Geotechnical Research Geotechnical Research, 2016, 3(3), 90–118

Volume 3 Issue 3 http://dx.doi.org/10.1680/jgere.16.00008

Paper 16.00008
Determining silt state from CPTu Received 17/08/2016; accepted 17/10/2016
Shuttle and Jefferies Published online 02/12/2016
Keywords: in situ testing/liquefaction/silts

Published with permission by the ICE under the CC-BY license.

(http://creativecommons.org/licenses/by/4.0/)

Determining silt state

from CPTu
Dawn Shuttle PhD, MICE, PEng Michael Jefferies PEng
Consulting Engineer, Surrey, BC, Canada ([email protected]) Consulting Engineer, Surrey, BC, Canada (corresponding author:
[email protected])

Tailings, produced when rock is crushed to recover metals, are normally discharged as slurry of predominantly silt-
sized particles into storage areas that are created using dams. These dams have a poor safety record with billions of
dollars in damages over the past decade alone, making reliable engineering of silt an important economic and safety
issue for the mining industry. Equally, engineering of silts is challenging, as understanding of soil behaviour relates
mostly to ‘sands’ or ‘clays’. Undisturbed silt samples suffer substantial densification between sampling, transfer to
element test and reinstatement of in situ stresses. Hence, silts require a sand-like approach that combines laboratory
tests on reconstituted samples with in situ cone penetration test (CPT) soundings. This paper presents calibrated
spherical cavity expansion in a general critical-state soil model to simulate the CPT in silt. The developed methodology
is numerical, accurately captures calibration data and allows determination of the in situ state parameter in silts from
CPT data. A validation is presented for a large tailing impoundment using stacked thickened tailings. Open-source
software implementing the methodology is provided on the journal website as supplementary material.

Notation qt CPT tip resistance: F/L2

A coefficient in elasticity model: F/L 2
R scalar measure of overconsolidation
Bq dimensionless excess pore pressure = (u2 − u0)/(qt − p0) su undrained soil strength: F/L2
Cc compression index u pore pressure: F/L2
CQ scaling factor = Qp/Qsph Vs shear wave velocity: L/T
Cs swelling index v specific volume = 1 + e
c1, …, c4 coefficients in solution of spherical cavity expansion d dilation angle (usually in NAMC model): °
ck, cm contributory coefficients to CQ G reference void ratio on critical-state locus,
D dilatancy, as strain rate ratio e_ v =e_ q conventionally defined at p0 = 1 kPa
plastic dilatancy, as strain rate ratio e_ v =e_ q
p p
Dp e1, e2, e3 principal strains (assumed coaxial with principal
e void ratio stresses)
F friction ratio e_ q shear strain rate measure work conjugate with s q ;
G elastic shear modulus: F/L2 e_ q ¼ ð1=3Þ½ðsin q þ 31=2   cos qÞe_ 1 − 2  sin q e_ 2 þ
Gmax elastic shear modulus measured geophysically: F/L2 ðsin q − 31=2   cos qÞe_ 3 
H plastic hardening modulus e_ v volumetric strain rate; e_ v ¼ e_ 1 þ e_ 2 þ e_ 3
Ir soil shear rigidity ð¼ G=s m Þ k slope of unload–reload line in e − ln(p0 ) framework
K elastic bulk modulus: F/L2 l, l10 slope of critical-state locus, for semi-log
K0 geostatic stress ratio prior to loading; K0 ¼ s h0 =s v0 idealisation; default is base e; subscript 10 denotes
k, m dimensionless coefficients in Qp = k exp(−my) base 10 logarithm (l10 ≈ 2·3l)
M critical friction ratio; equals hc at the critical state; h dimensionless shear stress measured as ratio of
varies with Lode angle, value at triaxial compression stress invariants; h ¼ s q =s m
(Mtc) taken as the soil property q Lode angle: radians
N volumetric coupling parameter (from Nova’s flow rule) n Poisson’s ratio
NKT undrained strength ‘cone factor’ relating su to qt r density (total): M/L3
P internal pressure of expanding cavity: F/L2 s1, s2, s3 principal stresses: F/L2
p0 mean effective stress ð¼ s m Þ: F/L2 sm mean effective stress; s m ¼ ðs 1 þ s 2 þ s 3 Þ=3: F/L2
Q, Nq dimensionless tip resistance or bearing capacity sq deviatoric stress invariant; s q ¼ ½ð1=2Þðs1 −s2 Þ2 þ
0
factor; Nq ¼ Q ¼ ðqt − sv0 Þ=sv0 ð1=2Þðs2 −s3 Þ2 þð1=2Þðs3 −s1 Þ2 1=2 : F/L2
Qp dimensionless cone penetration test (CPT) tip y state parameter; y = e − ec
resistance defined using mean stress ¼ ðqt − p0 Þ=p00 fc critical-state friction angle: °
Qsph dimensionless spherical cavity limit ¼ ðPlim −p0 Þ=p00 c state-dilatancy constant; value in triaxial
q triaxial deviator stress; q ¼ s1 − s3  ð¼ s q Þ: F/L2 compression taken as the soil property

90
Downloaded by [] on [13/08/20]. Published with permission by the ICE under the CC-BY license
 Geotechnical Research Determining silt state from CPTu
Volume 3 Issue 3 Shuttle and Jefferies

Dimensions condition. The sand fraction drops out of the slurry, usually forming a
F force beach near the spigot point, with these sands being slightly
L length contractive. The silt fraction moves down the TSF and often
T time accumulates under the reclaim pond, these silts being very weak
Subscripts (perhaps half the strength of a geologically idealised, normally
0 initial condition consolidated recent clay, a possible consequence of the very young
1, 2 alternative locations for pore pressure sensor on CPTu age of these silts). By way of comparison, Figure 1 also includes two
c critical state natural silts: one from a deposit on Vancouver Island near John Hart
lim cavity expansion limit dam, while the other is a natural silt that was sluiced from a borrow
min minimum value of parameter pit and then deposited as a slurry to form the core of Coquitlam dam
p variable defined on mean (as opposed to vertical) stress (located near Vancouver City, British Columbia, Canada).
ref reference stress level; by convention, pref = 100 kPa
sph spherical cavity symmetry A new trend in mining has been to process the tailing slurry to
tc triaxial compression condition (q = p/6) reduce its water content (‘thickened’ or ‘paste’ tailings), which
v vertical generally prevents hydraulic segregation during discharge and offers
Superscripts two advantages: a very substantial reduction in the potential for acid
e elastic drainage in the long term (near eliminating potential environmental
dot ‘·’ increment damage from this source) and a sufficient strength (see Figure 2) so
p plastic that the thickened tailings can be stacked above the crest of the
retaining dams/dykes (favourably affecting mine economics). In
Introduction short, hydraulically placed silts are now being used as, in effect,
Tailings are produced when rock is crushed to recover metals, and are structural fills within the TSF. Although these fills are at shallow
predominantly silt. Figure 1 illustrates some examples of gradations. slopes (say less than 6% vertical:horizontal (V:H)), the scale of tailing
Tailings are discharged as slurry into natural valleys closed with dams impoundments is such that some current mine plans envisage stacked
or by complete perimeter containment dykes, generically known as tailings more than 30 m above the crest of the original retaining
tailing storage facilities (TSFs). Tailings within a TSF are commonly dams/dykes. The situation is further complicated by the more
characterised using the cone penetration test (CPT), and some valuable ore bodies seemingly always being found in earthquake-
examples of CPT data are shown in Figure 2 to illustrate their in situ prone areas – liquefaction is a pervasive concern. And TSFs have

Sand sized Silt sized Clay sized

100

Silt_A
Silt_B
80 Silt_C
Silt_D
Proportion smaller than D: %

Silt_E
Silt_F
60 Silt_G

40

20

0
1000 100 10 1 0·1
Particle diameter, D: µm

Figure 1. Tailing gradations illustrating dominance of silt fraction

91
Downloaded by [] on [13/08/20]. Published with permission by the ICE under the CC-BY license
 Geotechnical Research Determining silt state from CPTu
Volume 3 Issue 3 Shuttle and Jefferies

1000

Demarcation between strain softening and strain hardening

Dimensionless penetration resistance, Q(1 – Bq) + 1

behaviour following initial liquefaction (Shuttle and Cunning, 2008)

Gravelly sands
100

Sands to
sand some silt
Sands segregated from
tailings stream and
deposited subaerially

10 Silty sands to sandy silts Natural sandy silt

(John Hart dam)

Thickened tailings
deposited in layers with
Clayey silts drying between lifts

Clays and other sensitive Silt tailings

soils deposited subaqueously
1
0·1 1 10
Friction ratio, F: %

Figure 2. Examples of in situ states of tailings from piezocone

penetration test (CPTu) data

less than a stellar safety record – for example, the Aberfan disaster of testing against laboratory tests on undisturbed samples (with clear
1966 (UK, 144 dead: summarised in Wikipedia (2016a); also see cultural bias – for example, UK practice and protocols are different
Bishop (1973)), Los Frailes dam failure in 1988 (Spain, billions of from those in the USA), but, overall, there is a good basis for
dollars: Boliden (2016); also see Alonso and Gens (2006)), Mount engineering clays in essentially any loading/situation.
Polley dam failure in 2014 (one of the largest environmental disaster
in Canadian history: summarised in Wikipedia (2016b); also see In the case of sands, undisturbed samples of sand are essentially
Mount Polley Review Panel (2016)) and last year the complete impossible – even samples cored into in situ frozen sand show
failure of Fundão dam (Brazil, billions of dollars in damage with at gross densification when thawed and consolidated at in situ
least 17 deaths: summarised in Wikipedia (2016c); also see FTDRP stress. In an effort to get better samples at less cost than freezing,
(2016)). Reliable engineering of silt is an important economic and the Japanese ‘gel push’ sampler has been used with silty sands (e.g.
safety issue in the mining industry – and this issue is a challenge for Taylor et al. (2012)) – but there remains a concerning issue of
geotechnical engineering, as nearly all the understanding of the investigators not comparing the as-recovered void ratio (when the
mechanical behaviour of soil, as found in the literature, relates to sampler first arrives at the surface) to the as-tested void ratio (after
either ‘sands’ or ‘clays’. There are good reasons for this situation. the sample is in the element test cell at the in situ stress level);
there is an apparent belief that void ratio does not matter with silty
In the case of clay soils, undisturbed samples can be recovered and sands, despite this easily being shown not to be the case when
then transferred to an ‘element’ test (such as triaxial compression or using reconstituted samples. Reliance on undisturbed samples also
simple shear) – there can be some discussion about the disturbance misses a key issue – general understanding cannot be gained from a
and how that disturbance can be dealt with, but there are existing single test. Rather, a series of tests on similar samples are required
protocols for measuring soil properties. This testing of undisturbed so that fundamental understanding of sand behaviour is based on
samples is easily extended to investigate the ‘fundamental aspects’ preparing reconstituted samples in the laboratory and then testing
of clay behaviour, as clay samples can be remoulded, across a those samples in element tests; protocols have been developed (e.g.
spectrum of water contents (void ratios), to investigate directly the moist tamping, air pluviation, wet pluviation), allowing repeatable
effect of void ratio and stress history on clay behaviour. Further, samples to be prepared across a range of void ratios. However,
clay soils are amenable to in situ tests that invoke few assumptions sand behaviour (under all loading paths) depends on the state
– vane shear is an example of this, where the measured torque is parameter, not void ratio itself, with small changes in gradation
directly reduced to undrained shear strength by assuming that the affecting the critical void ratio; simply measuring the void ratio
vane sweeps a right cylindrical failure surface in the clay. In situ forces the nearly impossible task of also measuring the critical state
measurements are readily compared to laboratory measurements. A for the soil gradation with that void ratio (a process that must be
range of views exist in the literature on the relative merits of in situ repeated for every void ratio measured). Accordingly, in situ

92
Downloaded by [] on [13/08/20]. Published with permission by the ICE under the CC-BY license
 Geotechnical Research Determining silt state from CPTu
Volume 3 Issue 3 Shuttle and Jefferies

penetration tests (originally the standard penetration test; nowadays extrusion, with careful selection and trimming of the soil tested. Yet,
almost always the electronic CPTu) are used to determine the in despite this level of care, rather large densification, averaging about
situ state of sands (some workers consider the pressuremeter and De ≈ 0·15, was found between the as-recovered void ratio and the as-
dilatometer as alternatives to the CPT). In situ penetration tests, tested void ratio after the sample was in the test equipment and with
however, measure a mechanical response, and there is no simple, the vertical effective stress reinstated (Figure 3). The soil, although
accepted formula for converting the measured penetration resistance classified as low-plasticity clay, exhibits a behaviour in response to
to state measures (state parameter, relative density) or related disturbance that is much like what would be expected with sand.
strength measures (peak friction angle, peak dilation angle). Rather, This behaviour is not so unusual for some Canadian clays (e.g.
the use of penetration tests relies on calibration chamber studies in Becker et al. (2006)), as their silt and clay fractions are rock flours.
which a systematic set of controlled, large-scale laboratory tests has
allowed development of mappings between void ratio, confining The importance of sample densification for this John Hart silt was
stress and penetration resistance. As there is no single mapping investigated by reconstituting two samples at different water
applicable to all sands, a basis is also needed to relate the particular contents and then identically consolidating them before testing
circumstances being considered to the range of published mappings them in cyclic simple shear under identical loadings. The results
(it is rare to develop a site-specific mapping). Overall, while, are shown in Figure 4. Although the void ratio difference
arguably, engineering sands have a greater range of views over achieved between the two samples was only De = 0·06, even this
appropriate practice/protocols than exists in the case of clays, there difference doubled/halved the number of cycles to initial
is also a substantial knowledge base for engineering sands. liquefaction. Patently, the void ratio shifts shown in Figure 3 will
have a profound influence on the liquefaction resistance assessed
So, what is the situation with silt? Silt can be sampled without for remedial engineering. Of course, the important question then
much more difficulty than clays; mud-rotary drilling with a thin- becomes, how should this effect be assessed? And the effect
walled unswaged Shelby tube is often sufficient, although some is general in its consequences – for example, including the
silts may require a fixed-piston sampler, much as for soft clay. important post-liquefaction residual strength.
Silts have a sufficiently low hydraulic conductivity so that these
samples can be extruded, trimmed and moved to element testing There are two alternative, and complementary, approaches to
equipment (triaxial, oedometer or simple shear). However, there is characterising silts: (a) measuring strengths in the laboratory and
densification between what is recovered to the surface in the correcting the results for the measured densification from
sampler and the same soil after transfer to the element test and ‘as-recovered’ (= in situ?) to ‘as-tested’ conditions and/or
reinstatement of the in situ stress on the sample. The disturbance (b) treating silts as sand-like, measuring the state parameter y on-
appears to arise, in part, because of plastic strains imposed on the site and using that state to compute in situ strengths from
sample as it is extruded and transferred to the test equipment; measured soil properties by using reconstituted samples. This
there is also the possibility that yield in unloading (see Jefferies paper primarily addresses approach (b), with some contribution
(1997)), which is intrinsic to extruding the sample, induces plastic from approach (a) to provide validation. An effective stress
strains – and thus that disturbance may be intrinsic, for even ideal approach is followed, looking to use both the CPT resistance and
samplers, as it is a consequence of stress change. the induced excess pore pressure. A further consideration as to
why the approach should be anchored to the CPT is that the
The magnitude of this disturbance is illustrated for a loose natural
clayey silt. Mohajeri and Ghafghazi (2012) reported on a 0·9 Natural silt (John Hart dam)
comprehensive investigation in connection with the John Hart dam Note: plotted stress levels slightly adjusted to show
upgrading. The interest here is on the near normally consolidated 0·8 each test individually
low-plasticity silt (10 < PI < 18) in the depth range 26–35 m at the
location known as Test Site 1. The gradation is Silt_G in Figure 1, 0·7
Void ratio

