Soils and Foundations 2012;52(4):668–681

The Japanese Geotechnical Society

Soils and Foundations

www.sciencedirect.com
journal homepage: www.elsevier.com/locate/sandf

On the compression behaviour of reconstituted soils

B. Shiptona,n,1, M.R. Coopb,1
a
Arup Geotechnics, Ove Arup and Partners, 13 Fitzroy Street, London W1T 4BQ, United Kingdom
b
City University of Hong Kong, Hong Kong

Received 31 October 2011; received in revised form 12 May 2012; accepted 19 June 2012
Available online 7 September 2012

Abstract

Transitional behaviour has been recognised in a diverse range of soils in the literature to date, from gap-graded soils to well-graded silts and
sands. It is typified by non-convergent compression paths and critical state lines that are non-unique and which are dependent on the initial
sample density. Many soil mechanics theories assume soils to have unique normal compression and/or critical state lines, which is not the case
for a transitional soil. For such a soil it would therefore be difficult to identify the effects of structure on the mechanics since there is no unique
behaviour of the soil when reconstituted. This paper describes series of oedometer tests that were performed to investigate in more detail when
and why non-convergent compression behaviour might be expected, as a first step to identifying which soils are transitional. The effects of
mixing soils of different grain sizes and mineralogies were explored, the tests revealing that convergent or non-convergent behaviour could be
brought about either by relatively small changes to the proportions of the soil particles or by changes to their nature. It was also found that in
some soils with non-convergent compression behaviour there was particle breakage while in others there was not. Since the factors that influence
the mode of compression behaviour were found to be complex, it is concluded that each new soil encountered must be assessed individually for
whether it is transitional, and that the accuracy of void ratio measurements is of particular importance in establishing this reliably.
& 2012 The Japanese Geotechnical Society. Production and hosting by Elsevier B.V. All rights reserved.

Keywords: Compression; Particle breakage; Void ratio; Sandy soil; Artificially mixed soils; Laboratory tests

1. Introduction physical-chemical forces between the particles control the

behaviour whereas in coarse grained soils the stresses are
Over the past two decades research in soil mechanics has transmitted through inter-particle contacts in strong force
shown that the behaviour of many clean sands can be chains that result in particle breakage if the applied stresses
described within a critical state framework similar to that are sufficiently large. Another key difference is that sands are
for clays (e.g., Been and Jefferies, 1985; Coop and Lee, deposited at much denser states than clays, and so as they are
1993; McDowell and Bolton, 1998). The main difference compressed to higher stresses they may meet a unique
between the frameworks for sands and clays is the Normal Compression Line (NCL) in v:logs0 v or v:lnp0 space
mechanism of plastic volumetric compression. In clays (v, specific volume; s0 v, vertical effective stress; p0 , mean
normal effective stress) (Coop and Lee, 1993; McDowell and
n
Corresponding author. Bolton, 1998; Pestana and Whittle, 1995). The increase of
E-mail address: [email protected] (B. Shipton). gradient or ‘‘knee’’ point in the compression paths as they
1
Formerly Imprial College, London reach the NCL is generally believed to correspond to the
Peer review under responsibility of The Japanese Geotechnical Society.
onset of major particle damage and can be correlated with
the strength of individual particles (e.g., Kwag et al., 1999),
although it is not the true onset of plastic strains. The specific
volume and stress level at which the compression paths reach

0038-0806 & 2012 The Japanese Geotechnical Society. Production and hosting by Elsevier B.V. All rights reserved.
http://dx.doi.org/10.1016/j.sandf.2012.07.008
 B. Shipton, M.R. Coop / Soils and Foundations 52 (2012) 668–681 669

a unique NCL will also depend on the initial density of the

sand because of the effect that has on the coordination
number. It is therefore typical that clean quartz sands reach
their NCL at specific volumes in the range 1.3–1.6 and
at stress levels typically of 10–100 MPa (e.g., Mesri and
Vardhanabhuti, 2009). In contrast other sands with higher
volumes and weaker particles can reach their NCLs at much
lower stresses. For example the Dogs Bay carbonate sand
tested by Coop and Lee (1993) reached its NCL at specific
volumes in the region of 2.2–2.6 and at stresses in the range
1–3 MPa.
More recent research has identified that there are many
intermediate graded soils which have behaviour that cannot
be described within a simple critical state framework. These
soils have been called ‘transitional’ in that they have a grading
and a mode of behaviour between that of clean sands and
plastic clays. The main feature of transitional soil behaviour is
that unique normal compression and critical state lines cannot
be identified (e.g., Martins et al., 2001; Ferreira and Bica,
2006; Nocilla et al., 2006). The compression paths of all soils
must eventually become asymptotic to the minimum specific Fig. 1. One-dimensional compression paths of a model quartz sand–
volume of 1 and in this sense they must all eventually kaolin mixture containing 75% sand and 25% kaolin (redrawn from
converge. For data of Ferreira and Bica (2006) this must be Martins et al., 2001).
beyond the maximum stress they used of about 25 MPa, but
between this stress level and the possible locations of the further research was conducted, it was realised that this type
horizontal asymptotes, it did not seem likely that there could of behaviour was likely also to be applicable to many
exist a unique NCL since the void ratios had already reduced intermediate graded soils (Nocilla et al., 2006), challenging a
to very low values in the region of 0.2 at 25 MPa. Further common assumption that most or all soils can be described
aspects of the conventional critical state framework also break within a critical state framework. Nocilla et al. (2006) searched
down for their shearing behaviour, such as the existence of for transitional behaviour in a well-graded Italian river silt, for
unique State Boundary Surfaces (Nocilla et al., 2006). which the clay content was controlled. As the clay content
Unfortunately, the term ‘transitional’ has already been used reduced the behaviour changed from that of a typical clay
to describe two other aspects of soil behaviour, firstly an mode, with unique NCLs and CSLs, to a transitional one
intermediate shearing mode between turbulent particle motion with non-unique NCLs and CSLs. Nocilla and Coop (2008)
and sliding at the residual state (Lupini et al., 1981) and observed features of transitional behaviour in natural samples
secondly the transitional or limiting fines content when soil of the same well-graded silt, confirming that the transitional
behaviour changes from coarse grained to fine grained behaviour they had observed was not simply a function of the
dominated (e.g., Pitman et al., 1994; Lade and Yamamuro, reconstituted fabrics created in the laboratory. More recently,
1997; Yang et al., 2006). As yet it is unclear exactly how each Altuhafi and Coop (2011) have found that as they changed the
definition is related. The type of behaviour associated with the grading of three sands of different mineralogies from poorly
most recent use of ‘transitional’, of concern in this paper, was to well-graded, their compression behaviour changed from
first identified in gap-graded soils and was initially thought to a ‘‘sand’’ type with a unique NCL associated with particle
be a feature exclusive to their behaviour. Martins et al. (2001) breakage, to a non-convergent one where there was no unique
identified non-convergent compression behaviour for the gap- NCL. In each case when the non-convergent behaviour was
graded Botucatu residual sandstone. Ferreira and Bica (2006) observed there was no measurable breakage during compres-
confirmed that even if the compression was taken to very sion, even up to 100 MPa.
high stresses (25 MPa), convergence still could not be found The work described in the paper presents a new series
and also that the location of the critical state line of the of oedometer tests and particle size analyses that were
reconstituted soil was similarly dependent on the initial conducted to investigate the role of grading, mineralogy
density, so that the soil might be defined as transitional. and particle breakage in the compression of a wide variety
Critical states were defined to be those states of constant stress of reconstituted soils. Various combinations of fine and
and volume at large strains in a triaxial apparatus, although it coarse particles were used to create different soils to help
was realised that the non-unique volumes would mean that identify which types of soil display non-convergent beha-
the fabric was not unique at that state. viour behaviour and to try to understand how the different
Non-convergent compression paths were also found for a particle types can influence the compression and particle
mixture of kaolin and a quartz sand (Fig. 1), which replicated breakage behaviour. Non-convergent compression beha-
the grading but with a simpler mineralogy. However, as viour may be seen as a first step in identifying transitional
670 B. Shipton, M.R. Coop / Soils and Foundations 52 (2012) 668–681

