Vous êtes sur la page 1sur 12

Li, X. S. & Dafalias, Y. F. (2000). GeÂotechnique 50, No.

4, 449±460

Dilatancy for cohesionless soils


X . S . L I  a n d Y. F. DA FA L I A S {

Dilatancy is often considered a unique function of the stress La dilatance est souvent consideÂreÂe comme une fonction
ratio ç ˆ q= p9, in terms of the triaxial stress variables q unique du rapport d'effort ç ˆ q= p9, en termes de variantes
and p9. With this assumption, the direction of plastic ¯ow is de contrainte triaxiales q et p9. Avec cette hypotheÁse, la
uniquely related to ç, irrespective of the material internal direction de l'eÂcoulement plastique est lieÂe uniquement aÁ h,
state. This obviously contradicts the facts. Consider two quel que soit l'eÂtat interne du mateÂriau. Ceci contredit les
specimens of the same sand, one is in a loose state and the faits de manieÁre eÂvidente. Prenons deux speÂcimens d'un
other in a dense state. Subjected to a loading from the same meÃme sable, l'un meuble et l'autre dense. Soumis aÁ un
ç, the loose specimen contracts and the dense one dilates. chargement du meÃme h, le speÂcimen meuble se contracte et
These two distinctly different responses are associated with a le speÂcimen dense se dilate. Ces deux reÂponses bien diffeÂr-
single ç but two different values of dilatancy, one positive entes sont associeÂes aÁ un seul h mais aÁ deux valeurs de
and the other negative. Treating the dilatancy as a unique dilatance diffeÂrentes, une positive et l'autre neÂgative. Le fait
function of ç has developed into a major obstacle to uni®ed de traiter la dilatance comme fonction unique de h est
modelling of the response of a cohesionless material over a devenu un obstacle majeur aÁ la creÂation de modeÁles uni®eÂs
full range of densities and stress levels (before particle de la reÂponse d'un mateÂriau non coheÂsif sur toute une
crushing). A theory is presented that treats the dilatancy as gamme de densiteÂs et de niveaux de contraintes (avant le
a state-dependent quantity within the framework of critical broyage des particules). Cet expose preÂsente une theÂorie qui
state soil mechanics. Micromechanical analysis is used to traite la dilatance comme une quantite deÂpendante de l'eÂtat
justify and motivate a simple macroscopic constitutive fra- dans le cadre de travail de la meÂcanique de sol aÁ l'eÂtat
mework. A rudimentary model is presented, and its simula- critique. Nous utilisons une analyse micromeÂcanique pour
tive capability shown by comparison with experimental data justi®er et motiver un cadre de travail constitutif macrosco-
of the response of a sand under various initial state and pique simple. Nous preÂsentons ensuite un modeÁle rudimen-
loading conditions. taire et nous montrons sa faculte simulative par une
comparaison avec les donneÂes expeÂrimentales de la reÂponse
d'un sable dans divers eÂtats initiaux et sous diverses condi-
KEYWORDS: constitutive relations; plasticity; sands. tions de charge.

INTRODUCTION Sulem, 1995). Taylor (1948) proposed ç ‡ d ˆ constant, based


The concept of critical state (Roscoe et al., 1958) has been on the hypothesis that a constant `effective' friction coef®cient
successfully applied to modelling the behaviour of cohesive exists. Rowe (1962) showed, based on the theory of least rate of
soils. However, sand modelling has not always been formulated internal work, that d could be expressed as a function of the
within the critical state framework, and this is because the stress ratio and the true angle of friction between the mineral
behaviour of sand is somehow different from clay. Consider the surfaces of the particles. Although the particular forms proposed
stress±strain response of a sand along a path of constant stress by Taylor & Rowe were different, both of them suggested that
ratio ç ˆ q= p9, where p9 ˆ (ó 19 ‡ 2ó 39 )=3 is the effective con- the dilatancy d was a unique function of the stress ratio ç:
®ning pressure and q ˆ ó 1 ÿ ó 3 is the deviatoric stress in a
triaxial setting. First, unlike clay, sand does not possess a d ˆ d(ç, C) (1)
unique relationship between the void ratio e and p9 for a
particular ç. In fact, the density of a typical sand in the pressure where C is a set of intrinsic material constants.
range before particle crushing cannot be altered considerably by Equation (1) worked quite satisfactorily for cohesive soils.
a constant ç compression, either isotropic (ç ˆ 0) or anisotropic For example, d in two versions of Camclay models (Roscoe &
(ç 6ˆ 0). Secondly, when the ç of a sand reaches its limiting Scho®eld, 1963; Roscoe & Burland, 1968) were given by
value M (the critical stress ratio) during plastic loading, it does d ˆ M ÿ ç and d ˆ (M 2 ÿ ç2 )=2ç, respectively, where M is
not necessarily follow that the sand is at a critical state. The the critical stress ratio, an intrinsic material constant. In agree-
stress path can actually move along the ç ˆ M line, as for ment with the concept of critical state, these models make sure
example in an undrained dilative shear path up to ultimate that the soil yielding at ç ˆ M is coincident with d ˆ 0; that
failure. These differences suggest that the well-established is, the material being modelled reaches its ultimate failure
framework for clay modelling should not be directly trans- whenever a plastic deformation takes place at ç ˆ M.
planted to sand without a careful examination. In contrast, it was soon found, based on experimental
One of the fundamental issues in modelling the stress±strain evidence, that the applicability of equation (1) to granular soils
behaviour of a soil is to correctly describe its dilatancy d, the depends on the density. Observing the divergence between the
ratio of plastic volumetric strain increment to plastic deviatoric proposed theory and the test results, Rowe (1962) pointed out
strain increment in the triaxial space: d ˆ dåpv =jdåpq j, where that a variable depending on the sample density and the stress
dåv ˆ då1 ‡ 2då3 , dåq ˆ 2(då1 ÿ då3 )=3, and the superscript ` p' history should be added to the stress±dilatancy relationship that
stands for `plastic' (Roscoe & Burland, 1968; Nova & Wood, he had derived earlier. Rowe attributed the divergence to a
1979; Wood, 1990; Wood et al., 1994; Vardoulakis & Sulem, rearrangement of particle packing, a fact that was ignored when
1995). The second law of thermodynamics shows that ç and d his stress±dilatancy relationship was derived.
are interrelated at a very fundamental level (Vardoulakis & Rowe's work was followed in many later investigations (e.g.
Nova & Wood, 1979; Pastor et al., 1990; Wood, 1990; Jefferies,
Manuscript received 18 October 1999; revised manuscript accepted 28 1993; Wood et al., 1994) on sand modelling, in which, however,
January 2000 the dependence of d on the material internal state was consid-
Discussion on this paper closes 26 November 2000. ered insigni®cant and thus dropped. This simpli®cation leads to
 The Hong Kong University of Science and Technology. the common practice that treats a sand with different initial
{ University of California at Davis. densities as different materials and results in multiple sets of
449