and gradationally it is actually a sandy silt, despite being classified as

a low-plasticity clay by Atterberg limits. Figure 2 shows the CPT 0·6 In situ void ratio shown
as open point
data adjacent to the sampled zone on a soil behaviour-type plot, with
the data indicating a sensitive clayey or silty soil that is distinctly 0·5 Tested void ratio in CSS
contractive according to the Shuttle and Cunning (2008) criterion. shown as solid point
The geology of this site is reasonably understood, with the horizon of 0·4
interest deposited in a complex sedimentary sequence because of
changing sea levels, isostatic rebound, waves, tides and other factors. 0·3
200 300 400 500 600 700 800 900
The clays range between thinly laminated beds separated by very Effective vertical stress: kPa
thin, fine sand seams to massive layers as much as 10 m thick. A
fixed-piston sampler was used, with extremely gentle handling while Figure 3. Measured silt densification from in situ to as-tested
transporting to the testing laboratory. Gamma ray imaging was used condition
to identify any potential defects inside the tubes prior to sample

93
Downloaded by [] on [13/08/20]. Published with permission by the ICE under the CC-BY license
 Geotechnical Research Determining silt state from CPTu
Volume 3 Issue 3 Shuttle and Jefferies

20 Test conditions
σv’0 = 580 kPa Test 2: e = 0·56
15 τs = 0 kPa Test 1: e = 0·50
τc = ±100 kPa
CSR = 0·17
10
Shear strain: %

0
0 1 2 3 4 5 6 7 8 9 10
Number of cycles
–5

–10

–15

–20

Figure 4. Effect of small void ratio change on remoulded silt in

cyclic simple shear

standard of practice for liquefaction assessments (e.g. Youd et al. 1·1

(2001)) presently mandates penetration tests. Thickened tailings
1·0 Silt some sand
Silt behaviour
Silts are fine-grained soils whose behaviour is controlled by forces 0·9
CSL
transmitted through contacts between the particles – in essence,
Void ratio

the same as sands but with much smaller hydraulic conductivity 0·8
because of the much smaller distances between particles. The
physical and mathematical representation of particulate soils 0·7
behaviour is gelling into the common understanding with, since
about 1990, the universal acknowledgement that soils exhibit a 0·6
critical void ratio that depends on the stress level (the critical-state
locus (CSL)); stress dilatancy links the ratio of strain increments to 0·5
10 100 1000 10 000
stress; dilatancy scales with the void ratio offset from the CSL (the Mean effective stress, p‘: kPa
state parameter y); and soils are work-hardening/work-softening
1·1
plastic materials within the framework of theoretical plasticity. CSL
Subaqueous silt
Silts fit this common understanding, as will now be illustrated. 1·0

Critical-state locus 0·9

Void ratio

Silts display well-defined CSLs. Figure 5 shows two examples

(Silt_B and Silt_C in Figure 1) determined using the usual 0·8
technique of reconstituted loose samples tested in drained and
undrained triaxial compression. The points shown in Figure 5 as 0·7
large dots are the identified critical state of each test. The CSL for
0·6
the stress range of 30 kPa < p0 < 1500 kPa is sensibly fitted with
the usual semi-log idealisation 0·5
10 100 1000 10 000
1. ec ¼ G − l10   logðp0 Þ Mean effective stress, p‘: kPa

Figure 5. Examples of CSL in silts (blue dots denote assessed

Table 1 gives the numerical values for the soil properties G and critical state of each test where that state was reached)
l10 for the silts whose gradations are shown in Figure 1. Broadly,
the property l10 represents a soil’s intrinsic compressibility and Stress dilatancy
the tabulated values are midway between clean sands (~0·05) and Alongside the CSL, stress dilatancy is a fundamental behaviour of
plastic clays (~0·3–0·5), an unsurprising situation for silts. particulate materials. Figure 6 shows the results of three drained

94
Downloaded by [] on [13/08/20]. Published with permission by the ICE under the CC-BY license
 Geotechnical Research Determining silt state from CPTu
Volume 3 Issue 3 Shuttle and Jefferies

Silt A Silt B Silt C Silt D Silt E Silt F Silt G

Critical state
G 1·076 1·423 1·115 1·305 1·900 2·120 1·420
l10 0·159 0·246 0·175 0·200 0·480 0·550 0·180
Mtc 1·25 1·35 1·40 1·20 1·37 1·26 1·31
Dilatancy
N 0·30 0·30 0·30 0·30 0·00 0·40 0·30
ctc 4·0 2·5 2·5 2·0 0·8 2·5 2·5
Plastic hardening
H 25 20–100y 30–200y 45 30 17–100y 50
Elasticity
a a a a
A: MPa 3·7 8·0 3·5
b — 0·50 0·60 — 0·40 — —
emin — 0·60 0·35 — 0·50 — —
Ir 100 60 — 100 — — 125
n 0·15 0·20 0·10 0·20 0·2 0·2 0·15
a
Only in situ data available; see Figure 8 for the trend used, which is only a function of stress

Table 1. Calibrated properties of silts

triaxial compression tests on a loose silt with the data transformed the further view that dilation is a kinematic behaviour. Thus, a
to stress-dilatancy space; although the stress-dilatancy theory better expression for the effect of state on soil behaviour is as a
involves plastic strain increments, here total strain increments are limiting dilation rather than peak strength
used in computing the dilation D – this is an approximation to the
true situation, but elastic strains are a small component of drained 3. Dmin ¼ cy
compression, and this simple algebraic transformation gives a ‘quick
look’ at the soil behaviour with minimal effort. Also shown in
Figure 6 is a trend line given by Nova’s (1982) stress–dilatancy rule where c is a soil property. Been and Jefferies (1985) used the
initial value y0 at the start of test, which is questionable
mechanics in general, as everything should be expressed in terms
2. D ≈ DP ¼ ðM − hÞ =ð1 − N Þ
of current, not initial, values – hence, the more general use of y
in Equation 3 (and which means a tad more processing when
reducing test data to soil properties). Figure 7 shows an example
where M and N are soil properties. Nova’s flow rule is an idealised
of data from drained triaxial compression tests plotted in this way,
dissipation of plastic work into heat, with Nova’s rule implying
inelastic energy storage in addition to the plastic work dissipated
1·6
by the plastic shear strain (Jefferies, 1997). The trend line was
fitted using M = 1·35 and N = 0·30; these are values that are
comparable to those found with recently crushed sands. The
1·2
original Cam Clay flow rule is simply Nova’s rule with the
Stress ratio, η

property N = 0. The mismatch between test data and Nova’s flow

rule for h < 0·4 also arises with sands and appears caused by the
sample reconstitution procedures inducing some overconsolidation, 0·8
Nova’s stress–dilatancy rule:
with the result that the initial loading is elastic. M = 1·36 and N = 0·3

State dilatancy 0·4

The trend relating peak friction angle f and y identified by Been
and Jefferies (1985) has been widely used. However, it is not the
best way of representing that aspect of soil behaviour because, as 0
–0·5 0 0·5 1·0 1·5 2·0
originally identified by Bishop (1950), the dilational component Dilatancy, D
of strength adds to the intrinsic soil friction – what today is called
the property Mtc. Mtc varies from one soil to another depending Figure 6. Stress dilatancy in loose silts
on, at least, the particle shape and particle mineralogy. There is

95
Downloaded by [] on [13/08/20]. Published with permission by the ICE under the CC-BY license
 Geotechnical Research Determining silt state from CPTu
Volume 3 Issue 3 Shuttle and Jefferies

ψ at Dmin must be determined through iterative forward modelling of test data,

–0·20 –0·15 –0·10 –0·05 0 illustrated shortly, and which does not have the precision of simply
0
plotting results from a wide range of initial states; it also requires a

Diltancy at maximum strength, Dmin

constitutive model, and that choice may affect the precision/bias of
–0·2 the determined value for the property. As quick comparison, silts
show about half the c values as those of sands, plausibly reflecting
the greater void space of silts and their greater compressibility.
χ –0·4
Plastic hardening
The soil properties G, l, M, N and c are all associated with soil
–0·6 behaviour that can be seen from plotting triaxial test data in various
ways – no constitutive model is involved, and the properties are
familiar and simply defined. Plastic hardening, however, is
–0·8
different in that there is no accepted property for describing this
behaviour. The Cam Clay variants avoid issues about plastic
Figure 7. State dilatancy of Nerlerk sand
hardening by linking that hardening to the slope of the CSL (i.e.
hardening ≈ 1/l). But all state-based models decouple their yield
with a clearly defined trend and little scatter. The proposition of surfaces from the CSL – which is how they gain the capability of
Been and Jefferies (1985) amounts to c ≈ 4 for a wide range of computing the spectrum of soil behaviour. This then leads to a
sands with less than <20% fines; today, this proposition remains need to define a plastic modulus; but the absence of a generally
acceptable for that range of soils but additional accuracy is accepted soil property leaves each model with its own choice about
gained, and the framework extended to all soils, if c is treated as how to characterise and represent this aspect of soil behaviour.
a soil property rather than a universal constant.
NorSand uses the dimensionless hardening modulus H as its
The dilation rate defined by Dmin is directly related to the familiar property, this property being defined from its role in the
maximum dilation angle d. For the case of triaxial compression hardening law – see the Appendix. Conceptually, H is related to
with its symmetry (e2 = e3) the initial slope of a drained, isotropic triaxial compression
stress–strain curve, with that initial stiffness normalised by the
4a. sinðd Þ ¼ −3 Dmin =ð6 − Dmin Þ confining stress to make it dimensionless.

Elasticity
or, on substituting Equation 3 into Equation 4a Elasticity is now commonly measured using geophysical methods.
The principle is to determine the velocity of an elastic shear wave
4b. sinðd Þ ¼ −3cy =ð6 − cy Þ by using polarity reversal techniques to distinguish the shear wave
from the faster-propagating compression wave. This elastic shear
wave velocity is then converted to the elastic shear modulus
Equation 4b emphasises the linkage between y and d, and in (generally referred to as Gmax, courtesy of the strain-dependent
some ways these two parameters are almost identities of each modulus viewpoint) by using the equations of isotropic elasticity
other. The leading minus sign is a consequence of dilation being a (i.e. Gmax ¼ rVs2 ). In the case of laboratory tests, transducers
negative strain rate for the usual compression positive convention (‘bender elements’) are mounted in the platens of the triaxial cell
of soil mechanics. to measure the wave propagation velocity. In the case of in situ
tests, it is now usual to mount the receiver as an additional
A small difficulty now develops that is particular to silts. The soil transducer in the CPT with a signal source at surface (a.k.a.
property c is ideally determined from drained triaxial compression vertical seismic profiling). The simplicity and repeatability of
tests on dense samples with a range of void ratios and confining shear wave velocity measurements has seen their rapid migration
stress – such data require no constitutive model, and it is simply a from research technique to routine engineering practice.
case of an algebraic conversion of measured strain increments into
dilation (elastic strain rates are zero at peak strength in drained Data on of the elastic shear modulus of silts is presented in
triaxial compression, making data processing for this soil property Figure 8 (the same silts whose gradation is in Figure 1). This
independent of elasticity). Figure 7 is for sand and was developed includes both in situ and laboratory measurements. Also shown
this way. The difficulty with silts is that the present procedures for for comparison is a trend for compact Ticino sand. Three things
reconstituting silt samples do not produce the range of void ratios stand out: silt modulus is affected by stress level and void ratio;
encountered on-site, producing silt samples that are both not as loose the silt trends are comparable to those of sands; and laboratory
and not as dense as those that naturally occur. Plots such as Figure 7 and field measurements produce similar values. The idea that silts
are difficult to produce for silts. This leaves c as a property that are just extra fine sand is reinforced.

96
Downloaded by [] on [13/08/20]. Published with permission by the ICE under the CC-BY license
 Geotechnical Research Determining silt state from CPTu
Volume 3 Issue 3 Shuttle and Jefferies

300
Dense silts
(e ≈ 0·7)
Elastic shear modulus, Gmax: MPa

Ticino sand e ≈ 0·73 ‘compact’

200
Loose silts
(e ≈ 0·9)

Data and test methods:

(I) in situ using seismic CPT
(B) laboratory bender tests
100
Silt_A (I)
Silt_B (B)
Silt_C (B)
Silt_D (I)
Silt_E (B)
0
0 100 200 300 400 500 600
Mean effective stress, p’: kPa

Figure 8. Elastic shear modulus of silts

Knowledge of Gmax alone is insufficient for modelling soils in the best fit to the measured stress path by using Equation 6. This
general. For incompressible pore fluid, using the elastic–plastic approach offers simplicity, albeit with some iteration between
strain decomposition, the undrained condition is drained and undrained fits.

