CIV5122Z Advanced Soil Mechanics Course August 2019

CIV5122Z Advanced Soil Mechanics:

Notes

By Dr. Mike Jefferies, PEng. and A/Prof. Denis Kalumba

([email protected] and [email protected])

Course Convenor: Ms Faridah Chebet

([email protected])

19th – 23rd August 2019

Postgraduate Seminar Room, 3.01
New Engineering Building
Department of Civil Engineering
University of Cape Town
Cape Town - SOUTH AFRICA

Page 1 of 49
CIV5122Z Advanced Soil Mechanics Course August 2019

OVERVIEW 4

SOFTWARE REQUIREMENTS 5

NOTATION 6

SOIL STRENGTH AND DILATION 8

SOIL IS PARTICULATE 8
INTERLOCKING AND TRUE FRICTION 9
FROM DIRECT SHEAR TO TRIAXIAL: INVARIANTS 11
TRIAXIAL COMPRESSION STRENGTH 12
STRESS-DILATANCY 14
CRITICAL VOID RATIO 15
STATE-DILATANCY 20

EXERCISE 1 22

EXERCISE 2 22

ELASTICITY 23
ISOTROPIC MODULI 23
ELASTIC STRAINS 23
GEOPHYSICAL MEASUREMENT 24
RELATIONSHIP BETWEEN G AND K 24
EFFECT OF STRESS AND VOID RATIO ON ELASTIC MODULI 24

PLASTICITY THEORY 26
SOIL IS PLASTIC 26
YIELDING AND YIELD SURFACES 27
FLOWRULE 29
HARDENING 29
CONSISTENCY CONDITION 31

A NUMERICAL WORLD 32
VISUAL BASIC FOR APPLICATIONS 32
EULER INTEGRATION 35

ORIGINAL CAM CLAY 37

KEY IDEAS 37
OCC FLOWRULE 37
OCC YIELD SURFACE 38
OCC HARDENING 39
OCC HARDENING VIA STATE 39
OCC HARDENING VIA PLASTICITY 39
COMPUTING TRIAXIAL COMPRESSION OF OCC 41
UNDRAINED LOADING 41
DRAINED LOADING 42

Page 2 of 49
CIV5122Z Advanced Soil Mechanics Course August 2019

EXERCISE 3 42

EXERCISE 4 43

EXERCISE 5 43

NORSAND 44
KEY IDEAS 44
NS FLOWRULE 45
NS YIELD SURFACE 45
NS HARDENING 47
COMPUTING TRIAXIAL COMPRESSION OF NS 47
ITERATIVE MODELLING FOR SOIL PROPERTIES 48

EXERCISE 6 49

EXERCISE 7 49

Page 3 of 49
CIV5122Z Advanced Soil Mechanics Course August 2019

Overview
Geotechnical engineering characterizes soil behaviour using , su, Cc and so forth, with
these parameters measured using insitu and/or laboratory tests. Advanced (or
‘theoretical’) soil mechanics seeks to understand these familiar parameters in terms of
fundamental properties. A feature of advanced soil mechanics is use of physical principles
rather than correlations; an immediate benefit is being able to compute, for example, how
changes in conditions (for example void ratio and/or stress level) affect the geotechnical
parameters such as .

Recent dam failures, with transitions from drained to undrained conditions in tens of
seconds (or quicker), have challenged conventional geotechnical engineering which
provides little in the way of explaining how such failure develop. But, this mechanism is
very simple in terms of physical principles. It is these sudden dam failures, possibly more
than anything else, that have created interest in what advanced soil mechanics offers
practical engineers – and thus this course.

Advanced soil mechanics applies the doctrine ‘if you cannot compute, you have nothing’.
This phrase is more important than it looks, as it derives from a most fundamental question
about knowledge (search the Church–Turing thesis on Wikipedia). For the practical
engineer it amounts to requiring that our theories ‘add up’; and the easiest way to find that
out is to use a spreadsheet – which is the approach followed in this course. Properties
are derived from data, but then these properties are used to formally compute the soil’s
behaviour in laboratory tests – so checking that what you believe to be your
‘understanding’ is actually consistent with the raw measurements you made.

Traditional views and teaching of advanced soil mechanics divides the subject into pore
water movement (‘consolidation’ theories) and strength (constitutive modelling). This
course only considers strength and stiffness, with pore water being treated as either
drained or undrained; the work hardening theory of plasticity is used, a theory commonly
known as critical state soil mechanics (CSSM) in the case of soils after the title of the book
that gave the first complete exposition of the approach. The models used are Original
Cam Clay and NorSand; these names do not denote a restriction to a soil type but are
best thought of as simply brand names for a particular collection of equations (plasticity
models in essence comprise three simple but simultaneous differential equations).

The course comprises four modules. The first module is about soil behaviour seen in
laboratory tests – there is little theory as such here other than the idea that soil comprises
particles. The second module looks to the ideas about how soil behaviour can be
represented in a computable manner albeit a little idealized. The third module is about a
proper generalization of the theory that works for all soils under all conditions. So far
everything has been focused on laboratory tests where we know the soil’s void ratio and
type – which you likely will not know insitu. The fourth module is about the piezometric
cone penetration test (CPTu) which as well as giving all sorts of stratigraphic information
is the most accurate measurement of insitu soil state and which allows laboratory-based
understanding to be used in practical situations.

Lectures are interwoven with tutorials, as that is the most effective way of learning. You
will need to be familiar with Excel and making graphs within Excel.

Page 4 of 49
CIV5122Z Advanced Soil Mechanics Course August 2019

Analytical methods (FLAC, Plaxis, ABAQUS etc) are not dealt with in this course. Rather,
this course aims to give you the tools to obtain realistic inputs for such analyses.

Software Requirements
The course is going to use Excel for nearly everything. Initially, we will using worksheets
(with some templates provided to save time in setting things up); this is perhaps the easiest
way to appreciate how things like stress-dilatancy come together with the hardening law.
However, when you start using the theory as an engineer you will find that you need to
quickly look at various tests to determine soil properties. And, as we move from ‘learning’
to ‘real soils’, we need to remove a few simplifying assumptions – and which causes very
cluttered, and difficult to read formulae, if you stay within a worksheet. Microsoft has
anticipated this situation and provides a proper programming language, Visual Basic for
Applications (VBA), that lies behind Excel and is accessed from the standard Excel menu
SOFTWARE)REQUIREMENT)for)VGS)WORKSHOP)16917/Jan/2015)
bar.
!
ThePlease!check!that!Excel!on!the!laptop!you!bring!to!the!CSSM!workshop!has!got!VBA!
course will use VBA for soil property determination and CPT processing. This
enabled.!!You!can!check!this!as!shown!below!under!the!Excel!menu:!
software will be provided, as by the time we switch to VBA you will have already
/tools/macro/VBA.!!!
understood how CSSM works. There are books on programming in VBA, but what you
!
need is really only to make yourself familiar with the provided code – and it is all written in
VBA!should!be!there!automatically!with!most!installations!of!Office,!but!just!in!case!
‘plain English’. If you would like to know more about VBA, the book Excel VBA
please!check!and!if!not!there!get!your!systems!folks!to!set!it!up!for!you!before!the!
Programming for the Absolute Beginner by Birnbaum & Vine is good and easily found on
workshop.!!
Amazon. You do not need it for the course, but it may be useful to you in the future as
VBA! is a truly wonderful feature within Excel for engineering.
The!first!day!will!be!done!in!familiar!worksheets,!but!we!need!to!switch!to!VBA!for!
day!two!(how!to!use!VBA!will!be!explained).!
Please check that Excel on the laptop you bring to the CSSM workshop has got VBA
!
enabled. You can check this as shown on the piccy below under the Excel menu.
!

!
VBA should be there automatically with most installations of Office, but just in case please
check and if not there get your systems folks to set it up for you before the course.

Page 5 of 49
CIV5122Z Advanced Soil Mechanics Course August 2019

Notation
Subscripts
c Critical state
i Image condition (occurs when D  0  D  0 )
p p

tc Triaxial compression condition ( = /6)

h Horizontal
v Vertical; volume
0 Initial condition
1, 2 ,3 Principal directions of stress or strain

Superscripts
e Elastic
p Plastic
Dot “.” Denotes increment

Stress Variables (bar over or  denotes effective)

1, 2, 3 [FL-2] Principal stresses
m [FL-2] Mean effective stress  m = ( 1 +  2 +  3 ) / 3
q [FL-2] Deviatoric stress invariant
 q = ( 1 2 ( 1 −  2 ) 2 + 1 2 ( 2 −  3 ) 2 + 1 2 ( 3 −  1 ) 2 )
1
2

p [FL-2] Mean effective stress (=  m )

q [FL-2] Triaxial deviator stress. q= 1 – 3 (=  q )
 [-] Dimensionless shear measure as ratio of stress invariants
 = q / m

 [Rad] Lode angle, sin( 3 ) = − 13.5 1  2  3 /  q3

 [Rad] Included angle between direction of major principal stress
(the “1” direction) and coordinate frame of reference
u [FL-2] Pore pressure

Strain Variables (dot superscript denotes increment)

1, 2, 3 [-] Principal strains (assumed coaxial with principal stresses)
v [-] Volumetric strain v = 1 + 2 + 3

q [-] Shear strain measure work conjugate with  q

(
q = 13 (sin  + 3 cos  )1 − 2 sin  2 + (sin  − 3 cos  )3 )
Dp [-] Plastic dilatancy, as strain rate ratio  v
p
 q
p

Page 6 of 49
CIV5122Z Advanced Soil Mechanics Course August 2019

State Variables
e [-] Void ratio
K0 [-] Geostatic stress ratio, K0 =  h  v
 [-] State parameter,  = e - ec
R [-] Overconsolidation ratio R = p' max p'
 [-] Scaling factor (soil property) for state dilatancy

Elasticity
E [FL-2] Young’s modulus
G [FL-2] Shear modulus
K [FL-2] Bulk (volumetric) modulus
Ir [-] Soil shear rigidity (= G /  m )
 [-] Slope of elastic line in e-ln(m) space
 [-] Poisson’s ratio

Critical State
 [-] Reference void ratio on CSL, defined at p = 1 kPa
 [-] Slope of CSL in e-ln(m) space for semi-log idealization
10 [-] Slope of CSL, but defined on base 10 logarithms
M [-] Critical friction ratio, equals c at the critical state. Varies
with Lode angle, value at triaxial compression (Mtc) taken as
reference.

CPT Parameters and Variables

qt [FL-2] CPT tip resistance after correction for unequal end area
fs [FL-2] CPT friction sleeve stress measurement
uc [FL ]
-2 Pore pressure measured by CPT during sounding at shoulder
location (sometimes denoted as u2 location in the literature).
Q [-] Dimensionless CPT resistance based on vertical stress.
Corresponds to standard usage within in situ testing
community, Q = (qt −  v ,0 ) /  v ,0
Qp [-] Dimensionless CPT resistance based on mean stress,
Q p = (qt − p0 ) / p'0
Bq [-] CPTu excess pore pressure ratio, Bq = (u c − u 0 ) (qt −  v 0 )

F [-] Stress normalised CPT friction ratio, F = f (q −  ) s t v0

k, m [-] Soil and rigidity specific coefficients in equation relating

Qp to 

Page 7 of 49
CIV5122Z Advanced Soil Mechanics Course August 2019

Soil Strength and Dilation

Soil behaviour, across the spectrum from gravels to clays, is controlled by effective stress.
The details of soil behavior are most directly seen in drained tests and which leads to a
preference for understanding the physics using sands simply because drained tests in
clay take a lot of time. Thus, this section largely looks to sand data supplemented by
some data on clays. Undrained response will be dealt with in the following section when
we involve theory.

