Computers and Geotechnics 98 (2018) 181–188

Contents lists available at ScienceDirect

Computers and Geotechnics

journal homepage: www.elsevier.com/locate/compgeo

Research Paper

Micro mechanics of the critical state line at high stresses T

⁎
John P. de Bono , Glenn R. McDowell
University of Nottingham, United Kingdom

A R T I C L E I N F O A B S T R A C T

Keywords: A critical state line is presented for a crushable numerical soil, which is parallel to the isotropic normal com-
Discrete element method pression line. A previous theory for the normal compression line, which correctly predicts the slope as a function
DEM of the size-effect on particle strength is extended to justify the slope of the critical state line. The micro me-
Particle crushing chanics behind critical states are examined, leading to a theory for a relationship between the volume of smallest
Critical state line
particles and mean effective stress. A unique relationship exists for crushed states, leading to a two-dimensional
State boundary surface
interpretation of the state boundary surface for soils looser than critical.

1. Introduction behind the CSL, in particular the co-ordination number and the im-
portance of the ‘smallest particles’, which must be rigorously defined.
Following Pestana and Whittle [1], McDowell [2] proposed analy-
tically that the high-stress normal compression line for a sand should be 2. DEM model
linear on log e–log σ axes. This was based on the kinematics of particle
crushing and the assumption of a fractal particle size distribution. This The simulation results presented here were performed using the
was investigated numerically using the discrete element method [3], software PFC [7]. For computational efficiency all particles are mod-
and it was found that the normal compression line (NCL) for a sand elled using spheres, and gravity is neglected. All simulations are per-
could be described by: formed using a cylindrical triaxial sample, enclosed vertically by rigid
platens and laterally by a flexible, faceted cylindrical membrane, shown
1 σ′
log e = log e y− log in Fig. 1. The membrane consists of 4320 facets and the vertices can
2b σy′ (1) move independently from one another. Essential simulation details are
where e is the current voids ratio, σ′ is the current stress, b describes the provided in Table 1. The initial sample (before isotropic compression)
hardening rate of the particles, and ey and σ′y are constants defining the consists of approximately 700 randomly packed particles of 2 mm
yield point. The parameter b defines the rate at which average particle diameter; although the sample is compressed and crushed (generating
strength increases with decreasing size: greater quantities of smaller particles) before any shear tests are per-
formed.
σav ∞d−b (2)
The work uses a simple crushing model and realistic particle
and as shown by Eq. (1), b determines the slope (=1/2b) of the NCL on strengths, which the authors have previously used to produce normal
logarithmic axes. This compression law in Eq. (1) was tested against compression lines [3,8,9], which exhibit the same slope as observed
experimental data for which normal compression lines and particle experimentally for the sand. This crushing model uses the average oc-
strength data (i.e. values of b) obtained from particle crushing tests tahedral shear stress as the particle fracture criterion:
were readily available in [4], and it was found to correctly predicted 1
the slope of the NCL for each case. q= [(σ1−σ2)2 + (σ2−σ3)2 + (σ1−σ3)2]1/2
3 (3)
It is a well-known phenomenon that the critical state line (CSL) for
soils is parallel to the normal compression line [5,6], at least in con- which is calculated from the average principal stresses in a particle
ventional critical state soil mechanics. In this work, DEM simulations (σ1,2,3), in turn calculated from contacts with neighbouring bodies.
are used to establish a critical state line, and after ascertaining whether The strength data used here (and previously) is for a silica sand.
it is parallel with the NCL, to clarify what determines the slope and From particle crushing tests [10], it was found that the size-effect on
position of the CSL. In additional, we examine the micro mechanics strength, i.e. the value of b in Eq. (2) was approximately 1 [10]. For a

⁎
Corresponding author.
E-mail address: john.debono@nottingham.ac.uk (J.P. de Bono).

