You are on page 1of 28

MECHANICS OF COHESIVE-FRICTIONAL MATERIALS, VOL.

2, 93–120 (1997)

Experimental analysis and theoretical interpretation of triaxial load


controlled loose sand specimen collapses

Claudio di Prisco1 and Silvia Imposimato1


1
Dipartimento di Ingegneria Strutturale, Politecnico di Milano, Piazza Leonardo da Vinci 32,
I-20133 Milano, Italy

SUMMARY
A series of triaxial load-controlled tests is performed.
Finite load increments are imposed. The single load-steps are followed by a time period during which the axial
load is kept constant.
At low stress levels the mechanical response is stable and characterized by a continuous decrease in strain rate
with time. At higher stress levels, the mechanical response changes and, subsequently, the collapse takes place.
The collapse is unexpected and occurs at a stress level less than that associated with the steady state,
experimentally observed by performing strain triaxial controlled tests.
In order to interpret such a behaviour, a theoretical discussion is introduced. This is based on a dynamical
reinterpretation of the micromechanical fabric rearrangement of granular assembly.

#
In particular, the role played by the kinetic energy of the system, as well as that played by the anisotropy of the
microstructure, have been analysed. 1997 by John Wiley & Sons, Ltd. Mech. cohesive-frictional mater. 2,
93–120 (1997)

KEY WORDS: loose sands; load-controlled triaxial tests; creep tests; instability; theoretical interpretation; strain-rate sensitivity

1. INTRODUCTION
The present paper is an attempt to experimentally describe and theoretically justify the mechanical
behaviour of loose sand in drained, load-controlled, standard triaxial tests. When this kind of
laboratory test is performed, sudden collapses take place. These specimen collapses, for the first time
shown experimentally by Begemann, Koning and Linderberg,1 occur at low stress levels, which are
much lower than those defining the steady state which may be reached by performing drained strain-
controlled tests.
In the first part of the paper some experimental results will be illustrated. The experimental tests
were performed by imposing finite load increments and by recording the strain versus time trend
during the time periods following the single load steps. The collapses are sudden and not anticipated
by a continuous hardening of the previous strains. They take place both when saturated and dry loose
sand specimens are tested.
In order to allow full water drainage, the experimental tests on saturated sand specimens were
performed slowly. The sudden increase in pore pressure was recorded not before, but during the
specimen collapses; this means that the presence of water is not necessary for instability to occur.
Similar experimental observations were presented by Eckersley,2 though by studying model lique-
fying slopes and not triaxial specimens.

#
CCC 1082–5010/97/020093–28 $17.50 Received 4 March 1996
1997 by John Wiley & Sons, Ltd. Revised 23 July 1996
94 C. DI PRISCO AND S. IMPOSIMATO

Recently, similar experimental results were published by Sasitharan et al.,3,4 Skopek et al.5
In the first papers, the authors interpreted the collapses as the consequence of an instantaneous and
undrained mechanical response of the material. Instead, in the second one, because the sand speci-
mens are dry, the authors are not able to theoretically justify the phenomenon experimentally
observed.
If a drained standard triaxial strain-controlled test is performed, the same material, up to the steady
state, manifests a stable, continuously hardening, mechanical behaviour. Consequently, it seems
evident that the same loose sand specimen may show two different mechanical responses, according
to the type of test performed: strain or load-controlled.
In the second part of this paper, a theoretical interpretation is proposed, which analyses the
unstable phenomenon from a microdynamic point of view. In order to investigate the dynamic aspect
of this unstable phenomenon, it is necessary to take into account the effects of the time factor on the
mechanical response of the granular assemblies. The concept of time dependency on the mechanical
behaviour of the material has already been experimentally demonstrated and theoretically considered
in di Prisco and Imposimato.6 In that paper, the authors presented the experimental results obtained
by performing drained standard triaxial tests, characterized by finite load increments and by different
time periods between two successive load steps.
As regards low stress levels, the authors showed that the asymptotical trend is not reached
immediately and that during a finite time period the strain rate is not nil, even though the load is kept
constant. Experimental evidence suggests that the microstructural evolution, in a single load incre-
ment, might be interpreted as a microdynamic, statistically determined phenomenon, which passing
through a transient condition, reaches a steady condition.
According to the authors, the collapses are caused by an unstable transient condition.
In the following section, the phenomenon considered and its theoretical interpretation will be
proposed as a useful tool in highlighting the static liquefaction phenomenon. The increase in pore
pressure will be considered as a consequence of a previous collapse, and not as the triggering cause.
By accepting this mechanical explanation, in boundary value problems, the pore pressure wave
becomes only a vehicle of instability, by making the instability propagation more rapid and prob-
able.7–9

