Вы находитесь на странице: 1из 18

Manzari, M. T. & Dafalias, Y. F. (1997). GeÂotechnique 47, No.

2, 255±272

A critical state two-surface plasticity model for sands

M . T. M A N Z A R I  a n d Y. F. DA FA L I A S {

Within the critical state soil mechanics frame- Dans le contexte de la meÂcanique des sols aÁ
work, the two-surface formulation of plasticity is l'eÂtat critique, on combine la formulation de la
coupled with the state parameter to construct a plasticite aÁ deux surfaces et le parameÁtre d'eÂtat
constitutive model for sands in a general stress pour construire un modeÁle constitutif pour les
space. The operation of the two-surface model sables dans un espace de tension geÂneÂrale.
takes place in the deviatoric stress-ratio space, L'exploitation du modeÁle aÁ deux surfaces se
and the state parameter is used to de®ne the produit dans l'espace deÂviateur des rapports de
peak and dilatancy stress ratios of sand. The tension et le parameÁtre d'eÂtat sert aÁ de®nir les
model is capable of realistically simulating rapports de pointe et de dilatance des tensions
stress±strain behaviour of sands under mono- du sable. Le modeÁle peut simuler reÂalistique-
tonic and cyclic, drained and undrained loading ment le comportement tension=deÂformation des
conditions. It includes features such as the sables dans des conditions de charge uniformes
softening of sands at states denser than critical et cycliques, draineÂes et non draineÂes. Il com-
as they dilate in drained loading and softening prend des caracteÂristiques comme l'ameublisse-
of sands looser than critical in undrained ment de sables aÁ des eÂtats de densite supeÂrieurs
loading, and the pore-water pressure increase aÁ l'eÂtat critique, quand ils se dilatent dans des
under undrained cyclic loading. Most important, conditions de charge draineÂes, et aÁ des eÂtats de
all these simulations are achieved by a unique densite infeÂrieurs aÁ l'eÂtat critique dans des
set of model constants at all densities and conditions de charge non draineÂes, ainsi que
con®ning pressures of engineering relevance for l'augmentation de la pression interstitielle dans
a given sand. The numerical implementation of des conditions de charge cyclique non draineÂes.
the model is particularly easy and ef®cient due Plus important encore, toutes ces simulations
to the very simple formulation. Calibration of sont reÂaliseÂes aÁ l'aide d'un seul ensemble de
model constants is done straightforwardly on constantes de modeÁle pour toutes les densiteÂs et
the basis of triaxial experiments and measure- pressions de con®nement utiles pour un sable
ments of well-known characteristics of sand donneÂ. La mise en oeuvre numeÂrique de modeÁle
stress±strain behaviour. Possibly the most at- est particulieÁrement facile et ef®cace en raison
tractive feature of the model is its simplicity and de la grande simplicite de la formulation.
its foundation on concepts and data which are L'eÂtalonnage des constantes du modeÁle se fait
well established and understood by the geotech- simplement aÁ partir d'essais triaxiaux et de
nical engineering community, with basic refer- mesures de caracteÂristiques bien connues du
ence to critical state soil mechanics. comportement tension=deÂformations des sables.
L'aspect le plus inteÂressant de ce modeÁle est
peut-eÃtre sa simplicite et le fait qu'il repose sur
des concepts et des donneÂes qui sont bien eÂtablis
et qui sont bien compris des geÂotechniciens, avec
KEYWORDS: anisotropy; bounding surface; critical mention sommaire de la meÂcanique des sols aÁ
state; plasticity; sands; state parameter. l'eÂtat critique.

INTRODUCTION interpreted as a general theoretical framework for


In this paper, critical state soil mechanics (CSSM) explaining the stress±strain behaviour of soils. The
(Roscoe et al., 1958; Scho®eld & Wroth, 1968) is CSSM can be used to study soil behaviour without
any reference to a particular constitutive model
(Roscoe et al., 1958; Wood, 1990). Alternatively, it
Manuscript received 7 August 1995; revised manuscript can provide a theoretical basis which, in combina-
accepted 2 February 1996. tion with concepts from general constitutive the-
Discussion on this paper closes 1 September 1997; for
further details see p. ii. ories, such as elastoplasticity, can be used to con-

George Washington University. struct pertinent soil constitutive models. Original
{ National Technical University of Athens and University efforts by Roscoe et al. (1963) and Roscoe &
of California at Davis. Burland (1968) are classic examples of applying the