https://doi.org/10.1016/j.compgeo.2018.02.016
Received 19 September 2017; Received in revised form 15 December 2017; Accepted 18 February 2018
0266-352X/ © 2018 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY license (http://creativecommons.org/licenses/BY/4.0/).
J.P. de Bono, G.R. McDowell Computers and Geotechnics 98 (2018) 181–188

further breakages, the simulation progresses to the following incre-

ment.
The shear tests are strain-controlled. The upper platen is gradually
accelerated (to 0.01 m/s) then decelerated, applying an axial strain
increment of 0.1%. During this process the membrane servo-control
acts to adjust the radial stress: the confining pressure is adjusted to
ensure either constant σ3, p′ or volume, depending on the stress path.
Once the strain increment has been applied, particles are allowed to
break, and the radial pressure is then reapplied if necessary. Once this
process is complete, the subsequent strain increment is applied.
To enable the extensive number of simulations required in an ac-
ceptable timeframe, the initial sample contained only 723 particles
(Fig. 1a). This is due to the crushing that occurs in nearly all shearing
simulations, resulting in a wide range of sizes and much greater
quantities of particles (Fig. 1c). A feature of the simulations is that there
is no comminution limit—i.e. no lower limit to particle size. This allows
Fig. 1. Initial sample (a), after isotropic compression (b) and after shearing to critical
the unadulterated evolution of a (fractal) particle size distribution. A
state (c).
consequence of this however is that the numerical timestep used in the
simulations (influenced by dmin) becomes very small. Thus, to enable
Table 1 the number of simulations required whilst not imposing artificial con-
DEM properties.
ditions, the initial number of particles is small, which helps minimise
General simulation properties the calculation time once extensive crushing has occurred.

Initial sample size: height × diameter (mm) 30 × 15 3. Simulation results

Initial no. of particles 723
Wall friction coefficient 0
Particle friction coefficient 0.5 3.1. Normal compression- and critical state lines
Contact model Hertz-Mindlin
Particle shear modulus (GPa) 28 Isotropic compression results and two unloading lines from 20 and
Particle Poisson’s ratio 0.25 30 MPa are presented in Fig. 2 on log e–log p′ axes, also showing the
Particle density (kg/m3) 2650
Damping coefficient 0.7
points from which the shear tests were performed. From most of these
Initial voids ratio 0.75 points, 3 types of triaxial test were performed: constant-σ3, constant-p′,
Initial particle size (mm) 2 constant-volume. However, it was not possible to perform constant-
Weibull modulus 3.3 volume tests at the lowest p′, due to dry liquefaction (i.e. without pore
Initial particle strength, q0,2 (MPa) 37.5
fluid); and it was not computationally feasible to perform all of the
constant-σ3 and constant-p′ tests at the highest stresses due to extensive
given size of particle, the strengths are distributed according to a crushing caused by very large magnitudes of q and p′.
Weibull distribution, defined by a characteristic value q0 and a modulus The critical states obtained from these tests are shown in Fig. 3.
m. The strength q0 is a value such that 37% of particles of that size are Triaxial simulations were run until reaching axial strains of between 25
stronger, and is numerically similar (and proportional) to the mean and 35%, typically 30%, at which point all of the tests exhibited no (or
value. The modulus m defines the variability. For the silica sand, the in a few cases, negligible) volume or stress change. The ultimate states
modulus was found to be 3.3 [10]. The size-effect on strength for the
simulations can therefore be written as:

q0 ∞d−1 (4)

where for example, q0 = 37.5 MPa for d = 2 mm.