Downloaded by [ National Taiwan University] on [19/11/23]. Copyright © ICE Publishing, all rights reserved.
450 LI AND DAFALIAS
parameters for a single sand and thus does not have a good particles; and â is the deviation of the tangent at the contact
control over changes in the material state during loading. points from the major principal direction. The quantity
Attempts have been made in recent years to tackle this issue D ˆ 1 ‡ d V_ =V å_ 1 is clearly a form of dilatancy measure.
from the perspective of dilatancy. With the concept of critical As shown in Appendix 1, resulting from the equilibrium
state as basis, Been & Jefferies (1985) introduced a scalar condition at rod contacts, the stress ratio R ˆ ó 19 =ó 29 is uniquely
quantity ø called the state parameter, which measures the related to â as follows:
difference between the current and critical void ratios at the ó 19
same p9. Kabilamany & Ishihara (1990) provided experimental Rˆ ˆ tan(öì ‡ â) tan â (3)
evidence showing that d ‡ ç increases as shear deformation ó 29
increases. Manzari & Dafalias (1997) presented a sand model in and resulting from kinematical compatibility condition, the void
which a linear dependence of the phase transformation, or ratio
dilatancy stress ratio (the stress ratio at which the response
8 sin â cos â
changes from contractive to dilative) on ø was introduced. Li eˆ ÿ1 (4)
(1997) investigated the response of sand at the ultimate stress ð
ratio and explicitly pointed out that the dilatancy d is not By combining equations (2)±(4) with â as an implicit variable,
related only to the stress ratio but is also a function of plastic a unique relation between R and D is established, with e or â
volumetric strain. More recently, Wan & Guo (1998) proposed a as an implicit dependent variable. That is, R and D are uniquely
model with its dilatancy modi®ed from Rowe's stress±dilatancy related, but the value of R (or D) depends on the volume of the
equation. The modi®ed dilatancy equation includes the density packing. This dependence is due to the equilibrium and kine-
dependence with the critical void ratio as a reference. matic constraints imposed by the given packing.
Cubrinovski & Ishihara (1998) also showed a dilatancy relation- Now consider a different packing B, as shown in Fig. 2. For
ship that depends on the material state represented by cumula- this packing equation (2) is still valid but equations (3) and (4)
tive plastic shear strain. Li et al. (1999) introduced a state- are not. As shown in Appendix 1, for this packing
dependent dilatancy into an existing hypoplasticity sand model
ó 19 2 sin â
(Wang et al., 1990), resulting in a successful simulation of the ˆ tan(öì ‡ â) (5)
responses of Toyoura sand to both drained and undrained ó 29 1 ‡ 2 cos â
triaxial loading over a wide range of densities and pressures.
In the present paper a number of issues on this subject are and
discussed, starting from some microscopic analytical considera-
8 sin â(1 ‡ 2 cos â)
tions and ending with the presentation of a simple macroscopic eˆ ÿ1 (6)
constitutive framework and modelling, the simulative capability 3ð
of which is shown by comparison with experimental data of the It shows again that there exists a unique relationship between R
response of a sand under various initial state and loading and D, with e or â as an implicit variable. However, the
conditions. relationship for this packing is different from that for the
packing A.
Rowe applied equation (2) to random mass of irregular
MICROSCOPIC OBSERVATION ON DILATANCY particles based on the hypothesis that the rate of internal work
Consider a pack of uniform rigid rods, packing A, as shown done is a minimum. This hypothesis yielded ⠈ 458 ÿ öì =2.
in Fig. 1. Rowe showed that Thus,
ó 19 tan(öì ‡ â) ó 19
ˆ (2) ˆ tan2 (45 ‡ öì =2) (7)
ó 29 (1 ‡ d V_ =V å_ 1 ) tan â ó 29 (1 ‡ d V_ =V å_ 1 )

where ó 19 and ó 29 are the major and minor principal stresses, Equation (7) is the well-known stress±dilatancy equation. It can
respectively; å1 is the strain in ó 19 direction (taking compression be seen that the minimization procedure makes the stress ratio
as positive for both stress and strain); V is the volume of the R uniquely related to D and independent of the packing of the
pack; öì is the angle of friction between the surfaces of the particles and the volume of the mass, in contradiction to the

L1 L1 L1 L1
L1 L1 L1 L1
β = 60˚ β = 45˚

L1/2
L2/2
L2 L2 L2 L2

l1 l1
r
r
L2 L2
L2 L2 l2
l2

L1 L1 L1 L1
L1 L1 L1 L1

Fig. 1. Regular packing A of a uniform rod

Downloaded by [ National Taiwan University] on [19/11/23]. Copyright © ICE Publishing, all rights reserved.
DILATANCY FOR COHESIONLESS SOILS 451
L1 L1 L1 L1
L1 L1 L1 L1
β = 60˚ β = 45˚

L1/2
L2/2
L2 L2 L2 L2

l1 l1
r
r
L2 L2 L2
L2
l2
l2

L1 L1 L1 L1
L1 L1 L1 L1

Fig. 2. Regular packing B of a uniform rod

exact analytical conclusion reached by the sets of equations material state changes, the theory also shows notable deviations
(2)±(4) and (2), (5), and (6), for the two examples of different from experimental observations. It is commonly observed that
packing arrangements. at low deviatoric strains both dense and loose samples show
Figure 3 shows the relationships between (ó 19 ÿ ó 29 )= contractive behaviour and the dilatancy at that stage is not so
(ó 19 ‡ ó 29 ) and ÿd V_ =V å_ 1 (a form of plots similar to ç versus d related to the material state. However, as shearing increases, the
plot in the triaxial setting) for the packings A and B, and the deviation between the dilatancy of loose and dense specimens
stress±dilatancy equation (equation (7)), respectively. Fig. 3 becomes increasingly pronounced. This phenomenon can also
indicates that the relationships between the stress ratio and the be seen in Fig. 3. As any microscopic constraints resulting in a
dilatancy depend on microscopic constraints. At a given stress deviation from the hypothetical sliding direction ⠈ 458 ÿ
ratio, a particular packing is associated with a particular void öì =2 will increase the rate of internal work, these constraints
ratio, re¯ecting the internal microscopic constraints. Therefore, tend to increase d. In Fig. 3, at the same (ó 19 ÿ ó 29 )=(ó 19 ‡ ó 29 ),
the dilatancy depends not only on the stress ratio, but also on the stress±dilatancy equation (equation (7)) yields the lowest
the void ratio. value of ÿd V_ =V å_ 1 .
This density dependence is not re¯ected in equation (7), this
being the result of an unconstrained minimization of the rate of
internal work. In the derivation, the rate of work was obtained
from microscopic observations on regular packings of particles PROBLEMS WITH UNIQUE RELATIONSHIP BETWEEN d AND ç
sliding in a given direction, but the rate minimization was done Although the dependence of dilatancy on the internal state of
by zeroing its derivative with respect to the sliding direction. the material was noted decades ago, it has been largely ignored
This approach implicitly treated a particulate system as a in subsequent developments. This fact is, however, not trivial,
continuum without considering the static and kinematical con- as a major obstacle to uni®ed modelling of sand behaviour
straints at the particle contacts. This treatment captured the arises from this ignorance.
main feature of dilatancy and led to a unique relationship When subjected to shear, loose sand contracts and dense sand
between the dilatancy and the stress ratio. However, as it does dilates. According to critical state soil mechanics, a loose or
not take into account the microconstraints, which vary as the dense state is de®ned not only in terms of density but also of
the con®ning pressure. This is because such a de®nition is
relative to the critical state line in the e versus p9 space. For a
1·0 given e, for example, the sand will behave like dense for a
suf®ciently low and like loose for a suf®ciently high p9.
φµ = 20˚ e = 0·188 e = 0·462 Furthermore, for a sand that initially is either in the loose or
0·8
dense state, there is an ultimate state of failure at which the
volumetric strain rate is zero. This ultimate state is the well-
(σ′1 – σ′2)/(σ′1 + σ′2)