255
256 MANZARI AND DAFALIAS

elastoplasticity theory within the framework of gether on the basis of modi®cations of the recent
CSSM. It can be said that, in general, the CSSM proposition by Wood et al. (1994) relating peak
plasticity models were able to capture well the stress ratio and state parameter.
stress±strain response characteristics of normally
consolidated clays, and with a lesser degree of
accuracy, that of sands under monotonic loading BASIC CONCEPTS
conditions. The basic underlying concepts of the model in
When the need to describe the cyclic response the triaxial q, p space (all stresses are effective)
characteristics of soils (such as the pore-water and the de®nition of some key quantities are
pressure development under cyclic undrained load- described below before the analytical form of the
ing conditions) arose, it became evident that what constitutive equations is presented.
was available was not enough. This, however, had CSSM is the basic theoretical framework. The
more to do with the classical isotropic hardening critical state, at which deformation continues for
plasticity framework used, and less with the ®xed stresses and zero volumetric strain rate, is
premises of CSSM itself. More advanced plasticity attained when the stress ratio ç ˆ q= p equals M c ,
modelling concepts, such as kinematic hardening, the critical value in triaxial compression, and
multisurface plasticity and bounding surface plas- simultaneously the void ratio e equals ec , the
ticity were introduced to model cyclic loading critical void ratio, which is assumed to be a unique
conditions in addition to monotonic ones. Plasticity function of p. The ec , p relation is usually taken
models proposed by Mroz et al. (1978), Prevost to be linear in the e, ln p space de®ning the
(1978) and Dafalias & Herrmann (1980, 1986) critical state line (CSL) of slope ë as shown in
were among the ®rst attempts, which were fol- Fig. 1.
lowed by many other works of a similar kind. The A typical material response is illustrated in Figs
CSSM concepts still remained in the background, 1 and 2. If a state initially denser than critical,
but often the price one had to pay for the represented by point a in Fig. 1, is subjected to
increased simulative capabilities of the models drained constant- p triaxial compression, it will ®rst
was greater complexity, more constants to calibrate consolidate, moving to point a9d , and then dilate
and greater dif®culty in numerical implementation. until it reaches point a9c on the CSL where e ˆ ec
In addition it appears that greater accuracy in the and critical failure occurs. Simultaneously, in Fig.
simulation of cyclic loading was achieved by 2, which is drawn in the ç, 1 space by normalizing
sacri®cing the accuracy of simulations for mono- both q and p by p, the point a moves upwards,
tonic loading, and vice versa. crosses point a9d where consolidation changes to
Particularly for sands, one additional negative dilation at a stress ratio ç ˆ M dc , moves up to
feature was that samples at suf®ciently different point a9b at a peak stress ratio ç ˆ M bc and then
densities were treated as different rather than the falls back to point a9c , at the residual or critical
same materials (e.g. Zeng & Arulanandan (1993), stress ratio ç ˆ M c where it fails. The description
which reports the calibrated parameters used for of the softening from a9b to a9c will be one of the
two different void ratios in the VELACS class A important aspects of the model. Under undrained
prediction exercise). In an effort to treat sands loading point a will ®rst move to point a d0 (Fig. 1),
using the same constitutive constants at all densi- as positive pore water pressure development due to
ties, Jefferies (1993) and Crouch et al. (1994) the initial contractive tendency reduces the effec-
introduced models based on CSSM from different tive stress p, and then upon dilative tendency will
perspectives. Both models, however, had no provi-
sion for suf®ciently accurate cyclic loading re-
sponse simulation. Crouch & Wolf (1994) extended
their previous work with Dafalias in a way which
looks promising but, perhaps, too complicated as
well.
The main objective of this paper is to develop a
comprehensive constitutive model for sands which
avoids the previously mentioned disadvantage and,
simultaneously, is both simple and easy to under-
stand. Two known concepts are used: (1) the
concept of a yield=bounding or two-surface plas-
ticity formulation (Dafalias & Popov, 1975; Krieg,
1975), and (2) the concept of the state parameter
(Been & Jefferies, 1985) as an essential variable Fig. 1. Schematic illustration of drained and un-
arising in a CSSM framework. In the process, drained paths in e, ln p space for a state denser than
these two independent concepts are coupled to- critical (point a) and looser than critical (point b)
TWO-SURFACE PLASTICITY MODEL FOR SANDS 257
CSSM provides a basis on which such consti-
tutive models can be built. It is a direct conse-
quence of the CSSM structure that the difference
e ÿ ec , with e and ec de®ned at the same p, plays
an important role in such model development. For
further reference the equation
ø ˆ e ÿ ec (1)
is stated, where the name `state parameter' for the
symbol ø introduced by Been & Jefferies (1985)
will be adopted (see also Fig. 1). Jefferies (1993)
developed the Nor-Sand critical state sand model
using ø explicitly in the formulation. In other
CSSM sand models (e.g. Crouch et al., 1994), ø
appears indirectly by way of the assumed shape of
the yield surface. In the present constitutive model
an explicit use of the state parameter ø will be
Fig. 2. Schematic illustration in the ç, 1 space of the made.
bounding (peak) stress ratio M bc , critical stress ratio The requirement to have a model which can
M c and dilatancy stress ratio M dc , the back-stress ratio describe the response under monotonic and cyclic
á, the width 2m of the wedge-type yield surface and loading conditions (i.e. a model which can describe
the constant-p drained paths of Fig. 1
reverse as well as forward loading), is addressed
by postulating the existence of a yield surface in
q, p space shown by the boundaries (Oc and Oe)
move to the critical point a c0 at the same e as the of the shaded narrow wedge with apex at the
initial value, where failure occurs. The correspond- origin O in Fig. 2. The equation of the wedge-type
ing undrained q, p path with the characteristic yield surface is given by
`hook' response is not shown in Fig. 2 for clarity. f ˆçÿá mˆ0 (2)
For a state looser than critical, as in point b of
Fig. 1, the response is different. Under drained where á locates the bisector of the wedge angle
constant- p loading the material consolidates and b shown by a discontinuous line, and 2m is the wedge
moves to the corresponding critical point b9c and `opening', á and m being stress ratio quantities.
fails (Figs 1 and 2), in general without dilation and The  sign stands for the top and bottom lines Oc
softening. It may, however, be the case that if the and Oe of the wedge, that is, a point is on the yield
state at b is not much looser than critical (an surface when the mobilized stress ratio is equal to
obvious example is the limit case b ˆ b9c initially ç ˆ á ‡ m (top) or ç ˆ á ÿ m (bottom). Equation
in Fig. 1) the constant- p path may cross point b9c (2) is a cone type stress-ratio yield criterion which
down to point b9d , denser than critical, and then introduces, for reverse loading, a mobilized stress
turn up and reach b9c again where it fails. This ratio in the reverse direction while ç is still greater
response will be accompanied by a rise to point b9b than 0. Corresponding concepts hold true for
(Fig. 2) followed by a small amount of softening triaxial extension. The effect of á is the equivalent
back to b9c . In a corresponding undrained loading, of a back-stress and ensuing kinematic hardening in
shown only in the e, ln p space of Fig. 1 for plasticity. The wedge-type yield surface will move
clarity, the state moves from point b to point bc0 together with the stress-ratio point in a typical stress
where it fails, and in fact most often it passes ®rst path of the type described earlier.
b0c and reaches bd0 before it turns to point bc0 where It is clear from the above that no plastic
it fails. In this latter case the characteristic `hook' deformation will occur if q and p increase at
shape of the undrained stress path in q, p space constant stress ratio ç, because in that case the
will appear, as for point a. stress path will not tend to cross the sides Oc or Oe
A constitutive model is required to simulate all of the wedge. One could introduce a `cap' closing
these features and the stress±strain relation q, å1 the open wedge at increasing p, as is done for
(not discussed here for brevity) in a quantitatively example in the double hardening sand model by
accurate way. It is also required to simulate the Vermeer (1978), and in the models by Yang et al.
response under cyclic loading if it is to be used in (1985), Lacy & Prevost (1987) and Wang et al.
problems involving such loading (e.g. earthquake- (1990). The complication of the formulation, how-
induced loading). In addition, a unique set of ever, resulting from the intersection of the cap with
model constants for all densities and con®ning the wedge does not warrant the little gain one
pressures is needed as well as a formulation for a would have. This is because under normal con®ning
general stress space. pressures of interest in geotechnical engineering,
258 MANZARI AND DAFALIAS