Table 1
Summary of test results from this paper and the existing literature.

Soil type Reference Origin/principal Mixes tested Observations on transitional and/or

mineralogy convergence of compression behaviour

Botucatu residual Martins et al. (2001) Natural origin, residual Tested in reconstituted Clear transitional behaviour for reconstituted
sandstone Ferreira and Bica (2006) aeolian. and intact state soil, but not for intact.
quartz/feldspar
Quartz sand/kaolin Martins et al. (2001) Artificial mix. 75% quartz sand, 25% Non-convergent compression behaviour
Quartz/kaolin kaolin.
Clayey silts Nocilla et al. (2006) Fluvial soils from river Reconstituted samples Reconstituted: transitional for clay contents
Nocilla and Coop (2008) embankments and sub- with controlled clay of 3.5, 8 and possibly 20%, not at 45%.
soil. quartz/calcite contents (3.5–45%) and Intact samples probably transitional for
intact samples with 13–16 and 20–26% clay contents, probably
variable clay contents not for 39–53%.
(13–53%).
Thanet sand Ventouras (2009) Predominantly quartz Reconstituted samples of Non-convergent compression behaviour in
Ventouras and Coop high fines stratum (15– high fines stratum but not in low fines
(2009) 30% fc) and low fines stratum.
stratum (5–10% fc)
Sands Altuhafi et al. (2010) Reconstituted single Leighton Buzzard quartz In all three sands evolution from convergent
Altuhafi and Coop mineralogy sands sand, Dogs Bay sand mode of compression behaviour to non-
(2011) carbonate sand and convergent mode as grading changed from
basaltic glacial sand poorly to well-graded
Carbonate sand/kaolin This paper Artificial mix. carbonate/ Mixes with 5, 10, 40 and Non-convergent compression behaviour for
kaolin 70% sand. 5, 10 and 40% sand. Convergent sand mode
of compression behaviour for 70% sand.
Carbonate sand/crushed This paper Artificial mix. carbonate/ Mixes with 0, 40 and Convergent compression behaviour for all
quartz silt quartz 60% sand. mixes.
Carbonate sand/quartz This paper Artificial mix. carbonate/ 0, 10, 30, 50, 70 and Non-convergent compression behaviour for
sand quartz 100% carbonate sand 10 and 30% carbonate sand. Convergent
sand mode of behaviour for all other mixes.
Quartz sand/kaolin This paper Artificial mix. quartz/ Mixes with 0, 25, 35, 50, Non-convergent compression behaviour for
kaolin 55, 60, 65, 70, 75, 90 and 75 and 90% sand. Clay mode of behaviour
100% sand. for sand r70%. Convergent sand mode for
100% sand.
Quartz sand/crushed This paper Artificial mix. quartz/ Mix with 75% sand. Non-convergent compression behaviour.
quartz silt quartz
Quartz sand/crushed This paper Artificial mix. quartz/ Mix with 25% sand, Non-convergent compression behaviour.
quartz silt/kaolin quartz/kaolin 25% silt, 50% kaolin.
Clays This paper Kaolin, London clay Reconstituted samples Slight non-convergent compression
(illite) and Oxford clay behaviour with very slow convergence
(illite) especially for kaolin and Oxford clay.

soil behaviour, although it is not necessarily diagnostic as complex effects that these might have on the mechanical
a soil may have non-unique compression paths but still behaviour of the soil, and that it facilitates experimental
have a unique CSL. The characteristics of the soils tested repeatability. Most of the soil mixtures tested were mixed or
and discussed, both in the existing literature and in this gap-graded in nature.
paper are summarised in Table 1. The grading curves and properties of the materials used are
given in Fig. 2 and Table 2. Dogs Bay sand, Thames Valley
2. Materials and fabricated mixtures sand and Redhill sand were used to form the coarse fraction
of most of the oedometer test samples. Dogs Bay sand is a
The behaviour of intermediate graded, mixed graded and biogenic carbonate sand from Ireland consisting mainly of
transitional soils is not well understood, mainly due to the unbroken angular mollusc and foraminifera shells, many of
complex interactions between the fine and coarse grains which have intra-particle voids. These features result in a
present in the soils and the broad range of soils which these poorly-graded soil with fragile particles and high specific
can include. To understand clearly the fundamental behaviour volumes. A comprehensive investigation of the behaviour of
of such soils it was therefore decided that much of the Dogs Bay sand was carried out by Coop (1990). Thames
laboratory research should concentrate on testing model soils. Valley sand is a poorly-graded river terrace sand from the
The main advantages of testing artificially created soils are UK that was chosen as a replacement for the extensively used
that the specimens can be designed to have controlled and Ham River sand (Bishop et al., 1965), and so has very similar
simple gradings and mineralogies, eliminating any more properties (Takahashi and Jardine, 2007). It is sub-rounded
 B. Shipton, M.R. Coop / Soils and Foundations 52 (2012) 668–681 671