2. SOME REMARKS ON STATIC LIQUEFACTION


2.1. Traditional laboratory test results
Liquefaction is an important phenomenon, causing dramatic effects. The phenomenon of sand
changing its behaviour from solid to liquid was recognized in the early stage of soil mechanics
development. The term spontaneous liquefaction was coined by Terzaghi10 to indicate the sudden
change of loose deposits of sand into flows, much like those of viscous fluid, triggered by a slight
disturbance.
Further studies on the liquefaction phenomenon taking place under static loadings11–15 concluded
that the material must be loose for liquefaction to occur, i.e., contractive during shearing at large
strain, and assumed that the loadings must not allow the water to flow prior to the collapse.
In order to clarify the liquefaction phenomenon, both static and cyclic, undrained, strain-con-
trolled, triaxial loading tests, on loose granular specimens, were performed by Konrad,16 Ishihara,17
and di Prisco, Matiotti and Nova.18 Over more than 20 years, a large number of monotonic undrained
test results on loose sand specimens have shown that the effective stress paths are characterized by a
peak in the effective triaxial plane, and by a peak of the deviatoric-axial strain curve, see Figure 1
(from Castro11).
LOOSE SAND SPECIMEN COLLAPSES 95

Figure 1. Standard undrained triaxial compression on a very loose sand specimen (from Castro11): (a) effective stress path; (b)
stress–strain behaviour

On the other hand, the undrained triaxial cyclic tests on loose saturated sand specimens17 are
characterized by a continuous increase in pore pressure, by an associated increase in stress level, and
lastly, if the test is stress-controlled, by a sudden collapse. All these test results are traditionally used
to clarify the liquefaction phenomenon, because undrained conditions are commonly assumed to be
physically realistic, even if boundary value problems and a material like sand are considered.
In order to justify the link between undrained laboratory experimental results and the physical
reality of the in situ conditions, the triggering collapse disturbances and the associated mechanical
response of the material must be assumed to be rapid.

2.2. Comments on undrained conditions


By considering various case histories of spontaneous liquefaction, it appears difficult to recognize
rapid triggering causes. Liquefaction commonly concerns local steepenings owing to erosion pro-
cesses and to seepage pressures during falling tide or rapidly accumulating sediments.19
Moreover, also the rapidity of the mechanical soil response needs to be discussed. In fact, the
authors have shown experimentally that the irreversible strains are delayed. With reference to loose
sand triaxial specimens, the time period required to reach the final strain is a few minutes. This effect
is irrespective of the presence of water; in fact, even if dry specimens are considered, the delay
phenomenon occurs.
In order to closely examine this particular aspect, a triaxial compression test on a loose iso-
tropically consolidated saturated sand specimen was performed. At the beginning, after the first
drained load controlled triaxial phase, the drainage valve was closed (point A in Figure 2(a)) and a
load increment was imposed.
In Figure 2(b) the trend of the pore pressure versus time is shown. A time dependency of the
increase in pore pressure is evident. This trend is also shown in Figure 2(a): point B0 corresponds to
the pore pressure value one minute after the load increment; point C0 to that recorded half an hour
later and D0 one hour later.
In undrained conditions, even though the load increment is applied almost immediately, the final
increase in pore pressure is reached only after a considerable time lag. Consequently, when the
drainage valve is open, the complete undrained increase in pore pressure (point D0 ) will never be
96 C. DI PRISCO AND S. IMPOSIMATO

Figure 2. Undrained axial load increment after a drained standard triaxial compression phase on a very loose Hostun sand
specimen: (a) total and effective stress path; (b) pore pressure versus time curve

reached: the value of the pore pressure increase will depend on the material permeability coefficient
and the strain rate. The main factor, in reality, is the ratio between these two variables.
With reference to particular boundary conditions and to particular hydromechanical characteristics
of the material, in time the pore pressure values may induce dangerous collapses. Nevertheless,
according to the authors, in some other cases, the instabilities may be caused by a completely
different mechanical mechanism, in which the pore pressure (i.e., the water presence) does not play
the role of the triggering factor.
LOOSE SAND SPECIMEN COLLAPSES 97