Soil is Particulate
Scoop up some sand on a beach and look at it in your hand – you can see particles, not
a solid material. If you use a microscope, the nature of the particles is even clearer. If we
move to finer soil the eye is no longer enough and we may resort to a scanning electron
microscope for silts, see Figure 1, and observe a collection of individual particles of
different sizes.

Figure 1: Particle shapes in a SILT seen with a scanning electron microscope

A collection of equal sized spheres sketched on Figure 2 was suggested as a first

idealization of soil more than a century ago (Reynolds, 1885). This idealization leads to
fundamental concepts of how and why soil behaves as it does. If a collection of equal
sized spheres is in its densest packing (body centered cubic, right hand side of Figure 2)
then any distortion of the assembly will cause the assembly to expand: dilatancy.
Conversely, if the collection is in its loosest packing (face centered cubic, left hand side of
Figure 2) then any distortion will allow the particles to pack closer together: also a form of
dilation, albeit in the opposite sense. These are ‘kinematic’ ideas associated with how the
(supposedly rigid) particles must move to accommodate distortion of the assembly, with
no theory as such. If we accept soil is particulate, and approximates spheres, as we hope
all can agree simply from looking at soil (using a microscope if needed), then we must
accept that dilation is going to be inextricably linked to soil behaviour.

Page 8 of 49
CIV5122Z Advanced Soil Mechanics Course August 2019

Figure 2: Reynolds (1885) kinematic explanation of dilation

The packing in either the left or right arrangement is controlled by the bounding spheres;
and it either case distortion causes a change in volume of the assembly

Interlocking and True Friction

Early strength testing used direct shear as the equipment was simple, Figure 3. Example
test results shown on Figure 3 illustrate the familiar Mohr-Coulomb strength criterion. The
interesting question that then follows is the physical nature of the ‘cohesion’ intercept,
since our photographs of soil particles (Figure 1) show there cannot be any bonds between
the particles; so, the ideas in the 1940’s developed primarily at Harvard (Casagrande) and
MIT (Taylor), were that the ‘cohesion’ was related to particle ‘interlocking’ – and
interlocking is the terminology you find in early publications about the nature of soil
strength (eg Taylor, 1948).

Figure 3: Shear box testing and example results (Skempton & Bishop, 1950)

The idea of strength alone is missing half the story. Looking at the results of a direct
shear test make it quickly apparent the Reynold’s idea about volume change during shear
are correct – dense sand samples expand with the top cap rising during shear (note the
micrometer dial in Figure 3). This led to the idea of ‘energy corrections’ since the top cap
rising against the confining stress was clearly doing work – and thus that not all of the
applied shear stress was being resisted by ‘friction’. The corollary was that workers in
the 1940-55 period developed interest in the nature of mobilized friction.

Page 9 of 49
 CIV5122Z Advanced Soil Mechanics Course August 2019

Taylor (1948, Sect. 14.9) provides an example of the importance of interlocking noting that
it contributed 26% of the strength measured in a direct shear test on dense Ottawa sand
(a standard laboratory soil). However, it was Bishop (1950) who first put the mathematics
to the behaviour deriving, using considerations of incremental work, the strength equation
(see Figure 4):
𝜏 𝜏𝑓 𝛿𝑣
= + [1]
𝜎𝑛 𝜎𝑛 𝛿∆

Figure 4: Soil strength related to rate of volume change in direct shear

(Bishop, 1950)

Contribution to Internal friction

strength from
rate of change
of soil volume

Figure 5: Evolution of soil strength components in direct shear (after Bishop. 1950)

May 22, 2019 33

Page 10 of 49
CIV5122Z Advanced Soil Mechanics Course August 2019

As interest was in separating ‘interlocking’ from ‘true friction’, Bishop used equation [1] to
compute f for a loose and a dense direct shear test with the results shown on Figure 5.
Notice how f rapidly develops to a near-constant value, which is rather similar whether
dense or loose. Also notice how the strength component from dilation evolves with shear
strain – the contribution from rate of volume change with shear is not something applying
at just peak strength. Sample expansion increases strength, as expected, but sample
volume decrease decreases the strength; mathematically, the same equation controls and
there is a complete spectrum of behaviour depending on how void ratio evolves. In
essence, this is the behaviour anticipated from idealizing soil as a collection of particles
as per Figure 2.
The maximum value of the stress ratio f/n was called the tangent of the angle of internal
friction; there was a tacit implication that this internal friction was a true soil property unlike
strength (strength depended on how much dilation developed).
These ideas of internal friction and dilation led to the ‘sawtooth’ idealization for soil
strength, Figure 6, where the internal friction is that on the sliding tooth face while dilation
is the tooth angle – a simple way of appreciating the basic behaviour.

Figure 6: Sawtooth model for soil strength

From Direct Shear to Triaxial: Invariants

The direct shear test is mechanically simple but has two deficiencies: it is difficult to control
drainage; and, only half the stresses acting on the soil are measured. So, on one hand
the test is not easy to use for exploring the spectrum of soil behaviour and on the other
hand, even if that limitation is overcome, there are insufficient measurements to know the
true stress state. The triaxial test suffers from neither of these problems and has become
the reference test for soils. The classic reference on the triaxial test is Bishop & Henkel
(1957) and that book still has much of value; however, today it is normal to use computer
control and data acquisition rather than mechanical methods for loading and manually
recorded gauges – have a look at the website www.gdsinstruments.com to see what is
now possible and used in commercial ‘good practice’ testing.

Page 11 of 49
CIV5122Z Advanced Soil Mechanics Course August 2019

As the triaxial test provides complete data on stresses and strains, it is now possible to
improve on our understanding of soil behaviour. Metals deform at constant volume and
confining stress has no effect on their behaviour; soils (= particulate materials) are more
complicated in that their behaviour depends on confining stress (= ‘frictional’) and they
change volume with changes in confining stress and changes in distortion (= dilatancy).
The mechanics of soil behaviour are simplified if we recognize these processes and
choose appropriate stress and strain measures.
The choice of stress and strain measures is also influenced by the principle that, for an
isotropic material with no intrinsic sense of direction, it must not matter if we are looking
from the left or from above when we define our stress and strain measures; this is the
notion that we should use stress and strain invariants that have no sense of direction.
The triaxial test uses cylindrical samples and thus two stresses and two strains are always
equal because of cylindrical symmetry. For the usual situation of standard triaxial
compression loading 2 = 3 and 2 = 3. The stress and strain invariants for this condition
of symmetry are shown on Table 1; note that all stresses are ‘effective’.

Table 1: Stress and strain invariants for triaxial compression

Distortion Volume Change

Stress q = 1 - 3 p’ = (1 +  3) / 3
Strain q = 2/3 (1 - 3) v = 1 +  3

The ‘2/3’ factor appearing in the deviatoric strain invariant q is there to make the stress
and strain invariants work conjugate, which is just a fancy phase that the incremental work
W on the soil being given by:

𝛿𝑊 = 𝑞 𝛿𝜀𝑞 + 𝑝′ 𝛿𝜀𝑣 [2]

As a further wrinkle, note that the ‘compression positive’ convention of soil mechanics
makes void ratio reduction correspond to positive volumetric strain: v = - e/ (1+e).
The invariants shown on Table 1 generalize to 3D, but triaxial compression is sufficient for
a practical engineer to understand soil behaviour and so we restrict this course to triaxial
compression. These invariants are sometimes called ‘Cambridge stress variables’ (or
equivalent phrase) in the literature, but that is a wrong attribution – these invariants were
proposed as part of the mathematical theory of plasticity which much predates their
adoption for soil mechanics.

Triaxial Compression Strength

The Taylor/Bishop strength relation developed considering direct shear, Equation [1],
generalizes to triaxial compression. We do this by changing our parameters from those
of a direct shear test to the invariants defined on Table 1 so that:

Page 12 of 49
CIV5122Z Advanced Soil Mechanics Course August 2019

=M–D [3]
…where, by comparison with, [1]:

 = q / p’ is used instead of /n [3a]

D = v / q is used instead of v/ [3b]

This transformation in [3] is not quite enough. Figure 7 shows the results from many
drained triaxial tests on Brasted sand, with the tests ranging from loose to dense, at peak
strength (ie plotting max vs D at max).

1.8
Brasted Sand
(data from Cornforth, 1961)

(1-N)

1.6
max

1.4

Mtc

SandDilation_r10.xls
1.2
0 -0.2 -0.4 -0.6 -0.8
Dmin

Figure 7: Drained strength of a sand in triaxial compression

Loose tests lie to the left of Figure 7 and show little dilation while those on the right are
dense and dilate strongly. As expected from the Taylor-Bishop idea of two strength
components a straight line is a good representation of the trends in this test data. Brasted
sand is not unusual, and similar results to Figure 7 will be found with all other sands.
However, there is a deviation from the Taylor-Bishop idea in that the trend line does not
have a one-to-one slope, being fitted by the equation:

𝜂𝑚𝑎𝑥 = 𝑀𝑡𝑐 − (1 − 𝑁) 𝐷𝑚𝑖𝑛 [4]

Maximum dilation, in the sense of maximum volume expansion, is actually Dmin because
of the compression positive convention of soil mechanics. Hence max occurs at Dmin. The
terms Mtc and N in equation [4] are soil properties. As D=0 at Mtc, Mtc corresponds to the
Taylor-Bishop internal friction; N represents the proportion of work going to volumetric,
rather than distortional, strain. Equation [4] was first stated by Nova (1982).

Page 13 of 49
CIV5122Z Advanced Soil Mechanics Course August 2019

The behaviour shown on Figure 7 was for a sand; there is nothing unusual about that sand
and others sands show similar behaviour (although the properties Mtc, N differ from one
sand to another). But this effect of dilation on strength has no knowledge of geology – it
derives simply from considerations of work being done; thus we would expect it to apply
to clays just as well as sands. Drained tests on clay take a long time, and are not routine
in normal engineering. But drained tests on clay have been used for research, in particular
during the 1950’s at Imperial College; Figure 8 shows data from tests on Weald Clay
(Parry, 1957). Exactly the same behaviour is seen on Figure 8 as Figure 7.