not causing crushing of the sand particles, a ˆ ÿø will begin to decrease from the positive
constant stress ratio loading induces only relatively value it had at a9b and, consequently, the peak or
small strains. Even if equation (2) is not entirely bounding stress ratio M bc will also decrease
accurate as an exact yield surface (Tatsuoka & Ishi- according to equation (3a). Since no plastic loading
hara (1974) detected experimentally mildly curved occurs, the yield surface will stay where it was, and
lines instead of straight), it may capture much of the stress ratio ç will become greater than the
that part of the sand response which is primarily reduced M bc . This aspect, which at ®rst seems
due to stress-ratio changes. contradictory, is inherent in the proposition made by
The relation between the peak and residual Wood et al. (1994), even for other shapes of yield
stress ratios at critical failure, that is, the soften- surfaces which allow for increasing p by neutral
ing will be addressed now. A modi®cation of a loading, but it was not examined because the
proposition recently made by Wood et al. (1994) authors considered only constant- p drained loading
will be adopted, according to which a `virtual' paths. In a general constitutive formulation, though,
peak or bounding stress ratio can be related to the all possibilities must be examined. Such a situation
critical stress ratio M c by way of ø. Denoting the brings the soil to a very unstable state where any
bounding stress ratio by M bc for compression and attempt to induce plastic deformation again in the
M be for extension, and with M e the critical stress same direction will result in an intense softening,
ratio in extension (the extension values are neces- leading the stress ratio back to M bc from above.
sary for the subsequent generalization to multiaxial This feature will be included in the multiaxial
stress space and will be taken positive), one can model's formulation by proper use of shrinking
write bounding surface concept associated with M bc ,
which will be allowed to be crossed by the current
M bc ˆ ábc ‡ m ˆ M c ‡ k bc hÿøi (3a) stress, not only during neutral or elastic loading as
M be ˆ ábe ‡ m ˆ Me ‡ k be hÿøi (3b) before, but during plastic loading as well, before
softening begins. Such allowance of crossing the
where k bc , k be are positive quantities and the Ma- bounding surface will consistently accommodate the
cauley brackets k l de®ne the operation hÿøi ˆ highly hypothetical but possible example given
ÿø if ÿø . 0 and hÿøi ˆ 0 if ÿø < 0. The use above and, more importantly, will provide the
of these brackets is the difference from the means to incorporate softening of dilatant soils
proposition by Wood et al. (1994). Their effect is during the operation of the model at plastic loading
to not allow M bc to become smaller than M c when as a matter of course.
ø . 0 (similarly for M be ). This is intended to re¯ect With reference to volumetric response, a dila-
the response characteristics of looser than critical tancy stress ratio M dc for compression is shown in
states. It can also be shown that this has a sta- Fig. 2, such that when ç , M dc the soil response is
bilizing effect on the numerical implementation. contractive, and when ç . M dc it is dilative. The
The word `virtual' is used because M bc or M be is not exact way of representing analytically this response
actually the peak stress ratio which will be reached will be discussed later. The line de®ned by M dc is
by the current stress ratio ç, but it changes with ø the phase transformation line described in Ishihara
until it is met by ç at a different value, where it et al. (1975), but perhaps the name `dilatancy line'
becomes the peak. Equations (3) introduce ábc and or `dilatancy stress ratio' for M dc better ®ts the
ábe , which are based on the consideration that if case since no phase is actually transformed in the
there is a `virtual' bounding or peak stress ratio, model. If M dc is constant, and since for dilatant
there is also a corresponding `virtual' bounding or soils M dc , M c , it follows that unrealistic unlimited
peak back-stress ratio for a given m, as a direct dilation will be induced at critical failure when
consequence of equation (2). Equations (3) de®ne ç ˆ M c . Hence, in order to be consistent with
the ábc and ábe in terms of ø, as much as they de®ne CSSM it will be required that M dc is variable and
the M bc and M bc . M dc ! M c as e ! ec . To this extent a proposition
There is a subtle point related to equations (3). similar to the one in equations (3) for M bc and M be
Assume that the stress point has reached the is adopted for both compression and extension,
position a9b (Fig. 2) and, therefore, the side Oc of whereby
the wedge-type yield surface is now along the peak
or bounding slope M bc (not shown for clarity). M dc ˆ ádc ‡ m ˆ M c ‡ k dc ø (4a)
Assume that a loading path along the top side of M de ˆ áde ‡ m ˆ Me ‡ k de ø (4b)
the wedge, that is, with a stress ratio ç ˆ M bc ,
begins from a9b with p increasing, causing no with k dc and k de positive quantities. Equations (4)
plastic deformation, that is, a neutral loading path. give a dilatancy ratio below M c or M e for ø , 0
Alternatively, the stress point could be entirely (denser than critical) and above M c or M e for
inside the wedge and move at constant ç with p ø . 0 (looser than critical), while M dc ˆ M c and
increasing. In either case it follows that ec ÿ e M de ˆ M e at ø ˆ 0. Also equations (4) introduce
TWO-SURFACE PLASTICITY MODEL FOR SANDS 259
the corresponding back-stress dilatancy ratios ádc hÿøi. The option to set k bc and k be ˆ 0 is always
and áde as a function of ø. available, in which case the peak stress and back-
It is now necessary to impose a restriction on stress ratios are always identical to critical, imply-
the dilatancy response, whatever may be the ing no softening in drained response. Many existing
analytical description, as long as dilation occurs models have this restriction built in, for example,
when ç . M dc (or M de ). The reason for a restriction Prevost (1985), Wang et al. (1990) and Manzari
can be understood if one refers back to the (1994). Similarly, the option k dc and k de ˆ 0 renders
hypothetical neutral loading path (considered after the dilatancy stress ratio identical to the critical
equations (3)) at a constant ç with p increasing, stress ratio. This added ¯exibility is useful for
starting from ç . M c and ø , 0 (denser than simpler model con®gurations.
critical). Eventually this stress path will cross the
CSL in the e, p plane and a state with ç . M c and
ø . 0 (looser than critical) will be attained. The GENERAL FORMULATION
value of ø may be such that according to equation The constitutive relations will be presented in
(4a) one can have ç . M dc . M c . Hence, a state two stages. The general equations of a classical
looser than critical has been reached but with a elastoplacity formulation with a cone-type yield
stress ratio ç higher than the dilatancy stress ratio criterion exhibiting isotropic and kinematic hard-
M dc . This implies that on plastic loading in the ening will constitute the ®rst stage in this section,
same direction (not reversed), causing a softening followed by the specialized equations of certain
response (see discussion after equations (3)), key aspects of the model in the next section.
additional dilation will occur for a looser than Tensor quantities are denoted by bold-face charac-
critical state. The ensuing change of ø towards a ters in a direct notation.
greater positive value will increase further the With reference to kinematics, the assumption of
value of M dc , and with ç decreasing owing to additive decomposition of strain rates is
softening, one will eventually have ç , M dc , thus,
consolidation will follow the initial dilation leading å_ ˆ å_ e ‡ å_ p (6)
to e ˆ ec and critical failure. It seems improbable, where å is the strain tensor, the superscripts e and p
though, that even a small amount of dilation can denote elastic and plastic components respectively
precede the consolidation for a state looser than and a dot over a symbol indicates its rate of change
critical. Hence, the restriction of zero dilatancy with time.
will be imposed whenever both conditions ø . 0 If e and s are the deviatoric strain and stress
and ç . M dc hold true. In this way the previous tensors respectively, and åv and p are the volu-
hypothetical case will result in no dilation at the metric strain and effective hydrostatic stress respec-
beginning when both previous conditions are tively, the isotropic hypoelasticity assumption is
satis®ed. It will then be followed by consolidation adopted, giving
towards ec once ç , M dc because of softening,
which will result in removing the zero dilatancy s_ p_
e_ e ˆ and å_ ev ˆ (7)
restriction. This restriction may be considered a 2G K
`loose end' of the model and it is a direct in terms of the elastic shear modulus G and bulk
consequence of the possibility to have neutral or modulus K. In a hypoelastic description one does
elastic loading at constant ç . M c , which can not account for the existence of a potential, and G
change the sign of ø, owing to the wedge-type and K are expressions of convenience. In particular,
yield surface. However, it is found to create no it will be assumed here that
problem of discontinuity or other analytical, phys-  a  a
p p