the technique used and the definition of particle size that is

adopted. The QicPic apparatus scans particles falling under
gravity with a laser that allows a randomly orientated
outline of the particles to be recorded, which is a very
different process to sieving. Of the various diameters that
the apparatus calculates, the Feret minimum diameter was
used as this gives particle sizes that are closest to those
measured by sieving (e.g., Altuhafi and Coop, 2011). The
limitation of its resolution means that it can only be used,
however, with the coarser fraction.
Because a memory of the initial void ratio of a soil can
only really reside in its fabric, a variety of preparation
methods was used so that it could be seen whether fabric
and the current void ratio was simply related to the current
stress and initial void ratio, or whether the preparation
method was also important. However, the majority of
samples were created as slurries at a variety of initial water
contents by mixing the dry components with deionized water
Fig. 2. Grading curves of soils tested. before carefully placing the resulting slurry directly into the
oedometer ring. Care was taken to ensure saturation by de-
airing the slurries under vacuum. The de-aired slurry was
then carefully spooned into the oedometer ring with great
Table 2 care taken to avoid entrapment of air. It was then carefully
Material properties (references in text). levelled across the top of the ring. Other samples were made
by either dry compaction or wet compaction, with the initial
Material f0 CS (deg.) Gs LL (%) PL (%)
void ratio being controlled by the compactive effort. Com-
Dogs Bay sand (Ireland) 40.3 2.71 – – paction was carried out by tamping, a greater number of
Thames Valley sand (UK) 33 2.67 – – tamps and/or a greater tamping rod weight being used for the
Redhill sand (UK) – 2.65 – – denser samples. For wet compaction a small initial water
Kaolin (UK) 23 2.61 62 32
Crushed quartz silt (HPF4) (UK) 33 2.65 – –
content was used to allow the soil to be placed in very loose
states, although it was compacted sufficiently to ensure that
there were no macro-voids.
to sub-angular and consists predominantly of quartz. The
In all cases, the samples were saturated through inundation
third sand, Redhill sand, is from the Cretaceous, Lower
of the water bath of the oedometer after an initial load
Greensand also of the UK, and has sub-rounded particles
increment, typically of 10–15 kPa, had been placed. This
and again is predominantly quartz.
initial load was to ensure good contact from the sample
The material used to constitute the fine fraction of the
through the platen to the loading arm so minimising errors
mixtures tested was either Speswhite kaolin or a crushed
due to bedding of the platens or lack of good contact with the
quartz silt known as HPF4. Both are commercial soils which
loading system, while also preventing swelling while the water
are readily available from UK suppliers. The crushed quartz
bath is filled. All of the oedometer tests were incrementally
silt has angular and elongated particles and its behaviour was
loaded, using sample diameters of 50 and 38 mm so that the
investigated by Zdravkovic (1996). Most of the fabricated
elevated stress levels required could be reached. The initial
mixtures created for testing were limited to two types of soil
heights were around 20 mm. For each mixture a check was
particle, one fine and one coarse.
made of the required time for each loading stage to ensure
that the continuing settlements from creep or undissipated
3. Experimental procedures pore pressure were not significant. Typically loading stages of
around several hours each were adequate.
The proportions, mineralogy and particle sizes of the fine
and coarse grains in the mixtures were varied and the 4. Calculation of the initial specific volume
samples created were tested in the oedometer at a range of
initial densities and to a variety of stress levels. Particle size For the identification of a transitional soil or at least
analyses using hand sieving and also a Sympatec QicPic confirmation of whether the compression behaviour is
particle analyser were conducted on selected samples before convergent or not, the accuracy of the calculation of the
and after testing to investigate the role of particle breakage initial specific volume, vi, is of paramount importance,
for the soils tested. For perfect spheres different techniques ensuring that the locations of the compression paths in
of measuring particle size should give the same result, but v:logs0 v space are accurate. This will determine whether it
for irregular particles the measured size will depend on can be conclusively determined whether the compression
672 B. Shipton, M.R. Coop / Soils and Foundations 52 (2012) 668–681

curves from different initial densities are convergent towards saturation at the start of the test was found to reduce the
a unique NCL or not. Fourie and Papageorgiou (2001) scatter in the values of vi and so improve its accuracy.
discussed the errors in the calculated void ratio from Despite de-airing under vacuum after mixing, the resulting
measurements of the height and diameter of samples using initial degrees of saturation of the slurries were typically
callipers. It was found that the resulting error in void ratio around 0.9, although the degree of saturation is likely to
was 0.003 to 0.006 for an aluminium dummy specimen but have improved through immersion during the test, which
the error was thought to be larger for an actual soil emphasises the need to account for this in calculating the
specimen. Because of the need to prove rigorously whether initial void ratio.
the compression curves converge or not, it was felt it was not For each oedometer test conducted, the values of vi
sufficiently accurate simply to estimate the likely accuracy in calculated from the four methods outlined above were
this way, but that actual comparisons of different means of compared and any anomalous values were discarded (value
measuring vi should be made. No.4 in Table 3). An average of the remaining values was
Table 3 shows an example of how the initial specific taken and the resulting vi was the value used for analysis. It
volume, vi, was calculated using four different equations. This is difficult to calculate a meaningful accuracy of the vi data
has been deliberately chosen as a typical case, and indeed one with only four values and so the approach adopted has
value of the four calculated is poor. The four equations used been to use the maximum difference between any indivi-
were chosen to be as independent of each other as possible, dual value and the mean (0.02 in this case), although this is
although there is of course some linkage between them. They likely to be an overestimate. The acceptable error for a
were based on the initial water content, wi, the initial dry unit given soil depended on how large the differences were in
weight, gdi, the initial bulk unit weight, gi, and the final water the location of the compression curves. For larger differ-
content, wf. In the case of the final water content method the ences in the compression curve locations larger errors
initial specific volume is derived from the final value by could be tolerated without jeopardizing the proof that
means of the overall volumetric strain measured during the the soil was transitional. For most soils tests were typically
test. Each method depends on and is sensitive to various discarded if they had estimated errors of greater than
measurements such as the initial height, the final wet mass about 7 0.05, and the typical mean error was around
and the specific gravity, Gs. They also ignore any possible 7 0.02–0.03. It is interesting that despite the great care
effect of the compressibility of the solids. Because there are taken with the testing techniques, the estimated accuracies
four methods to calculate vi, there is redundancy in the are rather poorer than those proposed by Fourie and
calculations and so there is no need for the methods based on Papageorgiou (2001), but they are similar to those of
wi, and gi to assume that the soil is fully saturated, even if it is others investigating transitional soils and using similar
likely for the samples made as slurries. An average value techniques (e.g., Nocilla et al., 2006).
of the degree of saturation, Sr, can be then calculated from
Eqs. (1) and (2). 5. Laboratory results
wi Gs
Sri ¼ ð1Þ 5.1. Carbonate sand–clay mixtures
vi 1