0·6 known critical state (Roscoe et al., 1958) characterized by a


unique combination of p9, q and critical void ratio ec in a
triaxial setting.
0·4
Packing A
Consider two specimens of the same sand. One is in a loose
e not specified
state and the other in a dense state, accounting for both density
Packing B e = 0·492 and pressure. Subjected to a shear loading increment from the
0·2 e = 0·261 same ç, the loose specimen contracts and the dense specimen
Equation (7) dilates, as shown in Fig 4 (data from Verdugo & Ishihara
0 (1996)) in terms of undrained stress path in q± p9 space. These
–5 –4 –3 –2 –1 0 1 two distinctly different responses are associated with a single ç
–dV/Vε1 but two different values of dilatancy, one positive and the other
negative. However, if d were a unique function of ç, the
Fig. 3. Deviation from the stress±dilatancy equation due to con- direction of plastic ¯ow, and hence of the undrained stress path,
straints at contacts would be uniquely related to ç, irrespective of the material

Downloaded by [ National Taiwan University] on [19/11/23]. Copyright © ICE Publishing, all rights reserved.
452 LI AND DAFALIAS
1600 2000
e = 0·833
D r = 37·9%
1400
Dense state, d < 0

1200 1500
Deviator stress, q: kPa

Deviator stress, q: kPa


1000 η = constant
η = constant

800 1000

e = 0·735 Loose state, d > 0


600 D r = 63·7%

400 e = 0·907 500


D r = 18·5%

Dense state, d < 0


200
Loose state, d > 0

0 0
0 200 400 600 800 1000 1200 0 500 1000 1500 2000
Effective mean normal stress, p ′: kPa Effective mean normal stress, p ′: kPa
(a) (b)

Fig. 4. Variation in dilatancy with material state (data from Verdugo & Ishihara (1996)). Undrained response of a sand with (a) different
densities and (b) the same density but under different con®ning pressures

state. This assumption obviously contradicts the fact as de- axis. Eventually, when the sand becomes too loose, the phase
scribed. transformation phenomenon totally disappears. The assumption
Now consider a sand in a dense state subjected to an that d is uniquely related to ç again contradicts the observation.
undrained shear. As shear proceeds, ç passes a so-called `phase Furthermore, undrained tests on dense sand often show that
transformation state' at which ç ˆ M d and d ˆ 0 (Ishihara et the q± p9 stress path eventually converges with a line of more
al., 1975) and then approaches the critical state at which or less constant ç ˆ M towards an ultimate state (Figs 5 and
ç ˆ M and d ˆ 0. If equation (1) held true, M d would be equal 6). At the ultimate state, both the stresses and the plastic
to M, because d is equal to zero at both the phase transforma- volumetric strain stop changing, as shown in Fig. 6 (see also
tion state and the critical state. As the critical stress ratio M is Figs 10±12). Since the stresses do not change, neither does the
considered an intrinsic material property, independent of the elastic volumetric strain. By de®nition, this is a critical state
initial material state, the logical outcome M d ˆ M from equa- where d p9 ˆ dq ˆ dåv ˆ 0 while dåq 6ˆ 0. Along the approxi-
tion (1) would render the phase transformation intrinsic too, mately ç ˆ M path and before the critical state is reached, the
resulting in a unique phase transformation line for a particular fact that åpv tends towards a constant value implies that the
sand at which the response of the sand would change from dilatancy d tends towards a zero value. If d were a unique
contractive to dilative, irrespective of its density and stress function of ç, however, along this path d would be essentially a
level. However, tests show that the phase transformation phe- constant, which means that, as shear proceeds, p9 would in-
nomenon can be seen only when the material is in a dense crease continuously as a result of the undrained constraint of
state, and M d is in general a variable quantity not equal to M. zero total volumetric change, and the critical state would never
As sand becomes `looser', M d becomes higher. This is clearly be reached. One may argue that along this path the stress ratio
corroborated by test data such as those shown in Fig. 5 ç only approaches M asymptotically, and correspondingly the
(Verdugo & Ishihara, 1996), where M d is identi®ed by the dark evolution of d from a non-zero value towards zero would be a
circles on the q± p9 path where the tangent is parallel to the q result of the tiny deviations in ç from M. However, this
argument only facilitates a mathematical description that barely
makes equation (1) not violate the concept of a critical state.
4000 Considering the uncertainties involved in soil testing, one would
Toyoura sand, e = 0·735, Dr = 63·7% not be able to physically identify and quantify such tiny
deviations, if any, in a meaningful manner. On the other hand,
with the hypothesis that d and ç are not uniquely related, such
Deviator stress, q: kPa

3000
Varied phase an argument becomes unnecessary.
transformation The above observations lead to the conclusion that a sand
2000
stress ratios model with its dilatancy following equation (1) works well only
when the change in the material internal state is minor.

1000
GENERAL EXPRESSION FOR STATE-DEPENDENT DILATANCY
Based on the aforementioned observations and accounting for
0
the critical state constitutive framework, one may propose a
0 500 1000 1500 2000 2500 3000 general expression for the dilatancy:
Effective mean normal stress, p ′: kPa
d ˆ d(ç, e, Q, C) (8)
Fig. 5. Variation in the phase transformation stress ratio with where Q and C, as collective terms, denote internal state
material state (data from Verdugo & Ishihara (1996)) variables other than the void ratio e (e.g. the evolving tensor of

Downloaded by [ National Taiwan University] on [19/11/23]. Copyright © ICE Publishing, all rights reserved.
DILATANCY FOR COHESIONLESS SOILS 453
(d = 0)
dεp

Ultimate/critical state
(d = 0)
q
p
q, εq

dεp

M
1

Ultimate/critical
(d ≠ 0)
states
dεp

η = Md

Phase transformation

p
p, εv εq
(a) (b)

Fig. 6. Illustration of the dilative shear on the failure surface: (a) stress path; (b) stress±strain response

anisotropy (Dafalias, 1986)) and intrinsic material constants, PARTICULAR EXPRESSION OF STATE-DEPENDENT DILATANCY
respectively. Equation (8) expresses the dependence of d on the To obtain d within the framework of equation (8) subjected
state variables, which consist of the external variable ç and the to the requirement of equation (9) for the critical state response,
internal variables e and Q. Hence, equation (8) de®nes a state- one needs to quantify the dependence on e and the variables Q.
dependent dilatancy. Strictly speaking, a dilatancy expressed by As the state of a material depends not only on its density (void
equation (1) is also related to material state via ç. However, as ratio e) but also on p9, the aforementioned quanti®cation should
discussed above, such a relationship is not unique because it is be able to describe adequately the physical conditions of a
not complete. The term `state-dependent dilatancy' introduced material, including both its density and its con®ning pressure.
here signi®es and emphasizes the need to de®ne state depen- Attempts have been made to describe the state, on which d
dence on both ç and e and Q. With this additional dependence, depends, with a single scalar quantity, which of course implies
d is now uniquely related to an existing state, a combination of the assumption of isotropy, since otherwise the use of tensor-
the external stress state expressed via ç, and the internal valued quantities is necessary. Been & Jefferies (1985) de®ned
material state expressed via e and Q. A subtle point here is a state parameter ø ˆ e ÿ ec, where e is the current void ratio
that, although it may appear that no explicit dependence of d and ec is the critical void ratio on the critical state line in the
on p9 is introduced in equation (8), the dependence on e and Q e± p9 plane corresponding to the current p9, as shown in Fig. 7.
may in fact introduce indirectly such p9 dependence. Here ø is a measure of how far the material state is from the
There are certain requirements to be satis®ed in formulating critical state in terms of density. Bolton (1986) proposed a
d within the framework of equation (8). First, the dilatancy scalar parameter I R , called the `relative dilatancy index', that
must be zero at a critical state; that is, d ˆ 0 when ç ˆ M and also combined the in¯uence of density and con®ning pressure.
e ˆ ec (the void ratio at the critical state) simultaneously: Ishihara (1993) introduced a scalar quantity I S, called the `state
index', that takes some characteristic states other than critical
d(ç ˆ M, e ˆ ec , Q, C) ˆ 0 (9) state in the e± p9 plane as references.