The particle breakage model replaces broken particles with 2
smaller fragments whilst obeying conservation of mass. The new frag-
ments are placed within the boundary of the broken particle, aligned in
the direction of the minor principal stress and therefore overlap. The
fragments therefore move apart immediately following breakage
(which requires a number of timesteps to be completed to allow energy
dissipation, via friction and mechanical damping), details of which are
given in [3], along with comparisons of alternative breakage models.
This initial sample is first isotropically compressed to establish a
compression line, and shear tests are performed from various points
along the loading path. The sample is also unloaded from different
stresses, and further shear tests performed from along overconsolidated
states.
During normal compression, the simulation proceeds by applying
isotropic stress increments (100 kPa) using the wall servo-controls.
Once a stress is applied, all particles are checked and allowed to break if
their strength is exceeded. At this point a number of timesteps are
completed to allow the overlapping fragments to move apart. If this
breakage results in a drop in the applied isotropic pressure, then the
Fig. 2. Isotropic compression and unloading lines.
stress increment is reapplied. Once a stress is sustainable without any

182
J.P. de Bono, G.R. McDowell Computers and Geotechnics 98 (2018) 181–188

Fig. 3. Critical states from triaxial tests: q–p′ axes (a) and log e–log p′ axes (b).

from all tests were taken as critical states. Additional details of these theory, a fractal distribution is infinite, however in reality a fractal PSD
tests, including select stress-strain results and stress-dilatancy plots are must be limited [15], bounded by an upper and lower particle size. The
available in [11]. largest particle size in all cases here is 2 mm, whilst the smallest fractal
Fig. 3(a) includes the phase transformation points [12] which are size depends on the extent of crushing, whereby more crushing leads to
often found to coincide with the CSL [5]. (The phase transformation a larger range of fractal sizes. Hence in Fig. 4(c), the smallest size in
points are the points at which in undrained tests, the behaviour changes each simulation is still emerging (a gradual process), and does not yet
from contractile to dilatant, and the rate of change of p′ reverses.) The adhere to the ideal fractal shown by the dashed trendline.
critical states reveal a clear CSL with M ≈ 0.7.
Fig. 3(b) shows the critical states on log e–log p′ axes, and at high 3.3. Co-ordination number
stresses (in this case, approximately > 10 MPa) there is a unique CSL,
parallel to the NCL. At lower stresses (< 10 MPa), the CSL appears non- One interesting observation from the simulations was that both the
unique, with the overconsolidated tests reaching a lower CSL, in normal compression line and critical state line appear to possess unique
agreement with experimental data [13]. coordination numbers at high stresses. The average coordination
The parallel nature of the CSL and NCL at high stresses suggest that number is plotted as a function of p′ in Fig. 5 for both normal com-
the principles that govern the slope of the normal compression line also pression and critical states. During compression but before yield (ap-
apply to the critical state line. The b value for the silica sand, as given in proximately elastic behaviour), there is a correlation between co-
Eq. (4), is 1. Using this b in Eq. (1), predicts a slope of −0.5 (using p′ as ordination number and stress, the coordination number then drops with
the stress variable). Both lines demonstrate agreement with this in the onset of crushing. Beyond yield, the average coordination number
Fig. 3(b). during compression then remains approximately constant at around
5.6. A similar trend can be observed for critical states—large variation
3.2. Fractal particle size distributions at the lowest stresses, before crushing becomes prominent after which
the average coordination remains approximately 5.1 for states on the
Fig. 4(a) shows progressive PSDs obtained from an example triaxial high-stress CSL, independent of stress level (and therefore PSD). This
shear simulation, which shows how the grading changes during suggests that the average coordination number on any plastic loading
shearing and that the rate of change reduces noticeably at large strains. line in log e–log p′ space is related to the stress ratio η (=q/p). What this
Fig. 4(b) gives the ultimate PSDs obtained from the end of all triaxial also shows is that for any shear test, shearing from any point on the NCL
shear tests. This conventional plot indicates the effects of stress level, causes the average coordination number to reduce from 5.6 to 5.1.
with example values of critical state mean stress annotated. It is clear
that shearing under increasing stresses leads to more developed grading 4. Micro mechanics of critical states
curves. The same data is replotted in (c) in terms of number on loga-
rithmic axes. For a discrete PSD (i.e. without continuous sizes), a fractal The kinematics behind the compression law in Eq. (1) are detailed
distribution will appear linear according to: in McDowell and de Bono [3]. The key assumptions were quantitatively
Ni ∝ d i−D (5) examined in [9], and the theory shown to be valid for a range of real
sands [4]. It will now be investigated whether the law is applicable to
where D is the fractal dimension [9,14]. It is evident from Fig. 4(c) that the critical state line. One of the assumptions was that a fractal PSD
for critical states under increasing stress, the PSDs evolve towards a emerges during compression, which also appears the case for critical
fractal distribution with a fractal dimension of around 2.5 (indicated by states under increasing stresses. The remaining assumptions will now be
the dashed line), just as occurs during normal compression [14]. In summarised, whilst full details and the kinematics behind the law can