ical or numerical malfunction. Furthermore it will G ˆ G0 and K ˆ K 0 (8)
be seen that the two relevant conditions, ø . 0 and pat pat
ç . M dc are easily generalized to a multiaxial stress
space. where pat is the atmospheric pressure used as a
Along the same lines as equations (3) and (4), reference pressure, for which G ˆ G0 and K ˆ K0,
the critical back-stress ratios ácc and áce for ø ˆ 0 and a is a properly de®ned exponent yielding the
are de®ned according to variation of G and K with p. Usually one has a ˆ 12.
As p ! 0 it is possible to introduce a cut-off value
M c ˆ ácc ‡ m (5a) pl , used instead of p for p , pl in K but not in G,
Me ˆ áce ‡m (5b) since at zero effective pressure one can still have
K . 0 but G ˆ 0.
Equations (3)±(5) interrelate the bounding (peak), The yield surface will be a direct generalization
dilatancy and critical back-stress ratios by way of the wedge-type shape in the triaxial space,
of ø, by elimination of m. For example, using which will appear as a circular cone in stress space
equations (3a) and (5a), one has ábc ˆ ácc ‡ k bc with its apex at the origin, given by the equation
260 MANZARI AND DAFALIAS
p
f ˆ [(s ÿ pá) : (s ÿ pá)]1=2 ÿ (2=3)mp ˆ 0 expressions for L in equation (12) is obtained by
(9) straightforward algebraic manipulations of the pre-
vious equations, starting with the ®rst expression in
where the back-stress ratio deviatoric tensor á _ which is its de®nition. In particular, the
terms of ó,
determines the position of the axis of the cone and last expressions of L in terms of e_ and å_ v was
the stress ratio scalar variable m the `size' of the obtained using in addition, equations (6) and (7) by
cone. The symbol : between two tensor quantities the procedure described by Dafalias (1986). The
denotes the trace of their product, that is, a : second expression for L shows that plastic loading
b ˆ tr (ab) and can be interpreted geometrically as depends entirely on the projection of the stress-ratio
an `inner' product. The variations of á and m rate on n, as expected.
portray kinematic and isotropic hardening respec- Equations (6), (7), (11) and (12) can be inverted
tively. Equation (9) can be expressed in terms of to yield
the deviatoric stress ratio r ˆ s= p only. In such a
stress-ratio space, for example, the ð-plane shown ó_ ˆ 2Ge_ ‡ K å_ v I ÿ hLi(2Gn ‡ KDI) (13)
in Fig. 3, equation (4) appears p as a Mises-type Equation (13) is the stress-rate expression in terms
circle
p centred at á with radius (2=3)m. The factor of the total strain rates e_ and å_ v . Recalling the last
(2=3) has been introduced for convenience of expression of L in terms of e_ and å_ v, equation (12),
interpretation in the standard triaxial q, p space, one can write equation (13) in the form ó_ ˆ E ep : å_
in which equation (9) becomes equation (2). The with E ep the elastoplastic tangent stiffness moduli,
scalar-valued á of equation (2) is related to the used in a numerical implementation.
tensor-valued á of equation (9) by á ˆ á1 ÿ á2, The last, but very important, quantity to be
where á1 and á2 ˆ á3 are the principal values of á speci®ed is the plastic modulus K p . With the two
in the triaxial space. The observation made in hardening variables á and m evolving according to
triaxial space about a possible cap closing the open
wedge also holds true for the open cone of equation _ ˆ hLim
m (14a)
(9). á_ ˆ hLiá (14b)
The normal to the yield surface, de®ning the
loading direction, is given by the consistency condition f_ ˆ 0 together with
equations (9), (12) and (14) implies
@f
Lˆ ˆ n ÿ 13 NI (10a)
@ó @f @f p
p Kp ˆ ÿ : áÿ m ˆ p(n : á ‡ (2=3)m)
rÿá @á @m
nˆp and N ˆá : n‡ (2=3)m
(2=3)m (15)
ˆn:r (10b)
Many models can ®t within the foregoing gen-
where I is the second-rank isotropic tensor and n eral framework. What will distinguish one model
is the unit deviatoric stress-ratio tensor, that is, n : from another is the speci®cation of three quan-
n ˆ 1. The plastic strain rate is given by tities: m for isotropic hardening, á for kinematic
hardening and D for the dilatancy response.
å_ p ˆ hLiR (11a)
R ˆ n ‡ 13 DI (11b)
where R is the `direction' of å_ p consisting of a SPECIALIZED FORMULATION
deviatoric part n and a volumetric part 13 DI with D From triaxial to multiaxial stress space
the dilatancy coef®cient. Equations (10) and (11) The objective of this section is to formulate the
indicate that while an associative ¯ow rule exists in specialized equations of the constitutive model in
the deviatoric stress-ratio space (observe the com- a general stress space, by combining the general
mon n in L and R), a non-associative rule exists in concepts and general formulation presented in the
stress space (observe N 6ˆ ÿD, in general). L is the previous two sections. Since certain quantities
loading index enclosed by the Macauley brackets k l related to the general concepts were presented in
to indicate loading=unloading=neutral loading, and a triaxial stress space, the ®rst objective here is to
is de®ned by specify a general rule by which the quantities can
be generalized to the multiaxial stress space. This
1 1 1 can be achieved using the modi®ed Lode angle è
Lˆ L : ó_ ˆ pn : r_ ˆ [n : s_ ÿ N p_ ]
Kp Kp Kp de®ned in terms of the effective stress ratio r as
2Gn : e_ ÿ KN å_ v p  3
ˆ (12) 3 3 S
K p ‡ 2G ÿ KDN cos 3è ˆ J and r ˆ r ÿ á (16a)
2
where K p is the plastic modulus. The sequence of J ˆ [1 tr r ]
2
2 1=2
and S ˆ [13 tr r ]
3 1=3
(16b)
TWO-SURFACE PLASTICITY MODEL FOR SANDS 261
Notice that s ˆ s ÿ pá can be used instead of r extension and other values of è in between by
 J of the second
since p is eliminated in the ratio S= introducing such è dependence in other parts of the
and third effective stress invariants. In Fig. 3 the è model.
de®ned this way is illustrated on the circle repre- For this purpose, the concepts introduced by
senting the yield surface, and it is totally different equations (3)±(5) can now be generalized from
from the è de®ned if r instead of r were used. triaxial to multiaxial stress space. In what follows,
Hence, in this model, the terms `triaxial com- instead of the M bc,e , M dc,e and M c,e (the subscripts
pression' and `extension' are interpreted in terms of c,e mean compression or extension), the ábc,e , ádc,e
r and not r. and ácc,e will equivalently be generalized to multi-
If Qc and Qe are the values of a quantity Q in axial stress space. Geometrically, the latter triplet
triaxial compression and extension respectively, an generalizes into three surfaces in a general devi-
interpolation by way of è for any other value atoric stress-ratio space, which is illustrated in the
denoted by Qè can be given by particular version of the ð-plane shown in Fig. 3.
The three surfaces are named the bounding,
Qè ˆ g(è, c)Qc and c ˆ Qe =Qc (17a) dilatancy and critical surfaces respectively, the ®rst
2c two shown by discontinuous lines (indicating their
g(è, c) ˆ (17b) change with ø) and the critical surface by a solid
(1 ‡ c) ÿ (1 ÿ c) cos 3è
line. Recall that they refer to back-stress ratios
For è ˆ 08 and è ˆ 608, equations (16) and (17) rather than stress ratios.
yield g(08, c) ˆ 1 and g(608, c) ˆ c, hence Q08 ˆ Their analytical description is achieved as
Qc and Q608 ˆ Qe respectively. follows. Consider a stress ratio r on the circular
The generalization of the yield surface from a yield surface with centre á (Fig. 3), de®ning the
wedge type in the triaxial space (equation (2)) to a effective stress ratio r ˆ r ÿ á, the corresponding
circular cone type in the multiaxial stress space deviatoric unit normal n along r ÿ á (the n is the
(equation (9)) was achieved without the interven- one introduced in equations (10) and (11)), and the
tion of the Lode angle è and associated third Lode angle è as shown. One can draw a straight
invariants of stress quantities. That is the reason line segment from the origin O along n (i.e. at the
why the cone is circular in the ð-plane (Fig. 3). same angle è), which intersects the three surfaces
This was done in order to avoid the introduction at three so-called `image' back-stress ratio points
of è in equation (9) which would have greatly ábè , ácè , ádè . Hence, each one of these image points
complicated the expression for L ˆ @ f =@ó (equa- is de®ned by
tions (10)) and consequently all related expressions p
áaè ˆ (2=3)áaè n (a ˆ b, c, d) (18)
in equations (12), (13) and (15) because of the
p
required derivatives with respect to è, and of è where the factor (2=3) has been added so that at
with respect to ó. Nevertheless, it is shown below triaxial compression and extension the scalar-valued
that the model can still describe differences of áa08 ˆ áac and áa608 ˆ áae with a ˆ b, c, d, that is,
values of many quantities between compression and one retrieves the back-stress ratio quantities intro-
duced in equations (3)±(5). Equations (18) are the
analytical expressions of the three surfaces in a
`polar' representation. Since n is de®ned given the
stress state, it remains to obtain the scalar-valued
áaè (notice the difference between the tensor-valued
áaè of the left-hand side and the scalar-valued áaè of
the right-hand side in equation (18)).
Referring to equations (3)±(5), one needs to
generalize the expressions for M ac,e (a ˆ b, d), M c,e
and k ac,e (a ˆ b, d) according to equations (17)
and then obtain the áaè with the help of the state
variable ø and m as was done for è ˆ 08 and
è ˆ 608 in equations (3)±(5). With the de®nitions