5. dev ¼ deve þ devp ¼ 0 The second approach to Kmax views isotropic compression behaviour
as the appropriate way of assessing the soil’s bulk modulus.
Isotropic or oedometer compression tests are used with
On substituting for the elastic volumetric strain increment by unload–reload cycles to distinguish the first loading response from
using the elastic bulk modulus Kmax, the change in excess pore swelling/reloading – giving the familiar soil properties Cc and Cs or,
pressure du during undrained shear is obtained equivalently as used in Cam Clay variants, l and k. Wroth (1984)
reported extensive tests on the strength of overconsolidated clays
6. du ¼ Kmax devp that were closely modelled using Modified Cam Clay with k/l =
0·2, this ratio then being viewed as a universally applicable constant.
The relation between k and Kmax is
Thus, accurate modelling of excess pore pressure, using any
plasticity model, requires accurate knowledge of Kmax. Note that 7. Kmax ¼ pð1 þ eÞ=k
in plasticity models, there can be no ‘strain-dependent elasticity’,
as this aspect of soil behaviour is captured by the plastic strains
themselves. This presents a further challenge, as Kmax can be The problem with both these approaches to determining Kmax is that
determined only by using seismic wave propagation methods for they often hide physically implausible (i.e. negative) Poisson’s ratios.
unsaturated soils because the compression wave velocity is
dominated by the stiffness of the pore water, not the soil matrix The third approach is more fundamental in that it treats the
elasticity. Broadly, three approaches are found in practice: geophysical measurement of Gmax as having independently
(a) using Kmax to fit the modelled stress path to data; (b) relating determined this property. The bulk modulus Kmax is then related
Kmax to l or (c) relating Kmax to Gmax and n. to Gmax through isotropic elasticity

The approach of using Kmax to fit the model to undrained tests 2ð1 þ uÞ

uses the fact that elasticity has a smaller effect when fitting a
Kmax ¼ Gmax
8. 3ð1 − 2uÞ
model to drained tests than it does in fitting undrained tests – and
which then allows fitting drained tests to determine the plastic
hardening. This plastic hardening cannot change between drained assuming that Poisson’s ratio (n) is constant for any soil. Thus, the
and undrained paths (because undrained paths are a consequence two elastic moduli (Gmax and Kmax) can depend on void ratio and
of a change in boundary condition), so that Kmax is then used as a stress, but they do so without violating physical laws (energy is
‘free’ parameter for modelling the excess pore pressure to obtain conserved around a loading loop). Practically, it is assumed that

97
Downloaded by [] on [13/08/20]. Published with permission by the ICE under the CC-BY license
 Geotechnical Research Determining silt state from CPTu
Volume 3 Issue 3 Shuttle and Jefferies

Poisson’s ratio can be measured on dry samples and its value is dilatancy; these are the two properties introduced earlier. The plastic
unchanged by saturation. The most comprehensive study is Bellotti hardening modulus H is needed because the yield surface is
et al. (1996), who investigated Ticino sand, and their data indicates decoupled from the CSL, although there is an approximate
n ≈ 0·2 (Jefferies and Shuttle, 2002). There appear to be no proportionality of H ≈ 2/l echoing aspects of the Cam Clay variants.
investigations into n for silts (in part because current reconstitution
procedures inevitably produce saturated samples). The value In terms of elastic shear modulus, the simplest approach uses just
n = 0·15 has been simply adopted as ‘not unreasonable’ given other two properties, Ir (= Gmax/p0 ) and n, but this is not particularly
issues about the magnitude of Kmax that will be discussed shortly. accurate, as soils generally have a non-linear dependence of
elastic shear modulus on the stress level. There is a further effect
Representing silt behaviour with NorSand of void ratio on elastic modulus. Equally, because all current good
Interestingly, despite what appears to be very different strengths and models are elasto-plastic, there is no ‘strain dependence’ of elastic
compressibilities, the physical and mathematical representation of modulus (a fact easily demonstrated by using bender elements
sand and clay behaviour is common. A range of constitutive models during triaxial shear). A physically plausible idea is that the
exist, differing in their details, but since about 1990, it is the models volumetric compressibility should tend to zero as a soil tends to
that are framed around the concepts of critical state soil mechanics its minimum void ratio (e.g. as a sand becomes so dense that it
(which is a lot more than either variant of Cam Clay) that have transitions into sandstone). Data suggest that quartz sands follow
become dominant. Such models are ‘good’ in the sense that adopting such an idealisation (Jefferies and Been, 2000). If the further
a CSL allows these models to predict the effect of void ratio and idealisation of constant Poisson’s ratio is adopted, this gives a
confining stress on soil behaviour for all stress paths and loadings. relation under isotropic elasticity that
From the practical perspective, only a few soil properties are needed,

0 b
which are both dimensionless and constant for any soil. A p
Gmax ¼
9. ðe − emin Þ pref
All such state-based models build on the critical-state concepts
and theory developed over the past 70 years and use the state
parameter y to relate strength and stiffness to the CSL. Simple where A, emin and b are the soil properties of isotropic elasticity.
insight into soil behaviour is offered, often with remarkable The parameter pref is the mean effective stress used to bring
accuracy. State-based models represent a long-established Equation 9 to a dimensionally convenient form, by convention
framework for understanding mechanical aspects of soil behaviour taken as pref = 100 kPa.
by using work-hardening plasticity with substantive contributions
from both the USA and the UK (see Jefferies and Been (2015) for Calibration procedures for determining sand properties are
a historical perspective). The key aspects of state-based models presented in Jefferies and Shuttle (2005), with further details in
are (a) the yield surface, (b) the ‘flow rule’ giving relative plastic Jefferies and Been (2015). Nothing unusual is needed with all
strain increments and which invokes stress dilatancy and (c) the sand properties being readily found in normal engineering
hardening law which controls how the yield surface responds to practice by using reconstituted samples. Silts present more
plastic strain. NorSand (Jefferies, 1993) was the first of these challenges, arising from the inability of the present laboratory
state-based models, and its equations are summarised in the procedures to reconstitute silt samples over the range of void
Appendix; NorSand accurately captures the entire spectrum of soil ratios found in situ.
behaviour from dilation of dense soils to extreme liquefaction of
very loose ones. Original Cam Clay (Schofield and Wroth, 1968), Silt samples consolidated from slurry readily give data sufficient to
with its very limited range of soil behaviours, exists as a special define a CSL – the paths shown in Figure 5 are from such samples,
case of NorSand by using a particular choice of soil properties and and examples of the measured stress–strain behaviour in drained
initial conditions (see appendix H of Jefferies and Been (2015)). and undrained triaxial compression are shown in Figure 9. The end
point of these tests also provides Mtc. The difficulty arises with the
The soil properties used in NorSand are largely familiar (i.e. most properties N and c, which are most clearly and accurately defined if
are not specific to NorSand), with some flexibility depending on the there are triaxial compression tests on dense samples – exactly the
level of detail sought. Some of these soil properties have already same dense samples that have proved difficult to create in the
been introduced in the earlier discussion of silt behaviour. All laboratory with present procedures. Although N is best found from
properties are determined by triaxial compression tests, with the a plot of hmax against Dmin for several dense drained tests, this
mathematics of NorSand handling the generalisation to three- property is embedded in the stress-dilatancy law that applies
dimensional (3D) stress conditions (Jefferies and Shuttle, 2002). In throughout the loading and regardless of void ratio; thus, fitting a
the simplest form, three properties (Mtc, G, l) are used to define the trend line through drained test data that have been transformed to
critical state (on current evidence, there is no need for anything other stress-dilatancy space gives an estimate of N (Figure 6). The
than a semi-log CSL with silts). Two properties are associated with property c is problematic as when y ≈ 0, the property has no effect
strength at peak dilatancy: N, the volumetric coupling following – see Equation 3. The procedure that is followed to refine N and to
Nova (1982), and c, determining the influence of y on maximum determine c is ‘iterative forward modelling’.

98
Downloaded by [] on [13/08/20]. Published with permission by the ICE under the CC-BY license
 Geotechnical Research Determining silt state from CPTu
Volume 3 Issue 3 Shuttle and Jefferies

400 400

300 300
Deviator stress, q: kPa

Deviator stress, q: kPa

200 200

100 100
NorSand
Silt_B
0 0
0 5 10 15 20 25 0 100 200 300 400
Axial strain: % Mean effective stress, p’: kPa
(a)
600

500 4
Deviator stress, q: kPa

400
3
Volumetric strain: %

300
2
200

NorSand 1
100
Silt_B

0 0
0 5 10 15 20 25 0 5 10 15 20 25
Axial strain: % Axial strain: %
(b)

Figure 9. Comparison of NorSand with reconstituted silt

behaviour in (a) drained and (b) undrained triaxial compression
(Silt_B, properties as per Table 1)

Iterative forward modelling is a procedure where the complete set compression tests using NorSand is a Supplementary Material
of parameters/properties is estimated (and some, such as the initial downloadable file from this journal’s website.
void ratio and test confining stress, will be known) and the entire
stress–strain behaviour computed. The computed behaviour is Iterative modelling requires a criterion as to what is an acceptable
compared to that measured, and the properties revised to improve fit of model to data. Although various statistical measures might be
the fit. Repeating the procedure (‘iterating’) to obtain a match adopted, not all parts of a stress–strain curve are of equal value –
between computed and measured values provides best estimates for example, peak strength is important, but the exact strain at
of the properties. In the case of NorSand (and other good soil which this is achieved might reasonably be regarded as less
models), it is easy to program them in a spreadsheet by using important than getting a reasonable match to the general shape of
numerical Euler integration (a simple method whose small step the stress–strain curve. It is also important to get a reasonable
size is no limitation when dealing with triaxial test data). Figure 9 match to the volumetric strain or, equivalently, the stress path for
also shows fits of NorSand to the drained and undrained triaxial undrained tests. Experience suggests that most engineers recognise
tests on silt, these fits being obtained by such iterative modelling. what is reasonable, and there is only a small difference in views
As it is slightly tedious to program constitutive models, over the chosen best fit; accordingly, formal statistical best-fit
the spreadsheet implementing iterative modelling of triaxial criteria are avoided in preference to engineering judgement.

99
Downloaded by [] on [13/08/20]. Published with permission by the ICE under the CC-BY license
 Geotechnical Research Determining silt state from CPTu
Volume 3 Issue 3 Shuttle and Jefferies

Iterative modelling works best when each of the parameters being correct critical state – the computed behaviour is nothing like that
iterated has a distinct effect on the computed behaviour. measured. Resorting to the first approach, where the elastic stiffness
Minimising the number of parameters being iterated also greatly is reduced to match the measured undrained stress path, produces a
improves the efficiency of iterating. In the case of any good rather good fit to all aspects of the actual test, as also shown in
model, G, l and M should be determined directly from test data Figure 10; this reduced elastic stiffness corresponds to K = Kmax/4,
and there is no difficulty in doing this (as shown earlier for silts). and G was reduced proportionally at the same time to maintain
This leaves H, and for silt c and N, to be found by iterating. In Poisson’s ratio at n = 0·15. It is emphasised that both of these
practice, NorSand is tolerably easy to fit to any test with one simulations had identical values for the other properties and which
interesting niggle – it is rare to find that peak dilation and peak were derived from the calibration discussed earlier. Interestingly, if
strength coincide in test data, as expected from stress-dilatancy this reduced K, and the associated n ≈ 0·15, is used to compute the
theory, and NorSand is usually best fitted by focusing on the volumetric elasticity by using Equation 7; the resulting ratio
volumetric strain rather than peak strength. k/l ≈ 0·25 for this silt, not far off the optimised ‘universal’ ratio
k/l = 0·2 for overconsolidated clays reported by Wroth (1984). The
Although these concerns about the iterating procedure might suggest presented undrained fit in Figure 9 had the reduced K.
that iterating leads to uncertain values for the soil properties, such a
concern misses an important attribute of the method: this formal Interestingly, the difficulties with Gmax and Kmax are not restricted
modelling is an optimisation of the soil properties across the whole to silts – the same troubles arise with sands. The implication is that
data set. The key offering of all good models is that their properties the elasticity of the soil skeleton is systematically different from the
do not change with void ratio or confining stress, which means that elasticity applying to the transmission of seismic waves through the
the same property values must be used for every test in the data set. soil. The systematic divergence appears, on current modelling, to
Iterative forward modelling forces recognition of just how well the be the same across the various silts and sands considered – which
chosen model replicates the entire set of measurements. makes it easy enough for the practical engineer to continue with
the simple, repeatable procedures of measuring shear wave velocity
As a final comment on fitting NorSand to triaxial data on silts, there and then to scale simply the data. Fundamentally, however, it does
is a general issue with Kmax. The three alternative methods for raise questions that ought to be investigated.
determining Kmax were discussed earlier. The most fundamental
approach appears to estimate Kmax from the geophysically measured CPT resistance in sands: calibration studies
Gmax by way of Equation 8 using a constant, plausible Poisson’s As will be seen shortly, there are several idealisations of
ratio; in the case of the silts considered, the Gmax data from Figure 8 questionable reasonableness in theoretical approaches to CPT
are adjusted for initial void ratio as per Equation 9. An example of resistance – and which correspondingly lead to uncertainties in
applying this choice for Kmax to an undrained test in silt is shown in the assessed soil state from CPT data. It is helpful to consider first
Figure 10, all other properties being at their best-fit values. Figure 10 the calibration of CPT to the sand state before looking at the
also shows the results of the undrained test being modelled for theory as these calibrations define, for their particular soil
comparison. As can be seen, using such a geophysically based Kmax properties and conditions, what should be expected. It is
produces a too-stiff initial response, which then develops into a necessary to look at the sand data, as there are no complete
sharp transient loss of strength followed by strong dilation to the calibration studies for the CPT in silt.