Gs þ Sr ðvi 1Þ Four different mixtures were created which had a coarse
gi ¼ gw ð2Þ fraction of Dogs Bay sand and a fine fraction of kaolin.
vi
The proportions by dry weight of sand were 5, 10, 40,
The value of vi used in Eqs. (1) and (2) is the mean of and 70%. The size range of the Dogs Bay sand was
the two values calculated from the gdi and the wf methods limited to the coarser gradings (0.3–0.425, 0.425–0.6 and
(Eq. (2) and 4 of Table 3), the latter assuming that the 0.6–1.18 mm), which, as will be discussed below, was to
sample is fully saturated at the end of the test, which is highlight any particle breakage that occurred.
more likely than at the start of the test. Not assuming The oedometer test results of the four mixtures are
shown in Fig. 3. Also included for comparison are the one-
dimensional NCLs for the pure kaolin and for Dogs Bay
Table 3
Example of calculation of the initial specific volume. sand, although the latter is for the natural grading
(unfortunately the coarser gradings of the Dogs Bay sand
a,b
Dogs Bay Test vi equation Chosen Estimated are only present in small quantities and so there was
sand : Kaolin vi c estimated
1 2 3 4
insufficient soil to determine the NCLs of the specific
(%) accuracy
gradings used). In some cases different sand particle sizes
5:95 5DBS-k 2.49 2.48 2.46 2.68 2.47 70.02 were used (Fig. 3a, b and d) but this has no clear effect
03 within the accuracy of the tests.
a
Calculated from: (1) vi ¼(wiGs/Sr)þ 1, (2) vi ¼Gsgw/gdi, (3) vi ¼ (Gs Sr)/
It is surprising how the addition of relatively small
(gi/gw  Sr), (4) vi ¼ (wfGs þ 1)/(1 ev). amounts of sand affect the compression behaviour quite
b
Values shown in bold italics not considered for ‘Chosen vi’. severely and already at 5 and 10% sand contents non-
c
Average of vis calculated from Methods 1–3. convergent compression behaviour is seen. The mean of
 B. Shipton, M.R. Coop / Soils and Foundations 52 (2012) 668–681 673

Fig. 3. Oedometer compression and swelling curves for Dogs Bay carbonate sand–kaolin mixtures. (a) 5% sand. (b) 10% sand. (c) 40% sand.
(d) 70% sand.

the estimated accuracies for all of the tests on the Dogs changing the fines content is a translation in the v:lnp’ plane
Bay sand: kaolin mixtures was 70.02 and the spacing rather than a change in gradient. Here the overall locations
between the compression paths is rather larger than this. of the compression curves do not however change that much,
The compression paths are not completely parallel and remaining slightly below the NCL of the pure kaolin, which
there is a gentle convergence, the denser samples having seems to form an upper bound to the data for the various
slightly lower compression indices, Cc, but the convergence mixtures. In contrast, the swelling index, Cs simply reduces
is far too slow for there to be any possibility of arriving at steadily with increasing sand content.
a unique Normal Compression Line before the compres- At 40% sand content there is very clear non-convergent
sion curves start tending towards their horizontal asymp- compression behaviour and a knee point starts to appear
totes as they approach v ¼ 1 at very high pressures. in the compression paths. This knee is much more apparent
Because the Cc values tend to change with the initial for the 70% sand content, but for this mixture the behaviour
density, it is not easy to make direct comparisons between the has returned to one of a unique NCL. It is interesting that the
different gradings, but as the sand content is increased, the mixed soil reaches its NCL at rather lower specific volumes
average compression index tends to decrease initially before than the original Dogs Bay sand, which might be explained by
increasing again at the higher sand contents. There have been the infilling of the voids with fines, but also the stresses at the
many investigations of how the gradient and location of point of reaching the NCL are much lower, which would not
NCLs and CSLs vary in the volumetric plane. Some have be expected for a denser soil. However, these comments are
seen this type of pattern of decreasing and then increasing somewhat speculative as the gradings of the Dogs Bay sand
gradient as the fines content increase (e.g., Carrera et al., used are slightly different.
2011) while others have only seen a continuously increasing For clean sands the steepening or knee of the compression
gradient (e.g., Nocilla et al., 2006; Yin, 1999). In some cases paths as the NCL is reached has been found to be due to the
(e.g., Thevanayagam et al., 2002) the dominant effect of onset of significant breakage (e.g., Coop and Lee, 1993;
674 B. Shipton, M.R. Coop / Soils and Foundations 52 (2012) 668–681

McDowell and Bolton, 1998) and so the straighter compres- 100

A B

sion paths at higher clay contents might be taken as

an indication that particle breakage did not significantly 80
0.074mm
influence the compression behaviour.