c ˆ M e =M c , cb ˆ k be =k bc and cd ˆ k de =k dc
(19)
one has
Fig. 3. Schematic illustration in the stress-ratio ð-
plane of the yield, bounding, critical and dilatancy ábè ˆ g(è, c)M c ‡ g(è, cb )k bc hÿøi ÿ m (20)
surfaces, and the `image' back-stress ratios ábè , ácè and
ádè related to á, n and the Lode angle è ádè ˆ g(è, c)M c ‡ g(è, cd )k dc ø ÿ m (21)
262 MANZARI AND DAFALIAS

ácè ˆ g(è, c)M c ÿ m (22) (11)), the equivalent deviatoric plastic strain rate
e_ p ˆ [(2=3)_ep : e_ p ]1=2 and equation (18) for ábè. It
These equations imply that, when ø ˆ 0 (critical
gives
void ratio), one has ábè ˆ ádè ˆ ácè for all è. For
è ˆ 08 and 608, equations (20), (21) and (22) yield p
á_ ˆ 32 h[(2=3)ábè e_ p ÿ e_ p á] (25)
equations (3), (4) and (5) respectively. Equations
(18)±(22) fully specify the bounding, dilatancy and
which is similar to the non-linear kinematic hard-
critical surfaces associated with the back-stress ratio
ening rule by Armstrong & Frederick (1966) with
á in stress-ratio space.
the added advantage of h depending on jb : nj.
The dependence of á_ on a `distance' between á
and its image ábè on a surface, which implies the
Isotropic hardening dependence of the plastic modulus K p on this
The elastic range is de®ned by m which distance, is the underlying concept of bounding
determines the `radius' of the circular yield surface surface plasticity theory (Dafalias, 1986), in par-
in the ð-plane. It is physically expected that m ticular the two-surface version of it slightly
increases as the void ratio decreases and vice modi®ed from its initial conception in Dafalias &
versa. Since it is considered that elastic volumetric Popov (1975) and Krieg (1975). A form of h
deformation does not in¯uence m, it is more which is used in subsequent applications is moti-
accurate to state that the rate of change of m is vated by earlier propositions (Dafalias, 1986), and
related to the plastic rate of change of void ratio reads
de®ned by e_ p ˆ ÿ(1 ‡ e0 )_åpv, å_ pv being the plastic
volumetric strain rate and e0 the initial void ratio jb : nj
h ˆ h0 (26)
(if the logarithmic volumetric strain measure is bref ÿ jb : nj
used, e0 ˆ e). Hence, recalling equations (11), one
can write where h0 is a positive constant and bref a reference
p distance . jb : nj. The last requirement
p can be
_ ˆ ÿcm e_ ˆ hLic m (1 ‡ e0 )D ˆ hLi m
m  (23)
satis®ed
p if one de®nes bref ˆ 2 (2=3)ábc or bref
where c m is a positive quantity and in general it is a ˆ (2=3)(ábè ‡ ábè‡ð ), the latter de®nition shown in
function of state variables (s, á, p, etc.). The last Fig. 3 as the `diameter' of the bounding surface
two members of equation (23) de®ne the term along n.
m ˆ cx m (1 ‡ e0 )D in equation (14a), which enters Equation (24) shows that when b ˆ 0, that is,
equation (15) for K p. In most applications, includ- the back-stress ratio has reached the bounding
ing the one to be presented here, a constant and surface and á ˆ ábè, one has á_ ˆ 0 and á remains
rather small value of m will suf®ce, in which case stationary momentarily, having attained its peak
m  0. A rational way to obtain the variation of m, value. Since the bounding surface continues to
that is, to de®ne c m , in order to re¯ect the physical contract for a dilatant soil according to equations
signi®cance of the loosest and densest state, is (18) and (20) as ÿø increases, the á will be
presented in Appendix 1. crossed by the contracting bounding surface and
be placed outside of it, contrary to the basic
premise of a bounding surface formulation which
Kinematic hardening bounds within it all stress quantities associated
The general equation (14b) for kinematic hard- with it. This departure from the classical bounding
ening will now be specialized. It will be assumed surface formulation is a direct consequence of
that á_ is directed towards the bounding image equations (20) and (24). It is a Gordian knot
stress ratio ábè , de®ned by equation (20), and that solution to the problem of having the stress
it depends on the `distance' b ˆ ábè ÿ á ˆ bìb , ratio ç crossing the peak or bounding stress ratio
where b is the (positive) norm of b and ìb the unit M bc during neutral loading as discussed after
deviatoric stress-ratio tensor along b (Fig. 3). equations (3). It will also provide the way to
Hence, one can write handle the subsequent softening response. Indeed,
as á is found outside the bounding surface, the
á_ ˆ hLih(ábè ÿ á) ˆ hLihb ˆ hLihbìb ˆ hLiá de®nition of ìb is such (from á to ábè ), that now á_
(24) will be directed inwards, opposite to n. Hence,
since jb : nj entering h in equation (26) remains
where h is a positive scalar-valued function of unchanged in sign, the á will start moving again
jb : nj ˆ bjìb : nj (the use of jb : nj instead of after its peak value following a path inwards
simply b is related to the form of dependence of according to equation (24) and converging with ábè
K p on b, as will be seen later). It is worth re- from outside, as the latter approaches the critical
writing equation (24) in terms of the deviatoric surface. But this inward motion of á is tantamount
plastic strain rate tensor e_ p ˆ hLin (recall equation to softening, since the change of the stress ratio r
TWO-SURFACE PLASTICITY MODEL FOR SANDS 263
follows the change of the back-stress ratio á, and conditions ø . 0 and ç . M dc are satis®ed, dis-
when á reaches its peak value and subsequently cussed for the triaxial space, can now be easily
`softens', so does r, as long as the isotropic hard- generalized. The ®rst condition ø . 0 remains the
ening is negligible. same. The second generalizes to ìd : n , 0, which
implies according to the foregoing, that á has
crossed the dilatancy surface and the direction of n
Dilatancy coef®cient is outwards so that D , 0. Hence, when both ø . 0
The dilatancy stress ratio M dc,e delineates the and ìd : n , 0, one sets D ˆ 0 until at least one of
ranges of ç for which contraction or dilation oc- the above inequalities is violated.
curs in the triaxial space. Following Rowe's stress-
dilatancy theory (Rowe, 1962) and its invariant
form suggested by Nova & Wood (1979), it is The plastic modulus
proposed to have the dilatancy coef®cient D It is now possible to have a speci®c expression
proportional to the difference M dc,e ÿ ç, that is, for the plastic modulus Kp obtained from equation
p (15) and the de®nition of m  , á and D from
D ˆ (2=3)A(M dc,e p ÿ ç) with A a positive quan-
tity. The factor (2=3) is introduced for conve- equations (23), (24) and (27) respectively. It gives
nience so that the dilation ratio in triaxial space p
p Kp ˆ p(hbìb : n ‡ (2=3)c m (1 ‡ e0 )Adìd : n)
is given simply by (3=2)D ˆ A(M dc,e ÿ ç). The
above proposition was for monotonic loading in (28)
triaxial compression or extension. It is not obvious
how one can generalize it to reverse loading, or in The quantity b : n ˆ bìb : n appears automatically
a multiaxial stress space. It is proposed to choose in the K p expression. Hence, its introduction in
M dc or M de according to where the change of ç equation (26) for h is justi®ed for the sake of
points (and not according to ç itself), and then use consistency. The dependence of K p on the stress
the difference M dc,e ÿ ç, such that the values of the ratio `distance' quantity hb, between conjugate
M dc,e and ç subtract if M dc,e and ç are both at points á and ábè, is a characteristic of two-surface
compression or both at extension, and add other- plasticity models. Recalling the second de®nition of
wise. This may seem complicated, but in fact the loading index L in terms of the stress-ratio rate
yields a direct and easy-to-understand general- r_ , one observes that if K p . 0 then one must have
ization in the multiaxial stress space, as shown in n : r_ . 0 for L . 0, that is, stress-ratio hardening,
Fig. 3. Recalling the de®nition of the `image' while if K p , 0 it follows that n : r_ , 0 for L . 0,
dilatancy back-stress ratio ádè (equation (21)), one that is, stress-ratio softening. Hardening or soften-
can de®ne the `distance' tensor d ˆ ádè ÿ á ˆ dìd, ing in terms of stress ratio is not the same as in
d being the norm of d and ìd a `unit' tensor along terms of the deviatoric stress; one may have ç_ , 0
d (Fig. 3) and write but q_ . 0 and vice versa. The sign of K p fully
speci®es the hardening or softening response, which
D ˆ A(ádè ÿ á) : n ˆ Ad : n ˆ Adìd : n (27) depends on the combined effect of ìb : n and
ìd : n in equation (28). If the isotropic hardening
with A a positive quantity. In a triaxial setting, say of the yield surface is neglected as very small (m
compression, one can usepequations (2), (4a), (18) is assumed ®xed), which implies c m ˆ 0, then
and its counterpart p á ˆ (2=3)án, topwrite equa- equation (28) yields the very simple expression
tion (27) as D ˆ p(2=3)A(ádc ÿ á) ˆ (2=3)A(M dc Kp ˆ phbìb : n. Initially, K p . 0 as á is within the
ÿ m ÿ (ç ÿ m) ˆ (2=3)A(M dc ÿ ç), QED. Observe bounding surface. When b ˆ 0, that is, á ˆ ábè
that the sign of D in equation (27) depends directly (Fig. 3), Kp ˆ 0 and this implies reaching the peak
on the sign of ìd : n. D will be always positive if á stress ratio. Immediately afterwards ìb : n , 0, in
is inside the dilatancy surface (contraction), because general, and softening begins with K p , 0, placing
then the de®nition of ìd guarantees that ìd : n . 0 á outside the contracting bounding surface until
for surfaces convex with respect to the origin. If á eventually á ! ábè ! ácè at critical failure, whereby
is outside the dilatancy surface, one may have again K p ˆ 0.
ìd : n _ 0 depending on both á and n. If n points
`outwards', and if ìd : n , 0, dilation occurs. That
is the typical case in monotonic triaxial loading The critical state line
when á has exceeded M dc,e . On the other hand, Fig. The critical state line (CSL) will be represented
3 illustrates a case where although á is outside the by a straight line of slope ë in e, ln p space
dilatancy surface, n is so de®ned as to yield (Scho®eld & Wroth, 1968). Hence
ìd : n . 0, inducing contraction. D varies continu- ec ˆ (ec )ref ÿ ë ln ( p= pref ) (29)
ously with ìd : n, and D ˆ 0 when either á ˆ ádè
(i.e. d ˆ 0) or ìd : n ˆ 0. requiring two constants, ë and (ec )ref at a reference
The restriction of zero dilatancy when the pressure pref . Usually, but not necessarily, one
264 MANZARI AND DAFALIAS

