finite element or finite difference analyses and they differ only in the
method of solving the equations of equilibrium, compatibility and constitutive model (see
table on numerical modelling in chapter 1, table 1.1, page 5).
Even for these 'finite' models, there are ranges of complexity…..e.g.
• simple or complex meshes (e.g. with 2 elements for a 1/4 space (fig. 2.2 left) or an
adaptive mesh for the whole sample and 229 or even 1791 elements depending on the
level of strain in the soil(fig. 2.2 right)!)
Undrained triaxial test on over-

consolidated clay 229 elements 1791 elements
Figure 2.2: Range of mesh complexity for a triaxial sample
• special purpose: calibration of soil parameters (back analysis of a specific event, e.g.
Fig. 2.3) or prediction of behaviour (part of a design) or fundamental generic (investiga-
tions into a specific class of problem)
• Class A prediction or validation of physical (centrifuge model) tests
→ e.g. primary focus on specific match to exact centrifuge model test, does the
behaviour agree (Fig. 2.3)?
→ parametric analyses are possible to match prototype more closely & to reveal
further trends
• numerical modelling is also quicker & cheaper than many forms of modelling, provided it
is appropriate and the modellers are competent!
Finite element mesh for analysis of

model embankment on soft clay
Centrifuge model test of stage-

constructed embankment on soft clay
after failure (Almeida 1984)
Computed and measured development

of settlement at clay surface with time
(Almeida 1984)
Figure 2.3: Calibration of soil parameters and match deformation response
2m
200mm
100mm
h
1g 2g 20g h increases as time increases
Model “grows” as ng increases in centrifuge Layers added in prototype
Figure 2.4: Modelling “construction” processes
2.2 Validation of the finite element analysis (bench marking)

Validation has also been mentioned in several formats already, in this and the first chapter.
For any user to be able to accept the results of a computer analysis, some form of
validation is necessary. Sometimes this is called „bench marking“ carried out against a
known theoretical solution.
Access to an exact solution is one of the most approved modes of checking the validity of a
particular mesh set up and analysis. Both Bransby (1995) and El-Hamalawi (1997) have
shown that the mesh design influences the end result.
Mesh refinement entails either increasing the number of elements, or changing the density
of the elements according to the areas where either the most shear strain or the greatest
pore pressure build up arises (e.g. adaptive meshing - see also on right pictures in
figure 2.2). This means that the elements are smaller and hence the assumptions which are
valid for each finite element represent a smaller space and therefore

→ greater variation is possible,

→ the model becomes closer to reality and
→ the solution becomes more exact.
answer
exact solution
increasing mesh refinement
Figure 2.5: Accuracy of result with mesh refinement
2.2.1 El-Hamalawi (1997): mesh for a strip footing on clay

Comparing the case for a strip footing, loaded vertically in plane strain space q under
undrained conditions on homogeneous isotropic rigid perfectly plastic soil with uniform
shear strength su, the exact solution is (2+π)su (Prandtl, 1921, developed this based on
classical plasticity theory for metals). El-Hamalawi modelled this using finite elements and
represented the soil by an elastic perfectly plastic constitutive model under drained condi-
tions; uy is the settlement and b is the footing width.
Initial mesh At start of yielding At failure
Mechanism at failure
Width of footing b
Figure 2.6: Foundation on clay

q/s u
0 1 2 3 4 5 6
u y/b
initial (5.444)
remeshing (5.147)
exact = (π + 2)
Figure 2.7: Effect of mesh status on load - settlement curve
2.2.2 Bransby (1995): mesh for lateral pressure on pile in clay

The ultimate solution for lateral pressure pu caused by homogeneous isotropic rigid
perfectly plastic soil with uniform shear strength su flowing around a pile in plane strain
under undrained loading conditions….. (Randolph and Houlsby, 1984).
( 6 + π ) ⋅ su ≤ pu ≤ ( 4 ⋅ 2 + 2 ⋅ π ) ⋅ su
smooth pile rough pile

Bransby's work can be used to check an undrained analysis in which soil under similar
conditions is moved past a fully rough stationary pile. Following manual mesh refinement
and development, the final agreement between the exact (11.94 su) solution and the
computed result is within 2%, which is certainly close enough for most engineering
analyses.
Finite element modelling of a

single pile in 2-d.
Figure 2.8: Mesh and boundary conditions

Load-transfer curve for a single pile in elastic-plastic soil.
Figure 2.9: Load - transfer curve for pile under lateral load

2.3 Prediction
Prediction has already been mentioned in several formats. Determination of styles of
prediction are based on an alpha-numeric code which is given in the table below.
Class Stage of prediction Status
A Calculation before or during design process or before event Results unknown
B Calculation during event (e.g. construction process) Results unknown
B1 Calculation during event (e.g. construction process) Results known
C Calculation after event (e.g. construction completes) Results unknown
C1 Calculation after event (e.g. construction completes) Results known
(back analysis)
Tab. 2.1: Classes of prediction
Poulos summarised the modes of modelling as shown in Table 2.2.
Analy- Characteristics and typical Advantages Disadvantages

sis example
class
C Simplified methods, using Easily applied, and allow Requires substantial ideali-
closed form solutions. Sim- rapid parametric studies. zation, and experience in
ple soil models used. assessing parameters.
B Methods using boundary Relatively easy data input. Requires some idealization,
elements, with simplified Familiar soil model parame- and experience in assess-
soil models. ters used. Relatively rapid ing parameters. Difficult to
to run and interpret. examine complex prob-
lems.
A Complex numerical meth- Can consider detailed and Requires experience in as-
ods (finite element, finite complex problems. Soil sessing soil parameters
difference). models can be more realis- which may be unfamiliar.
tic. Considerable effort to pre-
pare data and interpret out-
put.
Tab. 2.2: Classes of soil-structure interaction analysis
2.4 Styles of numerical analysis using a computer

It is worth considering WHAT a civil engineer might use a computer-based analysis for.
1. Reviewing results of someone elses' analysis (e.g. as a 'proof engineer')
a) Check validity of calculation model
b) Check input parameters as stated in the assumptions
c) Check answers are in the right zone to validate the work
d) Examine all data critically (deformations, stresses, strains etc.) and use as neces-
sary
E.g. Modelling in Geotechnics: Exercise 1: GeoCAL SSI (Done Runs).
2. Setting up and running simple analyses
a) Simple mesh and boundary conditions
b) Simple loading conditions

c) Simple constitutive model (e.g. elasticity)

d) ….then steps as above as in steps 1a-d
E.g. Modelling in Geotechnics: Exercise 1: GeoCAL SSI (Paint or Supermesh).
3. Setting up and running more complex analyses
a) As in 2a-c above
b) Development of more complexity in terms of mesh, loading conditions and constitu-
tive model
c) ….then steps as above as in steps 1a-d
E.g. Modelling in Geotechnics: Exercise 2 (or 3).
2.5 Idealisation for numerical modelling.....as before for physical

modelling
• geometry
• soil
• structure
• loading
• construction effects.
To reiterate, most of these remarks relate to continuum analyses - mainly by
• finite element method (FEM) or
• finite difference method (FDM)
.....based on principle of discretization (meshing) - see p. 2 - 4 to solve complex
boundary-value problems PLUS
• compatibility - kinematic conditions => geometry, displacement, strains must be
compatible
• equilibrium - static conditions => forces and stress must be in equilibrium
• stress-strain relationship - physical conditions => material-dependent relationship
between stress and strain must be specified at element level
2.5.1 Geometry
• try to represent 3-dimensional effects as 2-dimensional effects (cheaper, quicker)
→ may reproduce as a plane strain or axisymmetric problem by use of symmetry
or asymmetry
• consider and idealise boundaries: soil/structure
• draw outline section and plan with material/boundaries
→ (too much vertical deflection at rollers indicates boundaries may not be far
enough away?)
• create mesh → nodes and elements
→ avoid large jumps in element size to < 3x (FEM) or < 1.5x (FDM)
→ refine mesh in regions of high strain but beware infinite stress concentrations
→ limit number of nodes and elements according to complexity, typically...

* non-critical structure 150~200 elements

* dam with deep foundation 300~400 elements (or more)
Some finite element types:

→ 2 dimensional triangular constant strain
→ 2 dimensional quadrilateral linear strain
→ 3 dimensional hexahedral cubic strain
→ interface elements: relative movement between elements

(progressive slip on piles)
→ bar elements: capacity for tension (soil reinforcement)
/compression (props)
→ beam elements: capacity for axial force and bending moments
(structural inclusions)
→ infinite elements: models unbounded area e.g. in dynamics
where fixed boundary would reflect waves
2.5.2 Mesh design

This has been shown in the past to influence the results obtained and the major guidelines
will be presented in more detail in chapter 3. Several examples have been shown on pages
2 - 4.
ADAPTIVE MESH REFINEMENT (e.g. El-Hamalawi, 1997) can be used to enrich and
subdivide mesh as regions of high strain develop - so that mesh choice does not precon-
dition outcome of the analysis.
2.5.3 Structure
Material
• use 'drained' properties (not much pore pressure in steel or concrete!)
• linear elastic (although can use linear elastic-perfectly plastic if trying to 'fail' structure)
• much stiffer than soil so beware of numerical instabilities (sometimes need double
precision in FEM or need more time steps/finer mesh in FDM)
Equivalence
• row of piles as a sheet pile wall - equivalent bending rigidity
(EI)wall = n (EI)individual piles + (EI)soil between piles
• similar equivalence when modelling cylindrical sand drains as a 2-dimensional sand

drain wall.

(EI)soil between piles
+ n(EI)piles
= (EI)wall
Figure 2.10: Equivalence
2.5.4 Loading and construction effects

• in-situ stresses defined initially
• loads primarily - normal and shear forces (tractions) on elements
• excavation and fill: construction sequences for embankment and retaining wall
• superposition of layers of soil or concrete (for geometric purposes) for subsequent
removal
• displacement and rotation fixities (x, y, z, θ) - either the soil or structure can be moved
relative to rest of mesh (Bransby analysis p. 5)
• pore pressure fixities - can be use to set up excess pore pressures, drains or free water
surfaces.
Reality Model
10 kN/m2
5 kN/m 5 kN/m
1 kN/m
2 kN/m2 1 kN/m
1m
Figure 2.11: Normal surface loading
Excavation Fill
1 3
2 2
3 1
Figure 2.12: Excavation and fill
Principles of numerical modelling 2 - 10

Use pile response to various loadings as examples:

• axial loading: shaft friction and end bearing (3-dimensional to axisymmetry) see
page 12
e.g. the pile behaviour is a function of…..f ( Epile / G , l / ro , ν )
where
Epile is the pile Young's modulus
G is soil shear modulus
l is the pile length
ro is the pile radius
ν is the Poisson's ratio
• lateral thrust/loading due to embankment surcharge (3-dimensional to plane strain)
• piled abutment (3-dimensional to plane strain)
Axisymmetry
Driven pile installation? not so good numerically (unless dynamic analysis); better in the
centrifuge
• spherical / cylindrical cavity expansion
• remoulds soil around pile with massive strains
• changes stress history
• wish-in-place pile is normally adopted
• it is artificially possible to change soil properties adjacent to pile or use interface
elements
Bored pile installation? not so good for centrifuge modelling; better in numerical analysis
• remove soil elements and replace with bentonite (relax circumferential stresses)
• tremie concrete (heavy liquid) to reload excavated cylindrical hole circumferentially
• replace concrete as a heavy liquid by hardened concrete.

Axial loading on a long flexible cylindrical pile

Axial load
What is the vertical deformation pattern in the
soil and in the pile due to the axial load on the
pile?
Pile δzp?
δzs?
Soil
Shaft friction
Plan
Pile
θ
Soil
End bearing
r
Section
Plan axisymmetry r and θ plane: z common
Shaft friction component Mesh

Normal, uniformly distributed load
r
z
r
z
Pile Soil
End bearing component
r
z
Soil
Circular ‘footing’ load
Figure 2.13: Modelling an axially loaded pile

Plane strain
Piled full-height bridge abutment Piled full-height bridge abutment

δu
Embankment
arching δup
Soft
p?
Stiff
1 Ellis PhD, 1997 1
Figure 2.14: Piled full-height bridge abutment
δu
z
x
Lateral thrust
y
Pile Soil
CL
Plan Section
Select half space: PLANE STRAIN x and y plane: z common
When pile is displaced laterally relative to the soil, what is

the relative soil-pile movement?
Figure 2.15: Idealisation for lateral thrust on a single row of piles from embankment loading

2.5.5 Ellis (1997): Piled full-height abutment: 3D problem as 2D....

• behaviour of soil under embankment of most interest
• relative soil-pile displacement less critical than this
• additional component of lateral thrust caused by arching is critical
• model row of piles as a wall of equivalent bending rigidity
• overlay soil and 'pile' wall with interaction law with relative soil-pile movement
• soil may be displaced past 'pile' wall so lateral thrust on piles added to equilibrium
equation.
sand low piled

embankment h ig sress abutment
hs wall
tre
s s
kPa
soft clay
Figure 2.16: Finite element analysis: contours of horizontal stress (Ellis, 1997)
SAND
CLAY
SAND
Figure 2.17: The finite element mesh with vertical drains (Ellis, 1997)

2.5.6 Soil
Why can we not use simple back of the envelope calculations probably based on elastic
analyses?
• simple elastic models do not reproduce key aspects of soil behaviour
• we can select more appropriate soil models for design, to account for
→ pre-yield stiffness
→ yield and failure criteria
• OK for more complex analysis because computing power in design offices is growing
Must select?
• type of analysis
• model of soil behaviour - or - constitutive model.
Type of analysis
• steady state (time-independent)
→ steady state seepage
→ static load-deformation problems.
• transient (time-dependent)
→ consolidation
→ dynamic loading (earthquakes, wave action)
→ contaminant transport processes
→ creep.
Drained analysis
• no excess pore pressure - highly permeable soils
• all the loads will be transferred to the soil skeleton: effective stress
• long-term condition - mostly interested in displacements.
Undrained analysis - low permeability soils
• loads will be carried by both soil skeleton and pore pressure
• no volume change - very large bulk modulus K compared to shear modulus G: K>>G
• short-term stability - mostly interested in (total) stresses - undrained failure of clays?
• avoid using equal size elements if the solution is oscillating or use higher order elements,
or
• set νu= 0.49 with a short time step within a consolidation analysis.

Figure 2.18: Influence of the Poissons ratio on the settlement of a strip footing (Potts & Zdraykovic, 1999)
Consolidation analysis (Biot's equations) - more time consuming

• transition from undrained condition to drained condition
• check the movement of the system with time
time
undrained
settlement
Figure 2.19: Settlement of a footing in time
Which do you want to choose for your analysis?

2.6 References
1. M.S.S. Almeida, Stage constructed embankment on soft clay. PhD thesis, University of
Cambridge. 1984.
2. M.F. Bransby, Piled foundations adjacent to surcharge loads. PhD thesis, University of
Cambridge. 1995.
3. A. El-Hamalawi, Adaptive refinement of finite element meshes for geotechnical
analysis. PhD thesis, University of Cambridge. 1997.
4. E.A. Ellis, Soil-Structure interaction for full-height piled bridge abutments constructed on
soft clay. PhD thesis, University of Cambridge. 1997.
5. D.M. Potts, L. Zdravkovic, Finite Element Analysis in Geotechnical Engineering. Vols.
1 & 2. Thomas Telford, London.1999.
6. H.G. Poulos, Experiences with soil-structure interaction in the Far East. 2nd Int.
Conference on Soil Structure Interaction in Urban Civil Engineering. Zürich, 2002.
7. L. Prandtl, Über die Eindringungsfestigkeit (Härte) plastischer Baustoffe und die
Festigkeit von Schneiden, Zeitschrift für angewandte Mathematik und Mechanik, 1921,
1(1), 15-20.
8. M.F. Randolph and G.T. Houlsby, The limiting pressure on a circular pile loaded
laterally in cohesive soil. Géotechnique, 1984, 34, No. 4, pp. 613-623.

Numerical Modelling
Finite Element Method
Dr. Jitendra Sharma

3 Finite Element Method (FEM) in Geotechnical Engineering
3.1 Introduction
The importance of a carefully planned and executed experimental modelling can not be
overstated. However, experimental modelling can be expensive and time-consuming and is
normally used only for high-cost and high-risk projects. For “normal” projects, site investi-
gation is undertaken in combination with laboratory testing to obtain soil parameters as
accurately as possible. These parameters are then used as input to either limit equilibrium
based programs (e.g. slope stability, bearing capacity, etc.) to predict failure loads (ultimate
limit state) or a numerical analysis program (e.g. finite element method, finite difference
method, etc.) to predict the deformation under working load conditions (serviceability limit
state). In this chapter, we will focus on one of the most popular numerical analysis
technique used in geotechnical engineering – the finite element method or FEM. The aim of
this chapter is to learn how to apply the FEM in solving a geotechnical engineering problem.
The emphasis is on the application and not on the formulation of the FEM. A curious reader
may well consult one of the numerous books that deal with the mathematics and the
numerical techniques used in the FEM, e.g. Zienkiewicz and Taylor (1989).
3.2 Numerical methods used in geotechnical engineering
Solution of Geotechnical Problems
Solution of Geotechnical
Problems
Empirical, Based
“Exact” or on Experience
Closed Form Numerical
Limit
Equilibrium
Finite/ Boundary Finite Finite Discrete
Boundary Element Difference Element Element
Element
Figure 3.1: Various ways of solving a geotechnical engineering problem
As stated in the beginning of this course, there are several different ways of finding
solutions to a geotechnical engineering problem. These are summarized in Figure 3.1. In
this section, we will focus on the numerical methods. One of the characteristic features of
the numerical methods is that they usually involve solving a set of simultaneous partial
differential equations (PDEs). Since soil is essentially a non-linear elasto-viscoplastic,
three-phase material, direct solution of the set of PDEs is often impossible. Therefore, an
iterative numerical approach is used. There are five major types of numerical methods used
in geotechnical engineering – the finite element, the finite difference, the boundary element,
the discrete element and the combined boundary/finite element. The way the PDEs are
formulated and solved differs for each of these methods.
Finite Element Method (FEM) in Geotechnical Engineering Page 3 - 1

3.3 What is FEM?
Semi-infinite Continuum
A finite element
Discretization
Discrete Problem Continuous Problem
Figure 3.2: Discrete vs. continuous problem
Before introducing the concept of the FEM, let us first explore the difference between a
discrete and a continuous system. For a discrete system, an adequate solution can be
obtained using a finite number of well-defined components. Such problems can be readily
solved even with rather large number of components, e.g. the analysis of a building frame
consisting of beams, columns and slabs (Figure 3.2). For a continuous system, such as a
soil layer, the sub-division is continued infinitely so that the problem can only be defined
using the mathematical fiction of infinitesimal. Depending on the level of complexity
involved, there are two ways of solving such a problem. Simple, linear problems can be
solved easily by mathematical manipulation. Solution of complex, non-linear problems
involves discretization of the problem into components of finite dimensions (Figure 3.2) and
then using a numerical method such as the FEM.
The most distinctive feature of the FEM that separates it from other numerical methods is
the division of a given domain into a set of simple subdomains, called finite elements. Any
geometric shape that allows computation of the solution or its approximation, or provides
necessary relation among the values of the solution at selected points, called nodes, of the
subdomain, qualifies as a finite element. Such a subdivision of a whole into parts has two
advantages:
1. It allows accurate representation of complex geometries and inclusion of dissimilar
materials.
2. It enables accurate representation of the solution within each element, to bring out local
effects (e.g. large gradients of the solution).

3.3.1 Historical Background

The idea of representing a given domain as a collection of discrete parts is not unique to the
FEM. It was recorded that ancient Greek mathematicians estimated the value of π by noting
that the perimeter of a polygon inscribed in a circle approximates the circumference of the
circle. They predicted the value of π to accuracies of almost 40 significant digits by repre-
senting the circle as a polygon of finitely large number of sides. Searching for approximate
solution or comprehension of the whole, by studying the constituent parts of the whole is
vital to almost all investigations in science, humanities, and engineering. The FEM is an
outgrowth of the familiar procedures such as the frame analysis and the lattice analogy for
2- and 3-dimensional bodies. Its application is not exclusive to engineering. It has been
used in other fields such as mathematics & physics. One of the earliest examples of its use
was in mathematics by R. Courant who used it for the solution of equilibrium and vibration
problems (Courant, 1943). However, Courant did not call his method the FEM. It was R.W.
Clough who first coined the term finite element in 1960 when he applied the FEM to plane
stress analysis (Clough, 1960).
During the early days of the digital revolution, due to the excessive cost of using the bulky,
not-so-easy-to-use mainframe computers, the FEM remained in the hands for those “elite”
people of science who had access to this rather expensive computing power. Only after the
advent of the personal computer and the smaller, more manageable and efficient minicom-
puters, did it manage to break the barriers. Now, with tremendous amount of rather cheap
computing power at their disposal, FEM is the first choice for many engineers and scientists
embarking on the analysis of a wide variety of engineering problems – from designing a
new ergonomic shoe sole to designing a supersonic fighter aircraft. Its use in the field of
bioengineering, for example, the modelling of knee prosthesis or stress analysis of brain
oedema, is also fast becoming popular.
3.3.2 The fundamental steps of the FEM

The three fundamental steps of the FEM are:
1. Divide the whole into parts (both to represent the geometry as well as the solution of the
problem).
2. Over each part, seek an approximation to the solution as a linear combination of nodal
values and approximation functions.
3. Derive the algebraic relations among the nodal values of the solution over each part,
and assemble the parts to obtain the solution of the whole.
We will consider the example of the approximation of the circumference of the circle in
order to understand each of these three steps. Although this is a trivial example, it illus-
trates several (but not all) ideas and the steps involved in the finite element analysis of a
problem.
3.3.3 Approximation of the Circumference of a Circle

Consider the problem of determining the perimeter of a circle of radius R (Figure 3.3).
Ancient mathematicians estimated the value of the circumference by approximating it by
line segments, whose lengths they were able to measure. The approximate value of the
circumference is obtained by summing the lengths of all the line segments that were used.
Let us now outline the steps involved in computing an approximate value of the circum-
ference of the circle. In doing so, we will also learn about certain terms that are used in the
finite element analysis of any problem.

1. Finite element discretization: First, the domain (i.e. the circumference of the circle) is
represented as a collection of a finite number of n subdomains, namely, line segments.
This is called discretization of the domain. Each subdomain (i.e. the line segment) is
called an element. The collection of elements is called the finite element mesh. The
elements are connected to each other at points called nodes. In the present case, we
discretize the circumference into a mesh of five (n = 5) line segments. The line
segments can be of different lengths. When all elements are of same length, the mesh
is said to be uniform; otherwise, it is called a non-uniform mesh (see Figure 3.3b).
2. Element equations: A typical element is isolated and its required properties, i.e. its
length, are computed by some appropriate means. Let he be the length of the element
Ωe in the mesh. For a typical element Ωe, he is given by (see Figure 3.3c):
(3.1)
where R is the radius of the circle and θe < π is the angle subtended by the line segment at
the centre of the circle. The above equations are called element equations. Ancient
mathematicians most likely made measurements, rather than using (3.1) to find he.
Element
Node
(a) (b)
he
θe Approximation of the circumference
of a circle by line elements:
(a) Circle of radius R; (b) Uniform and
non-uniform meshes used to represent
the circumference of the circle;
(c) a typical element.
(c)
Figure 3.3: Approximation of the circumference of a circle by line elements
Assembly of element equations and solution: The approximate value of the circumference
(or perimeter) of the circle is obtained by putting together the element properties in a
meaningful way; this process is called the assembly of the element equations. It is based,
in the present case, on the simple idea that the total perimeter of the polygon (assembled
elements) is equal to the sum of the lengths of individual elements.

(3.2)
Then, Pn represents an approximation to the actual perimeter, p, of the circle. If the mesh is
uniform, i.e. he is the same for each element in the mesh, θe = 2π/n, and we have
(3.3)
3. Convergence and error estimate: For this simple problem, we know the exact solution:
(3.4)
We can estimate the error in the approximation and show that the approximate solution Pn
converges to the exact solution p in the limit as n → ∞.
In the summary, it is shown that the circumference of a circle can be approximated as
closely as we wish by a finite number of piecewise-linear functions. As the number of
elements is increased, the approximation improves, i.e. the error in the approximation
decreases.
3.4 Basic formulation of the FEM

In this section, the basic formulation of the FEM will be introduced using three simple
examples: (1) a system of interconnected elastic springs; (2) a one-dimensional plane truss
element; and (3) a constant strain triangular finite element.
3.4.1 Interconnected elastic springs
1
d1
a
Ta
2 2
d2 Tb W2 Td
b
Equilibrium at Node 2
3 d
d3
c
4
d4

Figure 3.4: A system of interconnected springs
1. In this system, linear elastic springs are the finite elements.