In other words, the condition ç ˆ M alone does not guarantee


d ˆ 0. It is important to emphasize in relation to equation (9)
Current state 2
that when ç and e attain their critical values M and ec it is not
necessary for Q to reach a corresponding critical value. It is
entirely possible to reach a critical state with different values of
Q.
ψ<0 ψ>0
Void ratio, e

Secondly, it is possible for sand to have a so-called `phase


(dilative) (contractive)
transformation state' at which d ˆ 0 but ç 6ˆ M and e 6ˆ ec, as
discussed earlier. Analytically this means that the equation
Critical state line
d(ç 6ˆ M, e 6ˆ ec , Q, C) ˆ 0 (10)
Current state 1
can be used to specify the combination of ç, e, and Q that
de®nes a phase transformation state. Conversely, one may use
an a priori experimental knowledge of phase transformation (p /pa)ξ
states to specify via equation (10), together with equation (9),
appropriate forms of the dilatancy function in equation (8). Fig. 7. Critical state line and state parameter ø

Downloaded by [ National Taiwan University] on [19/11/23]. Copyright © ICE Publishing, all rights reserved.
454 LI AND DAFALIAS
In the present study, ø was chosen to be the state variable Applying now equation (10) to the particular form of d,
that, in conjunction with ç, affects d. For improving the ®tting (equation (15)), one obtains the phase transformation stress ratio
with experimental data of certain sands, ø is represented by ç ˆ M d ˆ Me mø . This yields the following interesting interpre-
tation to the last member of equation (15) for d. It postulates
ø ˆ e ÿ ec ˆ e ÿ [eà ÿ ëc ( p9= pa )î ] (11) that the dilatancy d depends on the difference of the current
stress ratio ç from a reference stress ratio Me mø , which is
where eà , ëc and î are the material constants determining the similar to Rowe's stress±dilatancy theory but with the reference
critical state line in the e± p9 plane (Li & Wang, 1998), and pa stress ratio varying with ø instead of being ®xed. Based on
is the atmospheric pressure for normalization. The dependence equation (10) this reference stress ratio represents the variable
of d on e occurs via its dependence on ø ˆ e ÿ ec ( p9) (observe with ø phase transformation line. This is exactly the concept
introduction of explicit p9 dependence via ec ), while any other described by Manzari & Dafalias (1997), who used the linear
dependence on Q is suppressed. Hence, an equation of the dependence M d ˆ M ‡ mø. The same concept was recently
following form is proposed: used by Li et al. (1999) to improve the performance of the
d ˆ d(ç, ø, C) (12) hypoplasticity model proposed by Wang et al. (1990). In other
words, one could have started with the hypothesis that
which, according to equation (9), must satisfy the condition M d ˆ Me mø , corroborated by data, such as those shown in Fig.
d(ç ˆ M, ø ˆ 0, C) ˆ 0. 5 where the variation in the phase transformation line can be
To illustrate analytically the effect of dilatancy on the stress± clearly seen, and then de®ne d by d ˆ (d 0 =M)(M d ÿ ç), ac-
strain relationship, consider undrained triaxial loading during cording to the classical stress±dilatancy framework. It follows
which dåv ˆ dåev ‡ dåpv ˆ 0. With dåq ˆ dåeq ‡ dåpq and the elas- now that ø , 0 (dense states) implies M d , M, ø . 0 (loose
tic relations dåev ˆ d p9=K, dåeq ˆ dq=3G in terms of the elastic states) implies M d . M, and ø ˆ 0 (phase transformation
bulk and shear moduli K and G, respectively, the condition states) implies M d ˆ M.
dåv ˆ 0 yields dåq =dq ˆ (1=3G) ÿ (d p9=dq)=Kd, recalling that
d ˆ dåpv =dåpq . When the stress path reaches and moves along a
line of more or less constant stress ratio ç ˆ M (failure surface;
see Fig. 6(a)), one has d p9=dq ˆ 1=M, and hence the above A SIMPLE MODEL FOR TRIAXIAL COMPRESSION
relationship becomes (Li, 1997) Within the scope of this paper, it is assumed that plastic
  deformation occurs whenever ç exceeds its historic maximum
dåq 1 1 and a constant ç path induces no plastic deformation. This is,
ˆ ÿ (13)
dq çˆ M 3G KMd of course, only approximately true, but it is still a good
approximation in many cases since under normal levels of
Equation (13) portrays an analytical conclusion for a general con®ning pressures of interest a constant ç path induces only a
elastoplasticity class of constitutive setting, not related to a relatively small plastic volume change in sands, before grain-
speci®c model. This equation states that the dåq =dq, while crushing levels of pressures are reached as corroborated experi-
ç ˆ M, is controlled primarily by the dilatancy d ˆ dåpv =dåpq, mentally by Poorooshasb et al. (1966, 1967). However, for a
which de®nes the direction of plastic strain increment in the fully ¯edged model where the plastic deformations under con-
volumetric±deviatoric plane. Equation (13) also shows that stant ç are to be considered, additional mechanisms, such as a
(dåq =dq)çˆ M is independent of plastic hardening within the p9 controlling cap, can be added with ease (Wang et al., 1990).
approximation implied by setting d p9=dq ˆ 1=M. Under the above assumption, the yield criterion can be
Figure 6(b) illustrates a typical response described by equa- written as
tion (13) for medium-to-dense sand. The shear stress±strain
curve is characterized by a slope dq=dåq which, for lower q f ˆ q ÿ ç p9 ˆ 0 (16)
values, keeps increasing, but as q reaches higher values begins
to decrease as the stress±strain curve bends over, eventually By the theory of plasticity (Dafalias, 1986), a loading index
leading to the critical state at which d p ˆ dq ˆ dåv ˆ 0 while L can be de®ned as
dåq 6ˆ 0. Along the path to the critical state, while ç is constant,  
d changes tending towards a zero value. Recalling that the state 1 @f @f dq ÿ ç d p9 p9 dç
Lˆ d p9 ‡ dq ˆ ˆ (17)
parameter ø enters equation (12) for d, and that d is the main K p @ p9 @q Kp Kp
variable, which according to equation (13) can be used to
describe the curve of Fig. 6(b), one may propose a form of
equation (12) for ç ˆ M: where Kp is a plastic hardening modulus. With the dilatancy
mø d ˆ dåpv =dåpq , the plastic strain increment can then be written as
d ˆ d 0 (e ÿ 1) (14)
 p    
in which d 0 and m are two positive modelling parameters. dåq 1 p9 dç=K p
p ˆL ˆ (18)
Observe that d and ø have the same sign and d ˆ 0 when dåv d d p9 dç=K p
ø ˆ 0, which satis®es the relationship between the dilatancy
and the internal material state at the critical state.
While equation (14) may describe the dilatancy for ç ˆ M, Therefore, for L . 0:
one needs to generalize it for ç 6ˆ M. In this investigation, the  
speci®c form of the ø dependent dilatancy according to equa- dq p9 dç 1 1 ç
dåq ˆ dåep ‡ dåpq ˆ ‡ ˆ ‡ dq ÿ d p9
tion (12) subjected to yielding (equation (14)) when ç ˆ M is 3G Kp 3G K p Kp
obtained from a modi®cation to the dilatancy function in the
(19)
original Camclay model, as follows:
   