1.2
Weald CLAY
(data from Parry, 1957)

(1-N)
1.0
max

Mtc
0.8

SandDilation_r10.xls
0.6
0.2 0.0 -0.2 -0.4
Dmin

Figure 8: Drained strength of a clay in triaxial compression

Stress-Dilatancy
Taylor and Bishop’s ideas on two-components to soil strength developed from the context
of direct shear (shear-box) tests where the work components are obvious from the
arrangement of the test. But there is more to this work-based idea than just strength.
Looking back at Figure 5, it is obvious that the mobilized shear stress, throughout the
entire stress-strain behaviour, is strongly influenced by the evolution of the rate of change
of void ratio (dilatancy). This further development of the two-component idealization is
easiest seen if we take test data and plot the mobilized stress ratio  versus the current
dilatancy D.
Figure 9 shows two drained triaxial compression tests, one on dense sand and one on
loose sand. The measured data is shown on the left-hand side, and there is nothing

Page 14 of 49
CIV5122Z Advanced Soil Mechanics Course August 2019

unusual about these tests. The right-hand side shows the same data, but now
transformed using equation [3]. As can be seen, what looks at first glance to be very
different behaviour of the loose to the dense soil is actually a near common response of
mobilized stress to dilation: stress-dilatancy as the behaviour is known. Really, this takes
us back to Figure 2 – any deformation involves dilation, but it takes imposed stress to
cause that deformation. The coupling between ratio of deviatoric stress to mean stress
with dilatancy (= rate of dilation) is intrinsic to particulate materials, including soils whether
gravels, sands, silts, or clays (or any mixture of these); it is a direct consequence of
dilatancy being a work transfer mechanism, exactly as identified by Taylor-Bishop in their
assessment of how strengths developed in a shear box.

1.8
1.6
deviatoric stress ratio, q/p

1.4
OCC stress-dilatancy 1.8

Stress ra o h
1.2
1 1.6
Stress-dilatancy from Taylor-Bishop
0.8 dissipa on of plas c work
1.4
0.6 STATE-DILATANCY
0.4 determines how far 1.2
stress-dilatancy can develop
0.2 Mf 1
(this test)
0
0 5 10 15
Equation
20
[3] 0.8
axial strain: %
4 0.6 DENSE

2
0.4 LOOSE
volumetric strain: %

0
0 5 10 15 20 0.2
-2
0
-4
-0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8
-6 Dilatancy, D

-8

Figure 9: Drained triaxial compression data transformed to stress-dilatancy

Critical Void Ratio

It is apparent that the very loose packing of particles sketched on Figure 2 will contract
when distorted while the alternative a very dense packing will expand – the change of
volume with distortion will depend on the void ratio. Of course, soils comprise a distribution
of particle sizes but that does not change the basic insight from this simple idealization.
And, it is no great leap of the imagination to think that there may be a void ratio between
these two limits where soil deforms at constant void ratio.
Hydraulic fill dams were a common construction method during the late 1800’s to early
1900’s, an era when there were pumps but not earth moving machinery that we have
today; hydraulic filling was perceived as an inexpensive dam construction method.
However, these hydraulic-fill dams had a propensity to suddenly fail by liquefaction slides

Page 15 of 49
CIV5122Z Advanced Soil Mechanics Course August 2019

during construction (much as is a current problem with tailings dams). It is obvious that if
the fill is saturated (inevitable with hydraulic fills) then contractive response with cause
positive excess pore pressure. This understanding was reflected in engineering the first
liquefaction-resistant dam at Franklin Falls (New Hampshire) by the Corp of Engineers,
directed by Lyman (1938), although more widely associated with Casagrande (1936).
The canonical result of the testing is shown on Figure 10 as the evolution of sand strength
in a shear box with strain and its dependence on void ratio. A common end point was
found, denoted as the critical void ratio which was the balance between some soil particles
moving apart while others were falling into void space as the soil was deformed (in
essence a dynamic situation of birth-death processes in terms of the contacts between
the particles). And “critical” really meant what it said – it was the practical criterion of a
safe density in constructed engineering work, with the practical concern to avoid sudden
transitioning of drained construction with no excess pore pressure into an undrained
liquefaction failure.

Figure 10: Origin and meaning of the critical void ratio, Casagrande (1936)
(author’s emphasis)

Page 16 of 49
CIV5122Z Advanced Soil Mechanics Course August 2019

The critical state highlighted on Figure 10 comprises two conditions: void ratio change
with strain is zero; and that absence of void ratio change continues indefinitely. These
ideas are best expressed using dilatancy D as:
D=0 [5a]

D / q = 0 [5b]
If only condition [5a] is met, for example if you look closely at the volumetric strain behavior
of the dense test shown on Figure 9 D=0 occurs at about 0.5% axial strain (when the
volumetric strain trend reverses), the soil can be far from the critical state. D=0 can be,
and usually is, a transient condition. In undrained tests, D=0 corresponds to where the
stress-path reverses direction in those tests showing an ‘S’ shaped path, commonly called
the ‘pseudo steady-state’ or ‘phase-change’ or ‘quasi steady-state’. Relying on [5b] is
crucial to getting the critical state correct.
Triaxial testing is used to determine the critical void ratio in sands, silts, and clays. It is
much easier to work with loose (contracting) samples because dense samples tend to
form shear-bands (‘localize’) and where the void ratio of the overall sample does not reflect
the void ratio in the intense shear zone. Both drained and undrained tests are used.
Undrained triaxial tests on loose soils often reach their critical state at less than 10% axial
strain and well within the limits of the equipment; Figure 11 shows an example. Notice
how the sample continues to deform at constant p.q in the critical state.
350 350
300 300
deviator stress: kPa

250 250

200 200
critical state
150 150

100 100 critical state

50 50

0 0
0 5 10 15 20 0 100 200 300 400 500 600
axial strain: % mean effective stress: kPa

Figure 11: Example of critical state in load-controlled undrained triaxial compression

Casagrande appears to have viewed the critical void ratio (ec) as a constant, akin to emin
and emax; but that was quickly found to be wrong. Fort Peck dam failed shortly after the
work by the Corps at Franklin Falls. Taylor (working at MIT, just down the road from
Harvard in Cambridge, Massachusetts) applied Casagrande’s ideas about the critical void
ratio to the investigation of Fort Peck and found that the critical void ratio depended on the
effective confining stress (see Fig 14.12 of Taylor, 1948) – a dependence of critical void
ratio on mean effective stress that today is known as the critical state locus (CSL).
Taylor’s identification of the critical void ratio was not that used today because he focused

Page 17 of 49
CIV5122Z Advanced Soil Mechanics Course August 2019

on zero volumetric strain at peak strength rather than zero dilation rate, but this only
causes an offset in the estimate of the CSL and does not negate the importance of
identifying Casagrande’s error.
However, most investigation of this dependence on stress level was in the UK rather than
the USA, and was for clays not sands; part of the reason for this change in emphasis is
that void ratio (at least at that time) was a lot easier to measure with precision in clay
samples than in sand. Various workers contributed to the testing, with Parry (1958) giving
a very good synthesis and which supported a unique CSL, regardless of whether the tests
were drained or undrained: Figure 12. Parry’s results can be expressed as:
𝑒𝑐 = Γ − 𝜆10 log(𝑝𝑐 ′ ) [6]

…where ec is the critical void ratio and  10 are the soil properties (the subscript ‘10’
denotes the logarithm base). Parry also observed that the CSL and isotropic normal
compression line were parallel in the tested (remolded) clays; a corollary of which is that
10 = Cc where Cc is the familiar compression index.
This semi-log form of the CSL given by Equation [6] is usually adequate for engineering,
in sands, silts and clays, over the stress range 30 kPa < p’ < 800 kPa. Outside these
limits, an upper void ratio limit may be needed on one hand while on the other recognition
of constant compressibility may be appropriate at high confining stress; Verdugo (1992)
provides insightful comment. Neither of these factors in any way affects the validity of the
CSL as a limit condition in shear – rather, it is a matter of what level of detail is wanted in
the representation of the CSL.
Leaping a little ahead of the historical story, accurate determination of the CSL in sands
developed at Harvard in the late 1960’s (Castro, 1969) and gained considerable ease and
accuracy in the 1980’s with the development of post-test sample freezing (in Calgary, see
Been et al 1991). Nowadays, accurate CSL determination can be done by any good soil
mechanics laboratory – although engineering judgment is needed in the data evaluation.
An example of a CSL in sand determined with modern methods is shown on Figure 13;
the ‘blue dots’ on this figure are the identified critical state by inspection of the stress-strain
curve in undrained tests while the loose drained tests do not quite get to the CSL within
the limits of the equipment (and thus the CSL is inferred as a little lower void ratio than
attained in these loose drained tests). Overall, the CSL shows a power-law, rather than
semi-log, form; this kind of power-law seems most common for sands formed with a
significant fraction of non-quartz particles. Grain-crushing may be a further contributor.
There has also been much confusion about ‘uniqueness’ of the CSL. This generally
(invariably?) derives from experimental workers failing to understand that the critical state
is not a transient condition (commonly called the pseudo-critical state). More formally,
uniqueness simply means that any combination of the three effective principal stresses
(1, 2, 3) produces a single value for the void ratio; physical sense further requires that
increasing mean stress implies decreasing critical void ratio.

Page 18 of 49
CIV5122Z Advanced Soil Mechanics Course August 2019

Figure 12: Rates of changes in dilation (drained tests) and pore pressure (undrained
tests) at peak strength. Loose contractive samples are plotted in (a) and dense dilatant
samples in (b). The bounding line in each pair of plots is common, and inferred to be the
CSL. From Parry (1958).

1.05

CSL
0.95

0.85
void ratio

0.75

0.65

0.55
1 10 100 1000 10000
mean effective stress, p': kPa
Figure 13: Example of a CSL in sand determined using modern methods
(note the departure of the CSL from the semi-log form shown as dashed line)

Page 19 of 49
CIV5122Z Advanced Soil Mechanics Course August 2019

State-Dilatancy
Taylor and Bishop’s ideas on soil strength having two components provides an excellent
understanding of soil behaviour but, looking at Figures 7 and 8, begs the question: why
does a particular value of Dmin develop? There is a missing component.
The starting point to progress is to look at Figure 13, and the various tests shown on it.
Notice that the tests can start anywhere, both in terms of initial stress and initial void ratio;
let us call the combination of e,p at starting point state ‘A’. Wherever we start, we end up
on the CSL (the green line); where on the green line will depend on the loading path, but
let us call the end point e,p state ‘B’. In physics, if you are going from state ‘A’ to state ‘B’
then you expect the speed of movement to be proportional to the distance from the end
state. You will have encountered this idea in high-school physics when introduced to
radioactive decay, and you will know the same idea in geotechnical engineering from a
falling head hydraulic test (laboratory or insitu); in mathematical terms, both processes are
described by the first-order rate equation (check your undergraduate math text…). Soil
mechanical behaviour is no different.
A measure of deviation from the end state is needed. The end state is the critical void
ratio and, to keep things simple, we define that deviation measure at constant mean stress:

𝜓 = 𝑒 − 𝑒𝑐 [7]

… where  is the soil’s state parameter, the name reflecting that any soil exists over a
spectrum of void ratios and we need a state measure analogous to temperature of a gas.
Notice that the state parameter is not affected by the particular shape of the CSL or the
equation used to describe that experimentally determined shape.
That the critical void ratio is the endpoint of all loading paths is then given by requiring that
the deviation measure decreases with distortion:
𝑑𝜓 = −𝑔(𝜓) 𝑑𝜀𝑞 [8]

Physical sense requires that the function g() be ‘single-valued’ and return positive values
for positive , with the negative sign emphasizing that the deviation from the end state is
to be reduced with distortional strain. A little bit of mathematics leads to the elegant state-
dilatancy relation:

𝐷𝑚𝑖𝑛 = 𝛸 𝜓 [9]

…where  is a scaling coefficient and is a property of the soil (in critical state soil
mechanics it is a convention that all properties are upper-case Greek font; here the
property is upper-case ‘chi’). Putting equation [9] in [4] gives soil strength in terms of its
void ratio and the confining stress.
Does the mathematics work? Figure 14 shows some test data, and it is immediately
apparent that [9] is a very good representation of what was measured.