% passing
60
Throughout this paper the compression data have been
presented with conventional linear v: log stress axes, as are 40
typically used in critical state soil mechanics. Other authors initial grading
20 D
have pointed out that the Normal Compression Line may be C
after test
curved with these axes and have proposed alternatives. 0
Butterfield (1979) proposed using log v while Pestana & 0.01 0.10 1.00 10.00
Whittle (1995) used log e (void ratio). McDowell (2005) particle size (mm)
provided a micro-mechanical justification for linearity of the total breakage, Bt = area BCDB
breakage potential, Bp = area BCAB
Normal Compression Line with a log e axis for clean sands, relative breakage, Br = Bt/Bp
based on the fractal theory of particle crushing. The principle
purpose of the work presented in this paper, however, is to
observe whether convergence is seen in compression or not,
rather than to examine which axes provide the best linearity
of the Normal Compression Line. With such a wide range of
soils that are investigated here, for which the micro-mechan-
ical processes are also likely to be varied, it is also most
unlikely that a single set of axes would be optimum for all the
soils. A logarithmic void ratio axis was tried for Fig. 3(d) but
the resulting plot gave a Normal Compression Line that was
still not straight but convex upwards rather than concave.
The effect of changing the fines content observed here is
also somewhat in contrast to that seen by Nocilla et al.
(2006), who found that as they varied the clay content of a
well-graded silty sand they obtained a transitional type of
behaviour at low clay contents (8 and 3%) whereas a
unique NCL and CSL was found at higher clay contents
(40%). For these gap-graded soils a NCL of a sand type,
with a well-defined knee in the compression paths is seen
even with a clay content of 30% and no unique NCL of a
clay type could be found for a clay content of 95%. As for
Nocilla et al., a variety of sample preparation methods was
used, most of the samples being made by mixing the soil
into a slurry, but some being created by dry compaction
and some by compaction in a partially saturated state (wet
compaction). Again, no effect of the sample preparation
method could be found, so that the effect of the initial
density seems to be far more important than any influence
of the initial fabric created by the different techniques.
Fig. 4 shows an analysis of the breakage that occurs in the
sand fraction using Hardin’s (1985) relative breakage (defined
in Fig. 4a). There are several means available for quantifying
particle breakage, for example the total surface area of the
particles, as used by Miura and Yamanouchi (1977) and
Miura and O-Hara (1979). However this makes the assump-
tion that the particles are spherical, which would be a poor Fig. 4. Particle breakage of the sand fraction for the Dogs Bay carbonate
sand: kaolin mixtures. (a) Definition of relative breakage (Hardin, 1985).
assumption for the Dogs Bay sand. The key disadvantage of
(b) Influence of stress level. (c) Influence of sand content.
relative breakage is that it ignores any breakage in the fines.
This is not so much of a difficulty for these gap graded soils,
with the exception perhaps of the silt where some breakage
might have occurred, but as discussed below, in other tests on distribution of the coarse fraction was determined using
the pure silt no breakage could be measured. the QicPic or for the 40% sand content data by careful
For all of the tests the soil was wet sieved to separate dry sieving in the case of the 0.6–1.18 grading. The QicPic
out the fines passing 63 mm and then the particle size was especially useful for the very small amounts of sand
 B. Shipton, M.R. Coop / Soils and Foundations 52 (2012) 668–681 675

used in the low sand content samples, often as low as 1– 5.2. Carbonate sand–quartz silt mixtures
2 g, which would have been very difficult to sieve accu-
rately. For comparison the data for isotropic compression Oedometer tests were also conducted on mixtures of
of the Dogs Bay sand in its original grading (D50 ¼ 0.20 Dogs Bay sand and crushed quartz silt (HPF4) to inves-
mm) from Coop (1990) are included on the plots, which tigate whether the mineralogy and particle size of the fine
were again done by sieving. fraction had an effect on the compression and amount of
Fig. 4b shows the effect of the stress level and Fig. 4c particle breakage. The Dogs Bay sand grading used was
that of the sand content for tests taken to maximum 0.425–0.6 mm and the average estimated accuracy of the
stresses of about 8 MPa (7 0.5 MPa). For the isotropic specific volumes for these tests was about 7 0.01.
compression data and those for the 40% sand content of For the pure silt (Fig. 5a) the compression curves gently
the 0.6–1.18 mm grading a mean interpolated point was converge at high stress levels, although a QicPic analysis of
used as there was no specific data point near 8 MPa. The the tested soil was unable to identify any breakage. This is
effect of stress level for the 40% sand content data an especially interesting result because it means that in
resemble that which was determined by Coop (1990) for clastic soils the knee and large plastic strains in compres-
isotropic compression of the natural grading, with break- sion are not necessarily associated with any major particle
age starting at around a few tens of kPa. Although breakage. The QicPic has a resolution, however, of a few
the amount of breakage is very much less for the 10% mm, so although it would identify any significant change in
sand content, it is still significant, which is perhaps the major part of the grading curve, it may miss breakage
surprising since the granular void ratios {eg ¼ (e þ fc)/ of the finest particles.
(1  fc), where fc is fines content} of this mixture are in Convergence is also seen for the 40 and 60% Dogs Bay
excess of 20, so that the sand particles are most unlikely to sand mixtures (Fig. 5b and c) and the knee point in the
touch each other. It might have been expected that the compression curves tends to become sharper and move to
cushioning effect of the fines around the ‘‘floating’’ sand lower stresses and higher specific volumes as the propor-
particles would have prevented breakage and further tion of the Dogs Bay sand increases. It is clear from these
research is needed to determine the nature of this breakage tests that whether convergent compression behaviour is
and why it occurs. seen or not depends critically on the nature of the fines,
The much larger amount of breakage in the 40% sand especially whether they are plastic or non-plastic. For one
samples and the fact that the breakage starts at around sample of the 40% Dogs Bay sand: 60% silt mixture the Br
100 kPa, which is also about where the knee occurs in the value was measured for the sand fraction taken after a test
compression paths, tends to confirm the idea that at these to about 9 MPa (although for this test the sand particles
sand contents the compression behaviour is essentially of were between 0.6 and 1.18 mm). The data point is shown
a ‘‘sand’’ type, even if the values of eg are still in the on Fig. 4b and emphasises that the amount of breakage in
approximate range 1.8–7.8, which might indicate that the the sand–kaolin and sand–silt mixtures is fairly similar,
number of direct sand–sand particle contacts may be confirming that the mineralogy and particle size of the
limited, especially at the higher void ratios. matrix material does not affect greatly the amount of
It is also interesting that the amount of breakage particle breakage.
depends on the particle size of the Dogs Bay sand, even
if there was no discernable effect on the compression 5.3. Carbonate sand–quartz sand mixtures
curves, indicating that these small amounts of breakage
in a small fraction of the soil have no measurable effect on Mixtures of two sands of very different mineralogies and
the volumetric behaviour. This is highlighted in Fig. 4c, particle types were made; Dogs Bay sand and Thames
where it is clear that the coarser particles are more prone to valley sand. The carbonate Dogs Bay sand has weak shelly
break and also that the amount of breakage decreases as particles that are more angular while the quartz Thames
the proportion of sand reduces. The data in Fig. 4c seem to Valley sand has much stronger, more rounded and solid
show some incompatibility between the two data sets for particles. The Thames Valley sand was tested in its natural
the different techniques for measuring the particle size grading shown in Fig. 2 and for the Dogs Bay sand the
distribution, the data for size ranges 0.6–1.18 mm that 0.212–0.3 mm sieve interval was used due to constraints on
were done by sieving perhaps giving slightly higher Br the amount of sand available. The samples were all made
values than those for the 0.8–1.18 mm range that were by dry compaction at a variety of initial densities.
measured by the laser analyser. In total six different mixtures were tested, the propor-
The particle breakages seen here for gap-graded mix- tions of Dogs Bay sand being 0, 10, 30, 50, 70 and 100%.
tures of clay with a fragile carbonate sand are in contrast The pure Dogs Bay sand and the pure Thames Valley sand
to Nocilla et al. (2006) who found that for well-graded gave unique one dimensional NCLs with well-defined knee
soils with transitional behaviour there was no measurable points as expected from previous research on clean, poorly
breakage. It therefore appears that particle breakage may graded sands of single mineralogy (e.g., Coop and Lee,
occur in soils with transitional or non-convergent com- 1993; Pestana and Whittle, 1995). The compression curves
pression behaviour, but it is not a prerequisite. for two of the intermediate mixtures are given in Fig. 6.
676 B. Shipton, M.R. Coop / Soils and Foundations 52 (2012) 668–681