ç d0 d p9 d 1 dç
d ˆ d 0 e mø ÿ ˆ (Me mø ÿ ç) (15) dåv ˆ dåev ‡ dåpv ˆ ‡ d dåpq ˆ dq ‡ ÿ d p9
M M K Kp K Kp
(20)
It can be seen that the Camclay dilatancy d ˆ M ÿ ç is a
special case of equation (15) (m ˆ 0 and d 0 ˆ M). Note that at
ç ˆ M equation (15) is reduced to equation (14). More impor- Equations (19) and (20) establish the relationship between the
tantly, at a critical state, ø ˆ 0 and ç ˆ M simultaneously, and stress and strain increments. They can be inverted by a straight-
hence equation (15) yields a zero dilatancy, satisfying the forward algebraic manipulation, and expressed in a matrix form
requirement set by equation (9). as:

Downloaded by [ National Taiwan University] on [19/11/23]. Copyright © ICE Publishing, all rights reserved.
DILATANCY FOR COHESIONLESS SOILS 455
 
dq automatically satis®ed by equation (24). Condition (c) will be
ˆ
d p9 met at a particular material state ø , 0 (a dense state), for
"   #  which ç ˆ M b ˆ Me ÿ nø , depending on the parameter n. In
3G 0 h(L) 9G2 ÿ3KGç dåq addition to the above conditions, it also follows from equation
ÿ
0 K K p ‡ 3G ÿ Kçd 3KGd ÿK 2 çd dåv (18) that dåpv = p9 dç ˆ d=K p . As dåpv =dç ˆ 0 at ç ˆ 0 is nor-
mally observed, it is necessary to have d=Kp ˆ 0 at ç ˆ 0. By
(21) combining equations (15) and (24), it can be seen that this
condition is also satis®ed automatically, since K p ˆ 1 at
where h(L) is a Heaviside function with h(L) ˆ 1 for L . 0 and ç ˆ 0.
h(L) ˆ 0 otherwise. Note that the extension of the foregoing In equation (24), G serves as a normalizing factor of h. It
relationship to account for reverse loading requires some addi- was found that a variable h with density ®tted the experimental
tional mechanism such as a back-stress (Manzari & Dafalias, data better. In the present investigation, the simple linear
1997), or a memory of the reversal stress ratio point (Wang et dependence
al., 1990). These aspects are not addressed within the restricted h ˆ h1 ÿ h2 e (25)
scope of this paper.
For the model to be completed, in addition to the dilatancy d, was used, where h1 and h2 are two material constants, and e is
which is de®ned in equation (15), the moduli G, K, and Kp the current void ratio.
must also be de®ned. The elastic shear modulus G can be As shown in equation (19), the shear stress±strain response
expressed by the following empirical equation (Richart et al., is controlled by K p. Substitution of equation (24) into equation
1970): (19) with a constant p9 (a drained condition) and n ˆ 0 for
simplicity, yields
(2:97 ÿ e)2 p    
G ˆ G0 p9 pa (22) 1 1 1 q
1‡e dåq ˆ ‡ dq ˆ ‡ dq (26)
3G Kp 3G hG(qf ÿ q)
where G0 is a material constant, and e is used instead of initial
void ratio employed in Richart et al. (1970). Based on elasticity where qf ˆ Mp9 is the value of q at failure. Integration of both
theory, the elastic bulk modulus K is equal to sides of equation (26) yields
2(1 ‡ í)  ÿ1
KˆG (23) q 1 1 1
3(1 ÿ 2í) ˆ ÿ ÿ ln(1 ÿ r) (27)
åq 3G hG hGr
where í is the Poisson's ratio. In this model, í is considered as where r ˆ q=qf . Equation (27) can be converted into a normal-
a material constant independent of pressure and density. ized modulus reduction curve (secant shear modulus normalized
For the plastic modulus K p the following constitutive relation to its maximum value Gmax versus shear strain) with h as a
is proposed: parameter. Fig. 8 shows a family of such curves together with
  the curve based on the hyperbolic stress±strain relationship
M hGe nø q=åq ˆ 3Gmax =(1 ‡ å=åqr ), where Gmax and åqr are two material
K p ˆ hG ÿ e nø ˆ (Me ÿ nø ÿ ç) (24)
ç ç constants. It can be seen that the Kp function used here allows
more ¯exibility in calibrating the shear stress±strain response
where h and n are two positive model parameters; the state than does using the hyperbolic stress±strain response.
parameter ø is calculated from equation (11). Equation (24) is
a modi®ed version of the plastic modulus in a bounding surface
hypoplasticity model (Wang et al., 1990). The modi®cation is
intended to model the peak stress ratio response and softening CALIBRATION OF MODEL CONSTANTS
of dense sands, the lack of such a response for loose sands, and There are eleven material constants in the model, shown
the failure at a residual stress ratio M at the critical state for all according to their functions in separate columns in Table 1. A
densities. Based on the last member in equation (24), it follows systematic procedure can be followed to calibrate all these
that Kp depends on the difference of the current stress ratio ç parameters, based on triaxial data, as follows.
from a `virtual' peak stress ratio M b ˆ Me ÿ nø attainable at the The critical state constants consist of M, the critical state
current state de®ned by ø. Such peak stress ratio is variable stress ratio, and the parameters eà , ëc , and î of equation (11).
with ø in a way that yields M b . M for ø , 0 (dense states), These four constants can be determined by directly ®tting the
M b , M for ø . 0 (loose states), and M b ˆ M for ø ˆ 0
(critical states). The idea of having a virtual peak stress ratio
varying with ø in order to address the issue of peak stress and
subsequent softening of dense sand in drained conditions was
1·2
proposed by Wood et al. (1994). The idea was followed by
Manzari & Dafalias (1997), where a slightly modi®ed version Hyperbolic
of the Wood et al. (1994) linear relation M b ˆ M ÿ nø was
Modulus reduction, Gsec/Gmax

1·0
introduced instead of the present M b ˆ Me ÿ nø .
Observe that it follows from equation (18) that K p ˆ 0·8
h=6
p9 dç=dåpq , which requires the K p function (equation (24)) to
satisfy the following conditions: 0·6 h=3
(a) K p ˆ 1 at ç ˆ 0, because dåpq
ˆ 0 in response to a non-
h = 1·2
zero dç at ç ˆ 0 (the material is assumed isotropic without 0·4
previous loading); h = 0·6
(b) K p ˆ 0 at the critical states (ç ˆ M and ø ˆ 0) because at 0·2
a critical state dç=dåpq ˆ 0; h = 0·3
(c) K p ˆ 0 at drained peak stresses M b , because at those peaks 0
dç=dåpq ˆ 0. 0·001 0·003 0·01 0·03 0·1 0·3 1 3 10
εq /εqr or γ /γr(εqr = qf /3Gmax)
For the hardening and softening responses before and after the
peaks, K p is positive and negative as ç , M b and ç . M b, Fig. 8. Modulus reduction as a function of the hardening parameter
respectively. It can be seen that conditions (a) and (b) are h