183
J.P. de Bono, G.R. McDowell Computers and Geotechnics 98 (2018) 181–188

Fig. 4. PSDs at critical states: conventional semi-log plot (a) and on two logarithmic axes (b).

be found in [2,3,9]. the smallest particles:

In deriving the compression law in Eq. (1), it was assumed that the
p ∝ ds−1 (7)
strength of the smallest (strongest) particles is proportional to the ap-
plied stress: For a discrete particle size distribution, using the smallest particle
q0,s ∝ p size ds in Eq. (5) gives an expression for the number of the smallest
(6)
particles:
which accounts for the plastic hardening of the soil, as fragments be- Ns ∝ ds−2.5 (8)
come stronger with decreasing size. If a fractal distribution evolves
through the fracturing of the smallest grains, and those smallest grains and the collective volume of the smallest particles can therefore be
are in self-similar configurations as the smallest particles become expressed:
stronger and stress increases, then the strength of the smallest particles
Vsm ∝ ds0.5 (9)
loaded in those configurations must be proportional to the applied
macroscopic stress level. Recalling the particle size-hardening law in The fundamental assumption invoked in the derivation of the
Eq. (4) therefore provides a link between applied stress and the size of compression law given in [2] is that the volume of the smallest particles

184
J.P. de Bono, G.R. McDowell Computers and Geotechnics 98 (2018) 181–188

particles, and therefore is clearly not the smallest.