Fig. 6. Oedometer test data and relative breakage for mixtures of Dogs
Bay carbonate sand and Thames Valley quartz sand. (a) 10% Dogs Bay
sand. (b) 70% Dogs Bay sand. (c) Relative breakage of the two sand
fractions.

As might also be expected, as the Dogs Bay sand content is

Fig. 5. Oedometer test data for mixtures of Dogs Bay carbonate sand of increased, the initial specific volume increases, due to the
0.425–0.6 mm particle size with crushed quartz silt. (a) 100% silt. (b) 40% particle angularity. The knee point where the NCL is
Dogs Bay sand and 60% silt. (c) 60% Dogs Bay sand and 40% silt. reached tends to occur at lower stresses and higher specific
 B. Shipton, M.R. Coop / Soils and Foundations 52 (2012) 668–681 677

volumes and the gradients of the compression paths at giving higher coordination numbers, and hence lower
higher stress levels increase. However, at 10% Dogs Bay contact forces between particles. It is also possible that
sand content there is a clear non-convergent compression the quartz sand has a protective effect, providing a rigid
behaviour and a less distinct knee, while at 70% content matrix sheltering the weaker Dogs Bay sand particles.
the compression curves converge, with more distinct knee
points as the NCL is reached. The average accuracy of the 5.4. Quartz sand–clay and quartz sand–crushed quartz silt
specific volumes for these tests was again estimated to be mixtures
about 7 0.02. The 50% Dogs Bay sand mixture also gave
a unique NCL while the 30% mixture did not (these data Oedometer tests were carried out on mixtures of Thames
have not been included for brevity). The type of compres- Valley sand and kaolin. A range of proportions of quartz
sion behaviour seen is therefore quite sensitive to the sand were used from 25 up to 90% at intervals of 5–10%,
proportions of the different sands used. along with tests on the pure kaolin and pure sand. With the
The breakage of each of the sand fractions was mea- exception of the clean sand all of the samples were made as
sured by first taking the final grading of the whole sample, slurries at a variety of different initial water contents.
dissolving and washing out the carbonate sand with Data from selected gradings are shown in Fig. 7. They are
Hydrochloric acid and finally repeating the grading. presented in terms of granular specific volume (vg ¼ 1þ eg) for
Because the QicPic laser scanner destroys the samples it the sand dominated samples and fines specific volume
tests, the gradings were done in this case by careful wet (vf ¼ 1þ ef where ef ¼ e/fc) for the clay dominated ones. The
sieving. Clearly there is substantial breakage whether or estimated accuracy of the tests was about 70.03. The effect
not a unique NCL is found and as expected there is much of using either is merely to translate the compression curves
more breakage in the carbonate sand than the quartz. It is but not to change the convergence. For low sand contents
also interesting that the amount of breakage in the Dogs (25%, Fig. 7a) the effect of the addition of the sand is to
Bay sand actually decreases as the proportion of that sand translate the compression curve downwards in the v:logs0 v
reduces. This is possibly the effect of the increasing initial plane, maintaining the same gradient as for pure kaolin, but
density of the samples with greater quartz sand content, when the data are plotted in terms of vf the translation is

Fig. 7. Oedometer test data for Thames Valley quartz sand: kaolin and Redhill quartz sand: crushed quartz silt mixtures. (a) 0% and 25% sand. (b) 0%
and 50% sand. (c) 100 and 90% sand. (d) 75% sand with kaolin (‘‘plastic’’) and crushed quartz silt (‘‘non-plastic’’).
678 B. Shipton, M.R. Coop / Soils and Foundations 52 (2012) 668–681

removed, indicating that the coarse fraction essentially has no

influence on the compression behaviour and is just an inert
addition making the soil slighter denser. At higher sand
contents (50%, Fig. 7b) the effect of adding the fines can no
longer be normalised with vf and the compression curves of
the mixture plot above those of the kaolin when normalised
in this way. The coarse fraction is therefore starting to affect
the compression. The data are a little scattered for the 50%
sand content, but still within the estimated error of the values
of vf. For the pure sand (Fig. 7c) the data tends towards a
unique line at large stresses, although they were not quite
loaded far enough to reach it. However, with a 10% clay
content, although the compression curves plot in the same
area when vg is used, the data no longer converge so
convincingly and the knee is less clear. The same was found
for a 25% clay content, but from 30% upwards the
behaviour was of a clay type similar to that in Fig. 7a
and b, with no clear knee point and a unique NCL. These
tests may be contrasted with those on the carbonate sand:
kaolin mixtures, for which a much clearer non-convergent
compression behaviour was seen over a much wider range of Fig. 8. Oedometer test data for quartz sand: quartz silt: kaolin mixture.
sand contents, and at much lower sand contents, emphasising
how sensitive the mode of behaviour is to the sand type.
Data for the 75% sand content are shown in Fig. 7d where
they are compared with oedometer tests on a quartz sand:
crushed quartz silt mixture. The latter mixture is especially
interesting as it has a uniform mineralogy but a gap grading,
although problems with sand supply meant that Redhill sand
had to be used, which is slightly coarser than the Thames
Valley sand (Fig. 2). Although in both cases there is some
convergence of the compression curves, they clearly cannot
reach and form any unique NCL if a linear v axis is used
before the curves approach the limiting v¼ 1, at which point
the curves must converge as they approach their horizontal
asymptote. This is consistent with the findings of Martins
et al. (2001) for their quartz sand: kaolin mix. The data for
the ‘‘non-plastic’’ quartz sand: crushed quartz silt mixture
contrast strongly with the carbonate sand: crushed quartz silt
mixtures for which convergent compression behaviour was
found. The effect of the plasticity of the added fines was Fig. 9. Oedometer tests on reconstituted high fines content Thanet sand
(data from Ventouras, 2009).
therefore quite different in the quartz sand to the carbonate
sand, since for the carbonate sand transitional behaviour was
only seen for some mixtures with plastic fines (i.e., kaolin) Fig. 8 indicate that this well graded artificial soil is clearly has
and not for any grading with the non-plastic fines (i.e., the non-convergent compression behaviour, as Nocilla et al.
crushed quartz). Again in contrast to the similar mixtures (2006) had found for controlled gradings of a naturally
using the carbonate sand, no breakage was detectable in the well-graded silty sediment.
quartz sand fraction after the tests using the QicPic analyser. Ventouras and Coop (2009) examined the behaviour of
the two distinct sub-strata that are found within the Thanet
5.5. Sand–silt–clay mixtures sand; a low fines content stratum (5–10%o63 mm) at
shallower depths with a higher fines content stratum below
A short series of tests was carried out on a mix of quartz (15–30%o63 mm); the gradings curve for the high fines
sand, crushed quartz silt and kaolin, in the proportions content stratum is given on Fig. 2. The compression
25%: 25%: 50% as measured by dry weight. The quartz behaviour has been reanalysed from the original data and
sand used in this case was the original Ham River sand, the oedometer compression curves for the high fines content
gradings curve of which is also in Fig. 2. The samples were sub-stratum, which may be described as a clayey-silty-sand,
compressed from slurries and the estimated accuracy are shown in Fig. 9. The reconstituted samples were made
of these tests was again 70.03. The compression data in by dry compaction and they have an estimated accuracy of
 B. Shipton, M.R. Coop / Soils and Foundations 52 (2012) 668–681 679