350 350

300 300
Measured
K = Kmax/4
Deviator stress, q: kPa

250 250
Measured
200 200
NorSand using K = Kmax/4
150 150

100 NorSand using K = Kmax from 100 K = Kmax

Gmax through Equation 8 and v = 0·15
50 50

0 0
0 5 10 15 0 100 200 300 400
Axial strain: % Mean effective stress, p’: kPa

Figure 10. Effect of elastic bulk modulus on computed silt behaviour

100
Downloaded by [] on [13/08/20]. Published with permission by the ICE under the CC-BY license
 Geotechnical Research Determining silt state from CPTu
Volume 3 Issue 3 Shuttle and Jefferies

CPT calibration chambers are, in effect, a giant triaxial cells in chamber data are treated as the gold standard, the reality is that
which a CPT is pushed into a sand of known void ratio and under each data point involves an ‘eyeballed’ pick of the representative
known stress. Repeating the procedure with many samples of the qt (which is rarely the desired uniform tip resistance with depth) as
same sand provides a mapping between tip resistance, soil state well as the accuracy of determining the void ratio of 1–2 t of sand
and stress level. Most members of the in situ testing community (however carefully placed). The systematic bias arises because
regard chamber calibration of the CPT as the ‘gold standard’ for calibration chamber tests are carried out over a range of confining
interpreting CPT data. The only difficulty in this is that each stresses, and when qt is normalised by confining stress into Q, the
chamber test involves careful placement of as much as 2 t of sand, data at low confining stress tend to plot to the top of the data
so it is not commercially viable for most projects. Chamber tests bandwidth in Figure 11, with the high-stress data plotting to the
tend to be research programmes by universities, with just a few bottom of this bandwidth; this point was addressed by Shuttle and
contributions from the industry; the paper of Jefferies and Been Jefferies (1998) and arises because elastic shear modulus increases
(2015) is a convenient summary of all this calibration chamber data. less rapidly with confining stress than a simple linear trend. A
further accuracy reduction is that the CPT resistance is influenced
The starting point for determining the state parameter from the by soil extending as much as 1 m from the probe – most chamber
CPT was this worldwide set of chamber tests, with samples of the tests results need correcting for the finite chamber size, and these
sands used being then tested to determine their CSL (Been et al., chamber size corrections are not without their own controversies.
1986). Combining the void ratio of each calibration test with the
appropriate CSL for that sand produced a relation between The results of all existing chamber test programmes are described
normalised CPT resistance Q and the initial state parameter y0 by Equation 10 but with the coefficients k and m depending on
(i.e. the soil state before pushing the CPT); Figure 11 shows an the soil’s properties as well as Ir ð¼ Gmax = p00 Þ. The coefficients k
example of such a transformation of calibration chamber data for and m will be discussed again after theoretical approaches to the
Erksak sand (data from Been et al., 1987a). The trends shown in CPT resistance are described.
Figure 11 can be fitted with an equation
Qp ¼ k expð−my Þ Cavity expansion as analogue of CPT
Background
 or in its inverted form used to evaluate CPT data
Cavity expansion involves, as the name suggests, expansion of a
cavity by internal pressure in either finite or infinite media. As in
10. y ¼ − lnðQp =kÞ=m
many things, the initial impetus was military, with the cavity
being battleship gun barrels of the early twentieth century. But
where Qp ¼ ðqt − p0 Þ = p00 and is the dimensionless CPT the interest here is the mechanics, as the extreme symmetry of
resistance (the subscript ‘p’ is used to denote the use of mean, not cylindrical or spherical cavity expansion allows a 3D situation to
vertical, stress; the corresponding Q found in much of the in situ be described in terms of one variable (radius to a point in the
testing literature uses the vertical stress). domain), while the kinematic constraints enforce constant
direction of principal stresses and strains. Originally, this
There is some scatter apparent in Figure 11 and also a small symmetry allowed closed-form solutions for some constitutive
systematic bias. The scatter develops because, although calibration models under finite deformations (a.k.a. large strain); this work
underlies many design formulae for deep foundations and in situ
1000 testing, the CPT in particular. The basic idea can be expressed as
Erksak sand
(Been et al., 1987a)
11. Qp ¼ CQ Qsph

where Qp is the dimensionless CPT tip resistance, Qsph is the limit

100
Qp

pressure of the soil in spherical cavity expansion and CQ is the

mapping factor (ideally a constant, but in principle something
that might depend on dilation and/or the friction ratio measured
by the CPT).

10 Bishop et al. (1945) derived the limit pressure for expansion of

–0·25 –0·20 –0·15 –0·10 –0·05 0
State parameter, ψ0 spherical cavity in an infinite medium of Tresca material and
suggested that the pressure required for a punch to indent metal to
Figure 11. Normalised CPT resistance in calibration chamber test greater than three punch diameters was related to this cavity
of Erksak sand expansion limit pressure – this is the first statement of the idea
represented by Equation 11. Gibson (1950) suggested that since

101
Downloaded by [] on [13/08/20]. Published with permission by the ICE under the CC-BY license
 Geotechnical Research Determining silt state from CPTu
Volume 3 Issue 3 Shuttle and Jefferies

the undrained strength of clay sc approximates a Tresca material, dependence of yield on confining stress, are what distinguishes
the Bishop et al. (1945) result could be adapted for calculating particulate materials from metals. Arguably, the most fundamental
the end bearing capacity of piles in clay by allowing for the behaviour of soils is dilation with deformation (Reynolds, 1885);
additional shear force along the pile tip that acts in addition to Rowe (1962) showed by way of micromechanical considerations
the normal stress that is taken to be the spherical cavity limit that stress-dilatancy behaviour is intrinsic to an assembly of
pressure (Plim). This idea has become standard, and is also applied interacting particles (although this does not imply that Rowe’s
to end bearing of piles in sand. The general scheme is shown in stress-dilatancy rule must be used – there are alternatives, derived
Figure 12. The difference between clay and sands is that this from idealised plastic work dissipation models). The difficulty with
shear force on the tip is an additive component to the spherical stress dilatancy is that it is incremental, which in terms of obtaining
limit pressure for clays, while it is a multiplier for sands. a cavity limit pressure requires integration throughout the domain
Choosing the standard CPT geometry with a = 60° gives and throughout time.

The first satisfactory solution for cavity expansion in a non-

12a. ‘clay’ ðundrained, TrescaÞ: qt ¼ Plim þ 31=2 sc associated Mohr–Coulomb (NAMC) soil was that of Carter et al.
(1986), and this solution remains widely used today (it is used
later in this paper for verification of the adopted numerical
‘sand’ ðdrained, Mohr​−​CoulombÞ: q  t procedures). The attraction of NAMC is that the dilation angle (d)
12b. ¼ Plim þ ð1 þ 31=2   tan fc Þ ≈ 2Plim is different from the friction angle (f), allowing simulation of soil
behaviour over a wide range of soil densities. The NAMC soil is
elastic-perfectly plastic with four properties: fc, d, G and n. In
where the condition is added that soils at the interface with the practice, when NAMC is calibrated to drained triaxial tests, G is
CPT will be in their critical state (subscripted by ‘c’) because of often taken as a quasi-elastic secant modulus to peak strength.
large shear strains induced in this zone. The approximation given
for sand uses fc = 31°, a reasonable value for natural subrounded Solutions with CPT geometry
quartz sands. The approximate scaling of the cavity limit stress in Since the development of the finite-element method, purely
sand corresponds to CQ ≈ 2, which is consistent with the numerical approaches using actual CPT (or pile) geometry have
experimental results of Landanyi and Roy (1987). become possible. However, finite-element analysis of the CPT is
extraordinarily difficult – which is why so few studies exist. The
The solution for expansion of a cavity in a Tresca material challenge in finite-element studies is that, within geomechanics,
exploited the constant volume behaviour of this material, which finite-element analysis is formulated in terms of stress and (usually,
gives a simple relation between the strain increments everywhere small) strain. Whereas the CPT presents a ‘flow’ problem closer to
in the domain and cavity wall movement. Clearly, sands (and those found in hydraulics than geomechanics and where it would
more generally, all soils under drained loading) do not exhibit be clearer to discuss the situation in terms of velocities rather than
zero volumetric strain increments – these volumetric strains, and a displacements. Some work has indeed gone in this direction with
the ‘strain path method’ (Baligh, 1985; Teh and Houlsby, 1991),
but only for Tresca soil (a zero-volumetric strain material) and
with considerable approximations in the solution method.

There are, as yet, few studies for CPT penetration in NAMC soil.
Pile
Willson et al. (1989) provided some early results, but they
introduced a double-hardening cap to deal with compressibility,
making their results difficult to compare with cavity expansion
solutions to assess the consequence of real CPT geometry. A most
qb interesting study using actual CPT geometry in NAMC soil is that
of van den Berg (1994); Figure 13 compares some of van den
α α Berg’s results with the Carter spherical cavity solution for exactly
the same NAMC properties. Broadly, van den Berg’s results suggest
Rigid
soil τ that CQ ≈ 4, which is double that inferred in the previously
discussed work. However, there are two cautions about van den
plim
Berg’s results: (a) the elastic modulus used was remarkably soft for
sand-like behaviour, with the results shown in Figure 13 being for
Ir = 55, and (b) too much load was being introduced into the
Figure 12. Postulated relation between cavity limit pressure and shearing zone by excessive shaft friction with these simulations
pile resistance (Randolph et al., 1994) having 2·2% < F < 2·6%, which is about five times larger than
found in CPT penetration of clean to somewhat silty sands.

102
Downloaded by [] on [13/08/20]. Published with permission by the ICE under the CC-BY license
 Geotechnical Research Determining silt state from CPTu
Volume 3 Issue 3 Shuttle and Jefferies

80 NAMC Soil NAMC soils. However, as finite-element simulations with real

φc = 30º CPT geometry are so difficult, and with none having achieved
Ir = 55
Bearing capacity factor, Nq

ν = 0·3 realistic results for good soil models, it is attractive to return to

60 Actual CPT geometry
cavity expansion and use a good model in that. Such cavity
(van den Berg,1994)
expansion studies require the scaling factor CQ to be established
40 by calibration to physical tests, but at least they offer insight on
how the soil properties affect the CPT.
Spherical cavity
20 (Carter et al., 1986) The first contribution using a good model was that of Collins et
al. (1992), who adopted a CSL with the semi-log idealisation;
strength varying with y using the average trend of Been and
0
0 2 4 6 8 10 12 14 Jefferies (1985); Bolton’s (1986) stress dilatancy; and Gmax that
Dilation angle, δ: º depended on mean stress and void ratio. The spherical cavity
expansion was expressed as simultaneous differential equations
Figure 13. Scaling between CPT tip resistance from finite-element with the solution computed using a standard numerical solver.
analysis and spherical cavity limit pressure in NAMC soil The numerical results were approximated by

Cavity expansion with good soil models 13. Qp ¼ c1 pðc2 þc3 vÞ expðc4 vÞ
It is clear from calibration chamber studies that there is no unique
relation between dilation and normalised CPT resistance –
Figure 14 illustrates some examples of trends found in calibration where v is the initial specific volume of the sand (v = 1 + e0) and
chamber studies, this figure showing test data on ten normally c1 to c4 are coefficients that depend on the sand properties (G and
consolidated clean sands. Any particular sand follows Equation l in particular). Collins et al. (1992) provided values of the
12, but there is a wide difference in the values for the coefficients coefficients c1 to c4 for several of the sands used in the various
k and m from one sand to another. The trends are not parallel, and calibration chamber studies, including Reid Bedford, Ticino and
there is about a factor of 4 between the lowest resistance and the Ottawa sands that capture the trend and bandwidth of these
greatest at any particular state. Figure 14 is presented in terms of calibration studies (see Figure 14). Using the coefficients quoted
the state parameter, but it makes no difference if the calibration by Collins et al. (1992), and the reported e0 and p00 for each result
data are presented in terms of dilation angle, void ratio or relative of a calibration chamber test on that sand in the database, gives a
density: the large systematic differences continue to exist between comparison of this spherical cavity solution with CPT calibrations
the various sands used for CPT calibration. The bandwidth shown for these three sands, which is shown in Figure 15.
in Figure 14 is commonly attributed to ‘compressibility’ (e.g.
Robertson and Campanella (1983)), although in reality this is too There are two aspects of interest in Figure 15. First, the computed
simple. Modelling is needed to understand how and why soil spherical cavity results are close to the measured CPT resistance at
properties affect the mapping between CPT and soil state. low values of Qp – in essence, when the initial state of the soil is
close to y0 ≈ 0. But as the sand becomes denser, and thus more
The initial attraction of the cavity expansion analogue was that it dilatant, the computed results systematically diverge from
allowed closed-form solutions for Tresca, Mohr–Coulomb and calibration measurements with the spherical cavity Qp being a third
of that measured for very dilatant states. Second, the offset between
1000 spherical cavity limit pressure and measured CPT resistance varies
CPT in calibration chamber
Ottawa sand
Normally consolidated sands between the three sands shown with the most ‘compressible’ Reid
Erksak sand Bedford having less offset than the least compressible Ottawa sand;
the Ticino offset is intermediate between these two. The scaling
factor inversely mirrors the trends shown in Figure 14.
Qp

100
Ticino sand A scale factor that varies with Qp is not an overly concerning
Hilton mine sand issue, as it merely adds little bit of complexity. However, the
second observation of sand-specific scaling is more serious, as it
leaves substantial uncertainty when moving to a different soil for
which calibration chamber data do not exist – and avoiding that
10 uncertainty is the whole purpose of the theoretical method.
–0·25 –0·20 –0·15 –0·10 –0·05 0
State parameter, ψ
It is a reasonable proposition that the work of Collins et al. (1992)
Figure 14. CPT resistance trends against state parameter for ten sands was limited by their soil model rather than their approach to
spherical cavity analysis. A particular concern is their NAMC-like

103
Downloaded by [] on [13/08/20]. Published with permission by the ICE under the CC-BY license
 Geotechnical Research Determining silt state from CPTu
Volume 3 Issue 3 Shuttle and Jefferies

500 law dependence of G on p0 has been introduced together with an

Normalised penetration Q from Collins et al. (1992)

Ottawa sand effect of void ratio: Equation 5. This elastic law improves accuracy,
Ticino sand but does so by adding another two soil properties; the exponent of
400 Reid Bedford sand Line of the power law and the lower limit of accessible void ratio emin. What
equivalence follows uses this modified NorSand (as per the Appendix) in drained
spherical cavity expansion, which is compared to the CPT calibration
300 chamber data to establish the scaling factor, before moving on to
consider undrained spherical expansion.