The aim of this paper is, precisely, to introduce this new interpretation, by analysing the experi-
mental results briefly described in the following paragraph. The doubts about the mechanism, which
in situ triggers of liquefaction phenomena, as already clearly expressed by Eckersley,2 derives from
laboratory experimental observations. By analysing the collapses involving the laboratory model
slopes he monitored, Eckersley concluded that the flow slides can initiate under essentially static,
drained conditions. The liquefaction occurs subsequently to failure initiation with excess pore
pressures being generated in relatively thin shear zones.
Moreover, the collapses described below by the authors, although they concern drained tests, are
qualitatively very similar to those observed when undrained conditions are imposed. The collapses
are unexpected and may occur very rapidly. When the specimens are saturated, it is possible to
observe a sudden increase in pore pressure following the specimen failure.
The material, from a macroscopic point of view, is always uniformly deformed, even after the
collapse takes place. Consequently, according to the authors, the experimental results described
below, together with those recently presented by other researchers,1,4,6,20–23 lead directly to a re-
examination of the liquefaction problem.

3. EXPERIMENTAL OBSERVATIONS
3.1. Experimental programme and procedure
A series of triaxial load controlled tests was performed. All the tests were carried out on specimens
(140 mm high, 70 mm wide) of loose Hostun RF sand (Dr  20%). A more detailed laboratory
device and test procedure description may be found in di Prisco and Imposimato.6
Two different types of triaxial tests were carried out, and are drawn in Figure 3. The first is a
drained standard triaxial compression test, obtained by increasing only the axial load and by keeping
the cell pressure constant. The second is characterized initially by a standard drained triaxial com-
pression phase, followed by a q constant effective stress path. This second phase is obtained by

Figure 3. Drained load controlled tests: standard compression and q constant stress paths.
98 C. DI PRISCO AND S. IMPOSIMATO

keeping the axial load constant and either increasing pore pressure when the specimen is saturated, or
by directly decreasing the cell pressure, when the specimen is dry.
All the stress paths were obtained by imposing finite incremental stress steps (Dq or Dp0 ). The time
elapses between two consecutive stress increments and the magnitude of stress increments were set as
variable: every test has a particular time history.

3.2. Strain controlled and load controlled tests


In order to underline the difference between the mechanical response obtained by imposing strain-
control and load-control in drained standard triaxial tests, it is interesting to compare the experimental
results of two different 100 kPa consolidated saturated specimens.
The strain-controlled test is performed by imposing a constant axial strain rate ( ˆ 1 mm=min);
whereas the load-controlled test is carried out by imposing finite axial load-increments. The axial
load time history is illustrated in Figure 4.
In Figure 5 the experimental results are shown. The strain-controlled test is characterized by a
continuous hardening regime up to the steady state (f0 ˆ 32 ). On the contrary, the load-controlled
test at the stress level characterized by a mobilized friction angle of 25 suffers a sudden collapse.
Up to this point, the mechanical responses of both tests are roughly the same, but the sudden
instability is peculiar only to the load-controlled test.
In Figure 6, the collapse points of the various load-controlled tests24 are collected, and compared
with the ultimate state points obtained by means of strain-controlled loadings. In Figure 6, not only
the standard triaxial test, but also the q constant test results are collected.
In order to explain the large dispersion of the points corresponding to the sudden collapses, it is
possible to draw the same locus in a Dr–q–p0 space or more simply in a f0m –e plane (Figure 7). But, as
will be clarified below, the unstable phenomenon considered cannot be completely explained by only
taking into account the relative density and the effective stress state: the problem is more complex.
The phenomenon is caused by a structural collapse and is a result of progressive destabilization of
the grain structure. In order to explain the difference in the material’s mechanical behaviour, obtained

Figure 4. Drained standard triaxial tests [s0c ˆ 100 kPa (T100d)]: axial load–time history
LOOSE SAND SPECIMEN COLLAPSES 99

Figure 5. Comparison between two drained standard triaxial compression tests performed by controlling the axial strain or the
axial load: [s0c ˆ 100kPa] (T100d): (a) deviatoric stress versus axial–strain curves; (b) volumetric behaviour

by performing strain and load-controlled tests, it is necessary to outline the main aspects of the two
test procedures.
When a strain-controlled test at constant strain rate is carried out, the system is kinematically
controlled: no global axial acceleration is allowed. On the contrary, when a load-controlled (i.e.
creep) test is performed, the system is free to accelerate and evolve without any kinematical con-
straint.
In order to highlight the causes of instability, in the following section the load controlled test will
be described more accurately.
100 C. DI PRISCO AND S. IMPOSIMATO

Figure 6. Comparison between collapse points in load-controlled tests and steady-state locus

3.3. Load-controlled tests: stable and unstable mechanical responses


The load-controlled tests were carried out by imposing finite load increments. The load increments
are followed by time periods at constant stress level.
Among the 20 tests performed, the test T100a will be analysed. The relative load time history is
illustrated in Figure 8(a), and in Figure 8(b) the corresponding mechanical response is drawn. In order
to highlight the nature of the phenomenon, it may be interesting to compare the material mechanical
response to the load increments at different stress levels.