Considering random packing, the minimum contacts that a particle
may have whilst mechanically loaded is 4, and this was used as the
criterion to define the ‘smallest’ particles when investigating the above
assumptions during normal compression [9]. However, it is of course
possible for a small particle to have more contacts—the theoretical
maximum number of contacts a spherical particle may have with equal-
or-larger particles is 12 (hexagonal close packing); higher contact
numbers are only possible with smaller neighbouring spheres. For
dense random packing however (such is the case here), this number will
be much lower.
In the present analysis, it is found that using 5 contacts to define the
smallest particles (and calculating their cumulative volume) leads to
the best agreement in Eq. (10). It is anticipated that this difference is
related to the boundary conditions—the previous analysis [9] used a
rigid oedometer, with no lateral strain, and a non-uniform distribution
of radial stress, while the current work uses a triaxial sample with a
flexible membrane and uniform distribution of radial pressure. (None-
theless, using a criterion of 5 contacts in the previous analysis still
demonstrates acceptable agreement.)
The term ‘smallest particles’ will now be used to refer to all particles
in a simulation that have 5-or-fewer contacts. The ‘volume of the smal-
lest particles’ refers to the collective volume of all such particles with 5-
or-fewer contacts. Fig. 6(a) shows the voids ratio as a function of the
Fig. 5. CN against stress during normal compression and at critical states.
volume of these smallest particles, denoted Vsm. The volume Vsm used
here is the volume of smallest particles normalised by the total volume
is directly proportional to the volume of voids, and therefore the voids of solids (constant throughout all tests), to give the percentage by volume
ratio: of these smallest particles (avoiding unsightly small values). A direct
e ∝ Vsm (10) proportionality exists between e and Vsm, in accordance with and re-
affirming the assumption in the compression law, and showing that the
The rationale for this is that as the smallest particles become smaller same relation exists at critical states. This therefore appears to explain
and fill the available voids, the available voids become smaller and the why the CSL is parallel with the NCL, and demonstrates that the same
typical void space must be more closely related to the volume of one of principles apply to CSL—whereby the slope is a function of the size-
those particles, as opposed to, say, the largest particle volume or the effect on particle strengths. The voids ratio at any critical state is de-
median one. termined by the extent of the fractal PSD (i.e. the range of particle
Substituting (7) into (9), together with the above assumption (10), sizes), which itself results from crushing caused by the applied stresses.
is therefore what gives the slope of the compression line, which can be Fig. 6 shows e ∝ Vsm applies to all states that reach either the plastic
written in the form: normal compression or critical state lines, and this includes some
e ∝ p−0.5 samples sheared from ‘virgin’ states, all normally consolidated states,
(11)
and (lightly-overconsolidated) samples that are unloaded to lower
Considering that the CSL is parallel with the NCL, with a constant stresses and then sheared. So samples with completely different initial
critical stress ratio, it seems logical that the same relationship between PSDs reach the same e ∝ Vsm line through shearing and the resulting
the volume of smallest particles and voids ratio (10) also exists on the particle crushing. For example, the compacted and over-consolidated
critical state line. This will now be assessed, i.e. whether the voids ratio samples sheared from, say 15 MPa, which have initial different voids
is proportional to the collective volume of all the smallest particles on ratios and PSDs (and the latter on an unloading/reloading line), reach
the critical state line. the same critical state, with almost identical e and Vsm.
The theoretical derivation by McDowell [2] and substantiated by An interesting observation in Fig. 6(a) is that the two sets of soil
McDowell and de Bono [3] would work for any constant stress ratio. states have different coefficients of proportionality. It might have been
Thus, there should exist in log e–log p′ space, a set of parallel lines, each anticipated that, if the assumption e ∝ Vsm is true, then any change in
corresponding to a unique stress ratio, and the rate of separation of voids ratio resulting from triaxial shearing would simply be due to this
these lines as stress ratio increases governs the shape of the yield sur- volume of smallest particles changing. It was shown earlier that the
face in stress space. average coordination number during normal compression and at cri-
However, as shown in Fig. 4 the ‘absolute’ smallest particle size is tical states both appear to be independent of stress level and PSD, im-
not necessarily the smallest fractal particle size. (Soil does not obey a plying that on the CSL (lower coordination number), particles are
perfectly discrete hierarchical splitting model, with all particles of one packed less efficiently together, which would explain the different
size breaking at one stress; it is a gradual and continuous process.) So coefficients of proportionalities observed in Fig. 6(a). It is now worth
what counts as the smallest particles in Eq. (10) is non-trivial for a exploring this phenomenon further.
gradually evolving and expanding PSD. In general, there are only ever a The equation e ∝ Vsm allows the constant of proportionality to be
few (< 10) particles of the smallest size, in some cases only 2 parti- different for each constant stress ratio. For each macroscopic stress
cles—which would clearly be inappropriate [9]. Any definition of the ratio, the contact distributions will be different, giving different average
‘smallest’ particles must contain information about the overall pore co-ordination numbers, and the plastic strain increment vectors will be
space and its interconnectivity. different at different constant stress ratios on the state boundary surface
It was found previously [9] that a suitable definition of the smallest when crushing occurs. Fig. 6(a) implies that for each constant stress
particles are those which have minimal contacts—which should be a ratio in between the critical state stress ratio and the isotropic condi-
characteristic of any ‘locally’ smallest particles. For a particle to have a tion, there exists a line for each stress ratio. If there exists a unique
very high coordination number, it must be surrounded by smaller e ∝ Vsm line for each stress ratio, as well as a unique compression line

185
J.P. de Bono, G.R. McDowell Computers and Geotechnics 98 (2018) 181–188

Fig. 6. Voids ratio as a function of the volume of smallest particles Vsm (a), and Vsm as a function of mean stress.