2. From a structural mechanics point-of-view, the structure is statically indeterminate.
3. Let the stiffnesses of individual springs be ka, kb, kc and kd. Therefore, the tensions in
these springs are given by:
(3.5)
where ea, eb, ec and ed are extensions of springs a, b, c and d, respectively.
4. Let us now invoke three fundamental principles of structural mechanics: compatibility,
material behaviour and equilibrium for the calculation of the displacement of each
spring. These three principles are applied in the order of compatibility – material
behaviour – equilibrium.
5. The compatibility equations are:
(3.6)
where d1, d2, d3 and d4 are displacements of nodes 1, 2, 3 and 4, respectively. Here, we
are making sure that the system does not fall apart, i.e. springs remain connected with each
other.
6. Material behaviour can be expressed using spring stiffnesses as:
(3.7)
7. Equilibrium (at node 2, see Figure 3.4):
or
(3.8)
which on rearrangement, results in:
(3.9)
8. Similar equations can be written for other nodes, giving four linear simultaneous
equations in d1, d2, d3 and d4 that can be expressed in matrix form as:
(3.10)

The matrix on the left-hand-side is called the global stiffness matrix. Equation (3.10) can
be written in matrix notation as:
Kd = W
These simultaneous equations can be solved by elimination and values of displacements
can be obtained. From the values of displacements, the force in each spring can be calcu-
lated.
9. The global stiffness matrix K consists of the sum of matrices of the following form
(where ke is the stiffness of one particular spring):
10.
(3.11)
This matrix is called the element stiffness matrix. It relates the nodal displacements to the
forces exerted on each spring at nodal points. One of these matrices is added into the
global stiffness matrix for each spring in the system.
(3.12)
3.4.2 A plane truss element
dy2
x’
dx2
y’ Length = L 2
dy1 α
dx1
y 1
Figure 3.5: A plane truss element
In this section, we will apply the same principles of compatibility, material behaviour and
equilibrium to a one-dimensional plane truss element (Figure 3.5). The formulation is now
more complex than that for a simple system of linear elastic springs. You may have noticed
that in the case of linear elastic springs, each node was allowed to move in only y-direction,
i.e. up or down. Here, each of the two nodes of the plane truss elements has two degrees
of freedom, i.e. it can move in both x- and y-direction. However, as we shall see, the
general solution procedure remains the same regardless of the increased complexity.

In calculating the strains in this element, we are only interested in the displacements along
the direction of the element. It is, therefore, logical to define a system of axes x’-y’ that is
local to the element, with x’-axis coincident with the direction of the element.
1. Let us first apply the condition of compatibility, i.e. the element should not break in the
middle. Mathematically, it can be expressed in terms of the equation for displacement at
a distance x’ along the element
: (3.13)
2. To obtain the element stiffness matrix, we need to write this expression in terms of the
degrees of freedom dx1, dy1, dx2 and dy2. This is achieved by noting that
(3.14)
from simple geometric consideration.
3. Making this substitution, we obtain:
(3.15)
4. The strains inside the element can now be related to nodal displacements using a
matrix that is obtained by differentiating equation (3.15) with respect to x’. This matrix is
called the B matrix in the FEM formulation and is given by:
In the matrix notation, the strain matrix is now written as:

(3.16)
where ae is the vector of nodal displacements – right-hand-side matrix in equation (3.15).
5. Assuming the plane truss element to be linear elastic,

the stress inside the element can now be expressed in
terms of nodal displacements as:
(3.17)

where D is the matrix of material behaviour or constitutive matrix for the element. In this
case, it simply reduces to the Young’s modulus of the plane truss element, E.
6. The principle of virtual work can now be used to find the nodal forces Fe that are in
equilibrium with this state of internal stress. A set of virtual nodal displacements
applied to the element accompanies a set of virtual strains within the element
according to the relation:
(3.18)
The principle of virtual work gives:
(3.19)
7. Substituting for σ and ε̂ , we obtain:

(3.20)
T
From the above equation, â e can be cancelled out to give:
(3.21)
where K is the element stiffness matrix. For our plane truss element, it can be shown to be
given by:
(3.22)
where A is the cross-sectional area of the plane truss element, C = cosα and S = sinα.
8. For a typical plane truss problem, the forces acting on the nodes are known. Hence,
equation (3.21) can be solved by first inverting the K matrix and then solving the
resulting simultaneous equations for nodal displacements.
3.4.3 A constant strain triangular finite element

After having successfully formulated the FEM for the solution of two one-dimensional
problems, we move to the formulation of a two-dimensional constant strain triangular finite

element. Figure 3.6 shows the simplest triangular finite element used for two-dimensional
continuum analysis.
dy2
2 dx2
h
dy1
dy3
x
3 dx3 1 dx1
Figure 3.6: A constant strain triangular finite element
1. Each of its three nodes has two degrees of freedoms and the terms dx1, dy1, dx2, dy2,
dx3, and dy3 denote the nodal displacements. In this case, the unknown variation of the
displacement within the element adds to the complexity of the problem. Here, we are
going to assume that this variation is linear, i.e.
and
(3.23)
Since the strain is the first derivative of the displacement, it will be constant within the
element. Hence, the element is called a finite element.
2. The coefficients c0, c1, etc. in equation (3.23) are obtained by substituting the coordi-
nates of the three nodal points into these expressions. In this case, too, we assume a
local coordinate system with origin at node 3 and x-axis along side 3-1 and y-axis along
side 3-2. Solving the resulting sets of simultaneous equations, we obtain:
and
(3.24)

3. Equation (3.24) can be written in matrix notation as:

(3.25)
where N is the matrix of shape functions for the finite element and is given by:
(3.26)
4. Now, we can formulate the B matrix by partially differentiating the N matrix with respect
to x and y as:
(3.27)
Here, the first row denotes strain in x-direction, second row denotes strain in y-direction and
the third row denotes the shear strain in the x-y plane.
5. Assuming plane strain conditions, the element stiffness matrix D can be easily obtained
from Hooke’s law as:
(3.28)
where E is the Young’s modulus and ν is the Poisson’s ratio for the material.
6. Formulating the element stiffness matrix K is now a simple task of calculating the matrix
product BTDB times the area of the element (h2/2) since the terms of all these matrices
are constant. K is given by:
(3.29)
where a = 1 – ν; b = 0.5 – ν and c = 1.5 – 2ν.

Although the above three examples illustrate the basic idea of the FEM, there are several
other features that are either not present or not apparent from the discussion of these
examples. These are summarized below:
1. Depending on its shape, a domain can be discretized into a mesh that contains more
than one type of element. For example, in the discretization of an irregular two-dimen-
sional domain, one can use a combination of triangular and quadrilateral finite
elements. However, if more than one type of element is used, one of each kind should
be isolated and its equations developed. All the commercial FEM
software take this into account and therefore, it is not a problem
to mix element types during an analysis.
2. The governing (simultaneous) equations are generally more complex than those
considered in these three examples. They are usually partial differential equations. In
most cases, these equations cannot be solved over an element for two reasons. First,
they do not permit exact solution. Second, the discrete equations obtained cannot be
solved independent of the remaining elements because the assemblage of the
elements is subjected to certain continuity, boundary and/or initial conditions.
3. The number and location of nodes in an element depend on (a) the geometry of the
element, (b) the degree of polynomial approximation, and (c) the integral form of the
equations. This point is elaborated further in the section dealing with types of finite
elements.
4. There are three sources of errors in a solution obtained by the FEM: (a) those due to
the approximation of the domain; (b) those due to the approximation of the solution; and
(c) those due to numerical computations. The estimation of these errors is not a simple
matter. The accuracy and convergence of a FEM solution depends on the differential
equation, the integral form and the element used. Accuracy refers to the difference
between the exact solution and the solution obtained by the FEM whereas conver-
gence refers to the accuracy as the number of elements in the mesh is increased. This
point is discussed in detail later in the chapter.
3.5 Approximations, accuracy and convergence in the FEM

1. Engineers sometimes regard the finite elements in a mesh as being connected only at
the nodal points in the mesh. This is not a good conceptual picture of how the elements
behave. Straining of finite elements results in a deformation pattern similar to that
shown in Figure 3.7a rather than that shown in Figure 3.7b (i.e. there are no gaps that
open up at the element boundaries). This is because the polynomials or shape
functions that approximate the distribution of displacement are chosen in such a way
that there is a continuity of displacements within the elements as well as between the
adjoining elements.

(a) (b)
Figure 3.7: Continuity of displacements in adjoining finite elements
2. Although strains will be continuous within a finite element, there will usually be a discon-
tinuity of strains between adjacent elements. Some approximation (e.g. a smoothing
zone as shown in Figure 3.8) is necessary so that the terms being integrated become
continuous.

Displacement u
‘Smoothing’ zone
Strain du/dx
Rate of strain d2u/dx2
-∞
Figure 3.8: The use of smoothing zone at element boundaries
3. The stress field within an element will be continuous but may not satisfy the equations
of equilibrium. Except for very simple problems, stresses on either side of element
boundaries will not be equal. Equilibrium is satisfied, however, in an average sense
through the equilibrium equations at nodal points where the resultant forces equivalent
to internal stress field balance the resultant forces due to external traction and body
forces. The extent to which the local stresses appear not to be in equilibrium with the
external forces gives some indication of the accuracy of the solution.
4. Before applying the FEM to solve real problems, it is advisable to test its accuracy by
solving certain benchmark or validation problems for which an exact or closed-form
solution exists. An error-free, robust FEM program should be able to reproduce the
exact solution accurately. One of the most popular benchmark problem in geotechnical
engineering is the calculation of undrained collapse load (qu) of a circular foundation on
soft clay of uniform undrained shear strength (su) – qu = (π+2) su.

5. In addition to testing the accuracy of the FEM, a convergence test should be carried out
for a given problem for which we do not have an exact solution. It involves conducting
three or more FEM analyses with progressively finer mesh. Convergence is achieved
when further refinement of mesh does not result in a significant increase in the accuracy
of the solution (Figure 3.9).
qu
‘Exact’ solution
Mesh A - 24 Elements
Mesh B - 48 Elements
24 48 96
No. of Elements
Mesh C - 96 Elements
Figure 3.9: Testing the convergence by progressive mesh refinement
3.6 Geotechnical finite element analysis

Most of the commercially available FEM programs are written with structural/mechanical
applications in mind. These programs cater for materials that can be produced under
controlled conditions and therefore, have well-defined physical or mechanical properties,
e.g. metals, plastics, polymers, concrete, etc. The most important material in a geotechnical
analysis is the soil. A soil’s physical or mechanical properties have to be measured instead
of being specified or specially fabricated. These properties vary enormously from site to
site and can be profoundly affected by factors such as sampling techniques, specimen
handling and preparation, characteristics of the measurement and data acquisition
techniques. Therefore, the constitutive modelling takes the centre stage in a geotechnical
FEM program. The three phase (soil-water-air) nature of soil makes realistic constitutive
modelling of soil a formidable task. Since the shear strength of a soil at a given point
depends on the effective stress at that point, the stress-strain response of a soil is highly
non-linear. For a geotechnical finite element analysis, the FEM program should have the
following features:
1. Material models that are capable of modelling non-linear stress-strain behaviour and
that include options for undrained analysis (short-term behaviour), drained analysis
(long-term behaviour), most importantly, coupled consolidation analysis.

2. The ability to specify non-zero in-situ stresses.

3. The ability to add or remove elements during the analysis (for modelling the
construction or excavation, respectively).
3.6.1 Plane strain and axisymmetric problems

While a three-dimensional finite element analysis is frequently used in structural or
mechanical applications, it is rarely used in geotechnical engineering. Most of the geotech-
nical problems can be assumed to be either plane strain or axisymmetric without signif-
icant loss of the accuracy of the solution.
1. Plane strain problems: The characteristic feature of a plane strain problem (Figure 3.10)
is that one dimension – in this case the dimension along the z-axis – is considerably
greater than the other two dimensions. As a result, the strains in the direction of z-axis
can be assumed to be zero. Therefore, we only have to solve for strains in the x-y plane
and the problem reduces to a plane strain problem. For plane strain problems, the
numerical integration is performed for a unit section (1 unit length) along the z-axis.
Typical examples of plane strain geotechnical problems are embankments, retaining
walls, tunnels (at sections sufficiently away from the head of the tunnel).
z
x
Figure 3.10: A plane strain problem
Axisymmetric problems: For an axisymmetric problem, both the structure and the loading
exhibit radial symmetry about the central vertical axis (Figure 3.11). Consequently, the
circumferential strains can be ignored in the solution and the problem reduces to a two-
dimensional problem in a vertical radial plane. Keep in mind that the problem can only be
reduced to an axisymmetric problem when both the structure and the loading are symmetric
about the central vertical axis. If one of the two does not exhibit radial symmetry, either the
problem has to be treated as a three-dimensional problem or techniques involving Fast
Fourier Transforms (FFTs) have to be used. The numerical integration for an axisymmetric
problem is performed from zero to 2p, i.e. for the entire horizontal circular cross-section.
Typical examples of axisymmetic geotechnical problems are pile foundation subject to
vertical concentric loads, excavation of vertical shafts of circular cross-section, consoli-
dation around a vertical drain.

C
L
y
Figure 3.11: An axisymmetric problem
3.6.2 Different types of finite elements

There are many different types of finite elements available for use with a geotechnical FEM
program. These elements can be classified based on either the dimensions of the problem
or the order of the element. They can also be classified on the basis of whether the coupled
consolidation formulation is adopted or not.
1. 1-D, 2-D and 3-D elements (Figure 3.12): 1-D and 2-D elements are used mainly for the
plane strain and axisymmetric problems. 3-D elements are used only for the truly three-
dimensional problems.
• Typical 1-D elements include: (a) bar elements for the modelling of struts, geotextile
reinforcement, ground anchors and any other structural element that is not capable of
resisting flexure, and (b) beam elements for the modelling of retaining walls, tunnel
linings and any other structural element requiring flexural rigidity.
• Typical 2-D elements include (a) triangles and quadrilaterals for the modelling of soil and
structural components of significant dimensions, and (b) slip elements for modelling of
soil-structure interface behaviour.
• Typical 3-D elements are hexahedrons and tetrahedrons for the modelling of soil and
structural components. Some FEM programs also have 3-D slip elements for modelling
of soil-structure interface behaviour.

dy2 dy2
dx2 θ2 dx2
dy1 dy1
θ1
dx1 dx1
(a) Two-noded bar element (b) Two-noded beam element
dy2
dx2 Structure
dy1 dy3 t
Soil
L
dx1 dx3 (d) 2-D slip element
(c) 2-D element (e) 3-D elements
Figure 3.12: 1-, 2- and 3-D elements
2. First-, second- and fourth-order elements (Figure 3.13): The order of the element is
determined by the order of the polynomial used as the shape function.
• For a first-order element, a first-order polynomial, i.e. a straight line, is used as shape
function. The constant strain triangle in the example above is a first-order element. A
mesh containing only first-order elements requires a large number of elements for a suffi-
ciently accurate solution.
• For a second-order element, a quadratic or second-order polynomial is used as shape
function. As a result, the strain within the element is distributed linearly. Hence, these
elements are also called linear strain elements. Such elements usually have one or more
mid-side nodes in addition to the vertex nodes. One does not need to use a large
number of second-order elements in order to achieve sufficient accuracy.
• For a fourth-order element, a quartic or a fourth-order polynomial is used as shape
function. The strains, therefore, have a cubic variation within the element and the
element is often called a cubic-strain element. Such elements have several mid-side
nodes as well as nodes inside the element in addition to the vertex nodes. It is not
common to use such elements for a routine geotechnical analysis. Their use is limited to
special situations such as testing a new constitutive model, unit cell radial consolidation
problems.

Displacement Displacement
x x
(a) (c)
Displacement
(a) First-order element

(b) Second-order element
x
(c) Fourth-order element
(b)
Figure 3.13: First-, second- and fourth-order finite elements
3. Consolidation elements (Figure 3.14): These elements are required when the FEM
program adopts a coupled consolidation formulation. In a coupled-consolidation formu-
lation, the excess pore pressures are treated as unknowns. Any variation in the
magnitude of excess pore pressure at a given point is reflected simultaneously in the
magnitude of effective stress at that point. In addition to the standard displacement
nodes, consolidation elements have pore pressure nodes where the value of excess
pore pressure is calculated. For second-order elements, pore pressure nodes are
normally superimposed on vertex displacement nodes of the element. For higher-order
elements, pore pressure nodes also exist inside the elements.
+ =
Displacement Pore Pressure Consolidation

Element Element Element
Figure 3.14: Consolidation element

3.7 Techniques for modelling non-linear stress-strain response
σ σ
Fe = K ⋅ ae δF e = Ki ⋅ δae
ε ε
(a) Linear stress-strain response (a) Non-linear stress-strain response
σ σ
E4
ε Increment No. 1 2 3 4 5 6
ε
(b) Non-linear stress-strain response (b) Piecewise linear approximation
Figure 3.15: Linear and non-linear material Figure 3.16: Piecewise linear approximation of
behaviour non-linear material behaviour
The basic formulation of the FEM described in Section 3.4 is applicable only for materials
that obey linear stress-strain laws (Figure 3.15a). However, as mentioned above, the
stress-strain behaviour of a soil is highly non-linear (Figure 3.15b) and therefore, for
solution of geotechnical engineering problems the fundamental equation of the FEM
(equation 3.21) cannot be used in its present form. First, the non-linear stress-strain curve
should be approximated by a set of interconnected straight lines (i.e. it is made piecewise
linear) and then an incremental form of equation 3.21 is used. This approach is illustrated in
Figure 3.16. Depending on the degree of non-linearity, the imposed loading (or
displacement) is divided into sufficient number of increments and equation 3.21 is solved
for each increment in succession. This is the simplest way of modelling a non-linear
material. The trick here is to make sure that the piecewise linear approximation does not
drift from the true stress-strain curve by a certain tolerable amount. However, the appli-
cation of this method is limited to material models that have a well-defined yield function,
e.g. models based on critical state soil mechanics theory. This method is not suitable for
elastic-perfectly plastic models such as the Mohr-Coulomb model. The reason for this is
that the yield function and the failure criterion are one and the same for such models and
there is no other way of detecting the yielding of the material than to cross (and go out of)
the failure envelope (Figure 3.17a). Such a stress state is not admissible and will result in
internal forces that are not in equilibrium with external forces. Therefore, the stress state
must be corrected back to the failure criterion. This can be achieved in several different
ways. The following two methods are commonly used in a geotechnical FEM software:
1. Tangential stiffness approach with carry over of unbalanced load
2. Modified Newton-Raphson method

3.7.1 Tangential stiffness approach with carry over of unbalanced load

This approach is illustrated in Figure 3.17b. In this approach, the global stiffness matrix is
computed based on the tangential stiffness at the beginning of an increment, say from 0 to
a displacement d1 as shown in figure 3.17b. In other words, the stress-strain response is
now considered linear for this increment and is represented by the tangent drawn at the
starting point of the increment. The internal load at the end of this increment (∆P1) is no
longer in equilibrium with external load and this out-of-balance load (∆PC1) is re-applied to
the finite element mesh at the beginning of the next increment (from displacement d1 to d2).
It is obvious that the accuracy of the solution will suffer considerably if the magnitude of the
out-of-balance load is rather large. The accuracy of the solution can be assessed by
examining the global equilibrium error (percent difference between the sum of external
loads and sum of internal forces) at the end of each increment. For elastic-perfectly plastic
models, this error should never be allowed to go beyond 15 to 20%. To achieve this goal, a
sufficiently large number of increments should be used. Another alternative is to divide
each increment into 5 or 10 sub-increments (Figure 3.17c). This will ensure that the
magnitude of out-of-balance load for each sub-increment is small.
P
τ
yield
1 surface
∆τc1
∆P2
2
∆P1 ∆Pc1
σ
{∆Pc1 } = ∫ B T ∆τ c1d (vol) d
d1 d2
(a) Stress state correction (b) Tangential stiffness approach
P P
∆P1
d d
Sub-increment 1 2 3 d1
(c) Use of sub-increments to apply
(d) Modified Newton-Raphson method
out-of-balance load
Figure 3.17: Methods of modelling non-linear material behaviour
3.7.2 Modified Newton-Raphson method

It is also known as the quasi Newton-Raphson method. In this method, similar to the
tangential stiffness approach, the stiffness matrix is computed based on the tangential
stiffness at the beginning of an increment. However, the out-of-balance load is not carried
over to the next increment. Instead, an iterative procedure shown in Figure 3.17d is
followed. The out-of-balance load (∆PC1) is re-applied to the mesh and the resulting incre-

mental displacements are added to the current displacements. If further yielding takes
place during the application of ∆PC1 then a second set of out-of-balance load (∆PC2) are
calculated and the above procedure is repeated until convergence is reached, i.e. the
resulting incremental displacements or the out-of-balance load is less than a preset
tolerance. The main advantage of this procedure is that the stiffness matrix is computed
only at the beginning of an increment. However, rather large number of iterations required
to achieve convergence compensates the savings on computation time thus achieved.
Also, the method may fail to converge for some highly non-linear problems.
3.8 Techniques for modelling excavation and construction

3.8.1 Excavation
Geotechnical activities that involve excavation can be broadly classified into three main
categories: trenches, shafts and tunnels. Trenches can be rather small, e.g. for laying of a
drainage pipe (Figure 3.18a), or big and deep, e.g. for the construction of basement car
park (Figure 3.18b). The effect of excavating a small trench on surrounding soil and struc-
tures is not so great and, therefore, such a problem is rarely analyzed using the FEM.
However, a deep excavation can result in significant ground movements capable of
damaging the surrounding structures. It is, therefore, not surprising that its design almost
invariably involves conducting a few FEM analyses. The length and the width for a typical
deep excavation are comparable and hence, it is a 3-D problem. However, a 3-D FEM
analysis is rarely used and often, the problem is assumed to be a plane strain problem.
Excavation of a shaft is modelled similar to a trench or a deep excavation; the only
difference is that axisymmetric conditions are assumed.
C
L
Diaphragm Wall
Struts
6~12 m
2m
4~8 m
5~10 m
1m
(a) (b)
Figure 3.18: Trenches and deep excavations

Ground
Surface
Tunnel Face
(a) Vertical Transverse Section
Unsupported Heading
(b) Vertical Longitudinal Section
Figure 3.19: Excavation of a tunnel and its 2-D FEM approximations
The construction of tunnels is an urban necessity. It involves excavation of soil using a

tunnel boring machine (TBM) and installation of permanent lining for the excavated section.
Excavation of a tunnel causes ground loss as well as stress relief at the face of the tunnel,
resulting in significant surface settlements. These surface settlements can result in signif-
icant damage to nearby structures. Tunnel excavation is also a 3-D problem (Figure 3.19).
However, it is quite common to assume plane strain conditions (representing a vertical
transverse section sufficiently away from the face of the tunnel) and to use a volume loss
parameter that takes into account the 3-D effect in an approximate manner. To study the
ground movements ahead of the tunnel face, a vertical longitudinal section is considered
and plane strain conditions are assumed.
For both the deep excavation and the tunnel, the modelling of excavation is achieved in the
same way – by removing the elements from the mesh. Here, it is worth noting that the body
forces within the element are composed of both soil (effective stress) and water (pore
pressure) as shown in Figure 3.20. When an element is removed, both the soil and water
body forces are removed. For the deep excavation, it represents an excavation that is dry,
i.e. not filled with water. For an excavation in a clayey soil, this means that there are
negative pore pressures (pore suction) on the inner boundaries of the excavation. Unless
some support in the form of a retaining wall is provided, the soil will eventually lose its
suction and the excavation will collapse. However, such removal of body forces is not
realistic for certain situations, e.g. installation of a diaphragm wall. The trench for a
diaphragm wall is filled with either water or bentonite slurry. In this situation, one must either
re-apply the water body forces or apply body forces corresponding to the bentonite slurry
on the inner boundaries of the excavation.

Effective stress + pore pressure
(a) Excavation (b) Stresses acting before excavation
(c) Stresses after excavation (zero) (d) Net effect of excavation
Figure 3.20: Stress changes during modelling of an excavation
3.8.2 Construction
The word “construction” in geotechnical engineering usually means placing one or more
layers of soil over existing or made-up ground, e.g. construction of a highway embankment
on soft clay. The placing of a layer is modelled either by adding elements to the existing
mesh or by applying pressure at the boundaries (Figure 3.21). The latter approach gives
satisfactory solution provided the newly placed layers are not expected to undergo any
shear deformation. If this is not the case, the technique of adding elements to the mesh
should be used. An element that is added is assumed to be unstressed and the self-weight
of the element is the only contributor to the body forces of that element. For this reason, the
added elements must either have elastic properties or have a small non-zero value of
apparent cohesion c’ if elastic-perfectly plastic model is used. A constitutive model that
requires specification of a stress history, e.g. Cam-clay or other critical state models, is
unsuitable for modelling of added elements.
First Layer Second Layer

CL CL
Embankment
Soft Clay
(a) by adding elements (b) by pressure loading
Figure 3.21: Techniques for modelling layered construction of an embankment

Compaction is an integral part of a geotechnical construction activity and its effects should
ideally be included in the modelling. However, the effects of compaction are difficult to
quantify in terms of stresses. In addition, the soil that is being compacted is usually partially
saturated. These factors make the modelling of compaction activity quite complicated and
therefore, most commercial geotechnical FEM software simply ignore it.
3.9 Advantages and drawbacks of the FEM

3.9.1 Advantages
1. It is relatively easy to use and, therefore, it is one of the most popular methods for
advanced geotechnical modelling.
2. There are many commercial FEM programs available that are capable of geotechnical
modelling (discussed later).
3. Since each element’s properties are modelled and evaluated separately, it is quite easy
to incorporate non-homogenous ground conditions such as layers of different soils.
4. Any shape of domain can be modelled with the possibility of including holes, gaps, etc.
5. Boundary conditions can be applied easily.
6. It is possible to couple different physical phenomena such as diffusion and thermal
conduction within the same formulation. This is possible because all of these
phenomena can be described by the Laplacian equation.
7. Construction and excavation of soil layers in geotechnical engineering can be done
easily by adding or removing elements from the mesh (discussed later).
3.9.2 Drawbacks
1. While an FEM program is relatively easy to use, interpretation of its output can be a
formidable task and usually requires considerable expertise and experience.
2. It is not suitable for highly non-linear problems or problems that involve large strains,
e.g. cone penetration test, consolidation of a hydraulic fill or a clay slurry. For such
problems, a finite difference formulation incorporating fast Lagrangian analysis
procedure is more suitable.
3. It is also not suitable for the modelling of brittle materials that exhibit discontinuities in
the form of cracks, faults and fissures, e.g. rock. For such materials, a discrete element
formulation is more suitable.
3.10 Some popular commercial FEM programs

3.10.1 ABAQUS
ABAQUS is a general-purpose FEM program that contains many useful features:
• Static stress-displacement, transient dynamic stress-displacement, heat transfer, mass
transport and steady-state transport analyses.
• Coupled formulations that include: Biot’s consolidation theory, thermo-mechanical
coupling, thermo-electrical coupling, fluid flow-mechanical coupling, stress-mass
diffusion coupling, piezoelectric and acoustic-mechanical coupling. The most important
of these from a geotechnical point-of-view is Biot’s consolidation theory.