Downloaded by [ National Taiwan University] on [19/11/23]. Copyright © ICE Publishing, all rights reserved.
456 LI AND DAFALIAS
Table 1. Model parameters calibrated for Toyoura sand value of 2G0 (1 ‡ í)=3(1 ÿ 2í), and could be negative. A micro-
Elastic Critical state Dilatancy Hardening
mechanics study (Chang & Misra, 1990) has shown that the
parameters parameters parameters parameters Poisson's ratio of an assembly of particles is predominantly
controlled by the ratio of the shear stiffness to the normal
G0 ˆ 125 M ˆ 1:25 d 0 ˆ 0:88 h1 ˆ 3:15 stiffness at particle contacts. The value of the Poisson's ratio of
í ˆ 0:05 eà ˆ 0:934 m ˆ 3:5 h2 ˆ 3:05
ëc ˆ 0:019 n ˆ 1 :1
the assembly could be much lower than that for the particle
î ˆ 0:7 material itself. If this stiffness ratio is high, í could be negative.
Even though a negative í is theoretically justi®able and affects
nothing but the volumetric strain at extremely low strain level
(pure elastic range), if it is encountered and disliked, as an
alternative one may pick a í value ®rst and then calculate G0 .
test data for the critical stress ratio and the critical state line in As this alternative approach does not guarantee the accuracy of
the e± p9 plane. the elastic shear response, it should be used only when the
The parameter m can be determined by equation (15) at a shear stiffness in the elastic range is unimportant or when
phase transformation state, at which d ˆ 0. Hence, accurate values of G are unavailable. Once G0 has been
determined, h, and therefore h1 and h2, can ®nally be found
1 Md from equations (25) and (31).
mˆ ln (28)
ød M Last, but not least, one can obtain the undrained deviatoric
stress±strain response by substituting d p9 ˆ K dåev ˆ ÿK dåpv
where ød and M d are the values of ø and ç at the phase ˆ ÿKd dåpq into equation (32) and accounting for dq ˆ 3G dåeq.
transformation state, measured from drained or undrained test The relationship is as follows:
results.  ÿ1
The parameter n can be determined by equation (24) at a dq 1 1 dq
drained peak stress state, at which K p ˆ 0. Hence, ˆ ‡  ˆ K p ÿ çKd (33)
dåq 3G K p ÿ çKd dåpq
1 M
nˆ ln b (29) As all the model parameters have already been determined, the
øb M simulated undrained q±åq curves can be used against their
where øb and M b are the values of ø and ç at the drained peak experimental counterparts to examine the quality of the calibra-
stress state, measured from test results. tion based on equation (33). If the ®t is not satisfactory, one or
Next, consider the drained tests. Ignoring the small elastic more parameters are to be ®ne tuned and the calibration can be
deformations, repeated until an optimal result is obtained. Note that, after the
  undrained stress path approximately converges with ç ˆ M, the
dåv dåpv ç K p value is still not zero as long as ø 6ˆ 0 (equation (24)).
 p ˆ d ˆ d 0 e mø ÿ (30)
dåq dåq M Hence, the second member of equation (33) does not yield
equation (13) until K p becomes very small and is neglected.
The parameter d 0 can then be calibrated based on the åv ±åq This shows the approximate character of equation (13) in
curves. reference to an actual model such as the one presented here.
By combining equations (19) and (24) for the drained tests, It should be pointed out that the set of model parameters
with either the conventional test (d p9 ˆ dq=3) or the constant calibrated is for one material over a wide range of densities and
p9 test (d p9 ˆ 0), one has pressures. There is no need to do the calibration again for the
dq dq Kp same material when the initial state changes.
 ˆ
dåq dåpq 1 ÿ aç
( p ) SIMULATION BY THE MODEL
(2:97 ÿ e)2 p9 pa [(M=ç) ÿ e nø ] Verdugo & Ishihara (1996) presented a sequence of triaxial
ˆ hG0 (31)
(1 ‡ e)(1 ÿ aç) test results on Toyoura sand which are particularly suited to
demonstrating the simulative capability of a critical-state-based
in which the parameter a is either equal to 1=3 (for conven- sand model, since some of the data provide a de®nite identi®ca-
tional tests) or to zero (for constant p9 tests). As all the material tion of the critical state line in e± p9 space, which is of cardinal
constants in the brackets have been predetermined, the com- importance for the determination of ø. The sand is described as
bined parameter hG0 can be calibrated independently based on uniform ®ne sand consisting of subrounded to subangular
the experimental q ÿ åq curves. It may be found during calibra- particles. The maximum and minimum void ratios are 0´977
tion that the quantity hG0 varies with density. Fitting these and 0´597, respectively. Verdugo & Ishihara reported a total of
values of hG0 into equation (25) yields the constants h1 and h2 17 shear loading tests in their paper. The tests include both
(after G0 has been determined). drained and undrained triaxial compression tests. The density
Now let us turn to undrained (constant volume) tests. For ranged from e ˆ 0:735 (relative density Dr ˆ 63:7%) to
dåv ˆ 0, equation (20) yields: e ˆ 0:996 (Dr  0%). The initial con®ning pressure p9 for the
  tests ranged from 100 kPa to 3000 kPa. This set of test results
dq Kp 3(1 ÿ 2í) hG0 [(M=ç) ÿ e nø ] covers comprehensively the behaviour of the Toyoura sand
ˆçÿ ˆçÿ (32) under monotonic triaxial compression loading conditions.
d p9 Kd 2G0 (1 ‡ í) d
All 17 tests were simulated using the simple model described
As all the material constants in the brackets have been prede- earlier with the uni®ed set of model parameters listed in Table
termined, the combined parameter 2G0 (1 ‡ í)=3(1 ÿ 2í) be- 1. Figs 9±14 show the experimental results for all the 17 tests,
comes the only means at this stage of adjusting the undrained as well as the results of the simulations obtained with the
p9± q responses of the model. By matching these responses with model. It can be seen that the model simulations broadly match
their experimental counterparts, the value of 2G0 (1 ‡ í)= the experimental results, indicating the effectiveness of the
3(1 ÿ 2í) can be determined. critical-state framework in conjunction with the state-dependent
Finally, one needs to separate the parameter G0 from h and dilatancy.
í. If shear stiffness at small strains is important, G0 should be To compare with others, the discrepancies between the model
determined by independent small strain tests, such as resonant simulation and the test results for drained responses at very low
column tests or bender element tests, through ®tting the test densities (initial void ratios e0 ˆ 0:96 and e0 ˆ 0:996; or rela-
data into equation (22). However, if G0 is high, the value of tive densities Dr ˆ 4:5% and Dr  0%) are more notable. A
Poisson's ratio í is reduced based on the already calibrated simple modi®cation of equation (25) by replacing the void ratio

Downloaded by [ National Taiwan University] on [19/11/23]. Copyright © ICE Publishing, all rights reserved.
DILATANCY FOR COHESIONLESS SOILS 457
0·95 2500
Model simulation
2000 Test results
0·90

Deviator stress, q: kPa


1500
Void ratio, e

0·85

1000
0·80

Experimental 500
0·75
e = 0·934 – 0·019(p ′/pa)0·7
0
0 500 1000 1500 2000 2500 3000
0·70
0 0·5 1·0 1·5 2·0 2·5 3·0 3·5 Effective mean normal stress, p ′: kPa
Effective mean stress, p ′: MPa (a)