(e ∝ p–0.5) for each stress ratio in log e–log p′ space, then the only way of behaviour is generalised in Fig. 8(b). The compacted sample, which
satisfying both conditions would be if a unique relationship existed initially does not lie on the directly-proportional e ∝ Vsm line, but in-
between Vsm and p′ for all plastic states on the loose side of critical. This stead on the virgin loading line, increases in volume and moves up-
concept is tested in Fig. 6(b) and is shown to be absolutely the case. wards to reach the ‘compacted’ critical state line. The NC sample, at
What this, in effect means is that Fig. 6(b) represents all plastic states on high stresses on the NCL, decreases in voids ratio, moving downwards
the loose side of critical and is therefore a two-dimensional re- until reaching the critical state line.
presentation and micro mechanical interpretation of the state boundary During the 8 MPa test, in which there is negligible crushing, e in-
surface. creases as the sample dilates. Fig. 8(a) shows that Vsm increases slightly,
Fig. 6(a) only shows states on the linear, high-stress NCL and CSL. which might be expected from volumetric dilation—as the sample ex-
Fig. 7(a) provides a fuller picture by including additional states: com- pands, particles become less-closely packed, and in general will there-
pacted states before yielding, OC states on the two unloading-lines, and fore have fewer contacts. This means that more particles will be con-
states on the low-stress critical state lines. Fig. 7(b) provides a sche- sidered as the smallest, so Vsm increases. This is reflected in Fig. 5(b)
matic to illustrate the behaviour. Upon normal compression the state which showed that the coordination is lower on the critical states line.
moves along a virgin-loading line, where the voids ratio decreases only During the 20 MPa constant-p’ test, there is significant crushing and
slightly with the volume of the smallest particles, with no proportion- contraction, and as shown in Fig. 8(a) Vsm does not decrease sig-
ality observed between these two quantities. Upon reaching the yield nificantly. Ordinarily, particle crushing accompanied by a decrease in
stress there then appears a direct link between Vsm and e, and both voids ratio would lead to Vsm decreasing—just as occurs during normal
decrease in proportion with each other as the macroscopic stress in- compression (Fig. 7). The fact that Vsm remains constant despite
creases. Upon unloading, the state moves back along an unloading-line crushing and contraction suggests that there is a counteracting effect
where the voids ratio changes only slightly (shown by a dashed line), which contributes to Vsm. Considering that the coordination number
similar to the initial virgin-loading line. reduces during shearing (together with observations from the 8 MPa
A similar relationship exists for critical states—there also is a direct dilatant simulation), it would appear that particles are packed less-ef-
proportionality between Vsm and e, however this lies to the right of the ficiently at the critical state. In other words, although smaller particles
normal compression states. The relationship e ∝ Vsm only applies to are produced by crushing, which enable the sample to occupy a smaller
states on the high-stress CSL (parallel to the NCL in log e–log p′ space). volume, the particles are relatively looser, which is supported by the
Critical states at low stresses do not conform to the relationship e ∝ Vsm. observation that the average coordination number reduces from 5.6 to
These states form a separate, flatter line, above the corresponding 5.1.
virgin/unloading-line. Fig. 7(b) shows a remarkable mirror image to the The observation that Vsm remains constant during the 20 MPa si-
normal compression and critical state lines in log e–log p′ space. mulation is consistent with the unique Vsm ∝ p–0.5 relationship shown
Two constant-p′ simulations will now be examined more close- earlier in Fig. 6(b). This unique ‘plastic state line’ is replotted sche-
ly—one dilatant (compacted to 8 MPa), and one contractile (normally matically in Fig. 9. According to this figure, for a normally-consolidated
consolidated to 20 MPa). The paths of these simulations are shown in contractile sample, there should be no change in Vsm as the state moves
Fig. 8(a), in terms of e and Vsm. Both simulations start on the (black) from the plastic NCL to the CSL while p’ stays constant. The constant-p′
normal compression line and end on the (red1) critical state line. This simulation at 8 MPa however is not initially on the plastic NCL, and so
the initial state is not located on the unique (Vsm ∝ p–0.5) plastic state
line. This is also illustrated in Fig. 9, where elastic loading states are
1
For interpretation of color in Fig. 8, the reader is referred to the web version of this located below the unique Vsm-p line. As shown, when such dilatant
article.