• Dynamic stress-displacement analysis, determination of natural modes and frequencies,

transient response via modal superposition, steady-state response resulting from
harmonic loading, response spectrum analysis, and dynamic response resulting from
random loading. These features are mainly for earthquake or other dynamic applications.
• It has a huge library of all finite elements developed in the literature such as 1-D, 2-D and
3-D continuum elements, shell, membrane, pipe, beam and elbow elements, springs,
dashpots, joint, interface and infinite elements. User-defined elements can also be used.
• Similarly, it has an impressive collection of constitutive models including general elastic
(linear and non-linear), elasto-plastic, elasto-viscoplastic, hyper and hypo-elastic
models. Constitutive models that are useful for geotechnical analysis are von Mises,
Mohr-Coulomb, Drucker-Prager, Extended Drucker-Prager (non-associated flow), Cam
Clay, Modified Cam Clay, Capped Drucker-Prager (Cam Clay with Extended Drucker-
Prager for use in tunnel excavation), and strain-rate dependent plastic laws.
• User-defined constitutive models can also be incorporated with the help of a subroutine
interface.
• It is possible to simulate excavation and construction.
• It can deal with large strain and large deformations.
• Presently, it is the only commercial program except ZSOIL (described below) that can
deal with partially saturated soils.
• It can model seepage problems with phreatic surfaces and capillary effects.
• It even allows cracks and rock joints to be modelled and it can model creep, too.
• It can perform adaptive mesh refinement for undrained and drained problems only.
• Operating System: Windows NT, UNIX, Sun Solaris and a host of other systems running
mainly on multiprocessor or parallel computers.
• It is very expensive but a cheaper, educational version with limited capabilities is
available for teaching and research use.
• More information can be obtained from http://www.abaqus.com/
3.10.2 SAGE CRISP

SAGE CRISP has evolved from CRISP – CRItical State Program – developed by the
Cambridge University Soil Mechanics Group in the 1970s and 80s. CRISP was one of the
first FEM programs dedicated to geotechnical analysis. In the early 1990s, SAGE
Engineering Ltd., UK developed the pre- and post-processors for this program and began
marketing the program by the name SAGE CRISP. The main features of SAGE CRISP are
as follows:
• It can perform static stress-displacement and coupled consolidation analyses in one-,
two- and three-dimensions. At present, there is no facility to do dynamic analysis but the
developers of SAGE CRISP are in process of incorporating this facility.
• Its element library includes 1-D, 2D and 3D continuum, bar, beam and interface
elements.
• Almost all of its constitutive models cater for geotechnical applications. These include
general elastic (linear and non-linear), anisotropic elastic, elastic-perfectly plastic with
von Mises, Tresca, Drucker-Prager, Mohr-Coulomb failure criteria, Cam Clay, Modified
Cam Clay, Schofield, 3-Surface Kinematic Hardening (for small-strain modelling) and
hyperbolic (Duncan and Chang type) models.
• It can model excavation and construction.

• It does not include a large-strain formulation but it can deal with large strain problems in
an approximate manner by using large number of increments and updating of geometry
at the end of each increment.
• Operating System: Windows (older versions of CRISP run under MS DOS).
• More information can be obtained from http://www.crispconsortium.com/
3.10.3 PLAXIS
PLAXIS is a geotechnical FEM program developed by PLAXIS BV of the Netherlands. Its
name is a combination of PLane strain and AXISymmetric. As the name suggests, it can
only do 1-D and 2-D analyses although a 3-D version is being developed. Its features
include:
• 1-D and 2-D static stress-displacement and coupled consolidation analyses.
• The element library consists of 1-D and 2-D continuum, beam, spring and interface
elements.
• Its library of constitutive models includes general elastic (linear and non-linear) aniso-
tropic elastic, Mohr-Coulomb (associated as well as non-associated flow), Soft soil (Cam
Clay), Soft soil creep and Hardening soil (hyperbolic) models.
• It can model excavation and construction. In addition, it can do analysis of tunnel
excavation that incorporates a volume loss parameter that represents the contraction
around tunnel lining due to overcut by the tunnel boring machine and the loss of pressure
at the face of the tunnel.
• It can deal with large strain and large deformation situations and can also model creep.
• It is able to select the optimum number of increments needed for efficient convergence of
non-linear problems.
• It is able to model seepage problems involving phreatic surfaces and capillary effects.
• It allows for incorporation of safety factors into an analysis of, for example, foundations
or slopes.
• Operating System: Windows.
• More information can be obtained from http://www.plaxis.nl/
3.10.4 ZSOIL
ZSOIL is a geotechnical FEM program developed by Zace Services AG, Switzerland. Its
features include:
• 1-D and 2-D static stress-displacement and coupled consolidation analyses.
• Elements include 1-D and 2-D continuum, beam, spring, shell, cable and interface
elements.
• Constitutive models include general elastic (linear and non-linear), anisotropic elastic,
elastic-perfectly plastic with Mohr-Coulomb and capped Drucker-Prager failure criteria,
and Hoek-Brown models.
• It can model excavation and construction.
• It is able to deal with large strain and large deformation problems.
• It can model seepage problems involving phreatic surfaces and capillary effects.
• It can model partially saturated flow problems and problems involving creep.
• Operating System: Windows.
• More information can be obtained from http://www.zace.com/

3.11 Guidelines for the use of FEM in geotechnical engineering

There are no shortcuts for learning to use the FEM effectively. One becomes an FEM
expert by experience and a lot of hard work. However, the following guidelines will make
sure that one has a good start to the learning endeavour.
• Use smaller elements in regions where the rate of change of stress with distance is
greater. This happens, for example, near the edges of a loaded area, near a re-entrant
corner in the mesh or where adjacent parts of a mesh have significant differences in
stiffness (e.g. soil reinforcement, retaining wall, pile foundation) as shown in Figure 3.22.
Note that some of these situations result in stress concentrations where the stresses
tend to infinity. The smaller we make the elements near the concentration, the higher are
the stresses. Sometimes, a stress concentration will “spoil” the solution locally, leading to
oscillation of stresses. Here, it is worth remembering that infinite stress concentrations
are mathematical fiction that may be unimportant in describing real behaviour and
therefore, it is often advisable to ignore them.
• When increasing the element size from area of interest to the far boundaries, avoid
increasing the element size by more than a factor of 2 between adjacent elements.
• Wherever possible, make use of symmetry of the problem (if any) - it will save both yours
and the computer’s time.
• Keep the triangular elements as equilateral as possible and the quadrilateral elements as
square as possible.
Edge of the Interface between

Loaded Area pile and soil
Stress
Re-entrant Concentration
Corner
Zone of Interest
Figure 3.22: Areas of FEM domains that require finer elements
• Avoid using curved elements as interior edges between elements in a mesh - only use
them at external boundaries or internal boundaries (e.g. inside of a tunnel) if absolutely
necessary.
• Where you place the boundary of a mesh can make a big difference to the outcome of
the analysis. If you are unsure of the boundary effect, try two different meshes – one with
a close and the other with a far boundary.

• When modelling an axisymmetric undrained problem where collapse is expected (e.g.

cylindrical cavity expansion in a pressuremeter test), use only fourth-order (cubic strain)
elements.
• Always check that the in-situ stresses specified at the start of the analysis are in
equilibrium. If the in-situ stresses are not in equilibrium, either you have not input
consistent values of soil unit weight or have not applied correct fixity conditions to one or
more of mesh boundaries.
• When using elastic-perfectly plastic constitutive models with either a Mohr-Coulomb or a
Drucker-Prager failure criterion, specify a small value for c’ (0.1 or 1 kPa) even if the
material has c’ = 0 kPa. This will ensure that the initial state of stress for the material is
not on the failure surface.
• In order to model the incompressibility of a saturated soil under undrained conditions, a
Poisson’s ratio (ν) of 0.5 should ideally be used. However, if ν = 0.5 is input into an FEM
analysis, it will result in serious ill-conditioning of the equations. The reason for this is
that the bulk modulus (K) of the soil approaches infinity as ν → 0.5. In such situations, ν
= 0.49 usually gives satisfactory results.
• Treat pore pressure boundary conditions with respect. They are the most likely source of
disaster in a geotechnical FEM analysis. Before applying these boundary conditions,
make sure that you fully understand the ground water conditions for your problem. Most
FEM programs treat any mesh boundary as impermeable by default. Setting the excess
pore pressure to zero on a boundary means that the boundary is now able to drain.
However, the task is not complete by just “switching on” the pore pressure boundary. Its
effect must be felt by the adjacent elements in the next time step. Otherwise, oscillation
of pore pressures can occur. The minimum time step required for this purpose can be
computed based on the parabolic isochrone solution to the consolidation equation as
shown in Figure 3.23.
umax umax
u u
L
y y
L2 Oscillation of pore pressure

L = 12cv t or t min = due to insufficient time-step
12cv
Figure 3.23: Minimum time-step for dissipation of excess pore pressures
• When modelling excavation or construction by removing or adding elements to the

mesh, respectively, use several layers of elements and remove/add these elements layer
by layer, applying each layer over several increments. This will ensure that the stiffness
of the soil being removed or added is correctly modelled.

3.12 Concluding remarks

In the beginning of the chapter, we stated that site investigation and laboratory testing are
used to obtain the input soil parameters for an analysis using FEM. As a geotechnical
engineer, one must never forget that soil is an extremely difficult material to characterize.
Sampling disturbances, poorly controlled laboratory experiments, failure to interpret the
results from laboratory tests in a scientific manner are some of the factors that introduce
errors and uncertainty in the values of soil parameters. Therefore, the results of a FEM
analysis must always be critically examined by comparing them with the results of another
FEM analysis of a successfully completed project in similar ground conditions. Otherwise,
one is likely to fall victim to the simplest equation of them all: Garbage In = Garbage Out !!
3.13 References
1. Clough, R.W. (1960). The finite element method in plane stress analysis. Proc. Second
Conference on Electronic Computation, ASCE, Pittsburgh.
2. Courant, R. (1943). Variational methods for the solution of problems of equilibrium and
vibrations. Bulletin of American Mathematics Society, Vol.49.
3. Zienkiewicz, O.C. and Taylor, R.L. (1989). The Finite Element Method, Vol. 1, Basic
Formulation and Linear Problems, McGraw-Hill, London.

Centrifuge Modelling 1

4 Scaling laws and applications for centrifuge modelling
4.1 Introduction
4.1.1 Scaling laws
It is necessary to take the scaling laws into account before conducting any test series. The
scale effects need to be worked out as a basis for the planned modelling. The following
table shows the different scaling factors. A brief discussion of these has been given at the
end of chapter one. Some limitations to the modelling process and ways of solving some of
these problems arising from the associated errors are mentioned briefly in this chapter.

(model/prototype)
Acceleration m/s2 n
Linear dimension m 1/n

Stress kPa 1
Strain - 1
Density kg/m3 1
Mass or Volume kg or m3 1/n3

Unit weight N/m3 n
Force N 1/n2
Bending moment Nm 1/n3
Bending moment / unit width Nm/m 1/n2
Flexural stiffness/ unit width (EI/m) Nm2/m 1/n3
Tab. 4.1: Scaling laws
4.1.2 Scaling of time

As in fluid mechanics, it is not always possible to achieve correct scaling in all dimen-
sionless groups, and so choices must be made.
In dynamics, where acceleration in m/s2 scales as n in the model, and the linear dimension
is modelled at 1/n prototype, then time is modelled n times faster in the centrifuge.
But the scaling factor for modelling time in terms of diffusion may be demonstrated to be:
n2 faster in the centrifuge.
The non-dimensional time factor, Tv = f(time/depth2) = cvt/d2 , becomes independent of
gravity level for a depth of sample reduced to 1/n of the original, if the model time is also
reduced by 1/n2.
(1D Diffusion equation - saturated soil)
2 2
∂u ⁄ ∂t = c v ∂ u ⁄ ∂z
where u is excess pore pressure and time t scales with length z2 provided cv m = cv p .
Scaling laws and applications for centrifuge modelling 4-1

This offers a significant advantage because 27 years of prototype diffusion may be

modelled in 1 day using a centrifuge at 100 g, and is especially useful for environmental
problems or heat loss by conduction where diffusion is the main transport mechanism.
However, in offshore foundations or earthquake problems, the pore pressures are created
dynamically, with time scaling as: n times faster in the centrifuge and yet they decay in a
diffusive process where time is modelled as: n2 faster in the centrifuge.
Solution: use pore fluid in the model with a viscosity of n times that of the prototype (and
same density) or reduce the value of permeability of the soil (Attention: this will cause a
change in the properties).
(model/prototype)
Time: diffusion s n2
Time: inertia s n
Time: viscous s 1
Frequency 1/s n
Tab. 4.2: Scaling of time
4.2 Scale effects

The range and magnitude of possible shortcomings exposed by the scaling laws may be
described as scale effects. These must be reviewed to ensure that they will not affect the
outcome of the experiments and so the verification or modelling of models technique
may well be a useful way of checking this.
Typical scale effects are:
• non-uniform acceleration field (with depth (beam & drum) - & width of model (beam not
drum))
• particle size effects
• Coriolis acceleration
• boundary effects (due to being at small scale)

4.2.1 Stress distribution in centrifuge model: Depth

Since the inertial radial acceleration in the centrifuge is not linear with depth, but propor-
tional to centrifuge radius, the depth to radius ratio of the model is of major importance.
R r* r
z σv σv
aR
2/3rds
model under-
depth stress
δr
over-
stress
z error z
Model Vertical stress in Centrifuge Vertical stress in Prototype
Figure 4.1: Vertical stress distribution
2
ω 2 2
Equation 1 σ vm = ρ ------ ( r – r∗ ) σ vp = ρg ( nz ) ; ρ const with z.
2
density of soil =ρ radius to top of model = r∗

model gravity level = n g depth below surface = z = r - r*
In the centrifuge: since radial acceleration varies with radius the vertical stress at z = r - r∗ :
r 2 2 2
ρr ω r ρω 2 2
σ v = ∫ ρrω dr = ---------------
2
= ---------- ( r – r∗ ) r > r∗
2 r∗ 2
r∗
So distribution of vertical stress with depth in a centrifuge forms a parabola / quadratic.

Vertical stresses in the prototype are (nominally) linear.
We must select the best way of minimising the error... so we should set the
centrifuge (parabolic) and prototype (linear) stresses to be equal at some depth to
achieve this over the important section (depths) of the model.

where (Equation 1) => σvm = σvp at z = aR

Effective radius of model = R occurs at a depth below surface of model = a R
where total depth of model = 3aR/2
so stresses are equal at r = R where r∗ = R(1-a) and z = aR
σvp = σvm
2
ω 2 2 2
ρngaR = ρ ------ ( R – R ( 1 – a ) )
2
2
ω 2 2
= ρ ------ R ( 1 – 1 + 2a – a )
2
2
ω R
gn = ----------- ( 2 – a )
2
2
ω = 2ng ⁄ { ( 2 – a )R }
Check error (which will be close to the maximum understress) at z = aR / 2 :

where r = R - aR/2 r*=R(1- a)
2 2 2 2 2
σ vm = ρ ⋅ ω { R ⋅ ( 1 – a ⁄ 2 ) – R ⋅ ( 1 – a ) } ⁄ 2
2 2
ρ ⋅ n ⋅ g ⋅ R ⋅ {– a + a ⁄ 4 + 2 ⋅ a – a }
= ------------------------------------------------------------------------------------------------
(2 – a)
ρ ⋅ n ⋅ g ⋅ R ⋅ a ⋅ {1 – 3 ⋅ a ⁄ 4}
= --------------------------------------------------------------------------
(2 – a)
ρ ⋅ n ⋅ g ⋅ R ⋅ a ⋅ (4 – 3 ⋅ a)
= -----------------------------------------------------------------
4 ⋅ (2 – a)
a⋅R
Error at z = ⎛⎝ ------------⎞⎠
2
= ( σ vm – σ vp ) ⁄ σ vm
ρ ⋅ n ⋅ g ⋅ R ⋅ a ⋅ ( 4 – 3 ⋅ a )-
⎛ ---------------------------------------------------------------- ⎛ (------------------------
4 – 3 ⋅ a )-
– ρ ⋅ n ⋅ g ⋅ R ⋅ a ⁄ 2⎞⎠ ⎞
⎝ 4 ⋅ (2 – a) ⎝ 2 – a – 2⎠
= ------------------------------------------------------------------------------------------------------------------------ = ----------------------------------------
ρ ⋅ n ⋅ g ⋅ R ⋅ a ⋅ (4 – 3 ⋅ a) 4–3⋅a
----------------------------------------------------------------- --------------------
4 ⋅ (2 – a ) 2–a
(4 – 3 ⋅ a – 4 + 2 ⋅ a)
= ---------------------------------------------------- = – a ⁄ ( 4 – 3 ⋅ a )
4–3⋅a
understress

If you use the overall height of the soil for your model, taking the effective radius of the
centrifuge at z=2/3 rds is a good choice to minimize the error. Looking at the problems
relating to near surface (e.g. foundations or wave propitiation) other choices may be better.
Example: From the error which will occur at a depth of z = aR/2 in the model (see above) if
the effective radius R is taken at 2/3rds model depth, where z = aR, calculate the error for
a = 1/6, a = 1/8, a = 1/10, a = 1/12, a = 1/15. For R = 1m, calculate the total model depth.
Now calculate the overstress at z = 3aR/2 for the same range of values of a.
Answers a=1/6 a=1/8 = 1 / 10 a = 1 / 12 a = 1 / 15

z = aR / 2 -1 / 21 -1 / 29 -1 / 37 -1 / 45 -1 / 57
z=3aR/2 1 / 23 1 / 31 1 / 39 1 / 47 1 / 59
Model depth mm 250 182.5 150 125 100
Tab. 4.3: Example
4.2.2 Stress distribution in a centrifuge model
Width (Exaggerated scale)
1m
Figure 4.2: Stress distribution as a function of width in a centrifuge
States of equistress in the soil sample exist at common radii.

For a drum centrifuge, the surface of the model is at constant radius from the centrifuge
spindle, so this is not relevant.
However, soil models in the beam centrifuge (unless the ground surface is curved) will
have higher stresses at the package boundary. Likewise the ground water surface will also
be curved and higher at the boundary (and this will affect the total stress (but not the
effective stress) at depth) ⇒ e.g. + 20 kPa here. The larger the radius of the beam, the
lower is this effect. When investigating the response of a sheet pile wall, the structure
should be placed centrally in the mould.

4.2.3 Particle size effects

How can the centrifuge be used to model soil if the particles are not reduced in size by a
factor n?
• Model: clay particles of mean size d50 = 1µm. Prototype: sand at 100g with
100 x 0.001 mm = 0.1 mm particle size?
→ not sensible: clay and sand have different particle shapes etc. and hence they
will have different stress-strain-volume change characteristics.
• Model: sand particles of 0.1mm. Prototype: sand (with 1mm diameter) at 100g
with 100 x 0.1 = 10 mm particle size?
• not equivalent
D • OK provided soil grain size is not significant compared to model
dimensions and to the boundary effects
• D > 15 d50 (preferably > 30 d50)
d50
• Model: sand particles of 0.1mm. Prototype: gravel at 100g with 100 x 0.1 = 10mm
particle size?
→ correct modelling process, but check particle shapes (rounded/angular) and
particle hardness (crushing) which will affect dilatancy + also beware of any
boundary effects.
Conclusion
• model appropriate stress-strain-volume change characteristics
• particle size dimensions to be at least 1/15th but preferably >1/30th of the relevant
„model“ dimension
4.2.4 Coriolis acceleration

This may become relevant when particles are changing radius with some speed:
For example - building an embankment in-flight, but here the problem is really only relevant
to the positioning of the embankment in the hopper.
Hopper
??
Sand
2 r· ω
embankment
rω2
too far! Wooden
spacer
block
Clay
Figure 4.3: Positioning of an embankment in the hopper

·
2r ω
Consider error as a function of the nominal centrifugal acceleration: ---------- < 10%
ng
so for r = 4m, @ 100 g, then ω ~ 15 rad/sec and for error < 10 %: radial velocity r· must
be < 3.27 m/s
cf. angular velocity = r ω = 4 x 15 = 60 m/s (typically for example above), the range of the
Coriolis error may typically be ~5% in terms of radial velocity over angular velocity.
So, for an earthquake model which is subject to changes in velocity, additional terms will
become important.
·
horizontal shaking due to E/Q: x· ,x·· ( r ω term )
vertical shaking due to E/Q: y·· ( r·· term )
On the other hand, high velocity particles in blast loading are also less affected by error
because they will tend to move in straight (almost) lines, when the radius of curvature of
their particle trajectory rc ~ r (and r > 3m).
4.2.5 Boundary effects
q q
SAND
q
same
D
τ
σh’ >5D
CLAY
“stress bulb”
boundary
effects
z
Figure 4.4: ‚Plane strain‘ sand embankment on clay and pile installation or penetrometer penetration
NB the pile or penetrometer must be reduced in scale (where the diameter D will be approx.
10 mm and should be greater than 30 particle diameters) with a comparable size of the
"stress bulb" within which the value of q is increased. The load cell used to measure q will
'report' the hard layer 5-10 D below
• side friction
- consolidation
- soil movements from loading/unloading ⇒ DRAG
• stiffness of container (plane strain means ε2 = 0)
• base effects
• refraction and reflection of waves

Side friction
Interface Lubricant OCR adhesion, δ°
kaolin/kaolin Nil 1 18.8
“ “ 8 28.3
kaolin/perspex Nil 1 11.9
“ “ 8 18.8
“ Adsil spray* 1 6.3
“ “ 8 14.0
“ Silicone grease** 1 2.3
“ “ 8 5.1
*) transparent
**)most effective; not transparent
Tab. 4.4: Kaolin/perspex residual friction characteristics from shear box (Waggett, 1989)
Lubricants are applied to the strongbox walls (which have also been coated with a low
friction paint) to minimise side friction on the soil model. The principal stresses may then
become almost vertical / horizontal at the model boundary.
Similarly, latex sheets may be marked with a grid, greased and placed between a sand
embankment and strongbox walls or perspex face to reduce soil-wall adhesion to ~5° at
stress levels up to 300 kPa.
τ = σh' tan δ ∴ Allow for reduced "weight" in embankment.
Base effects
In-flight penetration of piles or penetrometers close to side boundaries or near the base of
the box or a stiffer layer will affect the data because the rigid surfaces will influence the
strain field. It is recommended that there should be 5 penetrometer or pile diameters to a
side wall. Likewise, for stiff soils, there should be 5 - 10 diameters below the pile tip to the
rigid layer.
Refraction and reflection of waves

Typically waves may be generated by blast loading or earthquakes, and these are reflected
or refracted at the boundaries. These boundaries can be rigid with vibration suppressing
materials or stacked ring (flexible) systems. Both have advantages and disadvantages.
However researchers are mainly interested in the first passage of the shock wave.
Example:
a) For a drum centrifuge with a radius of 1.1 m to the base wall (i.e. r = R + aR/2), cal-
culate the maximum depth of model which would limit the over or under stress to
+/- 5%.
b) Which other errors might be relevant and why?
c) If the centrifuge is able to achieve from 100 to 400 g, what range of rotation speeds
is required and what range of soil depths are possible?
d) How many years of diffusion may occur in 1 day (24 hours) in the drum centrifuge
during this same range of gravities?

a)
σv
errors +/- 5% ∴ a = 1/6 is o.k.
⇒ ex.: Radius to drum wall = 1.1 m
<+5%
R(1+a/2) = 1.1 m ∴ R(1+1/12) = 1.1 254 mm
⇒ R = 13.2/13 = 1.015 m centrifuge
Depth of model = 3aR/2 = (3/12)*1.015 m = 254 mm
prototype
z <+5%
b)
Other relevant errors in a drum:
• particle size
• possibly Coriolis acceleration from moving particles
• boundary effects
NOT: stress distribution due to width of model because top surface is curved
c)
Rω2 = 100 g = 1.015 ω2 ∴ω2 = (100*9.81/1.015)*(30/π)2
ω = 296.8 r.p.m.
Rω2 = 400 g = 1.015 ω2 ω = 593.7 r.p.m.
Depth: 100 g for 254 mm model depth ⇒ 25.4 m prototype depth
400 g for 254 mm model depth ⇒ 101.6 m prototype depth
(but reducing the scale by a factor of 400 can be very diffi-
cult to achieve physically in terms of size of structures etc.)
d)
Diffusion is modelled as 1/n2
prototype 1 day ⇒ 1 x 1002 days @ 100 g
⇒ 27.4 years
prototype 1 day ⇒ 1 x 4002 days @ 400 g
⇒ 438.4 years
(a) 254 mm: (c) 296.8 r.p.m., 25.4 m@ 100g: up to 600 r.p.m., 101.6 m @ 400g (d) 27.4 yrs:
438.4 yrs!