the specific volumes of 70.02. They therefore indicate an

apparently non-convergent compression behaviour with no
clear knee and unique NCL being identified even for stresses
up to 30 MPa. It is however possible, that if still higher
stresses were reached, some convergence might be seen. In
contrast the soil from the low fines stratum gave convergent
compression curves, by similar stresses.

5.6. Clays

Initially it was assumed that non-convergent compression

and transitional behaviour would be limited to intermediate
graded or gap-graded soils and that unique intrinsic proper-
ties would be seen for slurried samples of clays. Laboratory
tests were therefore carried out on three reconstituted
clays with different plasticities, London clay (from Unit B2)
(LL¼ 69% PL¼ 27%), Oxford clay (LL¼ 70% PL¼ 25%)
and Speswhite kaolin (LL¼ 62% PL¼ 32%). In each case
samples were made either as slurries at different water
contents or by dry compaction at different initial densities,
although no difference in behaviour could be found from the
preparation method, and so this has been omitted from the
graphs for clarity. The compression data for the kaolin are
shown in Fig. 10. The behaviour of the other two clays was
similar. The data are slightly scattered but they seem to
indicate some non-convergence. To check in more detail to
what extent the curves are converging, the relationship
between the initial specific volume, vi, and the value of v at
sv0 values of 0.1 and 1 MPa are plotted and are shown in
Fig. 11 for each clay. In this plot a horizontal line would
indicate full convergence and a line at 451 would indicate
completely parallel compression paths. It was found to be
much more difficult to achieve accurate specific volumes in
the very much more compressible clays, and the mean
accuracy of the specific volumes for these tests was only
around 70.07, although, as discussed above this is very
likely to be an overestimate. Even allowing for the poorer
accuracy though, it is clear that at 100 kPa the compression

Fig. 11. Variation of specific volumes at 100 and 1000 kPa with initial
specific volume for three clays. (a) Kaolin. (b) London clay. (c) Oxford clay.

paths for the Oxford clay and kaolin have not converged,
and even for the London clay the differences measured are at
the limit of what could be explained by the accuracy of the
measurements. At 1 MPa the differences are much smaller
and more accurate tests would be needed to confirm whether
Fig. 10. One-dimensional compression behaviour of reconstituted Kaolin. there is complete convergence or otherwise.
680 B. Shipton, M.R. Coop / Soils and Foundations 52 (2012) 668–681

Burland (1990) recommended that in order to define seems that non-convergent compression behaviour tends to
unique properties for the compression of a reconstituted occur in soils of mixed grading, whether gap or well-graded,
clay, slurried samples should be made at water contents and soils with mixed mineralogy or particle types. Even the
between 1 and 1.5 eLL (the void ratio at the Liquid Limit). identification of unique NCLs in reconstituted clays may not
His normalised void index is defined in terms of the void be as simple as previously thought. Particle breakage can
ratio at 100 kPa and 1 MPa for reconstituted samples created occur in soils with non-convergent compression behaviour but
in this range, assuming that the clays will have reached a it is not a prerequisite.
standard fabric and hence unique void ratio at 100 kPa. In Much early research on the application of the critical
each case, there is little evidence that the trends between 1 state framework to sands tended to focus on poorly graded
and 1.5 eLL are different to those outside the range. It can sands of single mineralogy, often for simplicity in creating
therefore be tentatively concluded that even clays may show repeatable samples or identifying the effects of particle
slight evidence of non-convergent compression behaviour breakage. Although the behaviour observed for clean
with the implication, as Ferreira and Bica (2006) have sands had some differences with that of clays, especially
identified, that if a unique intrinsic compression line cannot in the effects of the particle breakage, the applicability of a
be identified, it is not possible to quantify the effects of critical state type of framework to both has perhaps tended
structure in a natural soil objectively. to obscure the fact that there may be many soils of
intermediate grading and/or mixed mineralogy that have
transitional or at least non-convergent compression beha-
6. Conclusions
viour. Since it appears that there are no simple means to
predict whether transitional or non-convergent compres-
The factors that control whether non-convergent behaviour
sion behaviour will occur or not, each soil must be assessed
is seen in compression are clearly complex. Shipton (2010)
individually.
collated all of the gradings curves for soils showing non-
While for some of the soils investigated the pattern of
convergent compression and full transitional behaviour, as
convergent or non-convergent behaviour is clear even at
shown in Fig. 12, including data from the literature, the tests
modest stresses, in other cases very high stresses have been
data from this paper and other unpublished tests carried out
used in order to confirm whether convergence occurs or not.
at Imperial College. They cover a wide range of gradings from
This may seem to be an academic exercise, but the compres-
clayey-silts to well-graded sands and gap-graded soils, so
sion behaviour at lower stresses will be influenced by whether
obviously grading alone is no indicator of whether transi-
compression curves from different initial densities are tending
tional or non-convergent compression behaviour will be seen.
to converge to a unique line or not. Many constitutive models
Referring to the summary of the existing data from this paper
are also based on volumetric hardening which is linked to the
and the literature in Table 1, it is clear that relatively small
isotropic Normal Compression Line, to which that in oedo-
changes to the nature or quantity of the constituent particles
metric compression is also linked.
of a soil can change the behaviour quite dramatically. It also

Acknowledgements

The authors are indebted to the following for the help in

conducting the laboratory tests: Mr. T.Hui, Mr. T. Kwok,
Mr. J. Lau, Mr. E. Lee, Mr. T. Tam, Dr. K. Ventouras
and Ms. Z. Wilson.