Page 20 of 49
CIV5122Z Advanced Soil Mechanics Course August 2019

-0.7
SandDilation_r10.xls

Brasted Sand
-0.6 (data from Cornforth, 1961)

-0.5

-0.4

Dmin

-0.3

-0.2

-0.1

0
-0.25 -0.2 -0.15 -0.1 -0.05 0

state parameter  at Dmin

Figure 14: State-dilatancy of Brasted sand

One caution. The widely-cited Been & Jefferies (1985) was a ‘laboratory paper’ that
synthesized a large body of triaxial test and, as such, it was natural to work with the sample
void ratios at the start of loading – all very convenient, and requiring no assumptions or
models. But, in terms of mechanics, actually not acceptable. Thus, Figure 14 is given in
terms of the state parameter  at the instant of Dmin – a bit of data processing is needed
for proper mechanics.

Page 21 of 49
CIV5122Z Advanced Soil Mechanics Course August 2019

Exercise 1
Apply numerical differentiation to the supplied example data file to:
a) Recover Figure 9
b) Then invert the results to find the equivalent of Figure 5 (the Taylor-Bishop
strength model) for the supplied triaxial data

Exercise 2
Using the supplied triaxial data on Nerlerk sand
a) Find Dmin and max for each test and plot these results to determine the
properties M, N
b) Determine the CSL of Nerlerk sand using the undrained tests to give the soil
properties  
c) Using this CSL, determine the dilation coefficient .

Page 22 of 49
CIV5122Z Advanced Soil Mechanics Course August 2019

Elasticity
Elasticity is, possibly surprisingly, essential to understand how undrained strengths
and stiffness develop. Because elasticity is common to everything that follows, it is
helpful to present it now before dealing with plasticity. Elastic behavior used to be
difficult to measure (requiring very precise transducers) but the advent of geophysical
methods, both in the laboratory and insitu, has made elastic measurement routine in
geotechnical practice.

Isotropic Moduli
Soil is generally anisotropic with, say, its stiffness in a horizontal loading direction being
greater than for vertical loading. However, anisotropy is somewhat secondary to such
basic effects as the influence of void ratio; and there is the further point that there is no
point in adding the complexity of anisotropy before you have an isotropic idealization
working. Plus, the rather important practical point that we have enough trouble measuring
soil properties without adding the additional requirement that we must measure in every
direction as well. Geotechnical engineering practice is dominated by isotropic elasticity
and reasonably so.
You will have encountered isotropic elasticity as part of your engineering education, in
particular with the two elastic constants of Young’s Modulus (E) and Poisson’s Ratio ().
In the case of soils, it is very helpful to separate their behaviour into distortional and
volumetric aspects; doing so makes it at least convenient to represent elasticity in terms
of a shear modulus (G) and a bulk modulus (K). The relation between these alternative
representations is:

𝐺 = 𝐸/ (2(1 + 𝜈)) [10a]

𝐾 = 𝐸/ (3(1 − 2𝜈)) [10b]

Elastic Strains
Because we are using work conjugate strain invariants, it is convenient to also have elastic
strain increments in terms of these invariants. These are (for triaxial stress invariants p,
q):
Δ𝜀𝑣𝑒 = Δ𝑝′ /𝐾 [11a]
…and
Δ𝜀𝑞𝑒 = Δ𝑞/3𝐺 [11b]

Note the “3G” in [11b] which arises because of the definition of this invariant and which is
not the “2G” you might have guessed from this being a shear strain measure.

Page 23 of 49
CIV5122Z Advanced Soil Mechanics Course August 2019

Geophysical Measurement
The travel velocity of elastic waves is readily measured geophysically: the travel time of
an identifiable signal is recorded between a source and a receiver and converted to a
velocity using the length of the estimated ‘ray path’ for the elastic wave. The first-arrival
is the ‘P’ (=compression) wave; the ‘S’ (= shear) wave travels more slowly, but is polarized
– so ‘polarity reversal’ techniques are used to identify the S wave within other arrivals.
Thus, any measurement of S-wave velocity involves two measurements. The modern
technique is to make each of these ‘two’ measurements multiple times and ‘stack’ (= add)
the signals, as that enhances the true signal over background noise.
The P-wave has limited value in soils as it is dominated by wave propagation through the
pore water of the soil (if the soil is saturated, which is usually the case). But, the S-wave
is now widely used because the S-wave velocity (Vs) is unaffected by soil saturation and
directly related to the elastic shear modulus (Gmax):

𝐺𝑚𝑎𝑥 = 𝜌 𝑉𝑠 2 [12]
…where the ‘max’ subscript denotes that this is the limiting value for elastic shear
modulus. Common experience is that this geophysical shear stiffness is too stiff in terms
of soil behaviour at strain levels of interest to geotechnical engineers. A common ‘rule-of-
thumb’ is that the G for geotechnical applications is:

𝐺 = 𝐺𝑚𝑎𝑥 /3 [13]
Evaluate this rule-of-thumb for yourself during the tutorials.
The geophysical techniques used include vertical seismic profiling (VSP) or enhanced
CPT, both insitu, or ‘bender elements’ in the laboratory as an add-on to normal triaxial
testing. All techniques are cheap, giving a lot of data for very few $.

Relationship Between G and K

Most (all?) soils show a near constant Poisson’s ratio () even as the elastic moduli
change with stress level. And, Poisson’s ratio tends to have a rather limited range, say
0.15 <  < 0.25 if the slightly anisotropic moduli of the soil are approximated with an
isotropic idealization. This makes it very convenient to relate the bulk (K) and shear
modulus of the soil (G) through:

𝐾 = 𝐺(1 + 𝜈)/ (1.5 − 3𝜈) [14]

… and with the same dependence on stress level and void ratio between the two elastic
moduli.

Effect of Stress and Void Ratio on Elastic Moduli

Our starting idealization of soil as a collection of spherical particles immediately suggests
starting with the ‘classic’ Hertz (1882) theory of deforming contacts between elastic

Page 24 of 49
CIV5122Z Advanced Soil Mechanics Course August 2019

spheres. This theory indicates that K should depend on pressure with an exponent of one
third. Equally, there must be an effect of void ratio, as denser packings have more
contacts. Some data, predominantly on silt tailings, illustrating these aspects is shown on
Figure 15. Sands show comparable behavior.

300
Data & Test Method:
Dense Silts
(I) = insitu using SCPT
(B) = laboratory "bender" (e ~ 0.7)
NCS (B)
NCP (B)
CQD (I)
elastic shear modulus, G max: MPa

EKO (B)
200
RoC (I)
NTSF N04 (I) Loose Silts
NTSF TC1 (B) (e ~ 0.9)
LU-Sa1 (B)
LU-Sa2 (B)

100

Gmax_silts_r2.xlsx
0
0 100 200 300 400 500 600
mean effective stress, p': kPa

Figure 15: Examples of measured trends for Gmax in silts

Many workers have looked into how void ratio and stress level affects G, with various
equations suggested to represent the two effects. An acceptable relation, and which works
well across a range of soils, is:
𝑏
𝐴 𝑝
𝐺 = (𝑝 ) [15]
𝑒−𝑒𝑚𝑖𝑛 𝑟𝑒𝑓

…where pref = 100 kPa by convention (this is done to avoid A having odd units). A, emin,
b are all soil properties: A is a normalized modulus in the same units as G (typically MPa);
emin amounts to the transition void ratio from ‘particulate’ to ‘rock’ and is somewhat denser
than measured using the ASTM procedure (try emin about 0.1 denser than that measured
in a maximum density test); and, b is a modified form of the Hertz exponent, being typically
about double the theoretical value of one-third.

Page 25 of 49
CIV5122Z Advanced Soil Mechanics Course August 2019

Plasticity Theory
So far, everything has followed from the idea that the grains of soil you can see in your
hand (or under a microscope) can be represented as essentially rigid particles that
interact with each other as the soil mass is deformed. This idea has led to both stress-
dilatancy (which gives detail about the stress-strain behaviour) and state-dilatancy
(which gives the current strength). But there is still one important missing factor:
most soil deformations are irrecoverable most of the time. Understanding soil needs
adding this irrecoverable aspect to the stress and state dilatancy: plasticity theory.
There are always four elements to plasticity theory: yielding; plastic flow; work
hardening; and the ‘consistency condition’. These elements are discussed in this
section.

Soil is Plastic
Undergraduate soil mechanics courses always involve some shear tests, possibly a
simple shear box or perhaps a triaxial test. Loading tests on soil, whether clay or sand,
will show that most deformation is not recovered when the load is removed. Figure 16
illustrates an example of a triaxial test, and defines the irrecoverable and recoverable
components that are known as plastic and elastic deformation respectively. You can do
your own tests easily enough and you will always find behavior similar to Figure 16; never
doubt that plastic strains are an exceedingly important part of soil behavior.

1200

1000

800
deviator stress, q: kPa

600

400
elastic-plasic strain decomposition:
 = e + p
200
plastic elastic

0
0 2 4 6 8 10 12 14 16 18 20
axial strain, 1: %

Figure 16: Occurrence and definition of plastic strain (dense Erksak sand, drained)

Page 26 of 49
CIV5122Z Advanced Soil Mechanics Course August 2019

Plastic behaviour is not a recent subject, having been studied for some 150 years. Why
so much study ? Two basic reasons. First, because plastic behaviour is so common –
and in the case of metals the basis of many industrial processes (for example, designing
the form for stamping the sheet metal to make your car). Second, because elasticity gives
a one-for-one relation between stress and strain whereas plasticity is very different,
plasticity often coming down to strain indicating stress but not the opposite – you cannot
substitute plastic modulus for an elastic one; a new theory is needed.
The separation of displacements (and strains) into the two components illustrated on
Figure 16 is fundamental to representing the mechanical behaviour of materials at the
macro scale; this separation is known as the elastic-plastic strain decomposition and
applies to any strain increment d (i.e. be the strain principal, in a particular direction, or
an invariant):
𝑑𝜀 = 𝑑𝜀 𝑒 + 𝑑𝜀 𝑝 [16]
…with the superscripts denoting elastic and plastic. Equation [16] is universal to plasticity
theories whether for soil, steel, copper etc. Elastic strains are familiar, being the starting
concept of every undergraduate civil or mechanical engineering course. Plastic strains
are perhaps less familiar and connected to the idea of ‘yield’, the stress conditions that
allow plastic strains to arise.