chooses pref ˆ 1 in which case (ec )ref has been in a constant- p drained test and, with M c and ø
symbolized by Ã. Much has been discussed (e.g. known, calculate k bc . A problem is the elusiveness
Been et al., 1992) about the appropriateness of of the CSL and corresponding ec . Measurement of
equation (29) as well as the elusiveness of any the constants in equation (29) can be done only
equation for the CSL, mainly because it is experi- approximately, given the experimental impossibility
mentally dif®cult to reach it. The tendency, how- of reaching the critical state. The value of k bc is
ever, of a sand to reach a CSL is clearly exhibited obtained by examining the results of drained
(Been et al., 1991), and for stresses of engineering constant- p tests or undrained compression tests,
interest equation (29) is satisfactory. A two-slope preferably the latter for greater accuracy, in order
CSL was proposed in Crouch et al. (1994) as an to obtain the stress ratio M dc where dilation starts.
improvement when dealing with high values of p, The value of this stress ratio can be used in
and any other relation which achieves a better ®t to equation (4) to obtain k dc , given M c and ø. A
experimental data can be used instead of equation similar procedure applies to the determination of
(29) without affecting the operation of the present k be and k de .
model. Values of A and h0 can be obtained by a trial-
and-error procedure, although A can be directly
determined if proper stress-dilatancy data are
available, for example the results of volumetric
MODEL CALIBRATION AND PERFORMANCE strain as a function of axial strain in a constant- p
The calibration of the model constants can be drained compression test. Our experience is that a
done on the basis of triaxial compression and good ®t is usually attained within the ®rst three
extension tests. A list of model constants is given trials. The constant h0 may be obtained by
in Table 1. The constants which are independent of matching the model predictions to the ¯attening
the particular constitutive model used are called slope of the q, å1 curve in a drained compression
the `general purpose constants'. These are the test. Finally, a constant m (no isotropic hardening)
elastic constants G0 , K 0 and b and the critical is usually set equal to about 0´05M c .
state constants M c,e , (ec )ref and ë. The special To demonstrate the performance of the model,
constants which are particular to the present model a series of simulations under different stress paths
are the coef®cients k bc,e , related to the peak or is presented below. Laboratory tests conducted on
bounding stress ratio, k dc,e related to the dilatancy Nevada sand by the Earth Technology Corporation
stress ratio, the dilatancy constant A, the h0 related in the course of the VELACS project (Arulmoli et
to the kinematic hardening and the initial values of al., 1992) were used for comparison. Methods of
m and the coef®cient c m related to the isotropic sample preparation, saturation techniques and other
hardening. If c m ˆ 0, only a constant m is details of testing have been reported by Arulmoli
required. The k bc,e and k dc,e can be called the `state et al. (1992). Before presenting the results of the
parameter constants' owing to their association model simulation and their comparison to the
with ø. In the following, the calibration of only laboratory tests, simulations of three hypothetical
the special constants will be discussed, assuming undrained compression tests on this sand at three
that the general purpose constants are known. different void ratios for a con®ning pressure of
There is always a ®nal ®ne-tuning process for all 160 kPa, for which ec ˆ 0:80, are presented. Fig. 4
constants. shows that these simulations clearly re¯ect the
In order to determine the coef®cient k bc one can inherent coherence of the model with the concept of
use equation (3a), measure M bc (peak stress ratio) critical void ratio, as was explained for undrained
loading in Fig. 1. The distinction of the response
between the states looser than critical (e ˆ 0:81,
Table 1. Model constants 0:85) and denser than critical (e ˆ 0:79), the former
showing softening, conforms with experimental
Elastic G0 3´14 3 104 evidence.
K0 3´14 3 104 As to the actual triaxial tests, ®rst, a series of
a 0´6
drained monotonic compression tests were con-
Critical state Mc 1´14 ducted under three different constant con®ning
ë 0´025 pressures of 40, 80 and 160 kPa. The sand samples
(ec )ref 0´80 were prepared with two different initial void ratios
State parameter k bc 3´975 of 0´65 and 0´72. Following the procedure outlined
k dc 4´20 above, the model constants were determined using
Hardening h0 1200 these tests and are reported in Table 1. All con-
m 0´05 stants are dimensionless except G0 and K 0, which
are measured in kPa. The pref ˆ 160 kPa corre-
Dilatancy A 2´64
sponds to (ec )ref ˆ 0:80, according to equation
TWO-SURFACE PLASTICITY MODEL FOR SANDS 265