References

Altuhafi, F.N., Baudet, B.A., Sammonds, P., 2010. The mechanics of

subglacial sediment: an example of new ‘‘Transitional’’ behaviour.
Canadian Geotechnical Journal 47 (7), 775–790.
Altuhafi, F., Coop, M.R., 2011. Changes to particle characteristics
associated with the compression of sands. Géotechnique 61 (6),
459–471.
Been, K., Jefferies, M.G., 1985. A state parameter for sands. Géotechnique
35 (2), 99–112.
Bishop, A.W., Webb, D.L. and Skinner, A.E. (1965) Triaxial tests on soil
at elevated cell pressures. Proc. 6th Int. Conf. Soil Mech., 170–174.
Burland, J.B., 1990. On the compressibility and shear strength of natural
clays. Thirtieth Rankine Lecture. Géotechnique 40 (3), 329–378.
Fig. 12. Grading curves of soils which show transitional and non- Butterfield, R., 1979. A natural compression law for soils (an advance on
convergent compression behaviour from Shipton (2010). e–log p0 ). Géotechnique 26 (4), 469–480.
 B. Shipton, M.R. Coop / Soils and Foundations 52 (2012) 668–681 681

Carrera, A., Coop, M.R., Lancellotta, R., 2011. The Influence of grading Miura, N. and Yamanouchi, T. (1977). Effect of particle crushing on the
on the mechanical behaviour of Stava tailings. Géotechnique 61 (11), shear characteristics of a sand. Proc. Japanese Society of Civil
935–946. Engineers, 260, 109–118 (in Japanese), Trans. Japanese Society of
Coop, M.R., 1990. The mechanics of uncemented carbonate sands. Civil Engineers, 9, (1978).
Géotechnique 40 (4), 607–626. Nocilla, A., Coop, M.R., 2008. The behaviour of sub-soils from the Po
Coop, M.R., Lee, I.K., 1993. The Behaviour of Granular Soils at Elevated River Embankments: an example of transitional behaviour in natural
Stresses. Predictive Soil Mechanics. Thomas Telford, London 186-198. soils. Italian Geotechnical Journal 42 (1), 49–58.
Ferreira, P.M.V., Bica, A.V.D., 2006. Problems on the identification of Nocilla, A., Coop, M.R., Colleselli, F., 2006. The mechanics of an Italian
structure in a soil with a transitional behaviour. Géotechnique 56 (7), silt; an example of ‘‘Transitional’’ behaviour. Géotechnique 56 (4),
445–454. 261–271.
Fourie, A.B., Papageorgiou, G., 2001. Defining an appropriate steady Pestana, J.M., Whittle, A.J., 1995. A compression model for cohesionless
state line for Merriespruit gold tailings. Canadian Geotechnical soils. Géotechnique 45 (4), 611–631.
Journal 38 (4), 695–706. Pitman, T.D., Robertson, P.K., Sego, D.C., 1994. Influence of fines on the
Hardin, B.O., 1985. Crushing of soil particles. Journal of Geotechnical collapse of loose sands. Canadian Geotechnical Journal 31, 728–739.
Engineering, ASCE 111 (10), 1177–1192. Shipton, B. (2010) The Mechanics of Transitional Soils. Ph.D. Thesis,
Kwag, J.M., Ochiai, H., Yasafuku, N., 1999. Yielding stress character- Imperial College, London.
istics of carbonate sand in relation to individual particle fragmentation Takahashi, A., Jardine, R.J., 2007. Assessment of standard research sand
strength. In: Al-Shafei, Bahrain (Ed.), Engineering for Calcareous for laboratory testing. Q. J. Eng. Geol. Hydrogeol. 40 (1), 93–103.
Sediments, 1. Balkema, Rotterdam, pp. 79–86. Thevanayagam, S., Shentham, T., Mohan, S., Liang, J., 2002. Undrained
Lade, P.V., Yamamuro, J.A., 1997. Effects of nonplastic fines on static fragility of clean sands, silty sands and sandy silts. ASCE. J. Geotech.
liquefaction of sands. Canadian Geotechnical Journal 34, 918–928. Geoenviron. Eng. 128 (10), 849–859.
Lupini, J.F., Skinner, A.E., Vaughan, P.R., 1981. The drained residual Ventouras, K. (2009) Engineering behaviour of Thanet Sand. Ph.D.
strength of cohesive soils. Géotechnique 31, 181–213. Thesis, Imperial College, London.
Martins, F., Bressani, L.A., Coop, M.R., Bica, V.D., 2001. Some aspects Ventouras, K., Coop, M.R., 2009. On the behaviour of Thanet sand: an
of the compressibility behaviour of a clayey sand. Canadian Geotech- example of an uncemented natural sand. Géotechnique 59 (9),
nical Journal 38 (6), 1177–1186. 727–738.
McDowell, G.R., 2005. A physical justification for log e–log s based on Yang, S.L., Lacasse, S., Sandven, R., 2006. Determination of the
fractal crushing and particle kinematics. Géotechnique 55 (9), transitional fines content of mixtures of sand and non-plastic fines.
697–698. Geotechnical Testing of Journal ASTM 29 (2), 102–107.
McDowell, G.R., Bolton, M.D., 1998. On the micro mechanics of Yin, J.H., 1999. Properties and behaviour of Hong Kong marine deposits
crushable aggregates. Géotechnique 48 (5), 667–679. with different clay contents. Canadian Geotechnical Journal 36 (6),
Mesri, G., Vardhanabhuti, B., 2009. Compression of granular materials. 1085–1096.
Canadian Geotechnical Journal 46, 369–392. Zdravkovic, L. (1996) The stress–strain–strength anisotropy of a granular
Miura, N., O-Hara, S., 1979. Particle crushing of a decomposed granite medium under general stress conditions. Ph.D. Thesis, Imperial
soil under shear stresses. Soils Foundations. 19 (3), 1–14. College London.