4.3 Scaling under earthquake conditions
t
E/Q σy
τxy
y
acceleration x·· 2/fp
freq.: fp σx
amp.: ap
– y·· = ng
1/fp
prototype
X -ap ap xp
body forces x
Y
base shaking
σy t
freq.: fm
τxy
amp.: am
σx 1/fm = 1/ (n fp)
model
xm
-ap /n = -am
Figure 4.5: Scaling under earthquake conditions
Plane stress conditions, with gravity and base shaking, so resolving orthogonally:
→ δσx / δx + δτxy / δy + X =0
↑ δσy / δy + δτxy / δx + Y =0
For sinusoidal base motions, vertically propagating horizontal shear waves give:
prototype: model:
x p = – a p ⋅ sin ( 2πf p t ) x m = – a m sin ( 2πf m t )
x· p = – a p 2πf p cos ( 2πf p t ) x· m = – a m 2πf m cos ( 2πf m t )
2 2
x·· = a ( 2πf ) sin ( 2πf t )
p p p p x·· = a ( 2πf ) sin ( 2πf t )
m m m m
y·· p = – g y·· m = – ng
In the prototype, the body forces are inertial where:
2
X = ρa p ( 2πf p ) sin ( 2πf p t )
Y = – ρg
So, for correct inertial modelling at 1/n scale:

• the model amplitude is 1/n times the embankment amplitude ∴ am = ap ⁄ n
• the max model velocity = max embankment velocity ∴ am fm = a p f p
cos (2πfmt) = cos (2πfpt) = 1 ∴ f m = nf p
• the model acceleration is n times the embankment acceleration ∴ y·· m = ny·· p
Scaling laws and applications for centrifuge modelling 4 - 10

so that:
→ δσx / δ(x/n) + δτxy / δ(y/n) + n X = 0
↑ δσy / δ(y/n) + δτxy / δ(x/n) + n Y = 0 ∴ n cancels
• same equilibrium equations in model ⇒ same stresses at homologous points in the
embankment
Referring to additional components of acceleration: y terms refer to vertical shaking
and x terms refer to lateral shaking
Example
A prototype has 10 cycles of a 1Hz earthquake (duration 10 secs) with amplitude 0.1 m.
Table 4.5 shows the implications of modelling this in the centrifuge with very small ampli-
tudes and extremely short durations due to the high frequency.
Model gravity [g] Frequency [Hz] Duration [s] Amplitude [mm]

100 100 0.1 1
200 200 0.05 0.5
Tab. 4.5: Modelling in a centrifuge
Example
For a 50 litre model subjected to an earthquake as described in table 4.5 at 100g, calculate
the equivalent prototype size and the peak accelerations in model and prototype. At what
angle should the model be tilted to achieve the same % lateral acceleration? Now calculate
the model details for a scaling factor of 50 on the prototype quoted above.
1g
n = 100g 100g
Vol.:
50,000 m3
0.402 g
40.2 g
θ
Figure 4.6: Model (left & centre) and equivalent prototype (right)
model: prototype:
fm = 100 Hz fp = (100/100) = 1 Hz
am = 1 mm ap = 1 x 100 mm = 0.1 m
Volm = 50 litres Volp = 50 x (100)3 litres
n = 100 = 50,000 m3

x m = – a m sin ( 2πf m t ) x p = – a p sin ( 2πf p t )
x· m = – a m 2πf m cos ( 2πf m t ) x· p = – a p 2πf p cos ( 2πf p t )

2 2
x·· m = +a m ( 2πf m ) sin ( 2πf m t ) x·· p = +a p ( 2πf p ) sin ( 2πf p t )
–3 2 2 m 2 2 m
∴ ( x·· m )max = 10 ⋅ ( 2π ) ⋅ 100 ⋅ 1 -----2 ∴ ( x·· p )max = 0.1 ⋅ ( 2π ) ⋅ 1 ⋅ 1 ----2-
s s
= 395 m/s2=40.2 g = 3.95 m/s2 = 0.402 g
θ = tan-1 (40.2 g/100 g)

= 21.9 °
∴ For a model at 1/50 scale of this prototype:

Volume = 50,000 m3 / 503 = 400 litres ⇒ 8 x larger!
Amplitude = ap/n = 100/50 mm = 2 mm
Frequency = fp×n = 1×50 Hz = 50 Hz@ 0.2 secs duration

Some typical examples of applications for centrifuge modelling

(mainly at Cambridge and Bochum)
Italics show projects completed in a small / large drum centrifuge: others are in a beam
centrifuge.
Dams and slope stability problems formed early topics of investigation in the centrifuge.
Padfield (1978) exposed mechanisms responsible for ongoing and unpredictable slides in
the Mississippi levees. Endicott (1971), Beasley (1973), Goodings (1979) and Horner
(1979) used 'real' soils in their slopes and they failed them vigorously. Ma (1994) studied
the stabilisation of a kaolin dam by means of vertical concrete ribs. Schofield conducted
extensive investigations into the Teton Dam failure with assistance from Avgherinos (1970).
Subsequently Davies (1981), Almeida (1984) and Barker (1998) constructed sand embank-
ments on clay to failure. Jewell (1980) revealed mechanisms of reinforced soil behaviour
that have led recently to publication as a UK Design Code. Sharma (1994) reinforced
embankments on soft clay with various ground improvement measures to enhance short
term 'undrained' stability.
Basic footing problems have also provided excellent benchmarks for comparison between
fundamental methods (clay and sand - undrained or drained) Lau (1988), Shi (1988), Tan
(1990). Bakir (1994) investigated the influence of foundations near to slopes, which led to a
new french design code. Cyclically loaded footings have been studied e.g. by Laue (1996)
or Bay Gress (2000). Nater (2004) studied the effects of layering of the ground on the
bearing bahaviour. Deep foundations have also offered an excellent source of problems
owing to the ability to model long piles at small scale. Ah Teck (1982) and Barton (1982)
investigated lateral pile group behaviour in sand. Grundhoff et al. (1998) used the
centrifuge to evaluate the moment distribution down into a pile due to the dynamic impact
lateral of loads. (e.g. car colliding with a bridge pier). Gui (1995) reviewed pile and
penetrometer installation in sand, and there have been many examples of piling carried out
for offshore applications in clays. Gurung (1985) carried out pull-out tests on transmission
line foundations.
Combining foundation problems with soil structure interaction for bridge abutments on clays
was first investigated by Springman (1989) and Sun (1989) with subsequent work from
Stewart (1993 - UWA), Bransby (1995) and Ellis (1997). Norrish (1996) created integral
bridge abutments in sand and then applied a range of lateral cyclic deck movements to
model temperature effects on backfill.
Powrie (1986) followed by Stewart (1989) conducted seminal work on the stability of
retaining walls in clay and the latter author extended this to nailing of clay excavations. Mak
(1984) loaded a strip behind a retaining wall in sand. Kusakabe (1982) and Phillips (1987)
investigated the stability of excavations, trench headings and shafts in clay. Reinforced
walls in sand were recently investigated by Balachandran (1996).
Thin-walled buried pipes and culverts (Britto, 1979) and uplift mechanisms associated with
pipelines (Ng & Springman, 1994; White, 2001) on and offshore have been modelled in the
centrifuge.
The offshore and coastal engineering industries have benefited more than most from
centrifuge model testing in the development of current design methods (Craig, 1988). They
are operating at such large scales, these can only be replicated at huge cost at full scale,
and so the centrifuge provides a thoroughly economic alternative.

Mechanisms observed, in clay, for gravity platforms (Rowe, 1975), anchors, rupture due to
collapse of abyssal plains (Stone, 1988), the installation and response to a variety of axial
and lateral forces as well as moments on a 3-legged jack up rig mounted on spud cans
(Tan, 1990, Tsukamoto, 1994, Wong et al., 1993) other spud can investigations (Craig and
Chua, 1991, Springman 1991), monotonic and cyclic loading in an axial sense on tension
piles (Nunez, 1989) and laterally on a variety of large tubular open ended piles (Hamilton et
al., 1991), suction piles and anchors (Fuglsang and Steensen-Bach, 1991; Renzi et al.,
1991), seabed mechanics and many other applications have been investigated using
centrifuges around the world.
An offshore jack up platform has a complex interaction between the rig, leg and spud
geometry, depth of embedment in, and characteristics of, the founding stratum, and the
dynamic wave and wind loading. To identify the basic controlling mechanisms, a single leg
with a central column of an equivalent, scaled, bending rigidity, instrumented to allow an
evaluation of the shear force and bending moment distribution is used. To mimic the lateral
load transfer from the latticework to the surrounding soil, plates with an equivalent cross
sectional area are mounted to fixings attached to the central shaft. For the general case of
a jack up platform, the soil-structure interaction is controlled by the combination of strength
and stiffness of the soft soil. By replicating a typical strength profile in the centrifuge model,
the mechanisms which lead to failure in the prototype will be reproduced
Tunnels have also been investigated very successfully in the centrifuge. Mair (1979, 1984)
inserted semi-cylindrical rubber membranes into clay and inflated them before he increased
gravity in a centrifuge. Subsequently he reduced the pressure and observed the defor-
mation above the tunnels and the failure mechanisms which formed for different cover-
depth-tunnel diameter ratios, for varying lengths of unsupported heading, for both 2 and 3
dimensional effects. Design stability charts were produced to aid estimates of ultimate limit
state and serviceability limit state analysis. Taylor (1984), and Taylor, Stallebrass and Grant
(1996) have continued this work in clay, following on from preliminary tunnel investigations
in sand by Potts (1976) and Atkinson (1977). More recently Bolton and Sharma (1996)
succeeded in dissolving polystyrene tunnels (with various ‘tunnel’ linings) in sand to create
and record surface deformations. König (1998) has built a tunnel drilling machine to follow
the influence of the excavation process inflight. Compensation grouting and hydro-fracture
in relation to limiting settlements above tunnels has also been achieved by Chin (1996) &
Lu (1997).
Environmental effects are ideally suited to modelling in a centrifuge. Time for diffusion
processes is increased by a factor of n2 and so 27 years of pollution migration may be
modelled in one centrifuge day at 100 g and nearly 700 years in one day at 500 g! Clearly
boundary effects become very important but there is potential to model transport processes
Hensley (1989), flow in multi-layered soils Boyce (1994), hydraulic, density and electrical
effects on transport processes Hellawell (1994), Potter (1996), contaminant migration
through intact or damaged liners Evans (1994), isokinetic clean up Penn (1997). Gronow
(1984), Edwards (1986) and Price (1996) have also used the two drum centrifuges at
Cambridge to model transport processes.
Hot thermal influences, e.g. sub-seabed disposal of high level radioactive waste, Savvidou
(1984) are also well modelled in the centrifuge. A heat source buried in a deep clay layer
generated pore pressure and caused cracking due to the differential thermal coefficients of
expansion of both the soil and fluid phases. Her analysis considered the coupling of heat
and fluid flow.

Poorooshasb (1988) fired projectiles of differing shapes (nominally containing such a radio-
active waste) into a soft clay layer and then heated the environment around which the
projectile came to rest. The primary question was whether a gap would be left behind the
projectile, creating a preferential leakage path or whether the soil would deform in a suffi-
ciently plastic fashion to close that gap. The effect of heat on this region was found to
increase the likelihood of leakage along this softened zone due to thermal cracking.
Cold regions engineering will have applications in Switzerland. Previous work at Cambridge
focused mainly on the influence of ice forces on offshore structures (Vinson, 1982; Lovell,
unpublished, Jeyatharan 1991) or iceberg scour above pipelines buried below the seabed
(Lach, 1992).
Experiments in which the ground has been frozen have been concentrated on mainly
onshore applications. Smith (1992) investigated the thaw-induced settlement of pipelines
and Vinson (1983) discussed the effect of hot fluids flowing through pipelines in frozen
ground.
There are few studies on the response to earthquake excitation for structures built on clay.
The difficulties of reconciling scaling laws for diffusion time and inertial time necessitate the
use of a more viscous pore fluid, which is complicated in fine grained clayey soils. However,
work has been carried out on basic dynamic behaviour (Morris, 1979), embankments
(Kutter, 1982; Lee, 1985), slopes and dams (Habibian, 1987; Pilgrim, 1993), retaining walls
(Steedman, 1984; Xeng 1990), piles (Maheetharan, 1990), towers (Madabhushi, 1991).
Most of these have been in sand with a replacement pore fluid with a viscosity 'n' greater
than water. This ensures that pore pressure dissipation rates are modelled at n times faster
than prototype time, which is the same time scale followed for the generation of pore
pressures in a dynamic sense.
The basis of the use of the centrifuge for dynamic problems such as e.g. wave propagation
have been studied by Siemer (1996).
The centrifuge is not suited to examining chemical effects or long term time
dependent processes such as creep.


5 Practical considerations: mechanical
5.1 Beam Centrifuges

5.1.1 Capacity
Of major importance in categorising centrifuges is:

• the acceleration level possible (ng), and
• the maximum payload (tonnes)
This is frequently cited as the g-ton capacity of a centrifuge.

Cambridge (UK): (1972) 0.9 t x 125 g = 112.5 gton at 4 m radius
ISMES (Italy): (1983) 0.4 t x 600 g = 240 gton at 2 m radius
Bochum Z1 (Germany): Krupp (1985) 2 t x 250 g = 500 gton at 4 m radius
Nantes (France): 2 t x 100 g = 200 gton at 5.5 m radius
UWA, Perth (Australia): (1989) 0.2 t x 200 g or = 40 gton at 1.8 m radius
City University, London (GB) 0.4 t x 100 g
Bochum Z2 (Germany): (1992)
Tab. 5.1: Geotechnical centrifuges (international)
0.4
payload t
0.2
100 200
acceleration g
Figure 5.1: Acutronic 661 (UWA, City and Bochum Z2)
Practical considerations: mechanical Page 5 - 1

5.1.2 Swing platform, package and liner

The soil model must be prepared in a container (or liner) and may need to be transferred to
a 'strongbox' which can carry the loads applied due to enhanced gravity. There are some
centrifuges that have their model integral with the beam system, but allowing model making
in a separate liner. However, most centrifuges have a swinging platform which provides
support to the base of a strongly reinforced box. This platform is usually in the form of a
swing and must be mounted on the centrifuge arm. This is ideal because the model hangs
always in the resultant direction of acceleration (e.g. vertical at 1g and horizontal at ng).
Figure 5.2: Arrangement of centrifuge arm during test (Cambridge)
Figure 5.3: Plan view of centrifuge arm with camera mounting to shoot vertically downwards

Beam balance
It is important for continued good health of the centrifuge to protect the bearings by
ensuring that the beam remains in balance. The weights applied to the other end of the
beam are designed to achieve this. Therefore, careful calculations of the mass of all items
on the package times their centroidal heights are required to find the total mass and mean
centroidal height. For example, the Cambridge beam is limited to difference in mass x
radius of +/-5 kgm between each end. More modern machines generally have an automatic
balancing system, based usually on a vibration measuring system.
Users must take into account movements occurring during the test - e.g. passing large
volumes of fluid from a storage tank to the model, pouring sand from a hopper, installing/
penetrating a structural device. If necessary, two calculations should be carried out to
establish the range of centroidal heights to check the mass x radius is acceptable, so that
the counterweight may be set between the two limiting cases.
Plumbing: water, air, other fluids

It is possible to pass water, air and other fluids (e.g. a 'model' pollutant) through the slip
rings on the central spindle to reach the centrifuge model. This is useful in setting up basic
groundwater conditions, steady state flows within a model, providing water or air into a
rubber bag to provide a normal load, providing a supply of 'pollutant' to a landfill for a short
period of time, pressurising one side of a hydraulic jack to 'actuate' a device (installing a
pile, simple monotonic load, opening the hopper etc.).
Actuators
All actuators designed for 'fitness for purpose':
• can apply or measure suitable range of loads with sufficient accuracy,
• where and how are they mounted on the package,
• must be as light as possible, but robust & manoeuvrable,
• think about the strength of soil to be tested to design
→ size of 'motor' or pressure ranges to apply load etc.,
→ size of load cell to measure strength accurately (i.e. range of linear response
derived from calibration).
• NOTE: must remember that fixings should allow for Newton's 3rd Law.......for every
action is an equal and opposite reaction!
Load (static, monotonic, cyclic, construction)

Are the actuators load controlled or displacement controlled???
static: pressure bags - made from latex rubber to suit required size & shape - LOAD
monotonic: single thrust actuators (maybe hydraulic rams - lattice leg) - LOAD
cyclic: electrically (lateral loading on piles) or mechanically (abutment movements at
integral bridges) induced - DISPLACEMENT
hydraulic: LOAD and DISPLACEMENT but be aware of change of stiffness of the whole
system
construction: use of self weight and hydraulic actuation for installation (piles, lattice leg) -
LOAD, or opening of ports at the base of the hopper (e.g. electrical solenoid) also LOAD

5.2 Drum Centrifuges

5.2.1 Capacity
Date of Drum dimensions Payload Max g Capacity Location

commiss m level
-ioning t g gton
Depth x Perimeter x Width
1971 0.025 x (0.25 x π) x 0.12 ~ 0.003m3 0.006° 1000* 6.1° UMIST, UK
(Schofield,1976)
1979 0.13 x (1.2 x π) x 0.23 ~ 0.11m3 0.2 650* 145 Davis, USA
(Fragaszy &
Cheney, 1981)
1986 0.1 x (0.8 x π) x 0.3 ~ 0.075m3 0.195° 150* 43° Utsonomiya,
Japan (Kusakabe
et al., 1988)
1988 0.15 x (2 x π) x 1 ~ 0.95m3 ~1.7 400 ~ 675 Cambridge, UK
(Dean et al.,
1990)
1995 0.115 x (0.74 x π) x .185 ~ 0.05m3 0.13° 416 ~ 50° Hiroshima,
Japanformerly
Cambridge
minidrum Mk I
(Kusakabe &
Gurung, 1997)
1995 0.12 x (0.74 x π) x 0.18 ~ 0.05m3 0.13° 400 ~ 50° Cambridge
minidrum Mk II
(Barker, 1998)
1996 0.17 x (1 x π) x 0.25 ~ 0.13m3 0.2 450 90 COOPE, Brazil†
(Gurung et al.,
1998)
1997 0.15 x (1.2 x π) x 0.3 ~ 0.23m3 0.6° 484 290° UWA, Perth‡
(Stewart et al.,
1998)
1998 0.2 x (2.2 x π) x 0.8 ~ 1.1m3 ~3.7 440 ~ 1600 Toyo, Japan
(Miyake &
Yanagihara,
1999)
1999 0.3 x (2.2 x π) x 0.7 ~ 1.45m3 2.0 440 880 ETHZ, CH
(Springman et
al., 2001)
Tab. 5.2: Drum centrifuge capacity
* horizontal axis
° estimated for better comparison with a fill of density of 2600 kg/m3 as no detailed data was
available
† a similar machine is working at the MIT
‡ an identical machine is in operation by Kiso Jiban Consultants in Tokyo and another with the
same size drum at the TIT Tokyo

The ETH drum allows the model to be made in the drum, which is mounted on the outer of
two shafts and rotated by a 37 kW motor. An inner shaft is connected to a separate tool
platform on which actuators can be mounted (e.g. excavate soil, load foundations etc.).
These shafts can be linked to rotate together or to be operated independently of each other,
in which case the inner shaft is powered by a separate motor. If the drum is still rotating but
the tool platform is stationary, a safety shield may be lowered and locked in place, while
adjustments are made to the tools mounted on the platform.
Figure 5.4: ETHZ Drum Centrifuge, Ø 2.2m, ETH Hönggerberg

Full field or small package configuration
Figure 5.5: Cambridge Geotechnical Drum Centrifuge
The options exist in all drum centrifuges to fill the entire drum ring to create a large deposit
(e.g. equivalent to 40 m deep, 1 km long, 80 m wide in a minidrum or 100 m deep, 2.5 km
long, 250 m wide in the ETHZ drum), or to place a smaller model within a 'box' (e.g. equiv-
alent to 40 m deep, 60 m long, 60m wide or 10 m deep, 15 m long, 15 m wide). If the latter
was provided with a perspex face, then it would be possible to take lateral photographs of
the cross section with depth.

Balance
Balance is also a concern on the drum to prevent it from stressing bearings unequally and
to cause the drum shell to go out of shape. The maximum out of balance of the ETHZ drum
that is allowed will be 10 kgm at 440 g or 600 r.p.m. and proportionally greater as the gravity
level reduces. Accelerometer transducers have been installed to measure vibrations and to
help ensure that the drum remains within the balance limits. There is a similar limit to the
maximum out of balance, which is allowed on the tool table (1 kgm at 440 g).
(a) (b)
Figure 5.6: Tool table with horizontal actuators (a) and piecocone placed in the actuator (d =11.3mm) (b)
Plumbing: water, air, other

This is very similar to requirements for the beam centrifuge, but it is also necessary to
supply water/air to both:
• the central console or tool plate
• the drum ring
There are two porous toroidal pipes inside the drum, which distribute the water supplied
from above the centrifuge via an open channel on top of the drum. A series of switches link
to a standpipe and a set of valves below the drum to control the water flow or to maintain
the water at a specific level within the drum.
Actuators
These are similar in complexity to beam actuators but there is much more space to mount
them and for them to act since they are generally mounted on the tool plate (Fig. 5.6). It is
intended that:
• they should be able to rotate with the drum ring and hence carry out operations on the
model,
• disengage and become stationary so that activity could take place within a shielded
region around the tool plate,
• the spindle could move at differing speeds to the drum ring
→ either to locate to a new position relative to the drum or
→ to plough a furrow for example....or tow a drag anchor
• there are fewer drums, so more opportunity for development!

5.3 Site investigation devices (penetrometers, vane)

The main questions are:
• what mode of actuation is required?
• what speed of testing - given the centrifuge scaling laws?
• what are the geotechnical aspects of what we are measuring....
5.3.1 Vane:
• this device is for clay only
• it is necessary to drive the vane down to the required depth and then rotate it
• speed of rotation should be about 72°/min
• we are finding su,vane (torque is measured & su,vane deduced)
• separate the shaft friction, peak friction and residual values: a 15° slip coupling is used
5.3.2 Penetrometer:
• continuous penetration
• displacement rates between 0.6 - 12 mm/min
• consider whether this is into soft clay, stiff clay, dense sand?
• motor/actuation must be able to achieve this axial push
• for sand, need a stronger, stiffer load cell than for clay
• diameter likely to be 10 mm rather than area of 10 cm2 at full scale
• measuring tip resistance qc and sometimes the pore pressure u
Figure 5.7: Vane Figure 5.8: Penetrometer
5.3.3 Cylindrical T-Bar:

• continuous penetration
• displacement rates between 0.6 - 100 mm/min
• only for soft soils (clay)
• measuring the total force to deduce su
• interpretation based on plasticity calculation of soil around a cylinder (and shearing past
the end faces)

5.4 Post-test investigation devices

5.4.1 Photographic: camera & flash
Both still photographs and video films are useful in revealing soil deformation mechanisms.
These require a solid base on which to mount the cameras and sufficient light. Tiny video
cameras (15 mm diameter, 50 mm long) are now available with adequate definition to
record marker positions to the necessary degree of accuracy. They may also be able to
pass the signals by radio waves. When combined with novel methods of 'grabbing' video
frames or digitising these directly, it is possible to use these in preference to still cameras.
5.4.2 Digital Images and PIV analysis

Image capture using an inexpensive 2-megapixel digital camera provides a significant
increase in resolution and image stability compared with video. Particle Image Velocimetry
(PIV) is a velocity-measuring technique in which patches of texture are tracked through an
image sequence. Image processing algorithms are available to apply the PIV principle to
images of soil. The software has a precision of 1/15th of a pixel when tracking the
movement of natural sand or textured clay. The system allows displacements to be
measured to a precision greater than with still photographs or video films without installing
intrusive target markers in the soil.
5.4.3 X-ray
It is common practice to insert mixtures of lead, water and soluble oil into a clay stratum or
lead balls into a sand stratum, at the model making stage. Before testing at higher gravities,
exposure of a suitable film to X-rays reveals the initial location of these lead markers.
Subsequently, the soil deformations caused by the loading sequences will allow post-test
examination of these internal movements by identical radiographic techniques. Clearly this
requires a radiographic device with sufficient power to penetrate soil of certain thicknesses,
the aluminium liner and sometimes the strongbox. It also requires sufficient shielding to
comply with local safety requirements.