Fig. 9. Experimental data and theoretical ®tting of the critical state 2500
line in e± p9 space for Toyoura sand
2000

Deviator stress, q: kPa


p0′: 3000 kPa
1500
5000

Model simulation
4000 Test results
1000 p0′: 100 kPa
Deviator stress, q: kPa

p0′: 1000 kPa


500
3000 p0′: 2000 kPa

0
2000 0 5 10 15 20 25 30
Axial strain, ε1: %
(b)
1000
Fig. 11. Comparison between undrained triaxial compression test
results and model simulations for Dr ˆ 37´9% (e ˆ 0´833)
0
0 500 1000 1500 2000 2500 3000
Effective mean normal stress, p ′: kPa
(a)
5000 1200
Model simulation
1000 Test results
4000
Deviator stress, q: kPa
Deviator stress, q: kPa

800
3000
p0′: 100 kPa 600

2000 p0′: 1000 kPa


400
p0′: 2000 kPa
1000 p0′: 3000 kPa 200

0 0
0 5 10 15 20 25 30 0 500 1000 1500 2000 2500 3000
Axial strain, ε1: % Effective mean normal stress, p ′: kPa
(b) (a)
1200
Fig. 10. Comparison between undrained triaxial compression test
results and model simulations for Dr ˆ 63´7 (e ˆ 0´735) 1000
Deviator stress, q: kPa

800 p0′: 2000 kPa

e with the initial void ratios e0 (i.e. h ˆ h1 ÿ h2 e0 instead of p0′: 1000 kPa
600
h ˆ h1 ÿ h2 e) brought the model simulations much closer to
p0′: 100 kPa
the experimental counterparts (Fig. 15). This modi®cation also
400
slightly improved other drained simulations, but has no effect
on the undrained response because e ˆ e0 under undrained
conditions. This e0 dependence may be due to the in¯uence of 200
material fabric at very low densities, since the void ratio of a
sand without a signi®cant shearing history e0 is strongly 0
0 5 10 15 20 25 30
correlated to the packing structure of the sand (see Figs 1 and Axial strain, ε1: %
2). However, since the stress and material state corresponding to (b)
a given e0 are not always clearly and objectively de®ned,
introducing e0 into a constitutive equation as a general para- Fig. 12. Comparison between undrained triaxial compression test
meter needs further investigation. results and model simulations for Dr ˆ 18´5% (e ˆ 0´907)

Downloaded by [ National Taiwan University] on [19/11/23]. Copyright © ICE Publishing, all rights reserved.
458 LI AND DAFALIAS
350 CONCLUSIONS
Model simulation The classical stress dilatancy theory in its exact form ignored
300
Test results the extra energy loss due to the static and kinematic constraints
at particle contacts. While this hypothesis leads to a unique
Deviator stress, q: kPa

250
relationship between the stress ratio and dilatancy, it obstructs
200 uni®ed modelling of the behaviour of cohesionless soils over a
full range of densities and stress levels.
150 To remove this obstacle, additional dependence of dilatancy
on the internal state of a material is needed. The concept of
100 state-dependent dilatancy was introduced, in conjunction with
the basic concepts of critical-state soil mechanics. The general
50
expression and basic requirements for dilatancy were addressed.
0 It has been shown that the state parameter, the difference
0·8 0·85 0·9 0·95 1·0 between the current void ratio and the critical state void ratio
Void ratio, e corresponding to the current con®ning pressure, is an effective
(a) means of measuring how far the material state is from the
350 critical state. With the state parameter as the state variable, a
particular form of state-dependent dilatancy was proposed, and
300 was shown to be equivalent to an interpretation whereby the
phase transformation stress ratio is variable with the state
Deviator stress, q: kPa

250
parameter, an idea introduced by Manzari & Dafalias (1997).
200
With this form of dilatancy and a state-parameter-dependent
plastic modulus, for which an interpretation is again possible
150 e0 = 0·831 whereby the peak stress ratio depends on the state parameter as
proposed by Wood et al. (1994), a simple model in the triaxial
e0 = 0·917
100 space as well as a systematic calibration procedure was intro-
e0 = 0·996
duced. It was shown that this simple model has the ability to
50 simulate data successfully, with a single set of model constants,
0 for a suite of 17 triaxial tests, both drained and undrained, of
0 5 10 15 20 25 30 Toyoura sand over a relative density range of around 0±64%
Axial strain, ε1: % subjected to a con®ning pressure range of 100±3000 kPa.
(b)

Fig. 13. Comparison between drained triaxial compression test ACKNOWLEDGEMENTS


results and model simulations for p0 ˆ 100 kPa
The ®nancial support provided to X. S. L. by the Research
Grants Council (RGC) of Hong Kong through Grant
HKUST721/96E is gratefully acknowledged. Y. F. D. acknowl-
edges the support from the National Science Foundation, Grant
1600
Model simulation
No. CMS-9800330, of the programme directed by Dr Cliff
Astill.
Test results
Deviator stress, q: kPa

1200

APPENDIX. DILATANCY OF TWO REGULAR PACKINGS OF


UNIFORM RIGID RODS
800
Packing A (Fig. 1)
Following Rowe (1962), when sliding takes place, one has the ratio of the
loads
400
L1 =L2 ˆ tan(öu ‡ â) (34)
where Li is the load per rod in direction i, öì is the angle of friction
0
between the surfaces of the rods in contact, and â is the deviation of the
0·8 0·85 0·9 0·95 tangent at the contact points from the direction 1. The size of a basic cell
Void ratio, e in the packing is characterized by
(a)
l1 ˆ 4r sin â (35a)
1600 and
l2 ˆ 4r cos â (35b)
Therefore, the ratio of the major principal stress to the minor principal
Deviator stress, q: kPa

1200
stress is
ó 19 =ó 29 ˆ L1 l1 =L2 l2 ˆ tan â tan(öì ‡ â) (36)
800 The ratio of the strain rates in directions 1 and 2 is
e0 = 0·810 !
å_ 2 ÿd V_ =V ÿ å_ 1 d V_ _l2 l1
e0 = 0·886 ˆ ˆÿ 1‡ ˆ ˆ ÿtan2 â (37)
400 å_ 1 å_ 1 V å_ 1 _l1 l2
e0 = 0·960
Therefore, the ratio of the work done per unit volume by the major
principal stress to the work done on the minor principal stress is
0
0 5 10 15 20 25 30 ó 19 å_ 1 ó 19 tan(öì ‡ â)
ÿ ˆ ˆ (38)
Axial strain, ε1: % ó 29 å_ 2 ó 29 (1 ‡ d V_ =V å_ 1 ) tan â
(b)
The void ratio of the packing is
Fig. 14. Comparison between drained triaxial compression test l1 l2 ÿ 2ðr 2 16r 2 sin â cos â 8 sin â cos â
results and model simulations for p0 ˆ 500 kPa eˆ ˆ ÿ1ˆ ÿ1 (39)
2ðr 2 2ðr 2 ð

Downloaded by [ National Taiwan University] on [19/11/23]. Copyright © ICE Publishing, all rights reserved.
DILATANCY FOR COHESIONLESS SOILS 459
350 1600
Model simulation Model simulation
300 Test results Test results
Deviator stress, q: kPa