Figure 7. Mobilized friction angles versus void ratio values relative to collapse points
LOOSE SAND SPECIMEN COLLAPSES 101

Figure 8. (Test T100a), standard triaxial test, [s0c ˆ 100 kPa]: (a) load–time history; (b) stress–strain behaviour

In Figure 9(a) the first seven increments are considered. The axial strain versus time curves show
stable and clearly time dependent mechanical behaviour.6 The static disturbance induces a delayed
deformative response, which has been defined as delayed plasticity. The transient regime takes many
minutes, and the internal structure continues to evolve subsequently, too. The micro-structural
rearrangement, when it takes place, is not immediate but delayed.
When the mobilized friction angle of 22.3 is reached, the type of response illustrated in Figure
9(b) is observed for the first time.
102 C. DI PRISCO AND S. IMPOSIMATO

Figure 9. (Test T100a), standard triaxial test, [s0c ˆ 100 kPa]: (a) dq ˆ 2 kPa, stable mechanical behaviour; (b) dq ˆ 2 kPa,
unstable mechanical behaviour; (c) dq ˆ 2 kPa, collapse increment

Initially, the axial strain rate increases, even if the effective axial stress remains constant. However,
after a while, the asymptotical value of strain is reached. At a stress level, characterized by a
mobilized friction angle of 25 , the collapse takes place (Figure 9(c)).
Each experimental load controlled test, both standard triaxial and q constant, is characterized by
the same mechanical trend (Figure 10). At a low stress level, the axial strain rate continuously
decreases (curve a). By increasing the stress level, the system initially accelerates, but subsequently a
strain asymptotical stabilization takes place (curve b), and lastly collapse occurs (curve c).
LOOSE SAND SPECIMEN COLLAPSES 103

Figure 9. (continued )

By assuming the traditional mechanical definition of stability,25 it is possible to define the material
mechanical behaviour, described by curve c, as unstable. In fact, the continuity of the material
response is lost. Small stress disturbances cause large strain increments.
If no instability occurred, the steady state would be reached and, by increasing the stress level,
strain increments would continuously increase. On the contrary, when such an instability occurs, a
gap between the previous load step and the unstable one (Figure 9(c)) may be observed.
If curve B of Figure 9(b) is considered, together with the previous and the subsequent load steps (A
and C, respectively (Figure 11)), it appears evident that curve C is more rigid than curves A and B.
This means that the incipient instability (seen in curve B) induces a more stable micro-structure and
consequently a more rigid macroresponse in the subsequent load steps. Moreover, it is important to
observe that the incipient material instability recorded is associated with compactive volumetric
behaviour (see section 4.5).
In order to demonstrate that the instability phenomenon previously shown is not linked to the
rapidity of the load test considered, many different load-controlled tests were performed. Among
these the collapse obtained by means of a q constant test performed very slowly will be illustrated.
Moreover, this experimental result clearly shows that the onset of instability is not due to a rapid
increase in pore pressure.

Figure 10. Schematical mechanical responses to a generic axial load increment


104 C. DI PRISCO AND S. IMPOSIMATO

Figure 11. Comparison between the unstable mechanical behaviour and the mechanical behaviour corresponding to the
previous and the successive load increments

The collapse takes place very slowly (Figure 12); acceleration is continuously positive, though
initially almost nil. The collapse occurs some hours after the load increment has been applied. The
pore pressure increases when the collapse has already taken place. Before that point, no increase in
pore pressure can be observed. Therefore the phenomenon may be assumed to be unrelated to the
presence of water.

3.4. Axial strain rate and axial strain acceleration analysis


In order to highlight the actual nature of the phenomenon, it would be very interesting to describe,
during the single time periods between two consecutive load increments, the internal fabric.