6 Practical considerations: geotechnical
6.1 Introduction
What happens to soil as gravity increases in a centrifuge model?
σv , σv’
100 mm 10 m
u u
short long long
z
Model: 1g Model: 100g
σv = 1.5 σv’ = 11.5 u = -10 kPa σv = 150 σv’ = 50 u = 100 kPa LONG TERM
σv = 150 σv’ = 11.5 u = 138.5 kPa SHORT TERM
Figure 6.1: Stress distribution with depth
Following model making, there are often

suctions present in a clay (maybe up to -30
to -40 kPa). This causes extra effective
stress in the soil. Once stresses have been
increased to those caused by 100 gravities,
the total stress increases together with the
pore pressure. Subsequent dissipation to a
steady state, long term, stress condition is a
function of the non-dimensional group
designated by Tv:
Tv = cv t /d2
where cv is coefficient of consolidation, t is
the time and d is the drainage path. Dissi-
pation will often take between 2 - 20 hours
in the centrifuge at ng depending on the
thickness of the clay layer and whether the
clay can drain in one or two directions (d or Time (sec)
2d) and the permeability of the soil. Figure 6.2: Time since start of initial acceleration
of the centrifuge
The model must be in equilibrium before further perturbation commences.
Practical considerations: geotechnical Page 6 - 1

6.2 Design of soil model: real or laboratory

We aim:
• to describe a variable matrix of rock and soil in terms of zones and layers, and if possible
• appropriate properties are designed to suit the "prototype" soil conditions and to allow
measurement and analysis of performance under some loading regime.
Selection of soil for a centrifuge model:
• 'real' or ‘natural’ prototype soil with less classification data and knowledge of parameters
and their likely ranges
• specific laboratory soil - known parameters, classification details.
For clay deposits:
• undisturbed samples of ‘real’ or ‘natural’ prototype soils, local structure is maintained,
and naturally occurring preferential seepage paths remain (unsmeared),
• remoulded natural soil,
• most appropriate remoulded 'laboratory' soil (i.e. kaolin) with an extensive database of
properties.
For cohesionless deposits:
• disturbed, re-poured and compacted real soil, or
• known laboratory soil:
→ poured and compacted to a specific relative density
→ with particle size selected as function of some controlling dimension
→ this could be single size, uniform or well or gap graded.
su
Y o.c.
X
n.c.
Depth
Figure 6.3: Variation of shear strength with depth and influence of stress history
In the field:
• if field samples are taken from an overconsolidated deposit, they should be sampled
from the weakest stratum (lowest value of su ; point X for o.c. clays, point Y for n.c. clays)
• subsequent stress fields imposed in the centrifuge may lead to the best approximation of
the key field strength conditions, but in consequence, the strains will not be entirely
comparable.

Comparison between the use of real or laboratory clayey soils in (beam)

centrifuge model tests
Activity Field prototype soil Laboratory alternative soil

Aim To reproduce the exact soil behaviour from the To model the relevant prototype soil behaviour
field in the centrifuge model. using a known laboratory soil.
Sampling Inserting a cutter to extract a block sample, Mixed from slurry under vacuum, consolidated in
preparation for transport and subsequent liner/consolidometer before transfer of liner
and
trimming to fit a centrifuge strongbox, is likely containing the soil to the centrifuge strongbox.
disturbance to cause softening and loss of peak strength Sample is assumed to be uniform and
(extent depending on OCR of the soil). homogeneous in stress state and properties.
Disturbance will be significant. Disturbance is not likely to be a problem.
Local If 'fabric' exists, it is difficult to quantify & is No local cementation exists; critical states may
likely to confuse back analysis of model test. be adopted to explain soil behaviour.
structure
Aging Effect of local cementation, particle shape The repetition of 1g - ng loading cycles appears
variations, load cycling: soils will be disturbed to cause a slight increase in strength. Further
during sampling, and are difficult to model or research is underway at the moment.
interpret.
Particle For soil deposits which contain random large The size of any structure/probe acting on
particles, it should be remembered that these granular soil particles of mean diameter, d should
size
will be enlarged by a factor of n with respect to be > 15 d. The same field and model particle
the model dimensions. Therefore, shell size/void ratio/pore fluid will not affect
fragments in remoulded calcareous deposits permeability. Time taken for dissipation of excess
assume a reinforcing role out of proportion to 2
pore pressures (diffusion) will be n faster.
the rest of the soil layer, and should be
Sometimes pore fluid with n times higher
removed. Root systems in naturally occurring
viscosity may be used so that both diffusion /
clays may become major tunnel networks.
inertial velocity will be factored by n.
Permea- Some natural clays (e.g. montmorillonite) take Relatively permeable clays such as kaolin may
lengthy consolidation time at n g. be used to minimise consolidation time.
bility
Homo- Unless the sample is remoulded, it is difficult to This is completely controlled in the centrifuge
assess this until after the test. sample.
geneity
Hetero- The field sample is more likely to be 1d consolidation: no radial strain at sample
anisotropic. boundaries; stress field assumed uniform.
geneity
Properties Exhaustive sets of laboratory tests may be Extensive databases exist for. Classification and
necessary to establish soil parameters for extra laboratory tests using appropriate stress
design of test/subsequent back analysis. paths may be carried out.
Stress Undisturbed samples will retain the stress This may be designed to suit the requirement of
history and sampling effects at the nominal strength and stiffness with depth, remembering
history
depth of sampling +/- 200 mm. The centrifuge that the sample is consolidated at 1g under
model will reproduce stress history for n times uniform total stress. The effective stress profile
the depth of real samples, from ground surface may be manipulated by using upward /
downwards. The prototype may have been downward hydraulic gradients to create a linear
subjected to differing stress histories and pore pressure profile.
paths.
Global Samples taken from one/several number Realistic variation of some properties (e.g. OCR)
depths do not represent the global variation of with depth, but less realism in others (e.g.
structure
OCR, strength, stiffness with depth. anisotropy).
Results Distortion of global effects due to samples not Global effects modelled better except soil
representing fully the in-situ soils. properties may different from field soil.
Conclusion Direct scaling from model to prototype possible in rare instances only. Usually, model data
needs to be understood in an analytical framework, and then applied to the field situation.
Tab. 6.1: Comparison between ‘real/natural/field’ prototype soil and a laboratory alternative

Design of stress history

Similar arguments may be employed when considering whether to use natural or
remoulded soils in the drum centrifuge (Züst, 2000; Fauchère, 2000).
• future soil behaviour is a function of past stress history, recent & anticipated stress path,
• we can create a centrifuge model in the laboratory to control both stress history/stress
path due to 1D consolidation prior to model making and reconsolidation in the centrifuge,
• it is possible to design soil deposits to exhibit a chosen range of strength and stiffness,
although there are some anisotropic details which cannot be modelled effectively.
How do we design strength profile?

• decide strength profile required => su
• consider link between su & σv'
• design stress history to achieve required profile of σv'
undrained shear strength su [kPa]
Figure 6.4: Undrained shear strength versus depth, shown as f(g-level); for stress history see
Figure 6.10

What strength?
If:
• current effective vertical stress = σv'
• past maximum effective vertical stress in one dimensional normal compression
= σv,max’ = σvc'
• so that overconsolidation ratio OCR = σv,max' / σv'
su / σv' = a OCR b (1)
where a and b are constants: see Table 6.2 for data for speswhite kaolin. (For OCR = 1,
s u = a σv ' )
Researcher Date a b Comment

Nunez 1989 0.22 0.62 based on vane test
Phillips 1987 0.19 0.67 “
Springman 1989 0.22 0.71 “
Tab. 6.2: Speswhite kaolin
Figure 6.5: Correction factor, µ
(Bjerrum, 1973)
Does vane shear strength vary when compared against test methods or classification
parameters?
Strength ratio su /σv' or su /σv,max' (in Fig. 6.5) is dependent on plasticity index Ip for vane:
correct via factor µ to calculate su for design:
where: su design = µ su vane (2)
su design = µ σv' a OCR b (3)
so µ = 1, Ip = 20 % & µ = 0.9, Ip = 32 % (kaolin).
Vane data for the jack up 'lattice leg' sample are

compared with the expected values and in-flight
tests at 100 g (uncorr. by Bjerrum's factor µ). A
reasonable approximation has been achieved for
su,vane peak over the top 200 mm of kaolin,
assuming that average su,vane is represented by
Figure 6.6: Expected and deduced values of
the mid-depth su,vane (see Fig. 6.6). s u

Also: vane prediction may not relate to the actual su

mobilised in triaxial compression (TXC)/extension
(TXE), simple shear, plane strain compression/
extension etc. (see Fig. 6.7).
How does su from major laboratory tests vary with
vane shear test data ?
Laboratory and field tests (Bjerrum, 1973)
Figure 6.7: Strength ratio su/σvc' =

su/σv,max' versus plasticity
index Ip for vane tests
Equivalent su in triaxial compression (txc) / extension (txe) for normally consolidated clays,
together with vane data corrected for strain rate (Bjerrum, 1973) shows su,txc ~ 50% greater
than su,vane,corr, su,txe ~ 33% smaller then su,vane,corr (for Ip = 32%).
Results in simple shear, on normally consolidated speswhite kaolin (Ip = 32%) at constant
volume gave φ' = 21.8° (Airey, 1984) and su/σv,max' = 0.18, whereas Al Tabbaa (1984)
found φtxc' = 23° when OCR = 2.

6.3 Kaolin as a model soil

Notes: φtxc', φtxe', φpsp', φpsa', are values of φcrit' for triaxial (tx) compression (c),
extension (e) and plane strain (ps) passive (p), active (a) respectively.
Clay Method of Ip φtxc‘ (°) φtxe' (°) Source
consolidation or or
%
φpsa‘ φpsp’
Kaolin Isotropic, tx 20 24 24 Yong & McKyes (1971)
Kaolin Isotropic, tx 25 29.2 36 Broms & Casbarian (1965)
Spestone kaolin Isotropic, tx 32 22.6 20.5 Nadarajah (1973)
Spestone kaolin K0 , tx 32 20.8 28.0 Nadarajah (1973)
Spestone kaolin K0 , ps 32 20.9 21.7 Sketchley (1973)
Tab. 6.3: Properties of kaolin clays: frictional strength (Airey, 1984)
Clay Method of Ip (su/σv,max‘) (su/σv,max‘) Source

consolidation
% txc or psa txe or psp
Kaolin Isotropic, tx 25 0.43 0.34 Broms & Casbarian (1965)
Spestone kaolin Isotropic, tx 32 0.215 0.205 Nadarajah (1973)
Spestone kaolin K0 , tx 32 0.205 0.175 Nadarajah (1973)
Spestone kaolin K0 , ps 32 0.20 0.16 Sketchley (1973)
Spestone kaolin K0 , ps 32 0.30 0.18 Ladd et al. (1977)
Tab. 6.4: Properties of kaolin clays: undrained strength (Airey, 1984)
Since 1978, speswhite kaolin has been supplied for laboratory testing in place of spestone
kaolin, and Mair (1979) commented that both clays exhibited similar soil properties.
So given we can design a soil strength profile.... how do we fix the stress history?
• using preconsolidation at 1 g in the laboratory (e.g. as below for σvc’=100 kPa), there are
at least three possible methods:
→ soil in a liner under a hydraulic press ⇒ almost constant σv' with depth,
→ can apply a hydraulic gradient to allow σv' to vary with depth,
→ allow partial consolidation to create a stiffer crust (& base) as a second phase
1 g consolidation,
• allowing consolidation in the centrifuge
→ either after preconsolidation at 1 g, so some of sample will usually be o.c.,
→ from slurry, so sample will be entirely n.c.,
→ changing g-level for short periods of time,
→ adding and removing a surcharge,
→ applying a hydraulic gradient.

Consolidation at 1g, Centrifuge test at 100g

Vertical effective stress [kPa] Overconsolidation ratio Undrained shear strength [kPa]
100 1 2 10 0 10 20
at 1g
Overconsolidated
200 mm 200 mm
20 m 20 m
σv,max’
Normally consolidated
Depth Depth
Figure 6.8: Consolidated @ 1g, test @ 100g
Figure 6.9: (a) plane strain 1g consolidometer, (b) axisymmetric 1g consolidometer/downward hydraulic gdt.

What is consolidation with downward hydraulic gradient?
Figure 6.10: Development of stress history at 1g
su
Figure 6.11: Variation of su,vane and OCR with depth at 100g

Example: kaolin layer, 350 mm deep:

• initial σv,max' = 110 kPa after drainage to an excess pore pressure of zero,
• if we consolidated at 100 g in soil with γ' = 6 kN/m3 => σv' = 6 x 0.35 x 100 = 210 kPa at
the base of the sample: 0.35 x 100 = 35 m depth in the prototype, but before this,
• a 2nd stage consolidation may follow: σv is increased to 225 kPa, and
→ a seal in the circumference of top disk press allows additional pore water
pressure to be created, so downward hydraulic gradient with u = 165 kPa at
the top of sample, decreasing to 0 at the base,
• σv = u + σv' : σv' profile is assumed to be linear: 225 - 165 = 60 kPa at the surface and
225 kPa at the base,
• OCR profile is then given for equilibrium at 100g, together with the ideal profile of
su,vane ,
• σv' exceeds previous σv,max' below model depths of 170 mm (17 m prototype; point X in
Fig. 6.11),
• below this, OCR = 1, predicted profile of su,vane is linear with depth.
Example: Using the stress history from the example above and equation (1), verify
the undrained strength profile (su,vane) shown in Figure 6.4. What values of a and b
appear to have been used from the Table 6.2?

How can we create a stiffer crust?

• carry out first phase 1g consolidation at 100 kPa,
• apply second phase 1g consolidation at 200 kPa for a short period of time,
• parabolic isochrones indicate extent of excess pore pressure dissipation and show
increasing depth penetration from drainage boundaries with time, = (12 cv t )0.5 (shown
below in Fig. 6.12 for drainage in both directions).
• so total value of σv,max' increases with time at shallow depths,
• so OCR much higher above 100mmm / 10mp,
• and strength increases with time allowed for 2nd phase consolidation,
• more extreme values of su,vane in the crust are achieved with much greater temporary
values of σv,max'.
2nd phase consolidation => stiffer crust
100 200 1 2 40 0 10 20
1st at 1g
Stiffer crust
2nd at 2g Overconsolidated
200 mm 200 mm
20 m 20 m
σv,max’
Normally consolidated
Depth
Depth at 100g
Figure 6.12: Two phase consolidation (1st phase 100 kPa @ 1g, 2nd phase 200 kPa @2g)
What happens if slurry is consolidated directly in the centrifuge?
Normal consolidation in the centrifuge

Vertical effective stress [kPa] Undrained shear strength [kPa]
0 10 20
all depths normally consolidated
20 m 20 m
at 100g σv,max’
Depth Depth
Figure 6.13: Consolidation from slurry in the centrifuge

The g-level could also be increased for a period of time although this would tend to increase
the maximum effective stress at the base of the sample first (theory of parabolic
isochrones). If full consolidation (dissipation of excess pore pressure) was allowed then the
gradient of the lines describing dσ' v, max ⁄ dz or dsu/dz will double but remain triangular in
shape. Drainage of these excess pore pressures will develop first at the base.
An alternative in the ETHZ drum centrifuge will be to place temporary surcharge of
∆σv = γdhSand (e.g. sand) over the consolidating clay layer (Fig. 6.14 & Fig. 6.15a). Either
partial or full dissipation of excess pore pressures could be permitted prior to removal of this
surface layer (with a scraper tool) depending upon desired undrained shear strength.
assume ∆σv OCR 0 10 20

1
const. over
clay layer ∆σv = γdhSand
20 m 20 m
at 100g σv,max’
Depth n.c.
of clay Depth
Figure 6.14: Consolidation from slurry in the centrifuge

A hydraulic gradient could also be applied in the drum centrifuge by ponding water on top of
the clay layer and ensuring there was no pore pressure in the sand layer at the base of the
clay (Fig. 6.15b & Fig. 6.16).
hSand zw
z z
base sand base sand

layer layer
drain open:
clay surcharge u = 0 kPa clay water
layer γd layer
(a) (b)
Figure 6.15: Drum centrifuge test setup

(a) Sand surcharge
(b) Downward hydraulic gradient
Total vertical stress [kPa] Pore pressure [kPa] Effective vertical stress [kPa]
γwzw
γwzw
= + σ'v,max
Depth
of clay at 100g at 100g
z
γwzw
Figure 6.16: Application of a downward hydraulic gradient
These techiques are very much under development currently. Successful attempts have
been made in the Cambridge mini drum centrifuge to date.
On of the major questions remains the installation of pore pressure transducers and how
these are
- protected to ensure the ceramic filters remain saturated and
- that they are located at the expected depths.
How do we prepare our centrifuge models in the laboratory?

clay mixing
• using reconstituted laboratory soils, clay powder is mixed into a slurry with w > wL (e.g.
the Cambridge University Soils Group used to mix kaolin at w = 120%, University of
Bochum mix kaolin at w = 70-80% but for a longer period) under a vacuum for a period of
not less than 2 hours (so that the mixture is homogeneous/fully saturated),
• the slurry is then placed inside the liner/consolidometer or strongbox, trying to ensure no
air bubbles are included,
• porous strips and filter papers, saturated with de-ionised water are located above and
below the sample.
N.B. ETHZ are currently developing their own methods for application to clay placement in
the drum centrifuge.
NOTE: add rock flour to kaolin to change stress ratios at failure, φ'crit , permeability,
stiffness etc.
consolidation process in strongbox

• consolidation pressure will be applied to a rigid plate in contact with drainage layers / soil
deposit,
• pressure is regulated in stages => σv' on the consolidating deposit,
• cavitation effects due to suctions (negative values of u) from unloading from σvc' may be
minimised:
→ by using stress decrements ∆σv < 100 kPa with all drains open,
→ allowing for equilibration of the pore pressures at each stage,
→ possible air entry at the edges of the model is limited by permitting controlled
swelling,
• allow for any subsequent consolidation settlement or swelling in planning installation
depth of pore pressure transducers (PPTs), which are inserted between last 2 loading
phases via temporary unload,
• holes obtained following PPT insertion are backfilled with slurry injected through a
syringe to ensure that any air will be excluded from the soil matrix (procedure specified
by Phillips and Gui, 1992),
• transducer wires are taken out through special ports in the side of the 850 mm diameter
tubular strongboxes or pressed into the rear of the plane strain soil sample, with the
wires passing through a small slot in the piston,
• any possible reinforcing effect of the transducers and their cables must be considered,
• consolidation process in drum centrifuge.
How to calculate how much clay is required at model-making stage?

e.g. For fully saturated speswhite kaolin, with specific gravity γs = 2.61 × γw (Airey, 1984),
the following relationship may be used to calculate void ratio e following consolidation
under σ’v,max :
e = 2.767 - 0.26 ln (0.8 σ'v,max ) (Critical State Soil Mechanics) (4)
and this can be checked by taking samples for moisture content w determination:
e = (γs / γw)× w = 2.61 w (5)

The height of the sample is directly proportional to specific volume v = 1+e, the initial
volume (and height) of clay slurry may be determined.
h σ' v, m ax = h w=120% ( 1 + e σ' v, max ) ⁄ ( 1 + e w=120% ) (6)
Saturated unit weight may also be calculated:
γsat = ((γs / γw) + Sr e) γw / (1 + e) = (2.61 + e) 10 / (1 + e) (7)
Example: Calculate the compression expected in a fully saturated speswhite kaolin slurry
(model offshore soil in the jack up 'lattice leg' test) placed at a nominal w = 120% & consol-
idated to equilibrium σ’v,max = 110 kPa, with samples expected to show w ~ 60%. We need
a post-1g-consolidation model height of 300 mm.
e = 2.767 - 0.26 ln (0.8 x 110) = 1.603
w = 1.603 / 2.61 = 61.4%
ew=120% = 2.61 x 1.2 = 3.132
hw=120%= 4.132 x 300 / 2.603 = 476 mm
γsat = (2.61 + 1.603) 10 / 2.603 = 16.2 kN/m3
γ' = 6.2 kN/m3
Aim for at least 500 mm to allow for trimming/error in calculation, so a reduction in height to
~ 60% of original slurry was allowed to ensure sufficient depth of clay following consoli-
dation at 1g. (N.B. In a drum centrifuge, maximum depth of clay would be about 200 mm, so
clay will need to be placed slowly over a period of time to allow some consolidation to take
place because depth of the drum is only 300 mm).
Example: What depth of speswhite kaolin slurry (γs = 2.61×γw) to be placed at w = 120% in
the drum centrifuge so that after consolidating under a surcharge of ∆σv = 100 kPa at 100 g
there is a post-1g-consolidation model height of 60 mm. Calculate the respective weights of
water & kaolin needed to mix the slurry if the drum has dimensions 2.2 m (diameter) x
700 mm (width) x 300 mm (depth)? After consolidation under ∆σv = 100 kPa what is w at
mid-depth of the 60 mm layer and hence what is the saturated unit weight (assume initial
value)?
3
γ = 20 kN/m , ∆σv = 100 kPa
50 mm
100 kPa
30 mm x 100
Clay =3m
h
assume γ = 16.4 kN/m3
γ·z
Depth at 100g
of clay
z
Figure 6.17: Example (@ 100g)
At depth z = 30mm · 100 = 3m

e = 2.767 - 0.26 ln (0.8 x (16.4 x 3 + 100)) = 1.524

w = e / (γs / γw) = e / 2.61 = 58.4%
ew=120% = 2.61 x 1.2 = 3.132
hw=120%= 4.132 x 60 / (1+1.524) = 98.2 mm
γsat = (2.61 + 1.524) 10 /(1+1.524) = 16.4 kN/m3
also calculate values at 100 mm (1 m prototype) and 50 mm (5 m prototype):
1m 5m
e = 2.767 - 0.26 ln (0.8 x (16.4 x z + 100)) 1.588 1.472
w = e/2.61 60.8% 56.4%
hw=120%= 4.132 x 60 / (1+e) 95.8mm 100.3mm
γsat = (2.61 + e) 10 /(1+e) 16.2kN/m3 16.5kN/m3
Answer: 100 mm clay will be necessary for placement at w = 120% although some
“wastage” should be allowed for. Assume minimum of 120 mm clay depth.
Volume:
Tot: = π/4 · (2.22 - 1.962) · 0.7 = 0.549 m3
Water : Clay = 1.2 : 1
γsat,w=120% = 13.9 kN/m3
∴ Clay: 0.25 m3 → 347 kg (i.e. 14 x 25 kg bags)
∴ Water: 0.3 m3 → 30 l
Consolidation: time
Non-dimensional consolidation time factor Tv = cv t / h2 where cv is the vertical coefficient
of consolidation, t is time and h is length of the drainage path (for 2 way drainage, model
depth = 2h).
Clay Range of σv‘ e cv kv kv pred Source

kPa mm2/s 10-6 10-6
+ mm/s
= current σvc‘ mm/s Eqn (8)
Speswhite kaolin 256+- 450+ - 0.3 0.72 - Bransby (1993)
450+- 120 1.21 0.57 0.34 0.9

+
450 -60 - 0.58 0.35 -
Speswhite kaolin 100 - 200+
+ 1.30 0.18 0.95 1.17 Ellis (1993)
Speswhite kaolin 54+-91+ 1.54 0.25 2.87 2.03 Sharma (1993)
Speswhite kaolin 43+- 86 1.54 0.27 2.06 2.03 Springman (1989)

Tab. 6.5: Consolidation data derived from centrifuge models in a large 1 g consolidometer

Al Tabbaa (1987) quotes values of kv and horizontal permeability kh (from independent

falling head and consolidation tests) for speswhite kaolin with respect to void ratio e:
3.25 –6
k v = 0.5 ⋅ e ⋅ 10 mm/s (8)
2.09 –6
k h = 1.43 ⋅ e ⋅ 10 mm/s (9)
for normally and overconsolidated states with 0.98 < e < 2.2.
Data given above shows that values of kv,pred by equation (8) are:
• up to 3 times larger than those measured in the 1 g consolidometer for unloading incre-
ments, and
• in quite good agreement for virgin consolidation.
Example: If cv = 10-7 m2/s for unloading - reloading conditions, for a clay depth of 60 mm,
with top and bottom drainage, 90% consolidation.
Tv 90 = 0.848 and t90 = 0.848 x (0.06/2)2/10-7 = 2.1 hours;
Example: For a clay depth of 350 mm, draining to top and bottom of sample, σv' > 86 kPa
in the lower half of the model where the sample would be normally consolidated, calculate
time for 90% consolidation to occur. Adopt a conservative value of cv. How would you
reduce this time for consolidation?
Between 26-40 hrs: depends on cv selected (2.7 to 1.8*10-7 m2/s): place a thin sand layer
across most of the clay at mid-depth to aid dissipation of pore pressures. Link into the
drainage system at the top/bottom of clay to facilitate drainage: reduces consolidation time
by a factor of 4 (to ~ 7-10 hours).
Stage 1 of 2 stage consolidation process => mid-depth sand layer in a deep

model in clay (Nunez, 1989)
• drainage pipe to middle sand layer fixed in tub,

• bottom drain placed,
• slurry poured and first clay layer consolidated to
σv' = 90 kPa.
Figure 6.18: Stage 1

Stage 2
• pressure released and swelling allowed,
• middle sand layer poured with gaps to allow for pile /
penetrometer tests,
• filter fitted to top of drainage pipe,
• filter paper placed on sand.
Stage 3
• slurry poured and first and second layers consolidated
to σv' = 260 kPa,
• sand layer to settle onto filter and drainage pipe,
• final clay surface to be cut flush with the top of the tub.
A similar method could be adopted in the drum centrifuge (although placing filter paper
would be a bit complicated and fiddly)!