Deviator stress, q: kPa


250 1200

200
800
150

100
400
50

0 0
0·8 0·84 0·88 0·92 0·96 1·0 0·8 0·84 0·88 0·92 0·96
Void ratio, e Void ratio, e

350 1600

300

Deviator stress, q: kPa


Deviator stress, q: kPa

250 1200

200
800
150 e0 = 0·831 e0 = 0·810

100 e0 = 0·917 e0 = 0·886


400

50 e0 = 0·996 e0 = 0·960

0 0
0 10 20 30 0 10 20 30
Axial strain, ε1: % Axial strain, ε1: %
(a) (b)

Fig. 15. Simulations of drained tests with h varying with e0 instead of e: (a) p0 ˆ 100 kPa, h ˆ h1 ÿ h2 e0 ; (b) p0 ˆ 500 kPa, h ˆ h1 ÿ h2 e0

Packing B (Fig. 2) properties of granulates. J. Engng Mech. Div., ASCE 116, No. 5,
At each contact point, sliding takes place when the ratio of the loads 1077±1093.
L1 Cubrinovski, M. & Ishihara, K. (1998). Modelling of sand behaviour
ˆ tan(öì ‡ â) (40) based on state concept. Soils Foundations 38, No. 3, 115±127.
L2 =2 Dafalias, Y. F. (1986). An anisotropic critical state soil plasticity model.
The size of a basic cell in this packing is characterized by Mech. Res. Commun. 13, No. 6, 341±347.
Ishihara, K. (1993). Liquefaction and ¯ow failure during earthquakes.
l1 ˆ 4r sin â (41a) 33rd Rankine lecture, GeÂotechnique 43, No. 3, 351±415.
and Ishihara, K., Tatsuoka, F. & Yasuda, S. (1975). Undrained deformation
and liquefaction of sand under cyclic stresses. Soils Foundations 15,
l2 ˆ 2r ‡ 4r cos ⠈ 2r(1 ‡ 2 cos â) (41b) No. 1, 29±44.
Therefore, the ratio of the major principal stress to the minor principal Jefferies, M. G. (1993). NorSand: a simple critical state model of sand.
stress is GeÂotechnique 43, No. 1, 91±103.
Kabilamany, K. & Ishihara, K. (1990). Stress dilatancy and hardening
ó 19 2L1 l1 2 sin â laws for rigid granular model of sand. Soil Dynam. Earthquake
ˆ ˆ tan(öì ‡ â) (42)
ó 29 L2 l2 1 ‡ 2 cos â Engng, 9, No. 2, 66±77.
The ratio of the strains in directions 1 and 2 is Li, X. S. (1997). Modelling of dilative shear failure. J. Geotech. Engng
! Div., ASCE 123, No. 7, 609±616.
å_ 2 d V_ _l2 l1 2 sin2 â Li, X. S. & Wang, Y. (1998). Linear representation of steady-state line
ˆÿ 1‡ ˆ ˆÿ (43) for sand. J. Geotech. Geoenviron. Engng 124, No. 12, 1215±1217.
å_ 1 V å_ 1 _l 1 l2 (1 ‡ 2 cos â)cos â
Li, X. S., Dafalias, Y. F. & Wang, Z. L. (1999). State dependent
Therefore, the ratio of the work done per unit volume by the major dilatancy in critical state constitutive modelling of sand. Can.
principal stress to the work done on the minor principal stress is equal to Geotech. J. 36, No 4, 599±611.
Manzari, M. T. & Dafalias, Y. F. (1997). A critical state two-surface
ó 19 å_ 1 ó 19 tan(öì ‡ â)
ÿ ˆ ˆ (44) plasticity model for sands. GeÂotechnique 47, No. 2, 255±272.
ó 29 å_ 2 ó 29 (1 ‡ d V_ =V å_ 1 ) tan â Nova, R. & Wood, D. M. (1979). A constitutive model for sand in triaxial
The void ratio of the packing is compression. Intl J. Num. Anal. Methods Geomech. 3, 255±278.
Pastor, M., Zienkiewicz, O. C. & Chan, H. C. (1990). Generalized
l1 l2 ÿ 3ðr 2 8r 2 sin â(1 ‡ 2 cos â) 8 sin â(1 ‡ 2 cos â) plasticity and the modelling of soil behaviour. Intl J. Numerical
eˆ ˆ ÿ1ˆ ÿ1
3ðr 2 3ðr 2 3ð Anal. Methods Geomech. 14, 151±190.
(45) Poorooshasb, H. B., Holubec, I. & Sherbourne, A. N. (1966). Yielding
and ¯ow of sand in triaxial compression, Part I. Can. Geotech. J. 3,
No. 5, 179±190.
Poorooshasb, H. B., Holubec, I. & Sherbourne, A. N. (1967). Yielding
REFERENCES and ¯ow of sand in triaxial compression, Part II. Can. Geotech. J. 4,
Been, K. & Jefferies, M. G. (1985). A state parameter for sands. No. 4, 376±397.
GeÂotechnique 35, No. 2, 99±112. Roscoe, K. H. & Burland, J. B. (1968). On the generalized stress±strain
Bolton, M. D. (1986). The strength and dilatancy of sands. GeÂotech- behaviour of `wet' clay. Engineering plasticity, pp. 535±609. Cam-
nique 36, No. 1, 65±78. bridge: Cambridge University Press.
Chang, C. S. & Misra, A. (1990). Packing structure and mechanical Roscoe, K. H. & Scho®eld, A. N. (1963). Mechanical behaviour of an

Downloaded by [ National Taiwan University] on [19/11/23]. Copyright © ICE Publishing, all rights reserved.
460 LI AND DAFALIAS
idealized `wet' clay. Proc. Eur. Conf. Soil Mech. Found. Engng, Verdugo, R. & Ishihara, K. (1996). The steady state of sandy soils. Soils
Wiesbaden 1, 47±54. Foundations 36, No. 2, 81±91.
Roscoe, K. H., Scho®eld, A. N. & Wroth, C. P. (1958). On the yielding Wan, R. G. & Guo, R. G. (1998). A simple constitutive model for
of soils. GeÂotechnique 8, No. 1, 22±53. granular soils: modi®ed stress±dilatancy approach. Comput. Geo-
Rowe, P. W. (1962). The stress±dilatancy relation for static equilibrium tech. 22, No. 2, 109±133.
of an assembly of particles in contact. Proc. R. Soc., London, Ser. A Wang, Z. L., Dafalias, Y. F. & Shen, C. K. (1990). Bounding surface
269, 500±527. hypoplasticity model for sand. J. Engng Mech. Div., ASCE 116, No.
Richart, F. E. Jr., Hall, J. R. & Woods, R. D. (1970). Vibrations of soils 5, 983±1001.
and foundations. Englewood Cliffs, NJ: Prentice-Hall. Wood, D. M. (1990). Soil behaviour and critical state soil mechanics.
Taylor, D. W. (1948). Fundamentals of soil mechanics. New York: New York: Cambridge University Press.
Wiley. Wood, M. D., Belkheir, K. & Liu, D. F. (1994). Strain softening and
Vardoulakis, I. & Sulem, J. (1995). Bifurcation analysis in geomecha- state parameter for sand modelling. GeÂotechnique, 44, No. 2, 335±
nics. Glasgow: Blackie Academic & Professional. 339.

Downloaded by [ National Taiwan University] on [19/11/23]. Copyright © ICE Publishing, all rights reserved.

Vous aimerez peut-être aussi