186
J.P. de Bono, G.R. McDowell Computers and Geotechnics 98 (2018) 181–188

Fig. 7. Voids ratio as a function of smallest particles (a) and idealised schematic (b).

Fig. 8. Triaxial paths between states of isotropic compression and critical states (a) and schematic (b).

samples are sheared at constant-p′, the state moves upwards as Vsm analogous to emin decreasing, and means that in general particles will
increases, towards the unique plastic state line. have fewer contacts and therefore more will be classed as the smallest,
This diagram can also explain the observed behaviour of constant- causing Vsm to increase. This again is reflected in a reduction of the
volume tests. For contractile constant-volume tests, such as that at average coordination number, which reduces from 5.6 to around 5.1.
p′ = 25 MPa, the mean stress reduces, so according to this figure the Equally, for dilatant constant-volume tests, in which negligible
state moves leftwards up the plastic state line, and so Vsm increases. breakage occurs, the mean stress increases to suppress dilation. As
This behaviour can be visualised in Fig. 8 as a path at constant e directly shown in Fig. 9, this is represented by a path that moves rightwards and
right from the NCL to the CSL. Just how Vsm increases is due to particle down the Vsm ∝ p–0.5 line, and a reduction in Vsm. An example of this
breakage occurring at a constant volume—shearing induces stresses behaviour is the OC test at 5 MPa: p′ increases to over 13 MPa; and with
great enough to cause significant particle crushing, which causes p′ to only limited crushing, this means that the particles are more tightly
reduce to maintain constant volume. However, the fact that volume packed together and have more contacts. This means fewer particles
change is prevented indicates that the sample becomes relatively looser, have ≤5 contacts and are considered as the smallest, and therefore Vsm

187
J.P. de Bono, G.R. McDowell Computers and Geotechnics 98 (2018) 181–188

PSD. Using a minimal value of coordination number seems a logical

way of defining the (locally) smallest particles in the sample, and en-
ables an explanation from a micro mechanic point of view of the slopes
of the NCL and CSL as well as the observed behaviour of compacted,
normally consolidated and overconsolidated sands. This is a scientific
approach to defining the smallest particles, which includes information
about the surrounding pore space; more primitive definitions might use
the physically smallest particle(s), which may have no load-bearing
ability or there may only be 1 such particle, as soil does not crush
hierarchically at discrete stresses.
The proportionality between current voids ratio and the volume of
the smallest particles was valid for all states on the plastic, high-stress
normal compression or critical state lines, but the coefficient of pro-
portionality is different for each stress ratio. It was shown however that
a single, unique relation exists between the volume of the smallest
particles and the current applied stress, for any stress ratio. When
plotted graphically, this single relation represents all states on any
plastic loading line (e.g. CSL or NCL), and therefore could be used to
explain the behaviour of any sample subjected to triaxial shearing. This
newly established unique relationship between Vsm and p′ is effectively
a two-dimensional interpretation of the state boundary surface and
future work will examine the evolution of Vsm along various stress paths
at constant stress ratio.
Fig. 9. Schematic showing unique relationship between Vsm and p′.