Fig. 4. Model simulation of three undrained triaxial tests starting at three different
void ratios, two at a state looser and one denser than critical (critical ec ˆ 0:80 for
p ˆ 160 kPa)

(29). Figs 5 and 6 show the comparison between same model constants were used to simulate an
the model simulations and the data. Good agree- undrained compression test conducted on a sample
ment with the experimental results is achieved at of the same Nevada sand with a void ratio of 0´72.
both densities and for different con®ning pressures, The model simulations, along with the experi-
including an accurate description of the softening mental results, are shown in Fig. 7. Very good
response in the q, å1 plots. The data show slightly agreement between model and data is achieved.
greater dilatancy for higher p at ein ˆ 0:72, Two simulations of triaxial cyclic tests conducted
contrary to what is expected for a less dilatant on samples with e ˆ 0:65 and 0´72 are shown in
sample. The model exhibits the correct trend. The q, p space, along with the experimental results, in
266 MANZARI AND DAFALIAS

Fig. 5. Model simulation of three different constant-p tests, data for Nevada sand
(Arulmoli et al., 1992) starting at an initial void ratio ein ˆ 0:65

Figs 8 and 9. Less good agreement is achieved. emphasizes this fact by including the results of
The parameters related to extension, M e , k be and two additional simulations for the case with p ˆ
k de , were given the values 1´14, 2´0 and 0´07 160 kPa shown in Fig. 5, for which k dc ˆ 4:20
respectively, in order to achieve the best ®t to (Table 1). In one of the new simulations, k dc was set
these cyclic triaxial tests. If reliable extension tests equal to zero, rendering the dilatancy stress ratio
were available, similar procedures used for the M dc equal to the critical stress ratio M c ˆ 1:14. In
compression state parameter constants k bc and k dc the other simulation, M dc was kept constant at its
could be used to determine k be and k de more ac- initial value, which according to equation (4a) is
curately. M dc ˆ 1:14 ‡ 4:2(0:65 ÿ 0:80) ˆ 0:51 (recall that
Proper calibration of parameters k dc and k de is ec ˆ 0:80 for p ˆ 160 kPa). Model performance is
important in the performance of the model. Fig. 10 seen to deteriorate in the new simulations, particu-
TWO-SURFACE PLASTICITY MODEL FOR SANDS 267

Fig. 6. Model simulation of three different constant-p tests, data for Nevada sand
(Arulmoli et al., 1992) starting at an initial void ratio ein ˆ 0:72

larly in reference to volumetric response. The CONCLUSION


simulation with M dc ˆ 0:51 is better than the one The two-surface plasticity formulations in devi-
with M dc ˆ M c ˆ 1:14, but for larger strains atoric stress-ratio space is combined with the state
(occurring within shear bands, for example), parameter ø arising within the CSSM framework
M dc ˆ 0:51 will induce unrealistic unlimited dila- to yield a comprehensive multiaxial constitutive
tion as ç ! M c , defying the critical state concept. model for sands, capable of simulating the material
It must be mentioned, though, that the new response under monotonic and cyclic, drained and
simulations were performed by keeping all other undrained loading conditions. The link between
parameters unchanged. It is possible that a change the two-surface (yield and bounding) formulation
of A, for example, might improve the results and ø is based on a modi®cation of a recent
obtained in these simulations. proposition by Wood et al. (1994) relating the
268 MANZARI AND DAFALIAS

Fig. 7. Model simulation of an undrained triaxial compression test, data for Nevada
sand (Arulmoli et al., 1992) at e ˆ 0:72

peak stress ratio to the critical stress ratio by way critical states at different densities and con®ning
of ø, and on the identi®cation of a proposed peak pressures, using the same model constants for a
stress-ratio surface as the bounding surface. A given sand.
similar proposition was made here, relating the The generalization to a multiaxial stress space is
dilatancy stress ratio to the critical stress ratio by obtained by introducing in the deviatoric stress-
way of ø, such that at ø ˆ 0 both peak and ratio space three surfaces, the bounding, dilatancy
dilatancy stress ratios becomes critical. These and critical surfaces in direct correspondence to
simple suggestions allow the model to simulate the bounding (peak), dilatancy and critical stress
both hardening and softening characteristics, as ratios in triaxial space. The bounding surface con-
well as the volumetric response or pore water trols the kinematic hardening of the yield surface,
pressure development of denser- and looser-than- which appears as a simple Mises-type circle in
TWO-SURFACE PLASTICITY MODEL FOR SANDS 269

Fig. 8. Model simulation of an undrained cyclic triaxial test, data plotted in q, p


space for Nevada sand (Arulmoli et al., 1992) at e ˆ 0:65

deviatoric stress-ratio space, and as a moving models (Prevost, 1985), the present model can
circular cone in general space. The simplicity of achieve the same results with much less numerical
the yield surface results in simple general con- effort, and it includes additional characteristics,
stitutive equations of the standard type for such as softening and zero dilatancy at critical
numerical implementation. state. Another feature of the model is that its
The numerical ef®ciency of the model is good current state entirely determines the next step,
because only the yield surface must be updated without necessitating the memory of entities such
owing to kinematic and isotropic hardening. The as past stress reversal points and accumulated
other surfaces are fully determined by the value of strain measures. The model is totally void-ratio
the state parameter ø. Compared with multisurface determined, given the CSL in e, ln p space. Its
270 MANZARI AND DAFALIAS

Fig. 9. Model simulation of an undrained cyclic triaxial test, data plotted in q, p


space for Nevada sand (Arulmoli et al., 1992) at e ˆ 0:72

initialization is straightforward, accounting for the data which are well established and understood
fact that the initial stress ratio can determine the by the geotechnical engineering community, with
initial back-stress ratio, being almost identical to it basic reference to CSSM. For example, the deter-
if the elastic range is small. The number of model mination of the CSL in e, ln p space or the peak
constants is eight (without the elastic moduli) for and residual (critical) stress ratios (important
triaxial compression, increased to ten for a general quantities for the model) can be made by even a
case since two of them require their values in prospective user who is not familiar with plasticity
extension as well. Calibration is straightforward. constitutive modelling techniques, but who can
Possibly the most attractive feature of the model interpret the results of relevant experiments for the
is its simplicity and foundation on concepts and purpose of model calibration.
TWO-SURFACE PLASTICITY MODEL FOR SANDS 271