200
Numerical method
The finite-element method was used for the work reported here,
with the numerical approach for spherical expansion being similar
100
to that used by Shuttle and Jefferies (1998) for drained sand. The
updated flow chart for the numerics is shown in Figure 16. In
0 addition to the NorSand constitutive model, the Fortran program
0 100 200 300 400 500 includes the NAMC model to allow verification of the finite
Normalised penetration resistance Q measured in calibration chamber
displacement (a.k.a. large strain) component of the spherical
cavity expansion implementation.
Figure 15. Collins et al.’s (1992) spherical cavity limit pressure
compared to CPT tip resistance in calibration chamber tests for
NorSand was implemented using an incremental viscoplastic
three sands
formulation following the approach of Zienkiewicz and Cormeau
(1974). This is computationally simple, but there is no theoretical
assumption that yielding does not occur until the stress ratio h maximum time step for numerical stability comparable to that
reaches the value inferred from the Been and Jefferies (1985) trend; derived for NAMC plasticity (Griffiths, 1980). The time step
the stress–strain behaviour of sand in drained triaxial compression required for stability with NorSand is small, about one twentieth
cannot be reasonably captured with such an assumption, even of the NAMC value and varying within the simulation as the
though the assumption is good enough for ‘strength’. It is thought elastic shear modulus varies. This small time step was not a
that this deficiency underlies the mismatch evident in Figure 15, as particular constraint for the analyses considered here.
cavity expansion in an infinite medium is a very confined situation
that magnifies the effect of volumetric strains induced during shear. The spherical cavity expansion of NorSand simulations of
Shuttle and Jefferies (1998) allowed the initial elastic shear modulus
In short, the work of Collins et al. (1992) was an interesting first to vary as per the elastic idealisation, but within any simulation, that
step but is not sufficient for practical engineering, where there is a modulus was then constant. Here, that work is duplicated and then
need to compute CPT calibration from familiar soil properties. extended to use G that varies throughout the domain of each
simulation in accordance with the elastic idealisation.
Spherical cavity expansion in NorSand
The next substantive contribution was that of Shuttle and Jefferies Simulations of triaxial compression were used to verify the
(1998), who adopted the first variant of NorSand, a variant that implementation of the NorSand constitutive model. The viscoplasticity
captured the complete stress–strain behaviour across the spectrum component of the numerics was standard small strain, and the
of soil behaviour. Since then, NorSand has been modified in three numerical approach of this triaxial code used the general approach of
ways: (a) the soil property c was introduced to reflect the effect Smith and Griffiths (1988). A mesh of 16 (4 wide by 4 high) eight-
of wider ranges of gradation and mineralogy on the maximum node elements was used so as to identify any internal instabilities that
dilation shown by any soil; (b) it was recognised that the soil might be hidden with a single element.
property N might be better viewed as affecting the scaling of the
yield surface through an operating critical friction ratio Mi The simulations for spherical cavity expansion were more
(analogous to ff of Rowe (1962)) rather than being used to sophisticated and adopted the elements optimised for cavity
specify the shape of the yield surface; and (c) the representation expansion proposed by Yu (1990). Typically 70 elements were
of Mi was changed to a computationally simple form (Jefferies used, distributed logarithmically with radius so that there are more
and Shuttle, 2011). A further refinement has been elasticity. elements near the cavity where the strain gradients are greatest. The
finite-displacement approach used was ‘Lagrangian’ with the finite-
The early version of NorSand followed the spirit of critical-state soil element mesh being associated with material points – as strains
mechanics with the idea that elastic shear modulus should be develop, the mesh moves. A moving-mesh approach is near trivial
expressed as a dimensionless elastic rigidity Ir – this is a ‘clay-like’ to implement for the symmetric situation of cavity expansion and
idealisation with the elastic modulus scaling with mean effective its associated kinematic constrains – it just requires radial
stress. But no sand behaves like this (see Figure 8). Thus, the power coordinate updating at each step in the cavity expansion, which is a

104
Downloaded by [] on [13/08/20]. Published with permission by the ICE under the CC-BY license
 Geotechnical Research Determining silt state from CPTu
Volume 3 Issue 3 Shuttle and Jefferies

Large strain finite element analysis using viscoplasticity

Read in material properties (.dat file)
Define geometry and number of steps
Read in simulation specific inputs (.val file)
Set initial element stresses
Define geometry and initialise state
Loop number of displacement increments
Null loads, excess loads (bodyloads), incremental strain Dε,
incremental plastic strain Dεp, stiffness matrix
Update G (if needed) and define elastic stress–strain matrix, D
Assemble displacement component of stiffness matrix, K0 + KL
Add convected term Kθ to stiffness matrix
Set stiffness matrix K to incremental nodal displacement at cavity wall
Invert K matrix
Loop iterations
Fix displacement at cavity wall and outer boundary
Compute nodal displacements from {δ} = [K]–1{f}
Check whether converged
Loop elements
dε = B da total strain increment
dεe = dε – dεp elastic strain increment
Dσ = D Dεe elastic stress increment
σ = σstep-1 + Dσ ‘new stress’
Does element stress state exceed yield?
No: continue to next element
Yes:
calculate dυvp current viscoplastic strain rate
dεvp = dνvp dt increment of viscoplastic strain
εvp = εvp iter-1 + dεvp viscoplastic strain this increment
BTDεvp calculate body loads for increment

bodyloads = bodyloadsiter-1 + bodyloads this increment

Update (harden) yield surface
Next element
Did any element yield this iteration?
Yes: ...continue to next iteration
No:
Recover element stresses and strains for all elements
Project stresses to cavity wall
Update nodal coordinates
Update state variables in each element
Update yield surface for NorSand softening (if using)
Output element stresses, strains, void ratio and so on
Next displacement increment

Figure 16. Flow chart of large-strain finite-element (FE) code for

spherical cavity expansion

minimal overhead. What is less trivial is that moving the mesh The reference solution for the verification is the direct Euler
convects work, and this convected work must be included in the integration of NorSand that is implemented in the spreadsheet
solution; including convected work is further complicated by used for iterative forward modelling – denoted as ‘VBA’, as
changing void ratio as the mesh moves because of dilation. The Visual Basic for Applications is the environment it is programmed
derivation of the additional term required for convected work is in. Two examples of such verifications are shown in Figure 17,
given by Shuttle and Jefferies (1998); the additional term used one for a lightly dilating silt and one for a very contractive silt
differs from that found in the paper of Zienkiewicz and Taylor (quite extreme parameter combinations are used, as the purpose in
(1991) because of allowing for void ratio change during mesh this instance is to test the numerical procedures, not to best-fit real
movement – a necessary improvement for dilating soil. silt). The finite-element and VBA integrations were very close,
indeed indistinguishable, within the thickness of the plotted lines
Verification of numerics for the denser silt. No numerical instabilities were encountered.
In the case of NorSand, verification of the viscoplasticity
implementation have been reported for sands (Shuttle and Verification of spherical cavity expansion, including finite
Jefferies, 1998). The additional verifications carried out here check displacement, used a spherical NAMC formulation similar to that
that the parameter combinations associated with silts do not cause described by Shuttle and Jefferies (1998) so that the finite-element
some untoward behaviour, and also focus on undrained behaviour. results could be directly compared to Carter et al.’s (1986)

105
Downloaded by [] on [13/08/20]. Published with permission by the ICE under the CC-BY license
 Geotechnical Research Determining silt state from CPTu
Volume 3 Issue 3 Shuttle and Jefferies

200 200
180 NorSand FE
180
VBA
160 160
ψ = –0·2 ψ = –0·2
140
Deviator stress: kPa

140
120 120
100 100
80 80
ψ = 0·1
60 60
40 ψ = 0·1
40
20 20
0 0
0 5 10 15 20 0 20 40 60 80 100 120 140 160
Axial strain: % Mean effective stress: kPa

Figure 17. Verification of viscoplastic finite-element (FE)

implementation of NorSand in undrained triaxial compression of silt

closed-form solution. Some comparisons are shown in Figure 18, one et al. (1986) solutions are virtually indistinguishable throughout the
set for a stiff Ir = 1000 soil and another for a rather softer Ir = 250 entire cavity expansion, including the limit pressure. In the case of
soil. In the case of the stiff simulations, the finite-element and Carter the softer simulations, the two solutions diverge slightly.

4000
Drained calibration in sand
3500 δ = 12º
Radial stress at cavity wall: kPa

Although this paper addresses the response of silt to the CPT, a

3000 Finite element difficulty in doing this is that there are no calibration chamber results
δ = 6º in silt that can be used to develop the spherical-CPT mapping factor
2500
Carter et al. (1986) CQ. Hence, the approach adopted is to better understand this factor
2000 δ = 0º CQ in sand and then assume that the developed understanding can be
1500 transported to silts since nothing in the mechanics changes; this
NAMC assumption depends on relating CQ to the soil properties.
1000 φc = 30º
Ir = 250
ν = 0·2
500 Not all of the reported chamber test programmes are suitable for
determining CQ as some programmes have rather few tests – such
0
0 50 100 150 200 250 300 350 400 450 500 programmes are useful checks, but in themselves contain
Cavity hoop strain: % insufficient information to define trends. Hence, this paper focuses
8000
δ = 12º on the data for Ottawa sand (the least compressible sand of the
7000 available data and with well-defined trends), Ticino sand (by far the
Radial stress at cavity wall: kPa

largest data set) and Hilton Mine sand (by far the most compressible
6000 Finite element
and with a trace of silt). Erksak sand is also considered, as this was
5000 δ = 6º
in the largest chamber with the least boundary corrections. These
Carter et al. (1986)
4000 tests span the range of reported calibrations (see Figure 14) and
δ = 0º represent a range of properties, albeit those properties are skewed to
3000 the clean quartz sands of academia rather than the sands with some
2000 NAMC silt found in practical construction.
φc = 30º
Ir = 1000
1000 ν = 0·2
Each of the sands considered had a reasonable number of triaxial
0 tests to define their properties (commonly four to five drained and
0 50 100 150 200 250 300 350 400 450 500
Cavity hoop strain: % three to five undrained). Table 2 presents the properties of these
sands, and there is little ‘wiggle room’ with these values. For
Figure 18. Verification of large-strain finite-element code for example, the properties M and N are determined using Bishop’s
drained spherical cavity expansion in NAMC soil procedure of plotting hmax against Dmin with the properties defining
the fitted trend line; increasing M above its central value requires a

106
Downloaded by [] on [13/08/20]. Published with permission by the ICE under the CC-BY license
 Geotechnical Research Determining silt state from CPTu
Volume 3 Issue 3 Shuttle and Jefferies

Ottawa Ticino Erksak Hilton Mine

Particle size distribution

D50: mm 530 530 355 200
‘Fines’: % 0 0 3·0 2·5
Critical state
G 0·754 0·985 0·84 1·305
l10 (le) 0·028 (0·012) 0·055 (0·024) 0·050 (0·022) 0·160 (0·070)
Mtc 1·13 1·28 1·18 1·42
Dilatancy
N 0·25 0·34 0·20 0·35
ctc 4·70 4·00 3·40 2·70
Plastic hardening
H 130–1400y 55–850y 100–1000y 40–200y
Elasticity
A: MPa 9·7 9·5 9·5 4·3
b 0·50 0·47 0·47 0·70
emina 0·30 (0·49) 0·35 (0·58) 0·35 (0·53) 0·50 (0·62)
n 0·20 0·20 0·20 0·20
a
The value of emin was determined by regressing the fitted elastic model to minimise the deviation of data from the modelled trend. The value of
emin shown inside parentheses is that determined by the ASTM procedure

Table 2. Properties of sands used for calibration chamber tests

corresponding change to N to fit the data set. Without formalising, a In terms of the general trend for CPT behaviour in sand, the purpose
reasonable judgement is that l10 has a precision ±0·002; M, ±0·02; of using cavity expansion analogues is to quantify the effect of
N, ±0·03 and c, ±0·2 for these chamber sand data sets. Further details intrinsic soil properties on CPT resistance. The hypothesis is that the
on the gradational and other index properties of the various calibration effect of soil properties will be fully reflected in ksph and msph, with
chamber sands can be found in the paper by Been et al. (1987b). the scaling factor CQ in Equation 7 being a consequence of the
difference in deformation patterns between an actual CPT and the
The data file for each calibration chamber programme contains the spherical analogue – and which logically requires that CQ not depend
reported initial void ratio and stress conditions of each test, as on soil properties. This hypothesis was thoroughly examined, and
well as the ‘steady’ penetration resistance attained. The initial confirmed, for nine sands by Ghafghazi and Shuttle (2008). Here, that
void ratio and confining stress data were used as the starting point earlier work is updated to reflect the introduction of stress-dependent
for the numerical simulations, so that there was a spherical cavity elasticity within the spherical code and the further enhancements (and
limit pressure computed for each of the actual tests. Figure 19 simplifications) of NorSand since that time. A further important
shows an example set of results (for Ticino sand). The spherical difference is that Ghafghazi and Shuttle (2008) used geophysically
cavity solution is offset from the reference chamber results, and determined Gmax, whereas in this paper G = Gmax/4 is adopted across
that offset depends slightly on y. the cavity expansion simulations, both with drained sands and
undrained silts, for consistency with the fitting of undrained silt
When introducing the results of calibration testing earlier (Figure 14), behaviour presented earlier. Similarly to Equation 10 fitted to the
it was noted that these data sets are affected by elastic rigidity, with chamber data, the finite-element results can be expressed as
the high-rigidity data lying to the top of the results and the low-
rigidity data lying to the bottom. As the effect of Ir on k and m must 14. Qsph ¼ ksph expð−msph y Þ
be known when faced with inverting CPT data to recover y, it is
helpful to include explicitly the rigidity effect in the scaling factor
CQ. This is easily done using parametric simulations, varying Ir with where
the finite-element code and otherwise constant properties. Figure 20
plots the effect for the four calibration chamber sands considered – 15a. ksph ¼ ak þ bk lnðIr Þ
within reasonable accuracy, simple trends are evident for both ksph
and msph with log(Ir). This effect of Ir on CPT resistance is the
cause of the so-called stress level bias when expressing the results of 15b. msph ¼ am þ bm lnðIr Þ
calibration chamber tests in terms of y (Shuttle and Jefferies, 1998).