Inflight downward hydraulic gradient with intermediate drain
Figure 6.21: Inflight downward hydraulic gradient with intermediate drain
Clay: final model preparation

Having consolidated the clay sample in a strongbox, some final preparation is required prior
to loading on the beam:
• close drains, unload press, then the liner is removed from consolidometer,
• clay excavated/cut to size,
• deformation markers added (internal and external),
• structures, instrumentation and site investigation equipment installed,
• final assembly of strongbox and installation on swing,
• installation of swing on centrifuge arm,
• connection of all fluid/air supplies, electrical/electronic equipment.
Clearly this procedure is slightly different in the drum but the principles are the same.
How does this modelmaking procedure vary for granular materials?
Sand: placement
• dry sand will be placed to a uniform relative density D or (ID) by pouring from a hopper at
a specific height with a uniform flow rate,
• pouring may be halted at any time to allow placement of deformation markers, spots or
lines of coloured sand, for subsequent displacement or strain determination,
• known weight of sand poured into a specific volume allows determination of e, and
hence D (or ID),
• the sand will be saturated either by upward flow (but with low differential head to avoid
piping) or by using a vacuum under a restraining sheet,

• then, as for a clay model: structures, instrumentation & site investigation equipment
installed........etc.
•
Properties: strength
The peak angle of friction φ'max and the critical state angle of friction φ'crit may be based on
D (ID) as follows (Bolton, 1986):
ID = (emax - e) / (emax - emin) x 100% (D is f(n)) (10)
where:
emax = maximum void ratio
emin = minimum void ratio
φ'max = φ' crit + A Ir (11)
where:
A = 3 for triaxial and 5 for plane strain, and
Ir = ID (Q - ln p') - 1 (12)
where:
Q is f(particle crushing strength), (usually given as 10 for quartz,
i.e. crushing strength > 20 MPa),
p' is mean effective stress in kPa.
Note: Most quartzitic sands have φ'crit = 32 - 33°
(Dilatancy is f(φ'max - φ'crit))
Triaxial data @ failure
Figure 6.22: Dilatancy for quartz sands: φ'max - φ'crit v. p’ (Bolton, 1986)
This shows more easily how these equations may be used to determine possible dilatancy.

Stiffness
Initial stiffness Gmax may be found from
• lab testing,
• by empirical relationships, e.g.
Gmax = 50’000 (p‘)0.5 / (1 + e)3 (kPa)
Permeability
This can be found from simple lab permeability tests on soil samples prepared at equivalent
D (ID).
6.4 Preparation of soil samples in the DRUM centrifuge

Many of the techniques described for beam models are used on the drum centrifuge,
however generally modelmaking is done in the centrifuge under ng.
• sand may be sprayed from a nozzle near the central console or onto a spinning disk
which moves vertically to place a uniform sample across the vertical height of the drum,
• saturation is possible by varying the groundwater levels, using an elevating standpipe
driven by a small motor,
• clay samples may be placed as a slurry in the same manner (although there are
concerns about achieving full saturation), and sand may be placed on top to achieve
overconsolidation,
• downward hydraulic gradient consolidation is also possible by raising the water table
above the clay and draining to zero pore pressure at the base of the sample, or
• block samples may be prepared separately and placed in the drum at a later stage
following preliminary placement of a sand base.
N.B.: More will be added to this chapter as experience with the Zurich drum centrifuge
develops ...

(1) A Sand delivery pipe can move up and down

B Air blows sand out through nozzle onto cylindrical surface
C Berm at base of drum wall assists stability of layer later
(2) D Water delivery pipe is at base of drum
E Water table in sand layer after hydration
(3) F Water now discharged through pipe
G Water table in sand layer after de-hydration
(4) H Cutting frame attached to central column
I Drum stationary @ 1g. Central column rotates. Blade trims surface, moved
outwards as cutting process; debris collected from base
Figure 6.23: Placement of a sand sample in a drum centrifuge (Dean et al., 1990)

(1) A Upper buoyancy foam blocks fixed to drum

B Lower buoyancy foam blocks holding drain pipeworks
C Base drain for clay layer
D Water drains from clay as self-weight consolidation occurs
(2) E Valve opened when sand layer reaches level of second drain
(3) F Water drains from sand; clay continues to consolidate under higher load of
dry sand
(4) G Dry sand falls away when drum stopped
H Solid clay layer has apparent cohesion and remains standing @ 1g
I Rotating blade trims clay surface @ 1g
Figure 6.23: Placement of a clay sample in a drum centrifuge (Dean et al., 1990)


7 In-situ testing, instrumentation, data acquisition
7.1 Measurement of soil properties

7.1.1 Vane shear testing to determine su at discrete locations
In the field su,vane for a homogeneous clay varies according to:
• the vane aspect ratio L / D (should be reduced in the centrifuge),
• the rotation rate of the vane ω.
Peak and residual torque T, may be measured from a torsion load cell (which is calibrated
under torsional load) mounted on the vane shaft (above soil penetration level), and causes
a nominal failure surface of a cylinder of diameter D, and height L (and also potentially
round the shaft):
T = π su,vane ( D3 + 3D2 L) / 6 => peak and residual su,vane, (13)
This is derived from the shear surface x su x lever arm (integrating an annulus on top and
bottom surfaces and direct calculation for the cylinder).
Torque T
rotation speed ω su
dz
Measure T, ω & dz
L/Dcentrifuge ~0.77 (14/18)
L/Dfield ~2 (130/65)
su,vane
L
slow fast
ω
D
speed of rotation
Figure 7.1: Shear vane geometry and influence of speed of rotation
Modelling principles
• su,vane is averaged over 3 surfaces: 1 cylindrical (vertical) and 2 horizontal discs
• su,vane is also averaged over depth if su is not constant (OK if dsu / dz is linear),
• So if L/Dcentrifuge is 14/18 cf. L/Dfield = 130/65 (dimension mm) then:
→ more emphasis is placed on su in the horizontal plane cf. su (vertical) - (i.e.
horizontal slip planes),
→ can conduct more vane tests per depth of clay model (> 1 vane depth not
tested between each shearing event at a specific depth),
→ the effect of the shaft diameter: vane diameter ratio is reduced (relative shaft/
vane resistance is smaller and disturbance from soil displacement due to
insertion will be relatively smaller),
In-situ testing, instrumentation, data acquisition Page 7 - 1

• the shaft is greased and is fitted with a 15o slip coupling to separate the component of
friction from that due to the shear around the vane,
• at fast ω, viscous effects would cause the measured su,vane to increase,
• at very slow ω, significant consolidation occurs, causing higher values of su,vane,
• optimal rotation speed gives lowest value of su,vane.
Scaling effects:
• consider testing procedures with respect
to difference in scaling with respect to
time,
• time relative to the strain rate will be
identical in both cases,
• time taken for dissipation of u following
insertion of the vane (by diffusion), scales
as n2 faster,
• the centrifuge vane data will tend to
indicate higher strengths (greater
drainage) for the same rotation speeds in
prototype and model; therefore increase
model vane rotation speed.
Figure 7.2: Vane shear test data
Operation:
• vane is driven vertically (6-12V electric motor) at between 2 - 6 mm/min,
• a linear potentiometer reveals when depth for the next test has been reached,
• a (5V) rotary motor is then engaged, and shearing is begun at n g after 1 minute,
• in the field, this delay is usually 5 minutes:
→ for radial drainage at the vane circumference (cv t / (D/2)2 )m = (cv t / (D/2)2 )p,
→ for cv equivalent in both models, tm = 5 x 60 / (65/18)2 = 23 seconds,
• centrifuge model clay will have consolidated more during this pause of 1 minute, and
higher strengths would be anticipated,
• surface water should be prevented from entering the vane 'bore',
• minimum su,vane for the 14 x 18 mm vane was achieved in kaolin at 72°/min at 100 g,
• these data should be reviewed again - as a function of new equipment, building new soil
database etc.
Output:
• relationship between su vane , σv' & OCR (Eqn (1)) as used in design,
• must allow for Bjerrum's factor µ (Eqn (2)) when converting su,vane to design strength
for analysis.

7.1.2 Cone penetration testing (CPT): to determine a profile of su

Modelling principles:
• the cone penetrometer presents different modelling difficulties & advantages (some
relative to vane):
→ the tests are quicker to perform and
→ a continuous profile of soil resistance is obtained, and
• since content of most centrifuge models is generally known,
→ can be used to check consistency of each test sample and to compare with in-
flight vane test data,
• but extrapolation from empiricism at full scale to model scale prediction leads to some
uncertainty........
Scaling effects: geometry

• modelling a prototype 10 cm2 diameter field cone at 100 g implies a centrifuge probe of
0.36 mm diameter! - impossible to manufacture a working device, let alone strain gauge
a load cell,
• and so we model (effectively) a pile installation,
• it is general European practice to use a standard cone of diameter 10-11.6 mm, with a tip
load cell and usually total resistance from tip and shaft together,
• Cambridge have a piezocone of 12.7 mm diameter, with a porous sintered stone at the
tip of a 60o cone, and a rosette load cell located at the top of the shaft, but isolated from
the friction exerted on the shaft.
• the ETH cone has a diameter of 11.3 mm allowing separate measurement of the tip
resistance and total resistance from tip and shaft together.
Scaling effects: time

• faster insertion rates are likely to be subject to viscous effects,
• whereas slow penetrations will allow significant dissipation of pore pressures,
• penetration rates of between 3 and 26 mm/s have been investigated in clay, and these
were found to fall in the intermediate range, with less than 13% difference in qc (tip
resistance).
Output:
• for clay, soil strength:
→ f ( qc tip resistance, field cone factor Nc)
→ f (OCR, φcrit', u @ cone shoulder, location of u measurement, what is the
most appropriate penetration rate?):
→ Nc = (qc - σv ) / su (14)
→ can plot su / σv ' to reveal OCR,
→ pore pressure acting on shoulder of a piezocone should be allowed for,

→ depth should be corrected (gravitational field depth effect error) to zpc from zm
→ zpc = n zm (1 + (zm / 2Rs )) (15)
where zm is depth of cone below soil surface, and n is gravity level at

centrifuge radius
Rs at the surface of the soil.
Figure 7.3: Tip resistance profile in kaolin clay (Gui, 1995)
• for sand, tip resistance: (N.B. pc = crushing strength)

→ f (D, z/B (depth: cone dia), B/d (cone dia: particle dia), OCR, σv'/pc , φcrit')
→ Nq = (qc - σv ) / σv', (16)
→ using Equations (11) & (12),
→ p' = (σv' qc) 0.5, (17)
→ also allows determination of φ'max
→ normalised tip resistance, Q = Nq - 1 (18)
→ normalised depth, zm / B = Z (19)

Mechanism Relative density effect (Note: Id = ID)
ID effect
Figure 7.4: Mechanisms and influence of relative density (Gui, 1995)

Operation:
• drive continuously between 3-26 mm/s
in sand, or 3-12 mm/s in clay,
• dimensional limitations > 5 B (5 cone
diameters) from the location of another
important section of the model or from
the edge of the strongbox,
• the proximity of the base of the
strongbox or a stiffer granular founding
stratum will increase the measured load
for between 5 - 10 B above that
boundary,
• these interactions may be less
noticeable in soft clays.
Figure 7.5: Penetration rate and OCR effect

(Gui, 1995)
Grain size effect
Figure 7.6: Grain size effect on LB sand (Gui, 1995)
Figure 7.7: Grain size effect on LB sand (Gui, 1995)

Stress level effect Dilatancy & mobilised friction
Figure 7.8: Influence of dilatancy (Gui, 1995)
European cooperation => unified testing methods:

• recommendations regarding the operational and cone design standards have been
reviewed by a group of European Universities under an EC contract (Bolton et al., 1999),
• better agreement was achieved in sand samples (Fig. 7.18a) than for clay (Fig. 7.18b),
but there were some differences in sample preparation, strongbox shape and size, cone
diameters and radius of centrifuges.
a) b)
Figure 7.9: Cone penetration data from various European centrifuge laboratories (Renzi et al., 1994)
a) sand
b) clay

Nonetheless, the agreement confirms that the CPT is the main method of site investigation
in the centrifuge and that it can be comparable, within an Institution, and between many
Institutions, provided testing methods are similar.
7.1.3 T-Bar penetration testing: to determine su
Figure 7.10: Schematic diagram Figure 7.11: Variation of Nb with surface roughness
(Stewart and Randolph, 1991) (Stewart and Randolph, 1991)
Modelling principles:
• a semi-rough cross bar, of length approximately 5 bar diameters db (db = 7 mm), with
smooth ends, mounted on a shaft instrumented to read axial load (Fig. 7.9), and
• a classical plasticity solution in plane strain (Randolph & Houlsby, 1984) => the force
required to pull or push the cylinder through a rigid-perfectly plastic medium of shear
strength su (graphically shown in Fig. 7.10),
• a fundamental relationship is advantageous in finding su from a miniature SI device.
Scaling effects:
• the bar diameter must be small so that dsu /dzprototype is insignificant, and can be
assumed to be constant in the near field to the bar, and
• the end effects of pushing or pulling the bar vertically adjacent to the soil surface or an
interface with a stiffer layer must also be considered,
• time/rate effects should also be investigated as for penetrometer and vane tests.
Output:
• bar factor Nb obtained from measuring force / unit length P on the cylinder where
surface roughness is equal to α su and adhesion factor α = 0 (smooth bar), and α = 1
(rough bar mobilising friction up to su )
Nb = P/(sudb) (20)
• comparisons with su predicted from in-flight cone penetrometer, post-test vane shear
and independent triaxial tests (Stewart and Randolph, 1991) were fair - the bar gave an
su profile:

→ which was very similar to the triaxial prediction for normally & lightly overcon-
solidated samples,
→ about 20% greater for OCR > 3,
→ vane shear results are expected to be < equivalent values at 100g (ingress of
water during deceleration / prior to inserting the vane) - dependent on time
after deceleration & clay cv .
Figure 7.12: 1 ≤ OCR < 3 Figure 7.13: OCR > 3
(Stewart and Randolph, 1991)
Comparisons between su obtained from in-flight bar and cone penetrometer tests, post-test
shear vane tests and predictions based on triaxial data are given in Figures 7.10 - 7.13
(Stewart and Randolph, 1991).

7.2 Measurement of displacement

7.2.1 Spotchasing
• insertion of black marker 'bullets'
=> evaluation of relative
movements at various stages in
the loading programme by back
analysis of high quality photo-
graphs taken in-flight,
• these differential movements =>
appropriate values for input into
a strain evaluation computer
program to estimate contours of
shear strain, directions of lines
of no-extension etc.,
• in some cases it has been
possible to derive sufficient
information for use in the
analysis of geostructural mecha-
nisms (Bolton and Powrie, 1988;
Sun, 1989),
Figure 7.14: Vectors of outward horizontal movement post
• e.g. for installation of an offshore installation of spudcan (after Phillips, 1990)
jack up rig near to piles which
are intended to support an
operational platform (Phillips,
1990), vectors of outward
horizontal movement may be
obtained.
7.2.2 Digital Images and PIV analysis

Image capture using an inexpensive 2-
megapixel digital camera provides a signif-
icant increase in resolution and image
stability compared with video. Particle Image
Velocimetry (PIV) is a velocity-measuring
technique in which patches of texture are
tracked through an image sequence. Image
processing algorithms are available to apply
the PIV principle to images of soil. The
software has a precision of 1/15th of a pixel
when tracking the movement of natural sand
or textured clay. The system allows displace-
ments to be measured to a precision greater Figure 7.15: Measurement of deformation in a
than with still photographs or video films triaxial test (White et al., 2001)
without installing intrusive target markers in
the soil.

a) Plane strain calibration chamber for pile testing b) Selected node paths during pile penetration
Figure 7.16: Application: pile penetration in a calibration chamber (White et al., 2001)
7.2.3 Displacement measurements

Sometimes it is helpful to measure post-test displacements - often this should be done
immediately after the test while the model is still on/in the centrifuge so that it is not
disturbed. This might be particularly relevant for:
• the shape of an embankment or
• backfill behind an abutment wall, or
• the failed shape of a retaining wall or
• the collapse of an anchored wall in sand after an earthquake etc......
7.2.4 Radiography
Information may be obtained from post-centrifuge model test radiographs:
• of the location of rupture zones (via detection of diagonal lead threads) or
Figure 7.17: Forced subsidence of a model abyssal plain; dashes represent ruptures (Stone, 1988)

• of movement of soil between piles due to surcharge loading (Springman, 1989),
Figure 7.18: Internal soil-pile interaction for a row of Figure 7.19: Internal soil-pile interaction for a pile
free-headed piles (Springman, 1989) group (Springman, 1989)
7.2.5 Excavation
Much useful information may be obtained from careful excavation. In particular, it is
necessary to determine the actual position of the pore pressure transducers (allowing for
post test swelling or desiccation) so that back analysis of the pore pressure test data makes
sense!
An alternative to using lead threads is to create a small hole and to insert a piece of noodle
or spaghetti. This softens on contact with water. The noodle can be exposed and its position
measured (approximately) during excavation.
7.3 Electronic/electrical instrumentation

• should be small and rugged,
• instrumentation calibration ideally should be linear but always repeatable,
• beware of creating reinforcement or preferential drainage paths with cables,
• maybe bought off-the-shelf or custom-made.
Internal
Common types of instrumentation inserted in a soil matrix are:
• pore pressure transducers (clay),
• total pressure/stress cells (Garnier, 1999),
• for contamination problems, resistivity probes (Hensley, 1989; Hellawell, 1993),
• for dynamic applications (e.g. earthquake studies), accelerometers (Kutter, 1982),
• indicators of displacement for exposure to X-rays.
• matrixes to measure deformation ⇒ stress distributions (e.g. Tekscan; Springman et
al.,2002, Laue et al. 2002)

Instrument Input Voltage Amplify/ mV or V Comments

Filter output
PPT 0 to -5V 10x mV actually better
LVDT -5V to 5V - / Filtered V Linear over 60-80% of
range, typically +/- 15
mm
LP 0 to 10V - V 300 mm range
LC 0 to 2V (or 3V, 5V) ~100x mV
TC -12V to 12V - mV separate J-Box
ACC 10V charge 7 pico separate J-Box, convert
amplifiers Coulombs/g pressure change on
1000x => V piezoceramic crystal to
charge
pH sensor
2 or 4 electrode 1V square wave @ yes mV electrical conduction
Resistivity probe 100Hz approx
Tab. 7.1: Instrumentation: Input, output, errors, signal processing
External
Other forms of measurement are mounted externally:
• thermocouples (TC), thermistors or digital thermometers (and CCTV),
• strain gauge bridges (as bending moment transducers - BMT),
• load cells (LC),
• displacement measurement,
→ linear variable differential transformer (LVDT) say 30 mm range,
→ linear potentiometer (LP) say 300 mm range,
→ laser profilometer,
→ external measurement,
→ wave height gauges.
Air/pore pressure transducers (PT/PPT)

• in future, PPTs may be made in the form of minuscule chips, interrogated by a remote
trigger so that the difference in the measured and the true pore pressure (that would
have existed in the soil in the absence of the PPT) due to interference between the soil,
the instrumentation cabling and transducer will be reduced significantly,
• in the meantime, care is necessary when designing and installing PPT layouts,
• air pressure transducers (PT) can be subjected to several bar of pressure and are inter-
posed on air supply lines, mainly for actuator control.

• Pore pressure transducers (PPTs)

measure various magnitudes of pore
pressures (typically, a 350 kPa range).
The company DRUCK sells the
smallest ones used as standard trans-
ducers in centrifuge.
• cylindrical in shape, 6.4 mm in external
diameter, with a main body 13 mm
long,
• a porous stone protects the load cell Figure 7.20: Diagram of a Pore Pressure Transducer
face so that pore pressure and not total
stress is measured,
• the porous stone should be de-aired,
the PPT should be inserted prior to the
addition of the final loading increment
in the consolidometer, having
unloaded the clay, in stages of 100 kPa
or less, to zero excess load at 1 g,
Resistivity probes
• a typical 4 electrode resistivity probe
shown with calibration test data,
• electrical conduction occurs through
pore fluid,
• this is affected by soil porosity and
chemical composition,
• but the only variation tends to be the
chemistry,
• signals are generally multiplexed.
Figure 7.21: Diagram of a resistivity probe
Fibre optic pH sensor

• used for electrokinetic remediation,
• non-electrical method required,
• light passed from LED, via fibre-optic cable, through liquid, filter, ball lens & a non-bleed
pH paper,
• transmitted light collected by second fibre-optic cable and passed to a silicon photo-
diode,
• a logarithmic amplifier enhances the output signal,
• absorbance measured and according Beer's Law => concentration can be determined.

Overview of system Schematic design of the sensor
Fibre-optic pH sensor
Ball lens coupler
Single channel
yellow
sensor photodiode log fibre - light in
l.e.d. fibre - light out
non-bleed pH paper
Figure 7.22: Schematic of system and sensor (Lynch, 1998)

CUED Environmental Geotechnics
Experimental arrangement for Sodium hydroxide pollution

plume detection plume in sand
Light in Light out

pollutant
optical fibres
sensor sand
porous base
Strain Gauge Bridges and Load Cells

Strain gauge bridges: using properties of changing electrical resistance due to strain of thin
foil elements, suitably attached to host material - connected together in electrical circuits to
measure:
• shear
• normal load
• torsion
• bending moment

A
R4
R1
D V E
B
R3
R2
C
Figure 7.24: Wheatstone Bridge
Constant voltage energisation:

in ABC, current i = E / (R1 + R2)
in ADC, current i= E / (R3 + R4)
V(B) - V(D) = i(AB) R1 - i(AD) R4 = V
For R1 = R2 = R3 = R4 = R, V = 0
For R1 = R + dR1, R2 = R + dR2 etc., V may be calculated
Gauges positioned according to what force is to be measured.
Mode of operation:
Lengthen straight fine wire, length by d , resistance changes
Initially
R = ρ /A
where A = πr2, ρ is specific resistance, then
dR/R = δρ/ρ + δ / - 2δr/r
for cylindrical wire under uniaxial stress, σ, the longitudinal tensile strain, εl ,
εl = δ / = σ/E
and radial strain, εr
εr = δr/r = -νσ/E = -νδ /
giving
dR/R = δρ/ρ + (1 + 2ν) δ /
and if
δ/ = ∆ δ/
then resistance change due to straining:
dR/R = [ ∆ + (1 + 2ν)] δ / = k δ /
• generally thin foil, only 4µm thick is used - to increase electrical resistance and to speed
up heat dissipation with relatively large surface area to reduce glue stresses,

• backing is usually 25µm of epoxy sheet,

• strain sensitivity of a metal foil gauge is typically k = 2.1,
• (but semiconductors have more sensitivity and therefore require a constant current
energisation to eliminate non-linearities & temperature effects due to changes in power
consumption).
Two types of strain gauge bridges:
Fully active:
e.g. R1 & R3 in tension, +dR
R2 & R4 in compression, -dR
V = E dR/R = Eεk
1 active set of gauges, one dummy set of gauges:

• due to problems of space
• other gauges likely to be in junction box
• must be at same temperature
• even though gauges can be temperature compensating
e.g. R2 & R4 in compression, -dR
R1 & R3 are unstrained, dR = 0
V = E dR/2R (i.e. half output of fully active bridge)
Power Dissipation:
i2 R in each arm of active bridge
material must be capable of dissipating this without affecting accuracy of gauge
Typical requirements for high accuracy require power density between
0.78 - 1.6 x 10-3 Watts/mm2
e.g. (E/2R)2 x R with E = 10V then for 2.5V per gauge, R = 350W,
power = 6.25/(4 x 350) = 0.0045 Watts,
so gauge area needs to be
4.5/1.6 ~ 3 mm2
7.3.1 Other considerations:

Temperature effects:
• use temperature compensating gauges, + equivalent coeff. of expansion between
gauge / material,
• use Wheatstone bridge to eliminate resistance changes due to temperature changes,
• use similar lengths of supply leads etc.

Aging of glue:
• creep with load and time
• curing is a vital part of process
Generally Araldite epoxy strain cement is used by the Cambridge Group.
Waterproofing:
• necessary for gauges & leads in 'wet' environment,
• various coatings & shrinkfit tubings may be used,
• submersion in silicone oil is another alternative.
Recording Equipment:
high input impedance ~ 5000 MW.
Load Cells:
• Cambridge contact stress transducers 1961 -
present,
• designed to measure:
⇒ magnitude & direction of shear & normal
stresses,
• thin metal webs instrumented with strain gauges,
• generally aiming at:
⇒ around 2000 micro strain maximum,
⇒ to ensure linear elastic response,
⇒ appropriate stiffness,
⇒ metal easily machineable,
• usually use heat treatable aluminium alloy HE15W, Figure 7.25: „Stroud“ load cell
• Wheatstone bridges give a useful output to such small resistance changes,
• locked in hysteresis/machining stresses should be relieved by strain cycling.
Calibration:
• apply normal, shear and eccentric loading independently to set up a loading matrix:
N
[dV] = [a] M
S
• invert to give N, M and S as a function of the total change in voltage of each strain gauge
circuit,
• loads should be applied repeatably & consistently on face of cell in same way as testing
situation,
• supply voltage should be stable,
• at least 5 readings should be taken on a load-unload cycle to maximum level expected in
tests.