Acknowledgements
decreases.
Fig. 9 is also consistent with the observed behaviour for constant-σ3 This work was supported by the Engineering and Physical Sciences
triaxial tests (indicated in the figure), as well as normal com- Research Council [grant number EP/L019779/1].
pression—in which the state simply moves down the plastic state line.
References
5. Conclusions
[1] Pestana JM, Whittle AJ. Compression model for cohesionless soils. Géotechnique
A cylindrical sample of spheres was isotropically compressed to 1995;45:611–31. http://dx.doi.org/10.1680/geot.1995.45.4.611.
[2] McDowell GR. A physical justification for log e–log σ based on fractal crushing and
establish a normal compression line, and then a large set of shear tests particle kinematics. Géotechnique 2005;55:697–8. http://dx.doi.org/10.1680/geot.
were performed from various states to establish the critical state line, 2005.55.9.697.
which was parallel with the NCL at high stresses, in accordance with [3] McDowell GR, de Bono JP. On the micro mechanics of one-dimensional normal
compression. Géotechnique 2013;63:895–908. http://dx.doi.org/10.1680/geot.12.
general experimental behaviour. From analysis of the particle size P.041.
distributions at critical state, it was shown that the crushing resulting [4] de Bono J, McDowell GR. Validation of the log e– log σ normal compression law
from shearing contributed to the emergence of fractal PSDs with a using particle strength data. Géotechnique 2017:1–6. http://dx.doi.org/10.1680/
jgeot.17.T.007.
fractal dimension of around 2.5, similar to what occurs during normal
[5] Coop MR. The mechanics of uncemented carbonate sands. Géotechnique
compression. This observsation, coupled with the parallel nature of the 1990;40:607–26. http://dx.doi.org/10.1680/geot.1990.40.4.607.
CSL and NCL, suggested that the same principles behind the compres- [6] Schofield A, Wroth P. Critical state soil mechanics. London: McGraw-Hill; 1968.
sion law could be used to reveal the micromechanics behind critical [7] Itasca. PFC3D; 2015.
[8] de Bono JP, McDowell GR. An insight into the yielding and normal compression of
states. sand with irregularly-shaped particles using DEM. Powder Technol
An assumption of the compression law is that on the NCL, the 2015;271:270–7. http://dx.doi.org/10.1016/j.powtec.2014.11.013.
available void space is proportional to the cumulative volume of the [9] de Bono JP, McDowell GR. The fractal micro mechanics of normal compression.
Comput Geotech 2016;78:11–24. http://dx.doi.org/10.1016/j.compgeo.2016.04.
smallest particles. What constitutes the ‘smallest’ particles has pre- 018.
viously been examined, and the best measure was found to be using the [10] McDowell GR. On the yielding and plastic compression of sand. Soils Found
coordination number to define the smallest particles. In the simulations 2002;42:139–45.
[11] de Bono JP, McDowell GR. Micro mechanics of drained and undrained shearing of
presented here if the smallest particles are defined as those with 5 or compacted and overconsolidated crushable sand. Géotechnique 2017. p. 1–15.
fewer contacts, then the total volume of such particles appears to be http://dx.doi.org/10.1680/jgeot.16.P.318 [Published].
directly proportional to the voids ratio for both isotropic compression [12] Ishihara K. Liquefaction and flow failure during earthquakes. Géotechnique
1993;43:351–451. http://dx.doi.org/10.1680/geot.1993.43.3.351.
and critical states. Thus, the parallel critical state line is a result of the [13] Bandini V, Coop MR. The influence of particle breakage on the location of the
fractal crushing of particles, where the slope is a function of the size- critical state line of sands. Soils Found 2011;51:591–600.
effect on particle strength. The voids ratio at a critical state is de- [14] McDowell GR, Daniell CM. Fractal compression of soil. Géotechnique
2001;51:173–6. http://dx.doi.org/10.1680/geot.2001.51.2.173.
termined by the extent of the fractal PSD, caused by crushing due to the
[15] McDowell GR, Bolton MD. On the micromechanics of crushable aggregates.
applied stress. Géotechnique 1998;48:667–79. http://dx.doi.org/10.1680/geot.1998.48.5.667.
The value of 5 is, however, unimportant. What is important is the
dynamic definition of the smallest particle size in a constantly evolving

188