107
Downloaded by [] on [13/08/20]. Published with permission by the ICE under the CC-BY license
 Geotechnical Research Determining silt state from CPTu
Volume 3 Issue 3 Shuttle and Jefferies

1000
Ticino sand
Range of normalised penetration resistance Q
found in calibration chamber tests
Normalised resistance ratio, Q

100

Range of normalised spherical limit

pressure Qsph computed with e0 and p0 pairs
reported for the chamber tests
10
–0·35 –0·30 -0·25 –0·20 –0·15 –0·10 –0·05 0
Initial state parameter, ψ0

Figure 19. Comparison of numerical results for spherical cavity

expansion in Ticino sand with data from calibration chamber tests

The coefficients ak, bk, am and bm were determined by fitting

13
trend lines through the results from parametric simulations shown
in Figure 20; values of these coefficients for the four chamber 12 Ottawa
sands considered are given in Table 3. Ticino
11 Erksak
It is helpful for subsequent processing of CPT data in general (i.e. Hilton mine
ksph

when applying the work to field data rather than calibration 10

chambers) if CQ is treated as comprising two components that
9
reflect the offset and skewness between spherical and chamber
that are evident in Figure 19. An appropriate form is 8

16. CQ ¼ ck expð−cm y Þ 7
100 1000

9
The purpose of the work is to best-estimate y. Accordingly, values
8 Ottawa
of ck and cm were determined by minimising the difference between
Erksak
y estimated using Equations 11 and 16 across the entire 130 results 7
Ticino
available for these four chamber programmes; strictly, Ticino is
msph

6 Hilton mine
overrepresented in this minimisation as a soil type, but Ticino sand
also includes a wide range of conditions that appear under-
5
represented with the other sands. The result of this optimisation was
4
17a. ck ≅ 3  2
3
100 1000
Elastic shear rigidity, Ir

17b. cm ≅ 2  4
Figure 20. Effect of elastic rigidity Ir on computed CPT coefficients
ksph and msph
Figure 21 shows the match between y computed from the
reported qc values and y reported based on the test’s initial void state computed from a test’s reported e0 and p00 and that computed
ratio and confining stress. The results lie around the line of by the inversion from the reported qt of that test is Dy = 0·02
equivalence, and the standard deviation of the difference between (lines have been drawn at two standard deviations from

108
Downloaded by [] on [13/08/20]. Published with permission by the ICE under the CC-BY license
 Geotechnical Research Determining silt state from CPTu
Volume 3 Issue 3 Shuttle and Jefferies

deal of convenience is gained from approximating parametric

ak bk am bm trends, and a little accuracy is lost. This paper, therefore, provides
a downloadable source code as Supplementary Material on this
Ottawa 4·650 1·004 −2·658 1·728
journal’s website for any reader to use in computing k and m for
Ticino 6·442 0·467 −1·571 1·401
their soils (there is also a user manual explaining the input file
Erksak 5·110 0·797 −1·292 1·276
and how to transfer output to a spreadsheet).
Hilton Mine 6·092 0·344 0·881 0·571

See Figure 19 for fit of trend lines to simulations Undrained results for silts
Most generally, undrained behaviour arises from the inability of
Table 3. Computed coefficients for spherical cavity expansion pore water to move on the timescale of the loading. In the case of
triaxial tests, undrained conditions are simply a boundary
condition that is imposed by closing a drainage valve. Neither
situation can affect the soil’s properties. In the case of finite-
0·1
element programs, the change from drained to undrained loading
State parameter ψ0 computed from reported qc

is achieved by the specified bulk modulus of the pore fluid – a

very low pore-fluid modulus (typically zero) gives a drained
0 response, while using the modulus of water gives an undrained
situation. Thus, all the numerical formulation described, verified
and validated for sands is directly used for undrained loading by a
single change of pore fluid bulk modulus in the input parameters.
–0·1
The calibration of NorSand for silt properties was presented
earlier, with the values being shown in Table 1. Examples of the
fit of NorSand to silt are shown in Figures 9 and 10. Simulations
Ottawa
–0·2 were run for a range of void ratio and confining stress (for ease of
Ticino 4
Hilton mine
using the elastic idealisation) to define trends.
Erksak 355_3
While the drained cavity expansion computes a limit pressure,
–0·3
–0·3 –0·2 –0·1 0 0·1 undrained cavity expansion produces both this effective pressure
State parameter ψ0 reported in calibration chamber and a matching excess pore pressure. When normalised by the
initial conditions, these parameters correspond to the usual
Figure 21. Accuracy in recovering y from calibration chamber dimensionless groupings
data (dashed lines show bandwidth y ± 0·04)

18a. Qp ð1 − Bq Þ þ 1 ¼ ðqt − u1 Þ = p00

equivalence in Figure 21); most of this difference appears
attributable to the intrinsic precision of chamber testing.
Interestingly Hilton mine sand – the ‘outlier’ in most views of 18b. Bq ¼ ðu1 − u0 Þ =ðqt − p0 Þ
chamber testing – is now nicely captured and evenly distributed
around the line of equivalence. Slightly less satisfactory is that
while Ottawa sand is systematically biased more than one where u1 is the excess pore pressure at the cavity wall or CPT tip.
standard deviation from equivalence, equally, the Ottawa results Figure 22 provides the scaled cavity expansion results in terms of
are parallel to the computed trend and might be explained by a these two parameter groups. The coefficients of the scaled cavity
small error of DG ≈ 0·03 in the assessed CSL (even though every expansion in this silt (Silt_C in Figure 1 and Table 1) are k = 6·2
effort was made to test the same sands in triaxial as used in the and m = 11·7.
chamber, the two sets of testing were sometimes years apart with
all the possibility for gradational differences that implies). The Although it may seem remarkable, and unlike drained results in
optimised CQ is consistent with van den Berg’s (1994) results for sand, the effective CPT resistance seen in Figure 22(a) is
the effect of correct CPT geometry discussed earlier. independent of elastic modulus with a unique relation between
effective tip resistance and initial state parameter. This result is
It is possible to carry out parametric simulations for every soil actually not remarkable at all because for undrained loading De is
property to determine the effect of each property on k and m – effectively zero, and the soil at and near the cavity wall (or CPT)
Shuttle and Jefferies (1998) followed this route. However, the experiences very large shear strains and is corresponding in the
finite-element code runs so quickly – and both approaches depend critical state. And the critical state for undrained (= constant void
on measured (or otherwise known) soil properties – so not a great ratio) shear of any soil is simply given by

109
Downloaded by [] on [13/08/20]. Published with permission by the ICE under the CC-BY license
 Geotechnical Research Determining silt state from CPTu
Volume 3 Issue 3 Shuttle and Jefferies

 The CPT is usually configured to measure pore pressure at the

p0c −y0 ‘shoulder’ of the probe (the u2 location), as that has proved most
¼ exp
19. p00 le useful in distinguishing between soil types. However, the cavity
expansion analogue corresponds to conditions of displacement as
well as shear – and that is at tip (the u1 location). Penetrometers in
where Equation 19 follows from the definition of the state
engineering practice are equipped with a single pore pressure
parameter. What is seen in Figure 22(a) is the expected consequence
sensor, which is normally at the u2 location, so the data of Peuchen
of the numerical procedures nicely and stably bringing the soil to its
et al. (2010) are used to map between the two pore pressures, as
critical state and then continuing to deform it in that condition as
this mapping uses only the parameters measured by the CPT itself.
the limiting cavity pressure develops. The soil properties N, c, H
and Gmax have no effect in the critical state. Only l and M matter.
So far the CPT has been considered in terms of effective stresses.
However, it is usual to evaluate CPT data on a total stress basis
Of course, this situation does not continue to Bq, with the pore
for undrained penetration of clays using
pressure reflecting both total stress change and the shear-induced
excess pore pressure. All soil properties now matter. The computed qt − sv
Bq trends are shown in Figure 22(b). There is now an effect of
su ¼
20. NKT
elastic shear stiffness, with different trends for different rigidities.

100
Dimensionless CPT resistance, Qp(1 – Bq) + 1

p’0 = 100 kPa and Ir = 50

p’0 = 100 kPa and Ir = 100

With Qp = CQQsph p’0 = 100 kPa and Ir = 200

p’0 = 300 kPa and Ir = 50

10
p’0 = 300 kPa and Ir = 100

p’0 = 300 kPa and Ir = 200

Spherical cavity in Silt_B

1·0

0·1
–0·2 –0·1 0 0·1 0·2 0·3 0·4
(a)
1·8
Excess pore pressure ratio, Bq (for u1 location)

Ir = 200
1·6
Ir = 100
1·4
Ir = 50
1·2

1·0

0·8

0·6

0·4

0·2
–0·2 –0·1 0 0·1 0·2 0·3 0·4
Initial state, ψ0
(b)

Figure 22. Numerical results for undrained spherical cavity expansion

in silt B. (a) Dimensionless CPT effective tip resistance Qp(1 – Bq) + 1;
(b) excess pore pressure ratio Bq (at u1 transducer location)

110
Downloaded by [] on [13/08/20]. Published with permission by the ICE under the CC-BY license
 Geotechnical Research Determining silt state from CPTu
Volume 3 Issue 3 Shuttle and Jefferies

where NKT is the undrained cone factor (generally lying in the range berms that contain cells of thickened tailings. In practical terms,
of 11–20). As the peak undrained strength is available from this change will have weak silts used as engineered fill stacked
simulations with NorSand for the chosen soil properties, it is some 25 m above the perimeter dams and underlain by even
straightforward to take the computed cavity expansion results weaker silts (‘slimes’ in the mining vernacular) as their foundation.
previously presented separately in terms of effective stress and excess
pore pressure to determine NKT. The computed NKT factor is shown The engineering of this change to stacked tailings has been based
as a function of initial state parameter in Figure 23(a) for an elastic on extensive CPT soundings to characterise the original
rigidity Ir = 100 (both the excess pore pressure induced by the CPT subaqueous tailings (which include some sands as well as the
and the undrained strength vary with Ir as illustrated in Figures 10 predominant silts) and the new thickened tailings. The CPT
and 22(b)). In the case of lightly dilatant silt (y0 ≈ 0) with no soundings have been complemented by extensive laboratory
softening, illustrated in Figure 23(b), NKT ≈ 13·5; as the silt becomes testing on both reconstituted and undisturbed samples. The
progressively looser, with brittle loss of strength after attaining su, the principal concern has been liquefaction failures, whether triggered
NKT factor increases and reaches NKT ≈ 16·5 for the rather by earthquakes or simply from static loading during construction
contractive silt (y0 ≈ +0·2) that is also illustrated in Figure 23(b). (as happened in the recent Fundão tailing dam failure in Brazil).

Validation
Site investigation data
Background Silt tailings are weak soils and CPT transducers that comply with
Neves-Corvo is the largest mine in Portugal and the second largest the usual standards EN ISO 22476-1 (ISO, 2012) or ASTM D
copper mine in Europe, located near Castro Verde some 220 km 5778 (ASTM, 2012) are insufficiently accurate. The data
south of Lisbon. Tailings from the mine are discharged into a discussed here were obtained with 0·5% at 10 MPa tip sensors,
190 ha TSF, originally designed for subaqueous containment of the with both the tip and pore pressure sensors calibrated on site by
tailings using a rock fill dam across a natural river valley and lowering the probe through a 20 m column of water (easily done
supplemented by several smaller dams – the overall crest length of in the thickening plant).
the dams is some 3·5 km. The original tailing disposal capacity
would have been exhausted in 2011, but the discovery of An example of the CPT data from a weaker part of the TSF is
additional ore reserves extended the mine life. After consideration shown in Figure 24, this figure being annotated to highlight
of alternatives, a change to stacked thickened tailings within the several aspects. First, tailings are new deposits that have been
existing footprint was adopted – Lopes et al. (2013, 2015) placed quite quickly – some may still be consolidating, while
described the basis for this change to tailing disposal. Stacking is others may show underdrainage into original ground (fractured
planned currently at an overall slope of ~5%, achieved using rock at Neves-Corvo); the current piezometric regime needs

20 2·0 1·5
Ir = 100 Undrained cone factor, Nkt
18 (left axis) 1·8

16 1·6
Normalised deviator stress, q/p’0

14 1·4
1·0
12 1·2

10 1·0 ψ0 = 0·0 ® Nkt ≈ 13·5

Dimensionless tip resistance, Q
8 (left axis) 0·8
0·5
6 0·6
ψ0 = +0·2 ® Nkt ≈ 16·5
4 0·4

2 Undrained strength ratio, su/p’0 0·2

(right axis)
0 0 0
−0·10 −0·05 0 0·05 0·10 0·15 0·20 0·25 0·30 0 5 10 15 20 25
Initial state parameter, ψ0 Axial strain: %
(a) (b)

Figure 23. Computed NKT factor and associated silt stress–stain state y0; (b) examples of undrained stress-strain behaviour
behaviour in silt B. (a) Variation of undrained cone factor Nkt, tip associated with computed Nkt values
resistance Q, and undrained shear strength ratio su/p0¢ with initial

111
Downloaded by [] on [13/08/20]. Published with permission by the ICE under the CC-BY license
 Geotechnical Research Determining silt state from CPTu
Volume 3 Issue 3 Shuttle and Jefferies

qt: MPa Pore pressure, u2: MPa Friction ratio, F: % Bq State parameter, ψ
0 2 4 6 8 10 0 0·5 1 1·5 0 1 2 3 4 5 6 –0·2 0 0·2 0·4 0·6 0·8 1·0 1·2 –0·2 –0·1 0 0·1 0·2 0·3
0

Subaerial
thickened
tailings

5
Depth below ground: m

10

Subaerial
sand
beaches
Hydrostatic profile shown as light blue line

15

Subaqueous
silts (slimes)

20

u0 profile defined State parameter profile

Note high
by dissipation tests from soil-specific
Bq values
calibration of CPTu

25

Figure 24. CPT profile in massive silt tailing deposit

determination at the time of CPT soundings, conveniently done by until just prior to laboratory testing. Here interest is in strength
dissipation tests when adding pushrods, as at this site. Second, the data, and those data were obtained by computer-controlled triaxial
weak consistency of these silts is evidenced by the high Bq equipment with careful determination of the as-sheared water
values, which are noticeably larger than those found in most content using the ‘freezing method’ at the end of each test.
natural soft clays. Third, these Bq values fluctuate about the basic
trend in both the thickened tailings and the subaqueous silts, An example of the measured behaviour of the undisturbed silt is
indicating some vertical variation in state; in the case of the shown in Figure 25. The sample is moderately dilatant. It is
deeper silts, there may be an effect of changes in distance from dilatant because it is denser than it is in situ, displaying the usual
the spigot point from year to year as the tailings accumulated, densification of silts during handling and reconsolidation that
while the thickened tailings do not segregate but do experience were discussed at the outset of this paper (see Figure 3).
variable drying during placement. Correction of this densification to the stress–strain behaviour at in
situ void ratio depends on properties determined on reconstituted
The CPT campaign was accompanied by undisturbed sampling, samples.
with careful logging of the water content of the as-recovered
samples. Most of the undisturbed samples were in the subaqueous State determined from undisturbed samples
silts, and these sample locations were chosen to complement the The testing of undisturbed samples was accompanied by another
understanding of site conditions from the CPT. The water content series of tests on reconstituted samples using a single
logging was on a ‘whole-sampler’ basis. Samples were carefully representative gradation (obtained by blending several samples
stored at their in situ orientation and protected from moisture loss together to give enough single-gradation soil for ten triaxial tests).