Yielding and Yield Surfaces

Coulomb (1733) identified yielding as an important behaviour in soil, but it was Tresca
(1864) who carried out experiments on the punching and extrusion of metals and first
formalized a general yield condition. Subsequently, various workers followed two routes
although restricted to metals. Some investigated the yield condition with various
experiments. Others invoked physical or mathematical ideas, for example the Von Mises
yield criterion corresponds to a limiting value for the elastic shear strain energy.
Regardless of how derived/investigated, a yield surface is simply the limit of a region in
‘stress space’ within which all stress increments produce only an elastic response; you
can choose your stress space to be expressed by principal stresses, but it is generally
more useful to show (and express) the yield surface using stress invariants. The region
of elastic response is defined by a continuous curve: the yield surface.
Overall, by about 1935 there was a complete theory of plasticity for metals; in this context,
what is meant by ‘theory’ is a complete framework (set of equations) that broadly matched
experimental observations. This plasticity theory does not explain the underlying
behaviour of the metal that causes plasticity, which is actually the movement of
dislocations within the metal. Plasticity theory is an abstraction for representing the effect
of underlying processes; it gains a fundamental slant through invoking the Second Law of
Thermodynamics in considering plastic work (as plastic strains are permanent, work is
done and dissipated as heat – bend a piece of metal repeatedly and feel it warm up to the
touch).

Page 27 of 49
CIV5122Z Advanced Soil Mechanics Course August 2019

Although Coulomb identified plasticity as intrinsic to soil behaviour nearly three hundred
years ago, not much developed in terms of applying plasticity theory to soil until the 1950’s.
Then, much like metals, developments with soils included both experimental work and
theoretical studies. However, both approaches were complicated by two aspects where
soils differ from metals: soil yielding depends on the mean stress (which has no effect in
metals at all) and soil generally changes volume during plastic distortion (while metals
yield at constant volume). With the exception of the undrained strength of clay as
represented in a total stress paradigm (“=0” and so forth), which reasonably matches
Tresca plasticity if the soil is not brittle, something more than just adopting metal plasticity
was needed – which should not be a surprise as soil plasticity is about realigning particle
contacts and void space change rather than the movement of dislocations between
crystals in metals.

1.4

1.2

qp
qp

Stresses here
1 vp
define yield
shear stress, q

0.8
surface ‘size’ Yield
surface
0.6
Only elastic
0.4 response within
yield surface
0.2

0
0.0 0.5 1.0 1.5 2.0 2.5 3.0
mean effective stress, p v p

Figure 17: Example of a yield surface (shown using invariants to define the space)
8

Getting slightly ahead of things, Figure 17 shows an example of a yield surface for soil;
we will derive this yield surface later in these notes. Yield surfaces are expressed by
equations, the equations being in terms of the stress system adopted (principal or
invariant) and one or more parameters characterizing the size and shape of the yield
surface in terms of the chosen stress system – think of the yield surface as defining the
‘strength’ of the soil.

Page 28 of 49
CIV5122Z Advanced Soil Mechanics Course August 2019

Classically, all the terms defining the yield surface are bundled into one side of an
equation. Thus, the general form of the yield surface is:
F = f(1, 2, 3, ‘strength parameter’, ‘shape parameter’,…) [17]

…where with this arrangement:

𝐹 < 0 ⇒ 𝑒𝑙𝑎𝑠𝑡𝑖𝑐 [17a]
𝐹 = 0 ⇒ 𝑝𝑙𝑎𝑠𝑡𝑖𝑐 [17b]

Strain increments within a yield surface are always purely elastic, but that begs the
question: what happens on the yield surface ? By definition, plastic strains develop on the
yield surface, but in which direction and to what extent? There are two ideas here, which
are known as, respectively, the flowrule and the hardening law (roughly equivalent to
Poisson’s ratio and Young’s modulus in elastic theory).

Flowrule
Since plastic strains only develop on the yield surface, it is natural to indicate the relative
amounts of those strains in a diagram that includes the yield surface. This is done by
plotting both stresses and strain increments on the same axes – see Figure 17. You can
use 1, 1 etc but the combinations of stress and strain increment used must be work
conjugate as explained earlier in these notes; Figure 17 uses the work conjugate
invariants for a triaxial test.
In Figure 17, the strain increment vector defines dilatancy; the basic behavior of soil seen
in experiments simply translates to a vector in stress space. At this point we run into an
important idea (actually, a theorem; Drucker, 1954) which is that, for work conjugate stress
and strain increment measures, the strain increment vector will be perpendicular to the
yield surface: normality (also called an associated flow rule). This is the concept shown
on Figure 17.
At this point we know the relative amounts of plastic strain, but not their magnitude. We
need to consider how the yield surface responds to these plastic strain increments.

Hardening
Plastic theory may seem a bit of a steep learning curve at this stage, but you will already
be familiar with some aspects and what you are familiar with is now important. Think of
the pre-consolidation pressure during oedometer compression of a clay sample, which
pressure is no more than the yield stress for a particular loading path. And, as you
accumulate plastic strain (‘normal consolidation’) then the pre-consolidation pressure
increases: plastic work hardening. This behavior is generalized from the familiar
oedometer test to the behavior of yield surfaces.

Page 29 of 49
CIV5122Z Advanced Soil Mechanics Course August 2019

Yield surfaces generally evolve, expanding or shrinking with plastic strain. The way this
works can be seen on the drained triaxial test shown earlier on Figure 16. Consider the
first (left hand) unload-reload loop; clearly most strain is not recovered, so the stress-state
at the start of unloading is on the yield surface. Now consider the reloading. Initially,
everything is elastic until the yield surface is met (as per Figure 16) and at which point
plastic strains restart. But, notice that the stress-strain curve then shows increasing
strength with those plastic strains – this is the general form of hardening that you are
familiar with in an oedometer but now carried across into 3D. What is now needed is to
codify how yield surfaces change size with plastic strain: the hardening law.
The hardening law is as fundamental to plasticity as the yield surface. If the yield surface
always has the same shape, and that is the simplest starting point, we can specify the
yield surface size by a single parameter and there are two obvious choices to do this: the
mean stress at maximum deviator (pqmax) or the intersection of the yield surface with the
mean stress axis (= the pre-consolidation pressure under isotropic compression). Which
of the two options should be chosen ?
A tacit view in the ideas sketched out so far is that plasticity is about large deformations
and with the concept of ‘strength’ in the background. Conversely, oedometer (or isotropic)
compression is all about confined stiffness. So, it is no surprise that pqmax became the
preferred characterization of yield surface size for soils. And the linkage to the critical
state is then immediate: because pqmax D = 0, state dilatancy – a core concept from our
particulate idealization – becomes a natural control on yield surface evolution. More
formally as D = 0 is one of the conditions for the critical state we want to make this link
clear – so, we refer to this scaling stress as the ‘image’ condition as in a way it is like
seeing an appearance of the critical state but we are generally not there yet; this image
stress is denoted as pimg. The hardening law will then have the general form:

𝑑𝑝𝑖𝑚𝑔 = ℎ(𝜓, 𝜎𝑚 , 𝜎𝑞 , 𝑝𝑞𝑚𝑎𝑥 , 𝑝𝑟𝑜𝑝𝑒𝑟𝑡𝑖𝑒𝑠) 𝑑𝜀 𝑝 [18]

…where h() is the hardening function that will be specific to a particular idealization (or
model). Notice the hardening law is written in terms of plastic strain, consistent with the
theoretical framework. If you want a yield surface to change shape as well as size, then
more measures are needed; for example, if the yiled surface was taken as an elipse we
would need to have a property specify the ratio of major to minor axes as well as the
measure of overall size. In reality, we can represent a great deal of soil behaviour just
using a simple (= constant shape) yield surface like that illustrated in Figure 17 and there
is no reason to add unnecessary complexity.
If you move the plastic strain increment from the left-hand side of [18] to the right-hand
side, then h() can be viewed as a plastic stiffness or modulus. This is true of all work-
hardening plasticity: the theory gives you a current stiffness, not a stress-strain curve.
This stiffness has to be integrated in our stress analysis to get the overall behaviour we
are interested in – which is not difficult, and we will come to that shortly after dealing with
one more aspect of plasticity theory.

Page 30 of 49
CIV5122Z Advanced Soil Mechanics Course August 2019

Consistency Condition
Work hardening (and softening) plastic models change the size of their yield surface with
plastic strain. The consistency condition is simply that the stress state must remain on the
yield surface during plastic strain, so that the stress state evolves on a one-for-one basis
with the evolution of yield surface size. For the universal description of yielding as F=0,
the consistency condition is just:
dF = 0 [19]
The consistency condition is illustrated in Figure 18, which shows an initial yield surface
that has hardened after an increment of plastic strain. The question then is: how the stress
state has evolved? We could try working out the intersection of the imposed loading path
with the hardened yield surface – but, this is both tedious for an element test when we
know the path and essentially impossible in general as it would involve solving
simultaneous equations at every integration point. So, conventionally, [19] is exploited
with the yield surface differentiated to provide a further equation that is applied
incrementally. Don’t worry, this is simpler than it sounds as will be seen shortly.

Figure 18: Illustration of the consistency condition

(triaxial drained and undrained paths shown)

Page 31 of 49
CIV5122Z Advanced Soil Mechanics Course August 2019

A Numerical World
Some aspects of soil behavior lend themselves to being described by simple equations,
but most generally we need to work with numerical methods. This simply is not an issue
as all engineers use spreadsheets today rather than calculation pads and sliderules that
dominated our calculations fifty years ago – and your starting data, whether oedometer,
triaxial or CPT – will be digital. But there is a little more to things than ‘just use a
spreadsheet’. First, we need to use spreadsheets properly. Second, we need to use
numerical integration of the governing equations for soil behavior. Neither aspect is
difficult, and the needed techniques are discussed in this section.

Visual Basic for Applications

Spreadsheets (and here we take Excel as the default program) are now widely used in
geotechnical engineering and are ubiquitous for assembling and presenting the results of
laboratory tests. There will be differences between companies as to what is the house
style for graphs and the calculations, but most companies and academic faculty do a
terrible job in their use of spreadsheets: in our experience (and we have reviewed a lot of
spreadsheets) there widespread trend to put far, far too much on the worksheets with
complicated formulae using cell references that are almost impossible to read and
understand. It is like going back to the very early days of computing when we had to
program an Apple II or Sinclair Spectrum and very simple names and expressions were
all that were allowed. But, that is why computer scientists developed ‘high-level’
languages such as Fortran and which offer two key features for engineers: i) variables that
can be written using plain-English and descriptive names (eg we can use ‘lambda10’
rather than ‘$F$21’); and ii) arrays with the associated looping constructs.
Engineering has seen a range of high-level languages that changed (or rather become
obsolete) rather quickly. Our problem as engineers is that our industry has minimal
financial importance to the wider market that supports software development (just think of
the vast intellectual effort going into ‘apps’ for your cellphone); a second problem is that
the common expectation of everything having a graphical interface means the
programming environment is changing as fast as new chips and operating systems evolve
– yet all we want as engineers is the ability to calculate in a clear and structured way, and
we would like to be isolated from changing fashions as to how things should look (which
take time to fix when we would be rather thinking about our engineering). Those of us of
a certain age just loved the language QB45 that came soon after the IBM-PC became the
dominant computing environment in our offices; but QB45 died with Windows. However,
Microsoft has not left us completely in the lurch and a wonderful language comes
incorporated with their Office suite: Visual Basic for Applications (VBA).
VBA is a programming environment that works with the Microsoft Office applications, and
here we are interested in Excel so we limit ourselves to that. Components (worksheets,
graphs etc) are exposed within VBA for the programmer to use and manipulate – almost
anything you can do in a worksheet can be automated using VBA. As a rather simple

Page 32 of 49
CIV5122Z Advanced Soil Mechanics Course August 2019

language with a Fortran-like syntax (it actually looks close to QB45) VBA suits many
geotechnical needs. You do not need the substance of a complete programming language
to build professional-quality applications because VBA works within Excel. For example,
you can calculate using VBA and then plot the results as a graph within Excel.
This is not a course on VBA and you won’t need to program in VBA to learn CSSM.
However, as you move from ‘learning’ to ‘applying’ you will need to know a bit about VBA.
The ‘learning’ tutorials all start with standard worksheets as these allow the easiest
understanding of how soil plasticity works – this is fine for the basic triaxial test with the
simplest CSSM idealizations. But, when we start introducing evolving friction ratios or
over-consolidation (for example) we need to add in the details/nuances of soil behaviour
and that gets to be horrible in a worksheet while trivial in VBA (using functions). So, the
public-domain software for real application of CSSM is built around VBA.; you could ignore
the VBA and just use the Excel applications provided but we hope that you do not – CSSM
is not black-magic and if there is any one thing we hope you gain from this course it is just
that appreciation. Knowing how to look inside VBA is key to you being confident.
The VBA ‘integrated development environment’ is included with Excel. You get to this by
pressing the short-cut keys ‘Alt’+’F11’ or through the Visual Basic launch-point contained
in the Developer tab that shows up in Excel, circled in Figure 19 below.