Fig. 10. Comparison of three model simulations using kdc ˆ 4:2 (variable M dc ), kdc ˆ 0
(M dc ˆ M c ) and a ®xed M dc ,M c with the experimental results from a constant-p,
drained compression test on Nevada sand with ein ˆ 0:65

APPENDIX 1 ç . M dc,e ˆ mmax , dilation from the densest state can


It is assumed that m varies from its minimum value occur. It will be assumed that the variation of m according
mmin (almost zero) at the loosest possible state de®ned by to the foregoing can be described by
the maximum void ratio emax to its maximum value mmax
at the densest state de®ned by the minimum void ratio  
e ÿ emin n
emin . The value of mmax cannot exceed the dilatancy m ˆ mmax ÿ (mmax ÿ mmin ) (A1)
stress ratio at emin given by equations (4), so that when emax ÿ emin
272 MANZARI AND DAFALIAS

Taking the rate of change of both members of equation Jefferies, M. B. (1993). Nor-sand: a simple critical state
(A1) and approximating å_  å_ p , one has model for sand. GeÂotechnique 43, No. 1, 91±103.
  Krieg, R. D. (1975). A practical two-surface plasticity
mmax ÿ m p
m_ ˆ ÿn e_ (A2) theory. J. Appl. Mech., ASME 97, 641.
e ÿ emin Lacy, S. J. & Prevost, J. H. (1987). Constitutive model
for geomaterials. Proc. 2nd Int. Conf. on Constitutive
which in conjunction with equation (23) determines cm ˆ Laws for Engineering Materials, Tucson, pp. 1±12.
n(mmax ÿ m)=(e ÿ emin ). Manzari, M. T. (1994). Finite deformation analysis and
Normally a constant m will suf®ce. This can be ob- constitutive modeling of non-cohesive soils for lique-
tained by setting n ˆ 0. Alternatively, one could set n to faction problems. Dissertation submitted in partial
be a very small positive number, which, on the basis of satisfaction of the Degree of Doctor of Philosophy to
equation (A1), would keep m  mmin until e ˆ emin and the University of California, Davis.
then m ˆ mmax . The proposed isotropic hardening is Mroz, Z., Norris, V. A. & Zienkiewicz, O. C. (1978). An
totally determined by the current void ratio, and does anisotropic hardening model for soils and its applica-
not depend on accumulated deviatoric plastic strain or tion to cyclic loading. Int. J. Numer. Anal. Methods
plastic work of past history, which are unknown to a Geom. 2, 203±221.
current user who wants to initiate calculations. Nova, R. & Wood, D. M. (1979). A constitutive model
for sands in triaxial compression. Int. J. Numer. and
Anal. Methods Geom. 3, 255±278.
Poorooshasb, H. B. & Pietruszczak, S. T. (1985). On the
REFERENCES yielding and ¯ow of sand, a generalized two surface
Armstrong, P. J. & Frederick, C. O. (1966). A mathe- model. Comput. Geotech. 1, 33±58.
matical representation of the multiaxial Bauschinger Prevost, J. H. (1978). Plasticity theory for soil stress
effect. CEGB, Report RD=B=N731. strain behavior. J. Engng Mech., ASCE 104, No. 5,
Arulmoli, K., Muraleetharan, K. K., Hossain, M. M. & 1177±1194.
Fruth, L. S. (1992). VELACS: veri®cation of liquefac- Prevost, J. H. (1985). A simple plasticity model for
tion analysis by centrifuge studiesÐlaboratory testing frictional cohesionless soils. J. Soil Dynamics Earth-
program, soil data report. Earth Technology Corpora- quake Engng 4, No. 1, 9±17.
tion. Also available from University of Southern Roscoe, K. H. & Burland, J. B. (1968). On the
California web site, http:rccgOl.usc.edu=eqdata=home. generalized stress±strain behavior of wet clays. In
html Engineering plasticity (eds J. Heyman & F. A.
Been, K. & Jefferies, M. G. (1985). A state parameter for Leckie), pp. 535±609. Cambridge University Press.
sands. GeÂotechnique 35, No. 2, 99±112. Roscoe, K. H., Scho®eld, A. N. & Wroth, C. P. (1958).
Been, K., Jefferies, M. G. & Hachey, J. E. (1991). The On the yielding of soils. GeÂotechnique 8, No. 1,
critical state of sands. GeÂotechnique 41, 365±381. 22±53.
Been, K., Jefferies, M. G. & Hachey, J. E. (1992). Roscoe, K. H., Scho®eld, A. N. & Thurairajah, A. (1963).
Discussion on `The critical state of sands'. GeÂotech- Yielding of clays in states wetter than critical. GeÂo-
nique 42, 655±663. technique 13, No. 3, 211±240.
Crouch, R. S., Wolf, J. P. & Dafalias, Y. F. (1994). Rowe, P. W. (1962). The stress±dilatancy relation for
Uni®ed critical-state bounding surface plasticity static equilibrium of an assembly of particles in
model for soil. J. Eng. Mech., ASCE 120, No. 11, contact. Proc. Roy. Soc., Series A 269, 500±527.
2251±2270. Scho®eld, A. N. & Wroth, C. P. (1968). Critical state soil
Crouch, R. S. & Wolf, J. P. (1994). Uni®ed 3-D critical- mechanics. McGraw-Hill.
state bounding-surface plasticity model for soil incor- Tatsuoka, F. & Ishihara, K. (1974). Yielding of sand in
porating continuous plastic loading under cyclic triaxial compression. Soils Fdns 14, No. 2, 63±76.
paths. Int. J. Numer. Anal. Methods Geom. 18, No. Vermeer, P. A. (1978). A double hardening model for
11, 735. sand. GeÂotechnique 28, No. 4, 413±433.
Dafalias, Y. F. (1986). Bounding surface plasticity, I: Wang, Z. L., Dafalias, Y. F. & Shen, C. K. (1990).
mathematical foundation and hypoplasticity. J. Eng. Bounding surface hypoplasticity model for sand. J.
Mech., ASCE 112, No. 9, 966±987. Eng. Mech., ASCE 116, No. 5, 983±1001.
Dafalias, Y. F. & Herrmann, L. R. (1980). A bounding Wood, D. M. (1990). Soil behavior and critical state soil
surface soil plasticity model. Int. Symp. on Soils mechanics. Cambridge: Cambridge University Press.
under Cyclic and Transient Loading, Swansea 1, Wood, D. M., Belkheir, K. & Liu, D. F. (1994). Strain
335±345. softening and state parameters for sand modelling.
Dafalias, Y. F. & Herrmann, L. R. (1986). Bounding GeÂotechnique 44, No. 2, 335±339.
surface plasticity, II: application to isotropic cohesive Yang, B. L., Dafalias, Y. F. & Herrmann, L. R. (1985). A
soils. J. Eng. Mech., ASCE 112, No. 12, 1263±1291. bounding surface model for concrete. J. Engng Mech.,
Dafalias, Y. F. & Popov, E. P. (1975). A model of ASCE 111, No. 3, 359±380.
nonlinearly hardening materials for complex loadings. Zeng, X. & Arulanandan, K. (1993). Overview of
Acta Mechanica 21, 173±192. calibration of constitutive models and soil parameters.
Ishihara, K., Tatsuoka, F. & Yasuda, S. (1975). Undrained In Veri®cation of numerical procedures for the
deformation and liquefaction of sand under cyclic analysis of soil liquefaction problems, Vol. 2 (eds
stresses. Soils Fdns 15, No. 1, 29±44. Arulanandan and Scott). Rotterdam: Balkema.

Вам также может понравиться