112
Downloaded by [] on [13/08/20]. Published with permission by the ICE under the CC-BY license
 Geotechnical Research Determining silt state from CPTu
Volume 3 Issue 3 Shuttle and Jefferies

The blended gradation was needed as soil properties can be State determination from CPT
determined only by using several triaxial tests on the same soil The results of three CPT soundings near these undisturbed samples
with varying initial void ratio and confining stress, as well as have been processed into estimated in situ state parameters using
requiring both drained and undrained tests – work discussed the coefficients for this silt, k = 6·2 and m = 11·7, which were
earlier in the paper and which produced the soil properties given computed earlier. The results of this data processing are also
for Silt_C in Table 1. shown in Figure 26. Each plotted point corresponds to a nominal
25-mm depth increment of the CPT; there is considerable
The in situ state parameter can be computed from the void ratio of repeatability between the three CPT soundings.
the undisturbed samples if the CSL is known. The problem,
however, is that minor changes in soil gradation affect G, and Validation
there was quite some variation in gradation within the subaqueous Overall, the average state inferred from the CPT is broadly
silts. The approach adopted was to use iterative forward modelling aligned with state inferred from the triaxial tests on the
of the undisturbed samples with the initial state parameter treated undisturbed samples, with the CPT data clustering close to the
as a ‘freedom’ to best-fit the model to data. At this stage all soil central estimate from the undisturbed samples.
properties were kept the same as determined for the representative
blended sample of the remoulded tests. With a good fit achieved, The variability seen in the undisturbed samples appears much
the CSL property G was then adjusted to align the modelled void larger than indicated by the CPT data in the in situ silts.
ratio with the measured ‘as-sheared’ void ratio (this actually has Regardless of the CQ factor, the CPT indicates quite uniform Bq, F
minimal effect on the computed stress-strain behaviour). An and qt trends – there is some variability in these soils in situ, but
example of the achieved fit is also shown in Figure 25. The on- nothing like the range suggested by the undisturbed tests. In terms
site state of each undisturbed sample was then computed using the of the undisturbed tests, there is little wiggle room in the state
fitted G for that sample, the l10 determined on the remoulded parameter inferred from modelling their as-tested behaviour (see
samples, the ‘as-recovered’ void ratio of that undisturbed sample the fit in Figure 25), and their tested void ratio is reasonably
(from the water content logging mentioned earlier) and the precise (determined by the freezing method). Where there is
estimated mean effective stress in situ using K0 = 0·7. These considerable uncertainty is the as-recovered void ratio – this void
estimates of the in situ state parameter are shown in Figure 26. ratio was determined from the measurements on the whole 0·8 m
long sampling tube, which necessarily averages what the sampler
The in situ states cluster around y0 ≈ 0·20; by comparison, the contains. And it is the difference between the as-tested and as-
classic idealisation of normally consolidated behaviour without recovered void ratios that give the inferred in situ state.
brittleness (i.e. normally consolidated original Cam Clay)
corresponds to y0 = l. Thus, the in situ behaviour of all these It is also true, however, that the silt does vary in gradation both
silts is expected to be contractive with a potential post-peak laterally and vertically, whereas the data processing used
strength loss of about 50%. coefficients for a single soil. If the effect of gradation change

600 600
NorSand
500 2_D10-75-M-2 500

400 400
Deviator stress, q: kPa

300 300

200 200

100 100

0 0
0 5 10 15 20 25 0 100 200 300 400 500
Axial strain: % Mean effective stress, p’: kPa

Figure 25. Comparison of NorSand to undrained triaxial

compression of undisturbed silt using properties from remoulded
tests

113
Downloaded by [] on [13/08/20]. Published with permission by the ICE under the CC-BY license
 Geotechnical Research Determining silt state from CPTu
Volume 3 Issue 3 Shuttle and Jefferies

In situ state parameter, ψ

–0·1 0 0·1 0·2 0·3 0·4 0·5
0

C04E04-355-25

C10E09-220-30
5
C10E10-110-30

10
Depth below surface: m

15

20

Inferred state from modelling

25 triaxial tests on undisturbed
samples and correcting for
sample densification as per text

30

Figure 26. Comparison of silt state computed from CPT to silt state
inferred from ‘undisturbed’ samples corrected for densification

predominantly affects G, that use of a single calibration will be in Figure 8 and corresponds to Ir ≈ 90. Looking at Figure 22(b),
reasonable as G has minimal effect on the stress–strain behaviour an in situ state of y0 ≈ 0·2 at Ir ≈ 90 gives Bq ≈ 1·3.
that underlies the CPT calibration; equally, the variation seen in
the measured friction ratio (Figure 24) indicates that l may vary a Overall, these validation considerations suggest using CQ from
little as l often correlates to F (Plewes et al., 1992; Reid, 2014). drained calibration studies in sand because the effective-stress
If l varies, then k and m may vary a little too. Allowing for such component of undrained spherical cavity expansion in silt is broadly
variation may attenuate some of the variability in the estimate of consistent; there may be a slight bias, however, in that the predicted
y0, as there is a geologic tendency for the state to relate to in situ state of the loose silt that is about Dy ≈ 0·02 denser than is
depositional conditions across a spectrum of soil gradations actually the case (assessed from visually shifting the processed CPT
(Jefferies and Been, 2015). results to best align them to the undisturbed tests in Figure 24).

The CPT procedures themselves provide a second check on the Downloads

estimated in situ state. Within these deeper silts, and neglecting Numerical studies such as those presented in this paper can be
the more dilatant ‘spikes’ as sandier interlayers, the range for the very time consuming for readers to implement. Accordingly, we
excess pore pressure ratio is 0·95 < Bq < 1·15 (see Figure 24); have provided (a) a commented Fortran source code for the
after applying the relation proposed by Peuchen et al. (2010) to spherical finite-element analysis; (b) a ‘user manual’ on how to
map the data at the u2 sensor used to a u1 location, one finds 1·20 run the code and import results into Excel to derive CPT
< Bq < 1·30. The shear modulus data for the deeper silts is shown coefficients; and (c) the forward modelling environment for

114
Downloaded by [] on [13/08/20]. Published with permission by the ICE under the CC-BY license
 Geotechnical Research Determining silt state from CPTu
Volume 3 Issue 3 Shuttle and Jefferies

calibrating soil properties as an Excel application with commented (2008) confirmed the universality of scaling the spherical cavity
source code in VBA. analogue to reference chamber data in sands. This paper builds on
these computationally driven developments to consider undrained
The routine NorTxl in the VBA source code is easily viewed CPTu soundings, as required for silts, and with further reflection
from Excel, and this commented code may be a better explanation of scaling factors from Shuttle and Cunning (2007).
of NorSand than the equations in the Appendix. It is also useful
to quickly run a few simulations, as that will give a far better The derived relation for determining the state parameter from CPT
sense of what is captured by the model than what can be data in silts, based on scaled undrained spherical cavity expansion,
appreciated from figures in the written paper; some data files are validates nicely at the example site with deep deposits of young silts
included to assist (and the calibrations are documented). (<25 years old). The methodology is general, and calibrations can be
computed for other soils provided that their properties are measured.
The Fortran source code follows the general coding principles But a limitation in this study has been the adequacy of the calibration
expounded by Smith and Griffiths (1988). Many users may be chamber data, as these data are depended on to establish the ‘cone
happy to simply compile the code and run it, but it was thought factor’ CQ relating the results of the spherical cavity analogue to
essential to fully expose the numerical method – there are no the CPT. As noted earlier, such calibration chamber data are seen as
hidden factors. the gold standard but it is nothing like as accurate as commonly
perceived. The lack of Gmax data has been mentioned, which is to
These codes allow any reader to duplicate the results presented here – some extent a consequence of much of the chamber testing predating
either using the authors’ data or their own. The downloads are on the bender elements. Even Ticino sand, with its extensive Gmax data,
journal’s website as supplementary material – please do download actually relied on a second study using the same sand. And in the
and look at them as they are a part of the paper. All programs and case of Ticino sand data, there is an instance of two tests with the
data are subject to the GNU General Public License Version 2, in same e0 and p0 and yet a factor of 2 difference in the stable qt value
essence a use-at-own-risk license that encourages further developments is reported – this is not to criticise the data themselves, but simply
based on that free software (‘take it and run with it …’). highlights the experimental difficulty in conducting these chamber
calibrations and why a fair bit of scatter is found in the results of any
Much of the other laboratory and calibration chamber data used in calibration programme, hence the use of the calibration chamber data
this study are public domain (Jefferies and Been, 2015). to scale computational trends anchored in mechanics and a
preference for simple, plausible mathematical form rather than
Concluding remarks treating calibration chamber data as ‘absolute facts’ that must be
The practical importance of the bandwidth of the CPT calibration matched. However, it must be acknowledged that the spherical cavity
data, as illustrated in Figure 14, is poorly appreciated. Consider a expansion is not as close an analogue to the CPT as desirable.
CPT sounding in a sand which produced a measured normalised
penetration resistance of Qp = 70. If a calibration trend is adopted at One obvious missing factor in the spherical cavity analogue is the
the low side of the bandwidth, the inferred state is y ≈ −0·14, a state load from the CPT shaft. In sands, this shaft load is on the order of
that will correspond to a very acceptable soil behaviour under most 10% of the tip load (the friction sleeve is 15 times larger than the tip
loadings, with substantial dilation and so forth. Liquefaction will not area, and F ≈ 0·5% is common) increasing to perhaps 30% of the tip
be a serious concern for such dense soil. Conversely, if the load in silts with their greater F values. And that does not include
calibration trend adopted is at the high side of the bandwidth the shaft friction loads from further behind the sleeve. Quite how this
inferred state is y ≈ −0·04 for exactly the same penetration resistance neglected load (within the spherical cavity analogue) becomes
Qp = 70, such a state is vulnerable to liquefaction (static and/or distributed away from an actual CPT is an interesting question, and it
cyclic) as, for example, as occurred in the recent billion-dollar failure is certainly a plausible cause for CQ ≈ 4 rather than CQ ≈ 2.
at Fundão dam (Brazil). The entire engineering can depend on the
CPT calibration used – ‘high’ or ‘low’ trend – not the measured CPT Despite the precision of the numerics offered (see the verification
resistance (Qp = 70 in this instance). How should the calibration be plots in Figures 17 and 18), and the very reasonable matches of
chosen? And the situation becomes more troublesome when dealing NorSand to soil behaviour of interest (e.g. Figures 9 and 25), these
with silts, as there are no reference calibration studies in silts at all. numerics are for the spherical cavity analogue, not the CPT itself.
Many practical engineers will still prefer a physical calibration, as
The work of Been et al. (1987b) was a milestone, as it was the there may be a CQ anxiety. Indeed, it is a little surprising that one
first to offer an approach to interpolating between existing sees so few calibration chamber studies in the context of billion-
calibrations for any particular sand. Strictly what was offered was dollar liabilities for many projects – surely some calibration tests
a plausible basis for viewing test data using general principles of might have been useful at Fundão? Lower-cost physical calibrations
critical-state soil mechanics. The formal mechanics were have been achieved by doing CPTu tests in a centrifuge (Bolton
subsequently provided by Shuttle and Jefferies (1998) and broadly et al., 1999), with the advantage of a smaller sample and the fact
supported by Been et al.’s (1987b) work while adding some that each test covers a range of effective stresses at the same void
refinement and insight. Subsequently, Ghafghazi and Shuttle ratio. Scaling of the soil particle size to the CPTu diameter is a

115
Downloaded by [] on [13/08/20]. Published with permission by the ICE under the CC-BY license
 Geotechnical Research Determining silt state from CPTu
Volume 3 Issue 3 Shuttle and Jefferies

complication with sands, and it is typical to check the five or so appropriate transducer choice) of all geotechnical tests – surely the
centrifuge tests with a couple of full-scale calibration chamber tests. CPT deserves the effort to understand this test so that the derived
Other researchers are investigating whether an even smaller CPTu soil parameters are as accurate as the measured data? This paper has
test in a triaxial specimen with good control of boundary stresses pushed the spherical cavity expansion analogue a little more, but
might be an alternative, but this is a relatively recent development really the industry needs more calibration studies, including in silts,
(Damavandi-Monfared and Sadrekarimi, 2015). and a much more advanced finite-element analysis.

Part of the reason that engineers are forced to look to calibration Acknowledgements
chamber studies is that the flow of soil past a CPT is a very The Neves-Corvo mine has been operated since 1988 by Sociedade
difficult computational problem with variable dilation – it is Mineira de Neves-Corvo (Somincor), a subsidiary of Lundin Mining,
unclear what streamlines might look like in variably dilating soils. with geotechnical aspects being the responsibility of Dr M. Oliveira.
But the research funding has never been there for this level of
computational soil mechanics – which is disappointing when the The in situ and laboratory work for Somincor was carried out by
worldwide extent of CPT testing is considered, and even more Golder Associates (Portugal). Eng R Bahia directed this work,
disappointing when the billion-dollar liabilities of some with substantial contributions from Dr N. Raposo and Dr R.
constructed works based on CPT data are considered. Olivera, PEng.