Effect of transducer stiffness:

• transference of stress around a deflecting piston is a function of transducer/soil (kc/ks)
stiffness,
• but is transducer measuring true state of stress in ground?
kc/ks > 100 to give accuracy > 99%: typical Cambridge cells have 50-100 MN/m stiffness.
Accuracy:
• Errors may be caused by many factors:
→ recording device
→ calibration constants
→ hysteresis & non-linearity
→ random drift, temperature & input/output voltage
→ effects of loads other than those cell is designed to measure.
The largest error is generally due to drift.
Data acquisition systems

• multichannel - ? multiplex ?,
• 2 orders of magnitude faster logging than in real time,
• many types of tests with a wide variety of transducer,
• e.g. 2 main requirements,
• steady state logging e.g. 50 channels 0.01Hz for 2 days,
• dynamic events logging e.g. 16 channels 10 kHz for a few seconds,
• need redundancy in data acquisition - test expensive if data is all lost,
• disk (hard, CD or zip),
• (magnetic),
• (tapestreamer),
• manual,
• better to have a modular system for fault tracing.
7.4 Summary
Centrifuge model testing is particularly advantageous in investigations of the performance
of large scale structures. Appropriate idealisation may be adopted to reveal the key mecha-
nisms of behaviour.
It is possible to create centrifuge models in clay using 'laboratory' soils such as kaolin,
according to a prescribed design strength profile or coarse-grained „cohesionless“ soils
with the relevant relative density.
A combination of site investigation devices, empirical interpretations and comparisons
between in-situ and laboratory data allow determination of soil strength.
Both displacement and failure mechanisms may be observed, leading firstly to an under-
standing of the problem under investigation, secondly to an evaluation of strains in the soil
matrix and thirdly to data about the mean soil strength at failure.

7.5 References
1. Airey, D.W. 1984. Clays in the simple shear apparatus. Cambridge University PhD
thesis.
2. Almeida, M.S.S. 1984. Stage constructed embankments on soft clay. Cambridge
University PhD Thesis.
3. Al Tabbaa, A. 1984. Anisotropy of clay. Cambridge University MPhil thesis.
4. Al Tabbaa, A. 1987. Permeability and stress-strain response of Speswhite Kaolin.
Cambridge University PhD thesis.
5. Bakir, N., Garnier, J., Canepa, Y., 1994: Etude sur modèles Centrifugeuse de la
capacite portantes de fondation superficielle, Revue Francais de Géotechnique 25, pp
5-14.
6. Bay - Gress, C., 2000. Étude de l`interaction sol - structure - comportement non
lineaire sol-fondation superficielles. Thesis pour obtenire le docteur, Louis Pasteur/
ENAIS, Strasbourg.
7. Bjerrum, L. 1973. Problems of soil mechanics and construction on soft clays and
structurally unstable soils. Proc. 8th ICSMFE Moscow, Vol. 3, pp. 111-159.
8. Blight, G.E. 1968. A note on field vane testing of silty soils. Can. Geot. J. Vol. 5, No. 3,
pp. 142-149.
9. Bolton, M.D. 1986. Dilatancy of soils. Géotechnique 36, No. 1, pp. 65-78.
10. Bolton, M.D. 1991. Geotechnical stress analysis for bridge abutment design. TRRL
CR270.
11. Bolton, M.D. and Powrie, W. 1988. Behaviour of diaphragm walls in clay prior to
collapse. Géotechnique 38, No. 2, pp.167-189.
12. Bolton, M.D., Gui, M.W. and Phillips, R. 1993. Review of miniature soil probes for
model tests. 11th SEAGC, Singapore. Preprint.
13. Bolton, M.D., Gui, M.W., Garnier, J., Corte, J.F., Bagge, G., Laue, J., Renzi, R. 1999.
Centrifuge cone penetration tests in sand. Géotechnique 49, No. 4, pp. 543-552.
14. Bransby, M.F. 1993. Centrifuge test investigation of the buttonhole foundation
technique. Data Report. 33p.
15. Bransby, M.F. 1995. Piled foundations adjacent to surcharge loads. PhD thesis.
16. Broms, B.B. and Casbarian, A.O. 1965. Effects of rotations of the principal stress
axes and of the immediate principal stress on the shear strength. Proc. 6th ICSMFE,
Montreal, Vol. 1, pp. 179-183.
17. Cheah, H. 1981. Site investigation techniques for laboratory soils models. Cambridge
University MPhil thesis.
18. Corte, J.-F., Garnier, J., Cottineau, L.M. and Rault, G. 1991. Determination of model
properties in the centrifuge. Centrifuge '91, H.Y.Ko and F.G.McLean (eds). Balkema.
pp. 607-614.
19. Craig, W.H. 1988. Centrifuge models in marine and coastal engineering. Centrifuges in
Soil Mechanics, W.H. Craig, R.G. James and A.N. Schofield (eds). Balkema. pp. 149-
168.

20. Craig, W.H. and Chua, K. 1991. Large displacement performance of jack-up spud
cans. Centrifuge '91, H.Y.Ko and F.G.McLean (eds). Balkema. pp. 139-144.
21. Dean, E.T.R., James, R.G. and Schofield, A.N. 1990. Drum centrifuge studies for
EEPUK. Contract No. EP-022R TASK ORDER 1-022. Phase 1. Draft Report.
22. Ellis, E.A. 1993. Lateral loading of bridge abutment piles due to soil movement.
Cambridge University 1st year Report. 27p.
23. Endicott, L.J. 1970. Centrifugal testing of soil models. Cambridge University PhD
thesis.
24. Fuglsang, L.D. and Steensen-Bach, J.O. 1991. Breakout resistance of suction piles
in clay. Centrifuge '91, H.Y.Ko and F.G.McLean (eds). Balkema. pp. 153-160.
25. Garnier, J., Ternet, O., Cottineau, L.-M., and Brown, C.J. 1999. Placement of
embedded pressure cells.
26. Grundhoff, T., Latotzke, J. & Laue, J. (1998). Investigations of vertical piles under
horizontal impact. Proc. Int. Conf. Centrifuge 1998, Kimura et al. (eds). Balkema,
Rotterdam. pp. 569-574.
27. Gui, M.W. 1995. Centrifuge and mechanical modelling of pile and penetrometer in
sand. Cambridge Univ. PhD thesis.
28. Hamilton, J.M., Phillips, R., Dunnavant, T.W. and Murff, J.D. 1991. Centrifuge study
of laterally loaded pile behavior in clay. Centrifuge '91, H.Y.Ko & F.G.McLean (eds).
Balkema. pp. 285-293.
29. Hellawell, E. 1993. Resistivity probes: a review of current practices and an investi-
gation into the properties of different systems. CUED Report.
30. Hensley, P.J. 1989. Accelerated physical modelling of transport processes in soil.
31. Horner, J.M. 1982. Centrifugal modelling of multi-layer clay foundations subject to
granular embankment loading. King's College, London University PhD thesis.
32. König, D., 1998. An inflight excavator to model a tunnelling process, Proc. Int. Conf.
Centrifuge 1998, Kimura et al. (eds). Balkema, Rotterdam. pp. 707-712.
33. Kotthaus, M. 1992. Zum Tragverhalten von horizontal belasteten Pfahlreihen, Schrift-
enreihe des Lehrstuhls für Grundbau und Bodenmechanik der Ruhr-Universität
Bochum, Heft 18, Bochum.
34. Kulhawy, F. and Mayne, P. 1990. Manual on estimating soil properties. Report EL-
6800. Electric Power Res. Inst., Palo Alto, 306p.
35. Kutter, B.L. 1982. Centrifugal modelling of the response of clay embankments to
earthquakes. Cambridge University PhD thesis.
36. Kutter, B.L., Sathialingam, N. and Herrmann, L.R. 1988. The effects of local arching
and consolidation on pore pressure measurements in clay. Centrifuge '88, J.F. Corte,
(ed). Balkema. pp. 115-118.
37. Lach, P. 1992. Seminar on centrifuge model tests on iceberg scour above pipelines
buried in clay, and subject for forthcoming thesis at Memorial University of
Newfoundland, St John's, Newfoundland.
38. Ladd, C.C., Foott, R., Ishihara, K., Schlosser, F. and Poulos, H.J. 1977. Stress-
deformation and strength characteristics. 9th ICSMFE, Tokyo, Vol. 2, pp. 421-494.

39. Laue, J. 1996. Zur Setzung von Flachfundamenten auf Sand unter wiederholten
Lastereignissen. Schriftenreihe des Lehrstuhls für Grundbau und Bodenmechanik der
Ruhr-Universität Bochum, Heft 25, Bochum.
40. Laue, J., Nater, P., Chikatamarla, R., Springman, S. M. 2002. Der Einsatz von
„tactilse pressure sensors“ in geotechnischen Labor- und Feldversuchen. Proc.
Messen in der Geotechnik. Braunschweig
41. Lawrence, D.M. 1980. Some properties associated with kaolinite soils. Cambridge
University M. Phil thesis.
42. Lerouil, S., Kabbaj, M., Tavenas, F. and Bouchard, R. 1985. Stress-strain-strain rate
relation for the compressibility of sensitive natural clays. Géotechnique 35, Vol. 2, pp.
152-80.
43. Lynch, R. 1998. Private communication.
44. Ma, J., 1994. Untersuchungen zur Standsicherheit der durch Stützscheiben stabilisi-
erten Böschungen, Mitteilung des Institut für Geotechnik ; 38, Stuttgart.
45. Mahmoud, M. 1988. Vane testing in soft clays. Ground Engineering, Vol. 21, No. 7, pp.
36-40.
46. Mair, R.J. 1979. Centrifugal modelling of tunnel construction in soft clay. Cambridge
University PhD thesis.
47. Mayne, P.W. and Kulhawy, F. 1982. K0-OCR relationships in soil. Proc. ASCE, JGED,
Vol. 102, pp. 197-228.
48. Mayne, P.W. 1992. In-situ determination of clay stress history by piezocone. Wroth
Memorial Symposium. Oxford. pp. 361-372 (Preprint).
49. Meigh, A.C. 1987. Cone penetration testing. CIRIA, Butterworths. 137p.
50. Mesri, G. 1975. Discussion: New design procedure for stability of soft clays. Proc.
ASCE, JGED. Vol. 103 GT5. pp. 417-430.
51. Muir Wood, D. 1990. Soil behaviour and critical state soil mechanics. CUP. 462p.
52. Nadarajah, V. 1973. Stress-strain properties of lightly overconsolidated clays.
53. Nunez, I. 1989. Tension piles in clay. Cambridge University PhD thesis.
54. Phillips, R. 1988. Centrifuge lateral pile tests in clay. PR-10592. Task 2 & 3 - Final
Report to Exxon Production Research. 39p.
55. Phillips, R. 1990. Spudcan/Pile Interaction Centrifuge Model Test Spud 3 - Final
Report. A report to EPR Corp. Houston, TX, USA. Lynxvale Ltd. Cambridge.
56. Phillips, R. and Valsangkar, A. 1987. An experimental investigation of factors
affecting penetration resistance in granular soils in centrifuge modelling. CUED/D
TR210. 17p.
57. Phillips, R. and Gui, M.W. 1992. Cone Penetrometer Testing. Phase 1 Report, In-situ
investigation. EEC Contract: SC1-CT91-0676.
58. Poorooshasb, F. 1988. The dynamic embedment of a heat emitting projectile.
59. Powrie, W. 1987. The behaviour of diaphragm walls in clay. Cambridge University
PhD thesis.

60. Randolph, M.F. and Houlsby, G.T. 1984. The limiting pressure on a circular pile
loaded laterally in clay. Géotechnique 34, No. 4, pp. 613-623.
61. Renzi, R., Maggioni, W., Smits, F. and Manes, V. 1991. A centrifugal study on the
behavior of suction piles. Centrifuge '91, H.Y.Ko and F.G.McLean (eds). Balkema. pp.
169-177.
62. Renzi, R., Corté, J.F., Rault, G., Bagge, G., Gui, M.W., Laue, J. 1994. Cone
penetration tests in the centrifuge: Experience of five laboratories. Centrifuge ‘94.
Leung, Lee and Tan (eds), Balkema, Rotterdam, pp. 77-82
63. Rossato, G., Ninis, N.L. and Jardine, R.J. 1992. Properties of some kaolin-based
model clay soils. ASTM GTJODJ, Vol. 15, No. 2, pp. 166-179.
64. Rowe, P.W. 1975. Displacement and failure modes of model offshore gravity platforms
founded on clay. Conf. Offshore Europe '75, pp.218. 1- 17. Spearhead Publications,
Aberdeen.
65. Savvidou, C. 1984. Effects of a heat source in saturated clay. Cambridge University
PhD thesis.
66. Schmidt, B. 1966. Discussion: Earth pressures at rest related to stress history. Can.
Geot. J., 3, No. 4, pp. 239-242.
67. Schmidt, B. 1983. Discussion: K0-OCR relationships in soil. Mayne, P., Kulhawy, F.
1982. Proc. ASCE, JGED, Vol. 109, pp. 866-867.
68. Schofield, A.N. and Wroth, C.P. 1968. Critical State Soil Mechanics. McGraw-Hill.
309p.
69. Schofield, A.N. 1980. Cambridge University Geotechnical Centrifuge Operations. 20th
Rankine lecture. Géotechnique 30, No. 3. pp. 227-268.
70. Sharma, J.S. 1993. Construction of reinforced embankments on soft clay. Cambridge
University PhD thesis.
71. Siemer, T. (1996): Zentrifugenmodellversuche zur dynamischen Wechselwirkung
zwischen Bauwerk und Boden infolge stossartiger Belastung, Schriftenreihe des
Lehrstuhls für Grundbau und Bodenmechanik der Ruhr-Universität Bochum, Heft 27,
Bochum.
72. Skempton, A.W. 1957. The planning and design of the new Hong Kong airport. Proc.
ICE 7, pp. 305-307.
73. Sketchley, C.J. 1973. The behaviour of kaolin in plane strain. Cambridge University
PhD thesis.
74. Smith, C.C. 1992. Thaw induced settlement of pipelines in centrifuge model tests.
75. Springman, S.M. 1989. Lateral loading on piles due to simulated embankment
construction. Cambridge University PhD thesis.
76. Springman, S.M. 1991. Performance of a single lattice leg under lateral load. EEPUK
Report. Contract No. EP-022R, Task Order: 2-022. 23p.
77. Springman, S. M., Laue, J., Boyle, R., White, J., and Zweidler, A. (2001). The ETH
Zurich Geotechnical Drum Centrifuge. International Journal of Physical Modelling in
Geotechnics, Vol. 1 (1), pp. 59-70.

78. S.M. Springman, P. Nater, R. Chikatamarla, J. Laue. 2002. Use of flexible tactile
pressure sensors in geotechnical centrifuges. Proc. ICPMG. St. Johns. Canada.
79. Stewart, D.P. and Randolph, M.F. 1991. A new site investigation tool for the
centrifuge. Centrifuge '91, H.Y. Ko and F.G. McLean (eds). Balkema. pp. 531-537.
80. Stone, K.J.L. 1988. Modelling the rupture development in soils. Cambridge University
PhD thesis.
81. Sun, H.W. 1989. Ground deformation mechanisms for soil-structure interaction.
82. Tang, Z. 1993. Laboratory measurement of shear strength and related acoustic
properties. MEng thesis, Memorial University of Newfoundland, St John's,
Newfoundland.
83. Tovey, N.K. 1970. Electron microscopy of clays. Cambridge University PhD thesis.
84. Trak, B., La Rochelle, P., Tavenas, F., Leroueil, S. and Roy, M. 1980. A new
approach to the stability analysis of embankments on sensitive clays. Can. Geot. J. 17,
Vol. 4, pp. 526-544.
85. Vinson, T.S. 1982. Ice forces on offshore structures. Proc. Workshop on High Gravity
Simulation for Research in Rock Mechanics. Colorado School of Mines. pp. 60-68.
86. Vinson, T.S. 1983. Centrifugal modelling to determine ice/structure/geologic
foundation Interactive forces and failure mechanisms. Proc. 7th Int. Conf. on Port and
Ocean Engineering under Arctic Conditions. pp. 845-854.
87. Waggett, P.R. 1989. The effect of lubricants on the interaction between soils and
perspex. Cambridge University Part II Project Report.
88. Wilkinson, B.J. 1993. An investigation into the effects of adding a substantial granular
content to a kaolin based mix. Cambridge University Part II Project Report.
89. White D. J., Take W.A, Bolton M.D. and Munachen S.E. 2001. A deformation
measuring system for geotechnical testing based on digital imaging, close-range
photogrammetry, and PIV image analysis. Proc. 15th ICSMGE.
90. Wong, P.C., Chao, J.C., Murff, J.D., Dean, E.T.R., James, R.G., Schofield, A.N., and
Tsukamoto, Y. 1993. Jack-up rig foundation modelling II. Offshore Technology
Conference 7303.
91. Wroth, C.P. 1972. General theories of earth pressures and deformations. Proc. 5th
ECSMFE, Madrid, II, pp. 33-52.
92. Wroth, C.P. 1975. In-situ measurements of initial stresses and deformation character-
istics. CUED/D TR23.
93. Wroth, C.P. 1979. Correlations of some engineering properties of soils. Proc. 2nd Int.
Conf. Behaviour of Offshore Structures, London. Vol. 1, pp. 121-132.
94. Yong, R.N. and McKyes, E. 1971. Yield and failure of clay under triaxial stresses.
Proc. ASCE SM 1, pp. 159-176.

Numerical Modelling
Finite Difference Method

8 Finite Difference Analysis using FLAC (Fast Lagrangian

Analysis of Continua)
8.1 Basics
Finite differences may be used to solve partial differential equations (e.g. steady state flow
or Newton's Laws etc.).
Governing equations are substituted by finite differences in space, written in terms of field
variables at discrete points i.e. the NODES - these are the key geometrical feature.
Implicit solution: solution at every node is dependent on solutions at four
neighbour nodes - the solution at each node isn't known
until the entire solution is known.
Explicit solution: nonlinear solutions are produced in the same time as for linear
problems (cf. longer solution times for implicit solutions).
Mixed discretisation: accurate modelling: plastic collapse
plastic flow
Solutions are mainly iterative - aiming to reduce the error to an acceptable level.
8.1.1 Specific to Geotechnics via FLAC (Fast Lagrangian Analysis of Continua)
Partial differential equations

=> matrix equations for each node, using dynamic equations of motion.
Forces nodes = f ( displacements ) nodes
The contour integral formulation of finite differences is used. This formulation overcomes
difficulties often associated with mesh pattern and imposition of boundary conditions. The
mean value of the gradient of a field variable in a zone may be expressed using the Gauss
theorem with the contour integral performed on the boundary of the same zone.
Finite Difference Analysis using FLAC (Fast Lagrangian Analysis of Continua) Page 8 - 1
x 103
Figure 8.1: Example for unbalanced force vs time steps for two load increments
The method is conditionally stable. depending on damping coefficient and ratio between
mass and time step (see Fig. 8.1).
For example, strain energy in the system may be converted into kinetic energy and may be
allowed to radiate away and dissipate.
Iterative speed must be faster than wave propagation velocity.
Linear variation of the relevant quantities is assumed along the edges of the zone.
Standard quadrilateral elements used to generate the NODES as implemented in FLAC,

lead to a stiffness matrix analogous to four node quadrilateral finite elements with reduced
integration but the finite difference method is 'physically' more justifiable.
The dynamic relaxation method (for static analysis) is implemented to solve the algebraic
system of equations of motion. Successive integrations of these full dynamic equations
(even for static solutions) of motion lead to the steady state solution usually based on
advice in the Software User Manual for FLAC.
For a static solution, mass and damping are fictitious: choose these in order to achieve
stability and accuracy of the solution. Generally damping should be less than the critical
damping for the system (check using simple beam theory or by trial and error).
Advantages of the method:

• a relevant number of degrees of freedom can be analysed quickly and effectively using a
PC version of the code. Since matrices are not stored, memory requirements are low.
• solution timelarge strain will be slightly higher than solution timesmall strain,
• the method is competitive with standard implicit-oriented FE codes for highly non-linear
problems.
Disadvantages of the method:

• in spite of the advanced formulation, strict limitations on the mesh pattern still exist
(make the elements in the grid as square as possible), unless calculation efficiency is
reduced (especially for elements of different stiffness),
• slower than FEM for linear problems,
• ratio of longest natural period to the shortest natural period affects solution time (e.g. will
be influenced by beams, large stiffness differences and large changes in element sizes).
Note: Users need to know a set of abbreviated commands (actually quite quick to learn).
Solution procedure
The solution procedure can be summarized as follows:
1. Assuming known displacements, velocities and nodal forces at time ti compute new
nodal accelerations.
2. Integrate nodal accelerations => nodal velocities & displacements.
3. From nodal velocities/displacements, impose constitutive law, obtain new stress state.
4. Calculate new nodal internal forces, integrate stress state along the element boundary.
5. Impose a suitable limit on the unbalanced forces (the difference between the applied
external forces and the internal forces).
6. Check current unbalanced force and
* if it is less than limit, end calculation,
* if more, increment the time step, go to (1), until steady state solution
achieved.
8.2 Finite Difference

Many similarities to finite element analysis
• idealised geometry (Boundary value problem: closed domain)
• Material(s) (idealised constitutive model)
• idealised loads
However: The method of NUMERICAL SOLUTION for the PARTIAL DIFFERENTIAL
EQUATIONS which govern behaviour is different.
Further :
Whereas finite elements describe behaviour of the continuum in terms of ELEMENTS and
a finite difference grid may appear also to be an assemblage of elements, in fact it is the
NODES which dictate behaviour.
Finite Difference NODES and connecting GRID
y
∆x Solution Domain D(x,y)
jmax
j+1
Node
j
∆y
j-1
2
1 x
1 2 i-1 i i+1 imax
Figure 8.2: Simple finite difference grid
Usually ∆x is approximately constant

∆x need not to be equal to ∆y but ∆x = ∆y has some advantages.
The value of a function “ f ” at any of the nodes x = i, y = j for example can be written as fij.
∂f ij 2
∂ f ij
In the PDE’s, typically this might vary as a function of x: e.g. ------- or ---------2-
∂x ∂x
∂f ij 2
∂ f ij
of y: e.g. ------- or ---------2-
∂y ∂y
Typical PDE’s might include the LAPLACE equation:
∇2f = 0 (f = function describing variation in terms of x,y space..)
∂ 2 f- --------
--------
∂ 2 f-
+ = 0 (2 dimensional)
∂x 2 ∂y 2
which may (should!) be familiar because it describes the STEADY STATE flow such as
e.g.
* steady state seepage (figure 8.3)
* heat diffusion (conduction)
i.e. No change with time as would occur in consolidation for example (FICKS LAW)
∂f ∂2f ∂2 f
= constant ⎛⎝ --------- + ---------⎞⎠
∂t ∂x 2 ∂y 2
Flow lines
Equipotentials
e.g
Q=ki
Figure 8.3: Flow net
Typically one might assume: f = a0 + a1x +a2x2+a3x3+..........+anxn
= b0 + b1y +b2y2+b3y3+..........+bnyn
e.g Notation fij = value for function f at node x = i, y = j

partial derivative: first order
∂f i, j ∂f i
= partial derivative in x direction for constant “j” can be written as or fx (i,j)
∂x ∂x
2 2
∂ f i, j ∂ fi
= 2nd order partial derivative in x direction for constant “j” as or fxx|(i,j)
∂x2 ∂ x2
3 n
∂ f i, j ∂ f i, j
------------ = fxxx|i,j = fnx|(i,j)
∂x
3 ∂ xn
similarly for changes in y direction

2 2
∂f i, j ∂f j ∂ f i, j ∂ fj
= = fy ( i, j ) and = = f yy ( i, j ) etc.
∂y ∂y ∂ y2 ∂ y2
So what does ”Finite Difference” mean....

using a finite distance such as ∆x or ∆y, the variation is written as follows
∂f- fi + 1 – fi f i – fi – 1 fi + 1 – fi – 1
----- = ------------------- or ------------------ or --------------------------
∂x ∆x ∆x 2∆x
fi
∆x ∆x
i-1 i i+1 x
f i + 1 – fi
forward: ------------------
-
∆x
fi – fi – 1
backward: ------------------
∆x
fi + 1 – fi – 1
central: --------------------------
2∆x
Similar equations can be written in terms of changes of “f” in the y direction

PDE’s e.g. (LAPLACE)
2 2
∂ f ∂ f
--------- + --------- = 0
2 2
∂x ∂y
f xx ( i, j ) + f yy ( i, j ) = 0
exact solution f ij approximate solution f ij
i.e. “correct” solution e.g. using FINITE Difference
=> very difficult to achieve
Difference = Error:
e.g. f – f
function may be derived of the exact indivisual

from a “Taylor” series partial derivatives
e.g. f = exact solution
1 1 1
( i, j ) ∆x ( i, j ) ∆x ( i, j ) ∆x + …… + ----- f nx ( i, j ) ∆x
f i + 1, j = f i, j + f x + --- f xx 2 + --- f xxx 3 n
2 6 n!
1 1 1
( i, j ) ∆x ( i, j ) ∆x ( i, j ) ∆x
f i – 1, j = f i, j – f x + --- f xx 2 – --- f xxx 3 …… ± ----- f nx ( i, j ) ∆x n
2 6 n!
and similar in terms of y fi,j-1; fi,j; fi,j+1.
We should rewrite the above equations in shortened (approximation) form:

1 1 1
f i + 1, j = f i, j + f x ( i, j ) ∆x + ----- f xx ( i, j ) ∆x 2 + ----- f xxx ( i, j ) ∆x 3 + …… + ----- f nx ( i, j ) ∆x n (1)
2! 3! n!
1 1 1
( i, j ) ∆x ( i, j ) ∆x ( i, j ) ∆x + …… ± ----- f nx ( i, j ) ∆x n
f i – 1, j = f i, j – f x + ----- f xx 2 – ----- f xxx 3 (2)
2! 3! n!
If we add the above equations (1) and (2) we get
2 2
( i, j ) ∆x ( i, j ) ∆x + ……
f i + 1, j + f i – 1, j = 2f i, j + ----- f xx 2 + ----- f xxxx 4
2! 4!
f i + 1, j – 2f i, j + f i – 1, j 1 2
f xx ( i, j ) = ------------------------------------------------
2
– ------ f xxxx ( i, j ) ∆x ± ……
∆x 12
similarly in the y- direction
f i, j + 1 – 2f i, j + f i, j – 1 1 2
f yy ( i, j ) = ------------------------------------------------
2
– ------ f yyyy ( i, j ) ∆y ± ……
∆y 12
--------
∂ 2f
∂ 2 f- --------
- = f xx
+ ( i, j ) + f yy ( i, j ) = 0 exact
∂x 2 ∂y 2
∂2 f ∂ 2f
--------- + --------- = f xx ( i, j ) + f yy ( i, j ) = 0 approximate (e.g ignore higher fxxxx
∂x 2 ∂y 2
terms etc.)
so assume an approximate solution where:
∂2 f ∂ 2f f i + 1, j – 2f i, j + f i – 1, j f i, j + 1 – 2f i, j + f i, j – 1
--------- + --------- = ------------------------------------------------ + ------------------------------------------------
∂x 2 ∂y 2 ∆x
2
∆y
2
when ∆x = ∆y
∂ 2 f-
∂ 2 f- -------- f i + 1, j + f i – 1, j + f i, j + 1 + f i, j – 1 – 4f i, j
-------- + = ---------------------------------------------------------------------------------------- = 0
∂x 2 ∂y 2 ∆x
2
The accuracy of the solution is affected by the size of the
ERROR = f – f
Convergence occurs when error -> 0

as ∆x = ∆y ->0
and the solution at every point depends on the solution at all the other points.
i.e this is an IMPLICIT SOLUTION
(There are also EXPLICIT and MIXED DISCRETISATION solutions, which are not intro-
duced here)
Example: flat plate heated at one boundary

Temperature varies as function of f
steady: 100°C at top boundary
0°C at the other boundaries
at each node: f i + 1, j + f i – 1, j + f i, j + 1 + f i, j – 1 – 4f i, j = 0
select a 3 x 4 grid in the first instance:
f = 100 100 100 no flow in z-direction!