Figure 19: Accessing the VBA integrated development environment in Excel

There are books on programming in VBA, but what you need is really only to make yourself
familiar with the provided code – and it is all written in ‘plain English’. If you would like to
know more about VBA, the book Excel VBA Programming for the Absolute Beginner by
Birnbaum & Vine is good and easily found on Amazon. You do not need it for the course,

Page 33 of 49
CIV5122Z Advanced Soil Mechanics Course August 2019

but it may be useful to you in the future as VBA is a truly wonderful feature within Excel
for engineering. Despite saying this course is not about learning VBA there are three
features that are really important in what follows.
The original intent behind the Basic language was that is should be simple and this led to
not requiring ‘declaration’ of variables ahead of their use – which means if you make a
typing mistake that mistake will not be detected; and which leads to erroneous results
and/or crashing applications… you get around this issue by forcing VBA into a professional
mode by typing Option Explicit in the first line of the code. All the open-source code
provided does this.
Although VBA is integrated with Excel, there are different ways to do things. In our
applications we will take some data, for example soil properties and initial conditions, and
then compute the entire stress-strain behaviour as (say) 4000 points. We then use Excel
to produce the various graphs. If each point is transferred back to the worksheet for
plotting as it is computed everything becomes seriously slow and tedious. But, if the
computation puts all the results into an array then that array can be transferred back to a
worksheet very quickly indeed as a single item – which is the approach used. Thus, using
CSSM (or processing CPT data) looks very much like an old-style Fortran program that
operates when requested and dumps its results back into a worksheet for plotting. Easy
to QA or understand and quick to use. This operation is ‘hung off’ buttons on the
worksheets; check the ‘macro’ (which is actually a subroutine) associated with the button
(left-click if you are left-handed to get to the ‘assign macro dialog) and then jump to the
VBA code to see that sub-routine and follow where the code goes.
The final important feature is an item of programming style: ‘named constants’. In many
places we have options – drained or undrained, semi-log CSL or something else. We
want the VBA code to be easily read in plain English and you cannot achieve that if these
options are just referred to as numbers (1, 2 , 3 etc). You can use text strings to show
these choices but it is far more efficient and elegant to declare constants as, for example,
in this code fragment:
' CSL Mode
Const semi_log = False
Const curved = True
Dim CSL_choice As Boolean
…which allows CSL_choice to be used in code with its values shown as ‘semi-log’ or
‘curved’. If there are more than two choices, just list them as constants but now the
choice becomes declared as an integer. Easy to read and understand, and widely used
in software engineering.

Page 34 of 49
 CIV5122Z Advanced Soil Mechanics Course August 2019

Euler Integration
When the phrase “constitutive model” is spoken, people tend to think of integrals and fancy
mathematics. That is a mistaken view, as very few realistic constitutive models for soil
can be directly integrated to give a stress-strain curve for a laboratory test. And never
mind used in any analysis of engineering construction. Rather, what the constitutive
models give is the equivalent of a current stiffness that evolves – and which can only be
used numerically. Numerical integration is not necessarily difficult.
In the case of standard laboratory tests, if all is working ‘as advertised’, these tests all
have the sample under uniform stress conditions – no redistribution of stress develops
within the sample. In reality there are end effects (restraints) and strain localization
(commonly later in a test), but if we treat the test as uniform conditions then standard
laboratory tests have fixed stress or strain paths that allows simple numerical integration.
The easiest method to understand was introduced by Euler (1707-1783), a mathematician
perhaps without peer.
Euler’s method is shown on Figure 20. Starting at point ‘j’, we compute the current
gradient of the equation being integrated – shown as the tangent line. We then apply an
increment along the path to move to the new position ‘j+1’, which in our case will be an
increment of strain. Multiplying the gradient by the step length gives the change in value
between two positions, and thus the new value of the equation at ‘j+1’. This method can
be used in a spreadsheet, and that is what this course uses.
Euler basics – what your xls will do…
800

700 ‘j+1’
600
deviator stress, q: kPa

500 ‘j’
400

300

200

100
1748
0
0 1 2 3 4 5
axial strain: %

Figure 20: Principle of Euler’s method for32 numerical integration

June 12, 2019

Page 35 of 49
CIV5122Z Advanced Soil Mechanics Course August 2019

The deficiency of Euler’s method develops if the gradient changes quickly compared to
the step size – look closely at Figure 20 and you will see the end point (open circle on the
tangent line) is not quite on the curve we are trying to follow. It is easy to see a problem
developing if the gradient at the end of a step is much different from that at the start;
commonly, this shows up as the computed results “drift” away from the true solution.
There are more sophisticated numerical integration methods (eg Runge-Kutta), but they
always involve more calculations. More sophisticated methods also cloud understanding
of what is going on, a particularly undesirable aspect here as understanding how and why
soil behaves as it does amounts to solving three simple but simultaneous equations. As
the intent here is about clearly communicating ideas, not numerical efficiency, so a simple
method is better even if lots of small steps are needed (and it turns out something like
3000 steps is plenty for a triaxial test, which is easy enough to implement in a
spreadsheet). This course thus depends on Euler integration.
When you move from laboratory tests to stress-analysis of the ground in civil engineering
works (‘boundary value problems’) an entirely different set of numerical procedures are
needed – but, the derivation of the algorithms in for finite element analysis is a specialist
discipline and not something most geotechnical engineers do for themselves. At the time
of writing, modern CSSM methods are available as public-domain ‘user defined models’
for some of the standard geotechnical modelling software suites and with a trend for their
direct incorporation as ‘standard’ models in the software. However, do be careful as
Modified Cam Clay is nearly universally offered by all software suites – and it is less than
useless for most geotechnical engineering; you need Bounding Surface or NorSand
variants. But that is getting ahead of ourselves, as the first task is to understand how and
why CSSM works.

Page 36 of 49
CIV5122Z Advanced Soil Mechanics Course August 2019

Original Cam Clay

Original Cam Clay (OCC; ‘Original’ to distinguish it from ‘Modified’) was the first
complete model for soil strength and stiffness. There were several preceding papers in
Geotechnique but these were in the nature of setting out the ideas and how they were
developing; thus it is the Schofield & Wroth (1968) book that defines the framework and
the OCC model. This book (or rather chapters 2,5, and 6) is a classic of the geotechnical
literature and ought to be core-reading in any post-graduate course in our subject; the
book is now freely downloadable on the internet. A feature of this book was its emphasis
on linking void ratio to soil behavior, but this emphasis moved the OCC equations away
from a standard plasticity view (which is actually simpler) and arguably is why people
find OCC difficult to understand. Here we encapsulate the key ideas of OCC and present
them for use in a worksheet for drained and undrained triaxial compression. We derive
the model using both the ‘void ratio’ and ‘plasticity’ view, with the plasticity view being
used for developing the worksheet.

Key Ideas
Some frameworks for understanding soil look to the experimental literature for their
justification. OCC is very different. A much under-appreciated aspect of OCC is that it
starts by adopting a few key ideas (assumptions) and then seeing where those ideas lead
using formal mathematics; test data is only introduced at the end to see if the framework
is relevant to engineering (‘validation’). Of course, the ideas adopted were influenced by
test data but the power in this approach is you can’t reject OCC because you don’t like
the shape of the yield surface or the like. There are actually only three key ideas in OCC:
• The work dissipation assumption, which gives the stress-dilatancy rule
• Normality, which is used to derive the yield surface from stress-dilatancy
• All yield surfaces intersect the CSL, which gives the hardening rule
We will now look at these ideas in turn, seeing how they match our understanding of soil
behavior, before implementing OCC in a spreadsheet.

OCC Flowrule
The starting point is the idealization about how plastic work is dissipated, and that
idealization follows directly from the two-component strength model of Taylor-Bishop that
was discussed earlier. If you turn back to Figure 3, the idea of dilation being a work
transfer mechanism between the principal stress directions naturally leads to the idea that
only the ‘constant volume’ friction dissipates work. Since only plastic strains do work (the
elastic ones just store energy), the obvious step is to modify the stress-dilatancy rule that
was inspired by observing the particulate nature of soils so that the earlier Eqn [3] now
becomes:
Dp = M –  [20]

Page 37 of 49
CIV5122Z Advanced Soil Mechanics Course August 2019

…where M is taken as a constant and the superscript ‘p’ denotes plastic. If you look at
Bishop’s work from seventy years ago (Figure 3), it is apparent that M evolves a bit at
small deviator stress but M being invariant with strain, and independent of void ratio, is a
pretty reasonable starting point. That is the first idealization of OCC.

OCC Yield Surface

The yield surface is derived from the normality condition. Differentiating q =  p (which is
the definition of  rearranged) gives:
𝑑𝑞 = 𝑝 𝑑𝜂 + 𝜂 𝑑𝑝 [21]
Now think back to high-school math. If you have a line with gradient ‘x’ the gradient of a
line perpendicular to that has gradient ‘-1/x’. The term dq/dp is simply the tangent to the
yield surface (since the yield surface is in p, q space as per Figure 17); the line
perpendicular to that tangent to the yield surface is the plastic strain increment vector from
the normality condition (again, see Figure 17). Thus:
𝑝
𝑑𝑞 𝑑𝜀𝑣
=− 𝑝 = −𝐷 𝑝 or, re-arranged as 𝑑𝑞 = −𝐷 𝑝 𝑑𝑝 [22]
𝑑𝑝 𝑑𝜀𝑞

On substituting the stress-dilatancy rule [20] into [22] we get:

𝑑𝑞 = −(𝑀 − 𝜂) 𝑑𝑝 [23]
And finally combining [23] and [21] to eliminate dq:

𝑝 𝑑𝜂 + 𝜂 𝑑𝑝 = −(𝑀 − 𝜂) 𝑑𝑝 ⇒ 𝑑𝜂 = −𝑀 𝑑𝑝/𝑝 [24]

Equation [24] is a separated differential equation and solved simply by integrating each
term, which gives:
𝜂 = − ln(𝑝) + 𝐶 [25]
…where C is an integration constant. This integration constant is chosen as at the stress
ratio  = M, which is that of the critical state, and thus:
C = 1 + ln(pc) [26]
So finally we get:
𝜂 / 𝑀 = 1 − ln(𝑝/𝑝𝑐 ) [27]
The yield surface given by [27] is exactly that shown in Figure 17; the scaling stresses
indicated on that figure are exactly pc and q =M pc. And notice that we have not used any
test data in deriving the yield surface; rather, OCC is based on plausible physical
equations inspired by observations about the particulate nature of soil.