Overall, it is suggested that engineers reflect on this truth: the CPT The authors appreciated the opportunity to contribute to this most
is the most repeatable, operator-independent and accurate test (given interesting mine development.

Appendix

Internal model variables yi ¼ y þ l lnðs mi =s m Þ and Mi ¼ Mð1 − Nci jyi j=M
tc Þ 

2
Mtc 3q p
Critical state ec ¼ G − l lnðs m Þ and hc ¼ M ¼ Mtc − cos þ
3 þ Mtc 2 4



h sm s mx  
Yield surface and internal cap ðs mi ¼ s mx Þ ¼ 1 − ln with ¼ exp −ctc yi =Mi,tc
Mi s mi sm
Hardening rule On outer yield surface



s_ mi M s m 2 s mx s mi p h s_ mx
¼H i − e_ q þ S
s mi Mi,tc s mi sm sm hL s mx
where S = 0 or 1 (a Boolean switch) and hL = Mi (1 − ciyi /Mi,tc)
On internal cap
s_ mi H Mi  p 
¼− e_ 
s mi 2 Mi,tc q
Flow rule Stress dilatancy: Dp = Mi − h
p p
Dtc ¼ Dp Mi,tc =Mi and Dte ¼ Dp Mi,te =Mi

p p
2Dtc − 3 2Dte − 6
z3,tc ¼ p and z3,te ¼ p
6 þ 2Dtc 3 þ 2Dte
pffiffiffi pffiffiffi
a ¼ ðsin q þ 3 cos qÞ=3, b = −2 sin q/3, c ¼ ðsin q − 3 cos qÞ=3
Strain rate ratios consistent with idealised work dissipation postulate


p
e_ 3   3q þ 90
p ¼ z3 ¼ z 3,tc − z3,tc − z3,te cos
e_ 1 2
p
e_ 2 aDp − 1 þ ðe_ 3 =e_ 1 ÞðcDp − 1 Þ
p ¼ z2 ¼
e_ 1 1 − bDp
G 2ð1 þ nÞ
Elasticity Ir ¼ with K ¼ G, n ¼ constant at all times
sm 3ð1 − 2nÞ

Summary of monotonic NorSand

116
Downloaded by [] on [13/08/20]. Published with permission by the ICE under the CC-BY license
 Geotechnical Research Determining silt state from CPTu
Volume 3 Issue 3 Shuttle and Jefferies

REFERENCES FTDRP (Fundão Tailings Dam Review Panel) (2016) http://www.

Alonso EE and Gens A (2006) Aznacóllar dam failure. Part 1: field fundaoinvestigation.com/ (accessed 26/10/2016).
observations and material properties. Géotechnique 56(3): Ghafghazi M and Shuttle DA (2008) Interpretation of sand state
165–183, http://dx.doi.org/10.1680/geot.2006.56.3.165. from cone penetration resistance. Géotechnique 58(8):
ASTM (2012) D 5778-12: Standard test method for performing 623–634, http://dx.doi.org/10.1680/geot.2008.58.8.623.
electronic friction cone and piezocone penetration testing of Gibson RE (1950) Correspondence. Journal of the Institution of
soil. ASTM International, West Conshohocken, PA, USA. Civil Engineers 34(Supplement 8): 382–383.
Baligh MM (1985) Strain path method. Journal of Geotechnical Griffiths DV (1980) Finite Element Analysis of Walls, Footings and
Engineering 111(9): 1108–1136. Slopes. PhD thesis, University of Manchester, Manchester, UK.
Becker DE, Jefferies MG, Crooks JHA and Been K (2006) ISO (2012) ISO 22476-1: Geotechnical investigation and testing –
Geology, characterization and properties of Beaufort Sea clays. field testing – part 1: Electrical cone and piezocone penetration
Proceedings of the Second International Workshop test. International Organization of Standardization, Geneva,
Characterization and Engineering Properties of Natural Soils, Switzerland.
Singapore, pp. 1855–1891. Jefferies MG (1993) Nor-Sand: a simple critical state model for
Been K and Jefferies MG (1985) A state parameter for sands. sand. Géotechnique 43(1): 91–103, http://dx.doi.org/10.1680/
Géotechnique 35(2): 99–112, http://dx.doi.org/10.1680/geot. geot.1993.43.1.91.
1985.35.2.99. Jefferies MG (1997) Plastic work and isotropic softening in
Been K, Crooks JHA, Becker DE and Jefferies MG (1986) The unloading. Géotechnique 47(5): 1037–1042, http://dx.doi.org/
cone penetration test in sands: part 1, state parameter 10.1680/geot.1997.47.5.1037.
interpretation. Géotechnique 36(2): 239–249, http://dx.doi.org/ Jefferies MG and Been K (2000) Implications for critical state
10.1680/geot.1986.36.2.239. theory from isotropic compression of sand. Géotechnique
Been K, Lingnau BE, Crooks, JHA and Leach B (1987a) Cone 50(4): 419–429, http://dx.doi.org/10.1680/geot.2000.50.4.419.
penetration test calibration for Erksak (Beaufort Sea) sand. Jefferies M and Been K (2015) Soil liquefaction: a Critical State
Canadian Geotechnical Journal 24(4): 601–610. Approach, 2nd edn. CRC Press, Boca Raton, FL, USA.
Been K, Jefferies MG, Crooks JHA and Rothenberg L (1987b) The Jefferies MG and Shuttle DA (2002) Dilatancy in general
cone penetration test in sands: part II, general inference of Cambridge-type models. Géotechnique 52(9): 625–638,
state. Géotechnique 37(3): 285–299, http://dx.doi.org/10.1680/ http://dx.doi.org/10.1680/geot.2002.52.9.625.
geot.1987.37.3.285. Jefferies MG and Shuttle DA (2005) NorSand: features,
Bellotti R, Jamiolkowski M, Lo Presti DCF and O’Neill DA (1996) calibration and use. In Soil Constitutive Models: Evaluation,
Anisotropy of small strain stiffness in Ticino Sand. Géotechnique Selection, and Calibration (Yamamuro JA and Kaliakin
46(1): 115–131, http://dx.doi.org/10.1680/geot.1996.46.1.115. VN (eds)). American Society of Civil Engineers, Reston, VA,
Bishop AW (1950) Reply to discussion on ‘Measurement of shear USA, Geotechnical Special Publication no. 128, pp. 204–236.
strength of soils’. Géotechnique 2(2): 90–108. Jefferies MG and Shuttle DA (2011) On the operating critical
Bishop AW (1973) The stability of tips and spoil heaps. Quarterly friction ratio in general stress states. Géotechnique 61(8):
Journal of Engineering Geology 6(3–4): 335–376. 709–713, http://dx.doi.org/10.1680/geot.9.T.032.
Bishop RF, Hill R and Mott NF (1945) Theory of indentation and Ladanyi B and Roy M (1987) Point resistance of piles in sand.
hardness tests. Proceedings of the Physical Society 57(3): Proceedings of the 9th Southeast Asian Geotechnical
147–159. Conference, Bangkok, Thailand, pp. 6–29.
Boliden (2016) http://www.boliden.com/Press/News/2015/ Lopes R, Bahia R, Jefferies M and Oliveira M (2013) Paste
Background-on-the-1998-dam-breach-at-the-Los-Frailes-mine- deposition over an existing subaqueous slurry deposit of high
in-Spain/ (accessed 26/10/2016). sulphide content tailings – the Neves Corvo experience. In
Bolton MD (1986) Strength and dilatancy of sands. Géotechnique Proceedings of Paste 2013, Australian Centre for Geomechanics
36(1): 65–78, http://dx.doi.org/10.1680/geot.1986.36.1.65. (Jewell R, Fourie A and Pimenta J (eds)). Australian Centre for
Bolton MD, Gui MW, Garnier J et al. (1999) Centrifuge cone Geomechanics, Perth, Australia, pp. 21–36.
penetration tests in sand. Géotechnique 49(4): 543–552, Lopes R, Bahia R, Jefferies M and Oliveira M (2015) Upstream
http://dx.doi.org/10.1680/geot.1999.49.4.543. stacking of thickened tailings at Neves Corvo. In Paste 2015 –
Carter JP, Booker JR and Yeung SK (1986) Cavity expansion in Proceedings of the 18th International Seminar on Paste and
cohesive frictional soils. Géotechnique 36(3): 349–358, Thickened Tailings (Jewell R and Fourie A (eds)). Australian
http://dx.doi.org/10.1680/geot.1986.36.3.349. Centre for Geomechanics, Perth, Australia, pp. 521–534.
Collins IF, Pender MJ and Yan W (1992) Cavity expansion in sands Mohajeri M and Ghafghazi M (2012) Ground sampling and
under drained loading conditions. International Journal for laboratory testing on low plasticity clays. Proceedings of 15th
Numerical and Analytical Methods in Geomechanics 16(1): 3–23. World Conference on Earthquake Engineering, Lisbon, Portugal.
Damavandi-Monfared S and Sadrekarimi A (2015) Development Mount Polley Independent Expert Engineering Investigation and
of a miniature cone penetrometer for calibration chamber Review Panel (2016) http://www.mountpolleyreviewpanel.ca/
testing. Geotechnical Testing Journal 38(6): 878–892. (accessed 26/10/2016).

117
Downloaded by [] on [13/08/20]. Published with permission by the ICE under the CC-BY license
 Geotechnical Research Determining silt state from CPTu
Volume 3 Issue 3 Shuttle and Jefferies

Nova R (1982) A constitutive model for soil under monotonic and for Numerical and Analytical Methods in Geomechanics
cyclic loading. In Soil Mechanics – Transient and Cyclic 22(5): 351–391.
Loads (Pande GN and Zienkiewicz OC (eds)). Wiley, Taylor ML, Cubrinovski M and Haycock I (2012) Application
Chichester, UK, pp. 343–373. of new ‘Gel push’ sampling procedure to obtain high
Plewes HD, Davies MP and Jefferies MG (1992) CPT based quality laboratory test data for advanced geotechnical
screening procedure for evaluating liquefaction susceptibility. analyses. 2012 New Zealand Society for Earthquake
Proceedings of the 45th Canadian Geotechnical Conference, Engineering Conference, Christchurch, New
Toronto, Canada, pp. 1–9. Zealand, Paper 123.
Peuchen J, Vanden Berghe JF and Coulais C (2010) Estimation of Teh CI and Houlsby GT (1991) An analytical study of the cone
u1/u2 conversion factor for piezocone. In CPT’10, 2nd penetration test in clay. Géotechnique 41(1): 17–34, http://dx.
International Symposium on Cone Penetration Testing (Robertson doi.org/10.1680/geot.1991.41.1.17.
PK and Mayne PW (eds)). Omnipress, Madison, WI, USA, van den Berg P (1994) Analysis of Soil Penetration. Delft
Paper 1-03. University Press, Delft, the Netherlands.
Randolph MF, Dolwin J and Beck R (1994) Design of driven piles Wikipedia (2016a) http://wikipedia.org/wiki/Aberfan_disaster
in sand. Géotechnique 44: 427–448, http://dx.doi.org/10.1680/ (accessed 26/10/2016).
geot.1994.44.3.427. Wikipedia (2016b) http://wikipedia.org/wiki/Mount_Polley_
Reid D (2014) Estimating slope of critical state line from cone mine_disaster (accessed 26/10/2016).
penetration test – an update. Canadian Geotechnical Journal Wikipedia (2016c) http://wikipedia.org/wiki/Bento_Rodrigues_
52(1): 46–57. dam_disaster (accessed 26/10/2016).
Reynolds O (1885) On the dilatancy of media composed of rigid Willson SM, Ims BW and Smith IM (1989) Finite element analysis
particles in contact, with experimental illustrations. of cone penetration. In Penetration Testing in the UK. Thomas
Philosophical Magazine 20: 469–481. Telford, London, UK, pp. 157–160.
Robertson PK and Campanella RG (1983) Interpretation of cone Wroth CP (1984) The interpretation of in situ soil tests.
penetration tests – part I: sand. Canadian Geotechnical Géotechnique 34(4): 449–489, http://dx.doi.org/10.1680/geot.
Journal 20(4): 718–733. 1984.34.4.449.
Rowe PW (1962) The stress dilatancy relation for static equilibrium Youd TL, Idriss IM, Andrus RD et al. (2001) Liquefaction
of an assembly of particles in contact. Proceedings of the Royal resistance of soils: summary report from the 1996 NCEER and
Society of London A 269(1339): 500–527. 1998 NCEER/NSF workshops on evaluation of liquefaction
Schofield A and Wroth CP (1968) Critical State Soil Mechanics. resistance of soils. Journal of Geotechnical and
McGraw-Hill, London, UK. Geoenvironmental Engineering ASCE 127(10): 817–833.
Smith IM and Griffiths DV (1988) Programming the Finite Yu HS (1990) Cavity Expansion Theory and Its Application to the
Element Method, 2nd edn. Wiley, Hoboken, NJ, USA. Analysis of Pressuremeters. PhD thesis, University of Oxford,
Shuttle DA and Cunning J (2007) Liquefaction potential of silts Oxford, UK.
from CPTu. Canadian Geotechnical Journal 44(1): 1–19. Zienkiewicz OC and Cormeau IC (1974) Viscoplasticity, plasticity
Shuttle DA and Cunning J (2008) Reply to discussion: and creep in elastic solids: a unified numerical approach.
Liquefaction potential of silts from CPTu. Canadian International Journal for Numerical Methods in Engineering
Geotechnical Journal 45(1): 142–145. 8: 821–845.
Shuttle DA and Jefferies MG (1998) Dimensionless and Zienkiewicz OC and Taylor RL (1991) The Finite Element Method.
unbiased CPT interpretation in sand. International Journal McGraw-Hill, New York, NY, USA, vol. 2.

HOW CAN YOU CONTRIBUTE?

To discuss this paper, please submit up to 500 words
to the editor at [email protected]. Your contribution
will be forwarded to the author(s) for a reply and, if
considered appropriate by the editorial board, it will be
published as a discussion in a future issue of the journal.

118
Downloaded by [] on [13/08/20]. Published with permission by the ICE under the CC-BY license