4
f2,3
f= 0 3 0 ∆x = ∆y = 5 cm
Temperature:
0 2 0 f = 0,100 °C
f2,2
Nodes
0 1 1,2...
1 2 3 x
0 0 0
Figure 8.4: Grid for flat plate, heated at top boundary
At node 2,2
4 ⋅ f 2, 2 – f 2, 3 = 0 (because f 2, 1 = f 1, 2 = f 3, 1 = 0 )
At node 2,3
4 ⋅ f 2, 3 – f 2, 2 – 100 = 0 (because f 1, 3 = f 3, 3 = 0 )
or
4 – 1 f 2, 2 = 0 2 unknown nodes : 2 x 2 matrix
– 1 4 f 2, 3 100
Accuracy of solution
y (cm)
100 °C
15
80
60
40
10
20
10
5 5
0
0 5 10 x (cm)
Figure 8.5: Temperature distribution on plate from the 3 x 4 grid
The answers are now compared with those from the exact solution and those determined
from a 5 x 7 grid where
∆f = f – f
Node fexact f ∆f f ∆f
x y 3 x 4 grid 5 x 7 grid
cm cm °C °C °C °C °C
5 10 26.049 26.667 0.618 26.228 0.179
5 5 5.261 6.667 1.406 5.731 0.47
12 Nodes 35 Nodes
Tab. 8.1: Exact and calculated temperatures
• The number of nodes control the accuracy of the solution

• Accuracy (∆f) is higher with more nodes...
• But this will require longer solution times: e.g => the diagonal matrix grows from a [2 x 2]
matrix to a [15 x 15] matrix when the grid changes from 3 x 4 to 5 x 7
for a 5 x 7 grid:
y 100
3x 5
15 x 15 unknown
values at
Diagonal = 0 0 nodes
Matrix
0 x
fi,j Boundary
temperatures
f = 100 or 0°C
So what happens if we use more nodes?

ACCURACY SOLUTION TIME ALSO
8.3 Details of FLAC Program
Start
MODEL SETUP
1. Generate grid, deform to desired shape
2. Define constitutive behaviour and material properties
3. Specify boundary and initial conditions
Step to equilibrium state
Examine
the model response
Results unsatisfactory
Model makes sense
PERFORM ALTERATIONS
for example
- Excavate material
- Change boundary conditions
Step to solution
More tests
needed Examine
the model response
Acceptable result
Yes Parameter
study needed
No
End
Figure 8.6: FLAC general solution procedure (FLAC manual)
FLAC uses nomenclature that is consistent, in general, with that used in conventional finite
difference or finite-element programs for stress analysis. The basic definitions of terms are
reviewed here for clarification. Figure 8.7 is provided to illustrate the FLAC terminology.
Figure 8.7: Example of a FLAC model (FLAC Manual)
FLAC MODEL — The FLAC model is created by the user to simulate a physical problem.
When referring to a FLAC model, the user implies a sequence of FLAC commands that
define the problem conditions for numerical solution.
ZONE —The finite difference zone is the smallest geometric domain within which the
change in a phenomenon (e.g., stress versus strain, fluid flow or heat transfer) is evaluated.
Quadrilateral zones are used in FLAC. Another term for zone is element. Internally, FLAC
divides each zone into four triangular “subzones,” but the user is not normally aware of
these.
GRIDPOINT — Gridpoints are associated with the corners of the finite difference zones.
There are always four (4) gridpoints associated with each zone. In the FLAC model, a pair
of x- and y-coordinates are defined for each gridpoint, thus specifying the exact location of
the finite difference zones. Other terms for gridpoint are nodal point and node.
FINITE DIFFERENCE GRID — The finite difference grid is an assemblage of one or more
finite difference zones across the physical region which is being analysed. Another term for
grid is mesh.
MODEL BOUNDARY — The model boundary is the periphery of the finite difference grid.
Internal boundaries (i.e., holes within the grid) are also model boundaries.
BOUNDARY CONDITION — A boundary condition is the prescription of a constraint or
controlled condition along a model boundary (e.g., a fixed displacement or force for
mechanical problems, an impermeable boundary for groundwater flow problems, adiabatic
boundary for heat transfer problems, etc.).
INITIAL CONDITIONS — This is the state of all variables in the model (e.g., stresses or
pore pressures) prior to any loading change or disturbance (e.g., excavation).
CONSTITUTIVE MODEL — The constitutive (or material) model represents the defor-
mation and strength behaviour prescribed to the zones in a FLAC model. Several consti-
tutive models are available in FLAC to assimilate different types of behaviour commonly
associated with geologic materials. Constitutive models and material properties can be
assigned individually to every zone in a FLAC model.
SUB-GRID — The finite difference grid can be divided into sub-grids. Sub-grids can be
used to create regions of different shapes in the model (e.g., the dam sub-grid on the
foundation sub-grid in Figure 8.7). Sub-grids cannot share the same gridpoints with other
sub-grids; they must be separated by null zones.
NULL ZONE — Null zones are zones that represent voids (i.e., no material present) within
the finite difference grid. All newly created zones are null by default.
ATTACHED GRIDPOINTS — Attached gridpoints are pairs of gridpoints that belong to
separate sub-grids that are joined together. The dam is joined to the foundation along
attached gridpoints in Figure 8.7. Attached gridpoints do not have to match between sub-
grids, but sub-grids cannot separate from one another once attached.
INTERFACE — An interface is a connection between sub-grids that can separate (e.g.,
slide or open). An interface can represent a physical discontinuity such as a fault or contact
plane. It can also be used to join sub-grid regions that have different zone sizes.
MARKED GRIDPOINTS —Marked gridpoints are specially designated gridpoints that
delimit a region for the purpose of applying an initial condition, assigning material models
and properties, and printing selected variables. The marking of gridpoints has no effect on
the solution process.
REGION — A region in a FLAC model refers to all zones enclosed within a contiguous
string of “marked” gridpoints. Regions are used to limit the range of certain FLAC
commands, such as the MODEL command that assigns material models to designated
regions.
GROUP — A group in a FLAC model refers to a collection of zones identified by a unique
name. Groups are used to limit the range of certain FLAC commands, such as the MODEL
command that assigns material models to designated groups. Any command reference to a
group name indicates that the command is to be executed on that group of zones.
STRUCTURAL ELEMENT — Structural elements are linear elements used to represent
the inter-action of structures (such as tunnel liners, rock bolts, cable bolts or support props)
with a soil or rock mass. Some restricted material non-linearity is possible with structural
elements. Geometric non-linearity occurs in large-strain mode.
STEP — Because FLAC is an explicit code, the solution to a problem requires a number of
computational steps. During computational stepping, the information associated with the
phenomenon under investigation is propagated across the zones in the finite difference
grid. A certain number of steps is required to arrive at an equilibrium (or steady-flow) state
for a static solution. Typical problems are solved within 2000 to 4000 steps, although large
complex problems can require tens of thousands of steps to reach a steady state. When
using the dynamic analysis option, STEP refers to the actual timestep for the dynamic
problem. Other terms for step are timestep and cycle.
STATIC SOLUTION — A static or quasi-static solution is reached in FLAC when the rate of
change of kinetic energy in a model approaches a negligible value. This is accomplished by
damping the equations of motion. At the static solution stage, the model will be either at a
state of force equilibrium or at a state of steady-flow of material if a portion (or all) of the
model is unstable (i.e., fails) under the applied loading conditions. This is the default calcu-
lation in FLAC. Static mechanical solutions can be coupled to transient groundwater flow or
heat transfer solutions. (As an option, fully dynamic analysis can also be performed by
inhibiting the static solution damping.)
UNBALANCED FORCE — The unbalanced force indicates when a mechanical equilibrium
state (or the onset of plastic flow) is reached for a static analysis. A model is in exact
equilibrium if the net nodal force vector at each gridpoint is zero. The maximum nodal force
vector is monitored in FLAC and printed to the screen when the STEP or SOLVE command
is invoked. The maximum nodal force vector is also called the unbalanced or out-of-balance
force. The maximum unbalanced force will never exactly reach zero for a numerical
analysis. The model is considered to be in equilibrium when the maximum unbalanced
force is small compared to the total applied forces in the problem. If the unbalanced force
approaches a constant non-zero value, this probably indicates that failure and plastic flow
are occurring within the model.
DYNAMIC SOLUTION — For a dynamic solution, the full dynamic equations of motion
(including inertial terms) are solved; the generation and dissipation of kinetic energy directly
affect the solution. Dynamic solutions are required for problems involving high frequency
and short duration loads — e.g., seismic or explosive loading. The dynamic calculation is
an optional module to FLAC (see Section 3 in Optional Features FLAC manual).
LARGE-STRAIN / SMALL-STRAIN — By default, FLAC operates in small-strain mode:
that is, gridpoint coordinates are not changed, even if computed displacements are large
(compared to typical zone sizes). In large-strain mode, gridpoint coordinates are updated at
each step, according to computed displacements. In large-strain mode, geometric non-
linearity is possible.
Figure 8.8: Options within FLAC (FLAC manual)
There are ten basic constitutive models provided in FLAC Version 4.0, arranged into null,
elastic and plastic model groups:
8.3.1 Null model group

1. null model
A null material model is used to represent material that is removed or excavated.
8.3.2 Elastic model group

2. elastic, isotropic model
The elastic, isotropic model provides the simplest representation of material behaviour. This
model is valid for homogeneous, isotropic, continuous materials that exhibit linear stress-
strain behaviour with no hysteresis on unloading.
3. elastic, transversely isotropic model

The elastic, transversely isotropic model gives the ability to simulate layered elastic media
in which there are distinctly different elastic moduli in directions normal and parallel to the
layers.
8.3.3 Plastic model group

4. Drucker-Prager model
The Drucker-Prager plasticity model may be useful to model soft clays with low friction
angles; however, this model is not generally recommended for application to geologic
materials. It is included here mainly to permit comparison with other numerical program
results.
5. Mohr-Coulomb model
The Mohr-Coulomb model is the conventional model used to represent shear failure in soils
and rocks. Vermeer and deBorst (1984), for example, report laboratory test results for sand
and concrete that match well with the Mohr-Coulomb criterion.
6. ubiquitous-joint model
The ubiquitous-joint model is an anisotropic plasticity model that includes weak planes of
specific orientation embedded in a Mohr-Coulomb solid.
7. strain-hardening/softening model
The strain-hardening/softening model allows representation of non-linear material softening
and hardening behaviour based on prescribed variations of the Mohr-Coulomb model
properties (cohesion, friction, dilation, tensile strength) as functions of the deviatoric plastic
strain.
8. bilinear strain-hardening/softening ubiquitous-joint model
The strain-hardening/softening ubiquitous-joint model allows representation of material
softening and hardening behaviour for the matrix and the weak plane based on prescribed
variations of the ubiquitous-joint model properties (cohesion, friction, dilation, tensile
strength) as functions of deviatoric and tensile plastic strain. The variation of material
strength properties with mean stress can also be taken into account by using the bilinear
option.
9. double-yield model
The double-yield model is intended to represent materials in which there may be significant
irreversible compaction in addition to shear yielding, such as hydraulically-placed backfill or
lightly-cemented granular material.
10. modified Cam-clay model
The modified Cam-clay model may be used to represent materials when the influence of
volume change on bulk property and resistance to shear need to be taken into consider-
ation, such as soft clay.
There are also six time-dependent (creep) material models available in the creep model
option for FLAC.
8.4 Example analyses
1. Benchmarking : FEM v FDM v known solution Schweiger, Freiseder NUMOG (1995)

Flexible strip footing solution : undrained soil
E = 35 MPa, ν = 0.3, γ = 18 kN/m3, φ' =27.5o, su = 25 kPa, K0 = 0.65
Classical solution: failure load qmax = 5.14su = 128.5 kPa
• agreement very promising pre-failure, but
• modelling of failure dependent on method and mesh
• they should probably have used adaptive mesh refinement for the finite element analysis
1.5m
q 3.0m
Point1 Point 2
20m
y
30m
x
Figure 8.9: Geometry and loading for a strip footing on clay (Schweiger & Freiseder, 1995)
applied load q 10-1 (kPa)
applied load q 10-1 (kPa)
FEM-169 elements
FEM-169 elements
FEM- 676 elements FEM- 676 elements
FLAC grid 40 x 20 FLAC grid 40 x 20
FLAC grid 120 x 80 FLAC grid 120 x 80
settlement (cm) y- displacement (cm)
Figure 8.10: Results (Schweiger & Freiseder 1995)
2. Subsidence under a geotextile reinforced embankment: Lawson, Jones, Kempton 1995
• geotextiles used to prevent collapse of road due to unexpected subsidence (solution of

limestone / gypsum) in underlying geology,
• causes a trapdoor: soil arches over this, with large strain (deformation) above the void,
• ideal for finite difference analysis to find tension in geotextile (and also differential
surface deformation which should be < 1% principal roads, and < 2% lower class roads),
• here geotextiles are modelled as an elastic beam because of significant deflection,
• results compared to the existing theory which assumes no arching in soil (BS8006-
generally conservative) and full arching (Giroud et al., 1990- likely to be unsafe),
• also could be solution to covering /sealing former waste deposit and improving
foundation for subsequent building.
Figure 8.11: Design cross section through reinforced embankments, Lawson et al., (1994)
1% strain (BS8006)
Vertical deflection at base of fill D (m)
5% strain 1% strain (Giroud)

Reinforcement tension Trs (kN/m)
BS 8006 1% strain (FDM)
5% strain (BS8006)
5% strain (Giroud)
Giroud 5% strain (FDM)
FDM
Embankment height H (m)
Width of void d (m)

Comparison of the three models in deter-
mining the reinforcement tension for a void Comparison of the three models in deter-
span of d = 4 m (γ = 20 kN/m2, φ’ = 35°) mining the maximum vertical deflection
at the base of the fill spanning the void
Figure 8.12: Results, Lawson et al., (1994)
• predicted maximum vertical displacement at base of embankment agrees well with

BS8006 and Giroud’s theories for all void widths with 1% and 5% strain in the geotextile,
• reinforcement tension will be overestimated if no arching in the embankment is included,
• if a rigid base (e.g. rock) lies under the reinforcement, full arching is most likely to occur
and tension will be lower,
• if soil is placed under the reinforcement, result is midway between the arching and no
arching solution.
3. Lateral cyclic loading of an embedded wall Springman, Norrish, Ng (1995)
An integral bridge abutment is sometimes used to avoid problems with expansion joints and
bearings, especially where damage from e.g. de-icing salts will cause expensive mainte-
nance. The deck-wall becomes a full moment connection and any changes in deck length
due to temperature effects (daily and seasonally) will load the fill behind the abutment. This
means that an element of soil from just behind the retaining wall will be subjected to cyclic
lateral loading and the results is a compaction and loss of volume.
Problem: cyclic behaviour of soil - how to model this?
Centrifuge behaviour (and prototype)
Settlement profile after CWWN1 test Settlement profile after CWWN2 test
loose (Id = 23%)

dense (Id = 83%)
Figure 8.13: Geometry: All dimensions shown in mm (not to scale) (Springman et al., 1995)
Numerical analysis + small strain stiffness

(BRICK) model
Deformed mesh of spread base wall in response to aa 0.115°
0.115°
applied rotation
66mm
mm 6mm
6mm Active displacement
cycle 10
y(m)
y(m)
x(m)
x(m)
Active displacement of cycle 10
Active displacement at the top of the wall after 10th cycle
Figure 8.14: Deformed mesh (Springman et al., 1995)
Real data from cyclic triaxial tests (Jeyatharan, 1991)
• shows significant volumetric strain occurring during cycles of axial strain along a stress
path
=> q/p’ ~ 3 (triaxial compression)
=> between axial strain of 0.3% to -0.4%
Deviatoric stress, q (kPa)
Deviatoric stress, q (kPa)

Effective stress p’ (kPa) Axial Strain %
Volumetric strain (%)
Volumetric strain (%)
stress ratio, q/p’ Axial strain %
Figure 8.15: Cyclic behaviour of 100/170 (Fraction E) sand (after Jeyatharan 1991)
University of British Columbia (UBC) model after Beaty & Byrne (1999) compared to cyclic
triaxial tests (Figure 8.15)
• shows good agreement in

=> q, p’ space although q/p’~ 2
=> q, axial strain space (between axial strain of 0.3% to -0.4%)
=> volumetric strain, stress ratio q/p’ space
=> similarly for volumetric against axial strain (although 1st unloading cycle not so good,
so values of volumetric strain a little low)
Figure 8.16: Cyclic behaviour: UBC model (Springman et al., 1995)
• results also look quite good when plotted in normal characteristics (particularly for loose
sand samples) in terms of :
=> strength
=> volumetric
• although the dilation law displayed by the algorithm (Figure 8.17b) does not permit the
soil to reach a critical state where no further dilation occurs.
Volumetric characteristics Volumetric characteristics

loose sand medium dens sand
Model prediction:
1000 kPa
continuous line
Experimental results:
markers
Model prediction:
continuous line
Experimental results: 50kPa
markers
100kPa 200kPa
50 kPa
1000kPa
a) b)
Strength Characteristics
1000kPa
Model prediction:
continuous line experi-
mental results:
markers
50kPa
c)
Figure 8.17: Comparison between monotonic laboratory test data and the UBC model (Springman et
al.,1995)
4. Deformation of a 3 m high reinforced wall with 2 m long reinforcement (Jommi et al.,

NUMOG 1995)
End of construction
Spacing: d
x
a b
Deformed mesh Displacement vectors
Figure 8.18: Reinforced soil wall :deformed mesh (Jommi et al., 1995)
y/H
xd/H [%]
Figure 8.19: Normalised horizontal displacements d at the end of construction:
(a) front of the wall
(b) back of the wall (Jommi et al., 1995)
See also Springman, Balachandran, Jommi (1997) for an application relating to modelling
reinforced soil wall in the centrifuge.
8.5 References
1. Beaty, M.H. and Byrne, P.M. 1999. A synthesized approach for modelling liquefaction
and displacements, International FLAC symposium, Minneapolis, Minnesota.
2. FLAC Reference Manuals.
3. Giroud, J.P., Bonaparte, R., Beech, J.F. and Gross, B.A. 1990. Design of soil layer
geo-synthetic systems overlying voids, Geotextiles and Geomembranes, Vol. 9, pp. 11-
50.
4. Hoffman, J.D. 1992. Numerical methods for engineers and scientists. McGraw-Hill,
New York.
5. Jeyatharan, K. 1991. Partial liquefaction of sand fill in a mobile arctic caisson under
dynamic ice-loading, CUED PhD Thesis.
6. Jommi, C., Nova, R. and Gomis, F. 1995. Numerical analysis of reinforced earth walls
via a homogenization method. Proceedings of the fifth international symposium on
numerical models in geomechanics-NUMOG V, pp. 231-236.
7. Lawson, C.R., Jones, C.J.F.P., Kempton, G.T. and Passaris, E.K.S. 1994. Advanced
analysis of reinforced fills over areas prone to subsidence. Fifth International
Conference on Geotextiles, Geomembranes and related products, Singapore 1994, pp.
311-316.
8. Schweiger, H.F. and Freiseder, M. 1995. Some results from benchmark tests for
geotechnical engineering. Proceedings of the fifth international symposium on
numerical models in geomechanics-NUMOG V, pp. 675-680.
9. Springman, S.M. 1989. Lateral loading on piles due to simulated embankment
construction, CUED PhD Thesis.
10. Springman, S.M., Norrish A.R.M. and Ng C.W.W. 1995. Cyclic loading of sand
behind integral bridge abutments. TRL Report Cambridge University.
11. Springman, S.M., Balachandran S. and Jommi C. 1997. Modelling pre-failure defor-
mation behaviour of reinforced soil walls. Geotechnique, Symposium in print, Vol XLVII,
No.3, pp. 653-663.

Dokumen - Tips Modelling in Geotechnics

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Dokumen - Tips Modelling in Geotechnics

Uploaded by

Copyright:

Available Formats

Modelling in Geotechnics - Script

Prof. Sarah Springman

3.10.2SAGE CRISP ............................................................................................... 26

4 Scaling laws and applications for centrifuge modelling ............................ 1

5 Practical considerations: mechanical ....................................................... 1

7.4 Summary ........................................................................................................... 19

Prof. Sarah Springman

Figure 1.1: Design Process

So what are the relevant Limit States?

Figure 1.3: Heathrow Airport: tunnel collapse

Figure 1.4: Failure of the Teton Dam, USA (1976)

Figure 1.5: Teton Dam during failure

Figure 1.6: Teton Dam after failure

Development of engineering judgement is crucial and often underestimated and observing

Figure 1.7: Types of models

e.g. simple small scale 1g physical model….

• Domenico Fontana (in 1585)

1.1.1 Numerical modelling

Type Sketch of Key Advantages Disadvantages

empirical Calculation • quick and cheap • past data may not

finite Soil continua • general analytic tool • approx. solution+

1.1.2 Physical modelling

Type Sketch of Model/ σv at A Advantages Disadvantages

Field 100 • ‘the real thing’ • time (for diffusion)

1g, 1 • time (very quick) • stresses incorrect

100g, 100 • stress correct • radial ‘g’ field (in a

Tab. 1.2: Physical modelling

What concerns do we have about modelling 'effectively'?

Idealise any combination of:

Figure 1.8 contains an idealisation

The prototype problem

Inflight construction of embankment

Ellis PhD, 1997

1.1.3 Validation and calibration of models

Prototype Model 1 Model 2 Model 3 Model 4

1.2 Full scale (FS) 1g testing

Figure 1.9: ETH Big Box for 1g models

Figure 1.10: Dissertation Hertweck (1998)

Investigation of stress-deformation behaviour for two-phase manufactured diaphragm walls

Figure 1.11: Dissertation Brinkmann (2001)

1.3 Geotechnical centrifuge modelling

1.3.2 Types of centrifuge

1.3.3 Principles of modelling in the centrifuge

Figure 1.13: Accelerations of a rotated body

1.3.4 Scaling effects

Parameter Unit Scale

Linear dimension m 1/n

Mass or Volume kg or m3 1/n3

1.3.5 Verification of models

1. El-Hamalawi, A., Adaptive refinement of finite element meshes for geotechnical

Prof. Sarah Springman

2 Principles of numerical modelling

2.1 Why model numerically?

Figure 2.1: Understanding computer technology

Principles of numerical modelling 2-1

Undrained triaxial test on over-

Figure 2.2: Range of mesh complexity for a triaxial sample

Finite element mesh for analysis of

Centrifuge model test of stage-

Computed and measured development

1g 2g 20g h increases as time increases

Model “grows” as ng increases in centrifuge Layers added in prototype

Figure 2.4: Modelling “construction” processes