Page 38 of 49
CIV5122Z Advanced Soil Mechanics Course August 2019

OCC Hardening
The really important contribution of OCC was not the derivation of yield surface shape
from simple physical principles but rather the linking of yield surface size to the soils void
ratio. This provides the hardening rule and the linking is done via the CSL.
OCC idealizes the CSL using the widely-used, and quite reasonable, semi-log idealization:

𝑒𝑐 = Γ − 𝜆10 log(𝑝𝑐 ) [6, repeated]

…where  10 are soil properties that apply equally well to sands as clays (and despite
the word ‘clay’ in OCC). There are two ways of incorporating [6] into OCC, which we will
name the ‘state view’ and the ‘plasticity view’.
OCC hardening via state
The Schofield & Wroth approach, widely shared by other workers at Cambridge, UK, was
to emphasize the role of void ratio in soil behavior; in such a world-view the kernel
framework is a void ratio versus log(p) plot, commonly called a ‘state diagram’ as it is a
picture of where the soil sits in its void ratio space. Strictly, such state diagrams ought to
include emin but most workers just show the normal compression locus as this framework
is much conditioned by experience with the oedometer test. Thus, [6] was rearranged,
with a change from base-10 to natural logs (as appear in [27]) to give:
ln(𝑝𝑐 ) = (Γ − 𝑒𝑐 ) /𝜆 [28]
However, there is a further step as OCC includes volumetric elasticity which means that
e does not generally correspond to ec – you have to allows for elastic void ratio change
between the current stress p and pc; OCC has the idealization in terms of soil state that:
𝑒𝑐 = 𝑒 + Δ𝑒 𝑒 [29]
… where e is the current void ratio of the soil at the current stress p. The elastic behaviour
is idealized from unload-reload behaviour seen in oedometer tests by introducing a new
soil property  (which is the oedometer property Cs but expressed in terms of natural logs)
such that:
Δ𝑒 = − 𝜅 ln(𝑝 / 𝑝𝑐 ) during unloading or reloading [30]

…where the ‘−‘ sign arises because void ratio reduction is positive strain. Substituting [30]
in [29] and the result in [28] gives:
(𝜆 − 𝜅) ln(𝑝𝑐 ) = Γ − 𝑒 − 𝜅 ln(𝑝 ) [31]
Or on pulling everything together as a single equation showing how yielding and strength
evolution depends on void ratio:
𝜂 / 𝑀 = 1 − ln(𝑝) − ( Γ − 𝑒 − 𝜅 ln(𝑝 )) / (𝜆 − 𝜅) [32]
OCC hardening via plasticity
Equation [32] directly links soil strength and stiffness to void ratio, but it is hardly intuitive.
Worse, it does not look like a stiffness so it cannot be used in the finite element method in

Page 39 of 49
 CIV5122Z Advanced Soil Mechanics Course August 2019

this form. And, it obscures the role of plastic strain – a role that is the kernel of the theory
on which OCC is based. So, let us start with plastic strain.
OCC is inspired by an idealized oedometer test, so let us think like that and consider void
ratio change as stress changes, Figure 21. Starting at stress state pA let us impose
sufficient deviator stress as we increase p such that the soil consolidates along the CSL
to stress state pB. The confining stress is then reduced, back to pA, along with the deviator
stress, such that the soil unloads (swells) – some of the ‘normal’ consolidation is recovered
during this unloading. If we view this behaviour incrementally the plastic (irrecoverable)
void ratio decrease for loading along the CSL is:

𝑑𝑒 𝑝 = −(𝜆 − 𝜅) 𝑑𝑝𝑐 /𝑝𝑐 [33]

CSL
A
Plastic void
void ratio, e

ratio change

 B

ln(p)
Figure 21: Loading and unloading in state space illustrating plastic void ratio change

August 6, 2019 We want to work with incremental strains rather 22 than void ratio change, so noting that

de = − (1+e) dv we substitute this in [33] and rewrite as:

(1 + 𝑒) 𝑑𝜀𝑣𝑝 = (𝜆 − 𝜅) 𝑑𝑝𝑐 /𝑝𝑐 [34]
There is now one further step. Recall Casagrande’s original discovery of the critical state
was about large-strain strength (see Figure 10). Thus, we should work with dq rather
than dv – which is easy enough to since by definition dvp = Dp dqp. So, using this identity
we rearrange [34] as:

Page 40 of 49
CIV5122Z Advanced Soil Mechanics Course August 2019

𝑑𝑝𝑐 1+𝑒 𝑝
= 𝐷𝑝 𝑑𝜀𝑞 [35]
𝑝𝑐 𝜆−𝜅

Equation [35] operates directly on the yield surface (equation [27]) to change its size in
response to plastic deviatoric strain – easy to appreciate, and an elegant implementation
of the Theory of Plasticity. This elegance has come at the cost that the direct linking to
void ratio is slightly lost, although that linking survives in setting the initial value of pc before
applying plastic strain. However, we also gain further insight as the parameter grouping:
1+𝑒
𝐻= [36]
𝜆−𝜅
…can now be seen as a dimensionless plastic modulus analogous to the elastic shear
rigidity Ir (= G/p). The slope of the CSL is not just about the critical state in itself, but
rather reflects the overall stiffness of the soil in distortion.

Computing Triaxial compression of OCC

All work-hardening plastic models change the size of the yield surface in response to
plastic strains (that is what their hardening law does), except when we get to the critical
state where everything continues indefinitely at constant conditions. So what happened to
our stresses as the yield surface changes size? The consistency condition is that the
stress state must remain on the yield surface during yielding; this is true of every plasticity
model, not just CSSM. In essence, the yield surface drags the stress state with it as it
changes size. Practically, the consistency condition gives us a further equation and
produces the slightly paradoxical result that while we discuss yield surface shapes when
developing the theory such yield surfaces are rarely used. Rather, we take the differential
of the yield surface to work out where an increment of plastic yield has taken us; working
out the corresponding stress change from the consistency condition is a key step in
implementing any constitutive model.
In the case of unknown stress paths, such as general analysis with the finite element
method, the consistency condition is used to drive the solution algorithm. In the case of
laboratory tests, where the equipment controls the stress or strain path, the consistency
condition is used directly to compute the stress-strain curve. For the ‘bullet shaped’ yield
surfaces of OCC, differentiating the yield surface (equation [27]) gives:
𝑑𝑝 𝑑𝑝
𝑑𝜂 = 𝑀 ( 𝑝 𝑐 − ) [37]
𝑐 𝑝

…where the first term in the () is given by the hardening law (equation [35] above). The
second term in the () is test-specific and is what you have to work out; the steps to do this
are as follows.
Undrained Loading
The undrained condition is v=0 (= constant volume). Using the elastic-plastic strain
decomposition, this condition can be expressed in terms of incremental volumetric strains,
broken down into their elastic and plastic components, as:

Page 41 of 49
CIV5122Z Advanced Soil Mechanics Course August 2019

𝑝
𝑑𝜀𝑣𝑒 = −𝑑𝜀𝑣 [38]
… and on substituting the elastic soil model we immediately get the change in mean
stress:
𝑝
𝑑𝑝′ = 𝐾 𝑑𝜀𝑣𝑒 = −𝐾 𝑑𝜀𝑣 [39]
You can see from the above equation [39] quite why elasticity is so important to undrained
loading even though you are using a plasticity model – the excess pore pressure simply
reflects the balance between elastic and plastic processes, with each process being
equally important.
Computing the new stress state after an increment of plastic strain is now simple. The
above undrained condition gives us the change in mean stress which we simply substitute
into the consistency condition [37] (and including the plastic hardening) to get d. We
then apply Euler integration to get:

NEW = OLD + d [40]

…and then on to the next step after adding in elastic shear strain.

Drained Loading
In the case of drained triaxial tests the slope of the load path q/p = 3 (assuming that
you are using standard equipment and not trying to get a particular stress path...) because
of the arrangement of the test. We use this loading direction by first differentiating :
𝑝 𝑑𝑞−𝑞 𝑑𝑝 𝑑𝑝 𝑑𝑞
 = q/p’  𝑑𝜂 = = (𝑑𝑝 − 𝜂)
𝑝𝑝 𝑝

𝑑𝑝
=> = 𝑑𝜂 /(3 − 𝜂) [41]
𝑝

We can now substitute [41] in [37] to get:

𝑑𝑝𝑐
𝑑𝜂(1 + 𝑀/(3 − 𝜂)) = 𝑀 𝑝𝑐
[42]

…which now allows is to advance the solution using [40] just as in the case of undrained
loading. We just modify one column in the undrained spreadsheet.

Exercise 3
Use the supplied template to calculate the undrained triaxial behaviour of OCC.
Verify your numerical implementation by comparison to the closed-form solution given by
Schofield & Wroth (supplied in the template).

Page 42 of 49
CIV5122Z Advanced Soil Mechanics Course August 2019

Investigate the dependence of numerical accuracy on numerical step-size. What would

be a reasonable ‘rule of thumb’ for the needed step-size ?

Exercise 4
Modify your undrained OCC worksheet to compute drained triaxial behaviour
Calibrate drained OCC to the supplied example file (Erksak CIC-682); what values of M
and 10 give the best-fit ?

Exercise 5
Import supplied example file Erksak CID-667
Change pc initial from ‘spacing ratio’ form to equation [31] and then honour the tests
reported initial void ratio
Why does OCC predict very large strengths ?

Page 43 of 49
CIV5122Z Advanced Soil Mechanics Course August 2019

NorSand
Original Cam Clay is mathematically elegant but only matches reality for loose soils.
This begs the question: what has been missed ? It turns out, as will have been discovered
in the tutorial exercises, that the assumption of yield surfaces always intersecting the
CSL is generally wrong; rather, the CSL is the end-point and yield surfaces evolve with
distortional strain to that end point. Thus, while many aspects of OCC can be retained
its hardening law must be generalized. That generalization is NorSand (NS) and the
generalization uses the state parameter. NS was derived out of experience with large-
scale hydraulic filling in the 1980’s and has evolved somewhat since first published in
1993. Partly, this evolution is because there are choices within an otherwise derivable
framework. Partly, this evolution is because additional features have been identified
that both add detail while simplifying NS. The presentation here of NS is that evolved
as of late 2018 although limited to trixial compression for this course; the emphasis for
the course is illustration of how little needs changing in OCC to get to NS that works well
for all soils.

Key Ideas
The ideas of OCC continue into NS but are further formalized as two axioms:
1> A unique CSL exists. This need not be the familiar semi-log form, with whatever
equation fits the soil test data being acceptable (ie the CSL shape is viewed as a
modelling ‘detail’ for any particular soil)
2> Soil state is characterized by the state parameter  with the axiom being that 
reduces with increasing deviatoric strain; this is simply a mathematical statement
of Casagrande’s canonical plot (Figure 5).
With these Axioms established, NS further adopts:
• The Taylor-Bishop work dissipation idealization, but in the modified form
suggested by Nova that best-fits test data; this admits the possibility of plastic work
in volumetric strains, not just deviatoric strains.

• Normality, which is used to derive the yield surface from stress-dilatancy; this is
exactly the same as used in OCC.
• A hardening limit from the state parameter principle (equation [9], see Figure 14)
• A hardening rule derived from Axiom 2 (as opposed to just assuming the CSL is
the hardening rule)
We will now discuss these ideas in turn, seeing how they change our understanding from
OCC, before verifying that NS captures everything from very dilating dense soils through
to static liquefaction of loose soil. Perhaps the kernel idea is that, while OCC only deals
with plastic strain increments, NS requires continued reference to an evolving ; in terms
of your worksheet you need an extra column for this.

Page 44 of 49
CIV5122Z Advanced Soil Mechanics Course August 2019

NS Flowrule
The starting point is the idealization about how plastic work is dissipated, and that
idealization follows directly from the two-component strength model of Taylor-Bishop that
was discussed earlier. However, as noted at the beginning test data is usually best fitted
by Nova’s adaption of the idealization (see Figure 7):

𝜂𝑚𝑎𝑥 = 𝑀𝑡𝑐 − (1 − 𝑁) 𝐷𝑚𝑖𝑛 [4 repeated]

This is a ‘strength’ equation and not a stress-dilatancy rule; Mtc and N are experimentally
determined properties that, as far as established by present testing, do not vary with void
ratio of confining stress. The step made with NS is based on the ideas of Dafalias and
co-workers and which is that, while it has been known for at least 50 years that M evolves
(see Figure 3), that evolution could reasonably be taken to depend on the state parameter.
So, NS uses this idea with the stress-dilatancy rule:
𝐷 𝑝 = 𝑀𝑖 − 𝜂 [43]
…where Mi is now something that evolves; in many ways this goes right back to the original
views of Taylor-Bishop about the work dissipation function not being a fixed quantity. The
evolution is derived by noting that we still want to honour [4] so that [43] must give the
same as [4] at peak strength; so:

𝜂𝑚𝑎𝑥 = 𝑀𝑡𝑐 − (1 − 𝑁) 𝐷𝑚𝑖𝑛 = 𝑀𝑖 − 𝐷𝑝 [44]

At peak strength dp=0 and dq =0 for all loading paths; so Dmin = Dp as elastic strain
increments vanish at peak strength. Rearranging [44] and adding in equation [9] leads
to:

𝑀𝑖 = 𝑀𝑡𝑐 + 𝑁 𝐷𝑚𝑖𝑛 = 𝑀𝑡𝑐 − 𝑁 𝜒 |𝜓| [45]

Equation [45] is straightforward for dense soils with  <0 as it simply fits widely measured
trends in soil behaviour. It is more troublesome for loose soils as the data is unclear on
the soil behaviour in this situation. Logically, there are two choices: i) view Mi as something
that only arises for dense soils and thus revert to Mi=Mtc; or, ii) make Mi symmetric around
Mtc and which then adds the ‘modulus’ function on . At present the second option
appears to fit test data on loose soils better. But, certainly Mi < Mtc as the issue is the
proportion of plastic work not being dissipated by plastic distortional strain.
This topic of mobilized friction ratio is something requiring an open mind on your part and
something to investigate in the soils that you test. There is no ‘right’ or ‘wrong’ here, as
yet, and that is one reason that there have been variants within NS. But do document the
idealization you use when assessing your data.

NS Yield Surface
The yield surface is derived from the normality condition and corresponds to exactly that
as done for OCC. But, the NS yield surface differs from OCC in two important ways. The
NS yield surface is illustrated on Figure 22.

Page 45 of 49
 CIV5122Z Advanced Soil Mechanics Course
NorSand (basic) i ) e
pi = H (pmx - pAugust p
2019
q
300

250
Dmin =  Mi
deviator stress, q: kPa

200 Image condition: pim

150

100

  p
50 pmx= p exp(− /Mi) 
F = 1 − − ln
M  pc 
OCR
0
0 50 100 150 200 250 300 350 400 450
mean effec ve stress, p': kPa

August 7, 2019 16
Figure 22: Example of NS yield surface

First, while the yield surface has the ‘bullet like’ shape of OCC, because the yield surface
does not intersect the CSL the scaling parameter is now something different: the image
stress pim. At this condition Dp = 0 but this need not be the end point and usually is not; it
is a bit like an appearance of the critical state but not quite – hence the name ‘image’.
Thus, the basic equation for the NS yield surface looks the same as that of OCC except
that we have switched out pc for pim:
𝜂 / 𝑀𝑖 = 1 − ln(𝑝/𝑝𝑖𝑚 ) [46]
This switch from pc to pim may appear to be just a change of notation but it is actually
fundamental. In OCC there are a very limited range of plastic behaviours with dense soil
being viewed as over-consolidated (and easily massively so). In NS, any combination of
e, p, q can be on a yield surface – this means soil can yield anywhere in its accessible
states, exactly as seen when you test soil. Of course, you can unload from a yield surface
too so over-consolidation also exists – see Figure 2.
Second, the NS yield surface differs from that of OCC by having an inner cap (the vertical
line shown on Figure 22). Recall that state-dilatancy sets a direct relation between the
soil’s y and its limiting dilatancy dmin (equation [9]). Since we have normality, this means
we can only go so far around a yield surface before we violate the dilatancy limit – and
which can also be viewed as limiting how far a yield surface can expand at any mean
stress. Thus, we must introduce a hardening limit pmx for pim to control how much dilatancy
develops.

Page 46 of 49
CIV5122Z Advanced Soil Mechanics Course August 2019

The hardening limit is derived by rearranging the NS yield surface equation to get:
𝑀𝑖 ln(𝑝/𝑝𝑚𝑥 ) = 𝑀𝑖 − 𝜂
… and then substituting the limiting dilation:
𝑝
𝑀𝑖 ln (𝑝 ) = 𝐷𝑚𝑖𝑛 = Χ 𝜓
𝑚𝑥

… and then inverting to get:

𝑝𝑚𝑥 = 𝑝 exp(−Χ 𝜓/𝑀𝑖 ) [47]

NS Hardening
The hardening rule in NS comes directly from Axiom 2 of CSSM: everything moves to the
CSL with distortional strain; this requires that the ‘image condition’ on the NS yield surface
moves onto the CSL with deviatoric strain. In NS this is done as ‘double hardening’ by
first moving pim to its limiting condition pmx and then tracking pmx as that evolves to pc. The
second aspect of hardening develops naturally because pmx is based on state-dilatancy
(equation [9]). So, it is sufficient to invoke:
𝑝
𝑑𝑝𝑖𝑚 = 𝐻 (𝑝𝑚𝑥 − 𝑝𝑖𝑚 ) 𝑑𝜀𝑞 [48]

…as the NS hardening law where H is the hardening modulus. Because the essence of
NS is decoupling of the yield surface from the CSL, the slope of the CSL no longer directly
acts as the plastic modulus; generally, an effect of  on the hardening modulus is also
found. Further, many soils show an effect of softening as the deviator stress ratio
increases. Accordingly, the hardening modulus used in NS is:

𝐻 = (𝐻0 − 𝐻𝑦 𝜓) 𝑝 / 𝑝𝑖𝑚 [49]

…where H0, Hy are experimentally determined soil properties. These are the only
properties in NS that are specific to NS; determination of their values will be discussed
shortly after first describing how to compute NS behaviour for triaxial tests. But, aspects
of plastic hardening derived for OCC continue across to NS with an approximation of
equation [36] often being a reasonable first-estimate for H0:
𝐻0 ≈ 2 / 𝜆 [50]

Computing Triaxial compression of NS

Computing drained and undrained triaxial compression behaviour of NS is near identical
to the calculations used for OCC other than to replace dpc/pc of OCC with dpim/pim.for NS.
For learning purposes that substitution will be sufficient; for accuracy, an additional term
is needed in the consistency condition because Mi evolves:
𝑑𝑝 𝑑𝑝 𝑑𝑀𝑖
𝑑𝜂 = 𝑀𝑖 ( 𝑝 𝑐 − )+𝜂 [51]
𝑐 𝑝 𝑀𝑖

Page 47 of 49
CIV5122Z Advanced Soil Mechanics Course August 2019

Iterative Modelling for Soil Properties

The essence of all critical state soil mechanics is that soil properties do not change with
void ratio or stress level (this may need a little modification, as there is some evidence for
an effect of void ratio on Mtc with very loose soils). Consistent with this, the properties
used in OCC and NS are simply obtained by plotting data from triaxial compression tests
with one exception: plastic hardening in NS. There is also the idea that the elastic shear
modulus is so sensitive to the details of particle to particle contacts that ‘Gmax’ might be
reasonably considered as a ‘state variable’ (which can change from test to test and vary
from place to place insitu) rather than as a ‘property’. These aspects are easily
investigated by iterative forward modelling (IFM).
In IFM we estimate the properties of interest and them compute the behaviour for the test
in question. We then compare the computed behaviour with that measured, then revising
our estimates based on the mismatch between computed and measured. The process is
then repeated (“iterated”) as many times as needed to achieve good fits to data.
The beauty of IFM is that it optimizes the fit of theory to reality, and while perhaps needing
a little more work that determining a property by plotting some data a particular way, it is
actually very easy to do using a laptop computer. Indeed, many people find IFM slightly
‘sucking them into’ the data and continue the modelling looking for ever more refinement;
it is actually as involving as a video game.
IFM is used to determine H0, Hy for NS. However, although this could be done in a
worksheet environment, it needs professional programming approach as we want to
include things like over-consolidation (very tedious in a worksheet) and the ability to easily
jump between the various tests on a soil (almost impossible with worksheets). Thus, the
open-code Excel worksheet provided and which you will be guided through as part of
Exercise 7.

Page 48 of 49
CIV5122Z Advanced Soil Mechanics Course August 2019

Exercise 6
Take the worksheet of Exercise 5 with the dense sand test Erksak CID-667 and convert
the worksheet from OCC to NS by adding three columns and updating the hardening law.
What value for H gives the best-fit to this test ?

Exercise 7
Open NorTxl_Nerlerk.xlsm (this is the xls holding the Nerlerk sand data used in Exercise
2).
Move to ‘param & plots’ worksheet and learn how to view the different tests by using the
‘Plot Data’ button.
Check that the NorSand properties are set to those you calibrated.
Systematically model the drained tests, varying H as needed, to best-fit all tests; record
your simulation parameters as you do so.
Plot your results to estimate the plastic hardening modulus for Nerlerk sand. What is this
hardening function ?
How sensitive is this function to your eyeballed fit ? Should one test be tried a little stiffer,
and if so does that improve the fit in your view ? Develop a reasoned judgement for your
fits across all the drained tests.
With this hardening function determined, model all the undrained tests varying Gmax to
best fit each test. What is your estimate of Nerlerk sand elasticity ? Go back to your
drained tests – does this estimate of Gmax change your fits ?
Choose a very loose undrained test. Move into VBA and change the Mpsi function to keep
Mi = Mtc for loose soil. Does this improve your fit to the loose test ?

Page 49 of 49