Figure 12. Collapse occurrence after a long time period


LOOSE SAND SPECIMEN COLLAPSES 105

Unfortunately, using ordinary geotechnical laboratory devices, this cannot be done. Consequently,
the following analyses will be made only by interpreting the external measures (the phenomenon
consequences) and not by considering the microstructural causes.
As previously observed, when a load-controlled or creep test is performed, the problem must be
dynamically analysed: the trend of the kinetical variables such as axial-strain rate and axial-strain
acceleration must be described.
In order to compare the stable and unstable material responses, the first seven load steps are shown
in Figure 13. While in Figure 14, the unstable response is illustrated together with the previous load
step (Figure 9(a)).
In Figure 13, it is evident that the load-increment causes a strain rate increase (points Ai). This is
followed by a continuous decrease up to the subsequent load increment.
The peak velocities, which are recorded a minute after increasing the axial load, will continue to
grow gradually. With reference to increasing stress levels and by keeping the time period elapsing
between two succeeding load increments constant, the material strain response becomes more rapid.
Nevertheless, the single material response to the load disturbance remains stable, in fact the loose
sand specimen mechanical behaviour is continuous and characterized by typical exponential decay.
In the same manner, if we draw the acceleration trend relative to the load steps of Figures 9(a) and
11, the curves of Figures 15 and 16 are obtained. In Figure 16 the unstable load step and the previous
one are compared.
From Figure 16, it appears evident that when instability takes place, the system accelerates over a
significant period of time.

3.5. Dry sand specimens: further experimental observations


Recently, by conducting a series of q-constant, load-controlled, triaxial tests, on dry loose Ottawa
sand specimens, Skopek et al.5 concluded that the structural collapse of very loose dry sand is a result
of progressive destabilization of the grain structure.
The same conclusions on dry loose sand specimens may be obtained, by analysing the experi-
mental results of Figure 17(c). The relative stress path is illustrated in Figure 17(a), while the p0

Figure 13. Axial strain-rate versus time: load increments corresponding to those illustrated in Figure 9(a)
106 C. DI PRISCO AND S. IMPOSIMATO

Figure 14. Axial strain-rate versus time: load increments corresponding to the unstable response and the previous one

versus time curve is shown in Figure 17(b). Similar results were obtained by performing load-
controlled, standard compression, triaxial tests on dry loose sand specimens. The collapse point of
Figure 17(a) corresponds with a mobilized friction angle of 23 . The axial strain versus time curve,
illustrated in Figure 17(c), is relative to the phase of the test characterized by a constant value of q
(from point B to C of Figure 17(a)). The points of the curve are relative to the instant of time
corresponding to the load increment application.
The particular strain versus time behaviour shows an initial rigid mechanical response. This is
followed by more considerable strain increments. This may be clarified by interpreting the material
mechanical behaviour in the light of elastoplasticity. In fact, initially the response may be assumed to
be elastic and, subsequently, to be elastoplastic.

Figure 15. Axial strain acceleration corresponding to the load increments of Figure 9(a)
LOOSE SAND SPECIMEN COLLAPSES 107

Figure 16. Axial strain-acceleration versus time: load increments corresponding to the unstable response and the previous one

The loose sand specimens designated as dry, are not actually completely dry, because of the
preparation method. In fact, to obtain the desired relative density, the traditional method introduced
by Bjerrum, Kringstad and Kummeneje26 was used. Before the specimen preparation, a small per-
centage of water (2% in weight) is added. Consequently, the specimens are not completely dry and
obviously, because of suction forces, this may cause a different mechanical response in comparison
with the saturated specimens.
By taking into account Figure 17, it is evident that the material behaves qualitatively as if it were
saturated. However, the stress level at which the collapses take place is quantitatively different. In
particular, if we compare the collapse points of the saturated specimens and the ‘dry’ ones, it is
possible to conclude that suction makes the microstructure more stable, and this effect is highlighted
by the increase in the mobilized friction angle at which sudden collapses occur.

4. THEORETICAL INTERPRETATION
The phenomenon shown experimentally in section 3 may be defined as delayed instability and,
consequently, may be interpreted as an unstable creep. Generally, delayed instability is classified as
tertiary creep and considered as peculiar to cohesive materials; in fact, they are studied within the
framework of fracture mechanics and are assumed to be the ultimate result of unstable fracture
propagation.
On the contrary, as regards granular assemblies, a theoretical framework, capable of justifying the
instability considered, does not exist. Therefore, in order to conceive constitutive models which are
able to reproduce such an unstable phenomenon, the problem of microstructure evolution from a
probabilistic point of view has been tackled.
The aim of the following paragraphs is to extend the applicability of the traditional elastovisco-
plastic theory to dynamic problems. In fact, according to the authors, when load-controlled tests are
performed, the microstructure evolution cannot be considered as quasi-static, but must be interpreted
dynamically. Within the framework of strain-hardening elastoplasticity, this implies the redefinition
108 C. DI PRISCO AND S. IMPOSIMATO

Figure 17. Load controlled triaxial q constant test on dry very loose Hostun sand specimen: (a) stress path; (b) mean pressure
time history; (c) axial strain versus time curve relative to the q constant phase

of the hardening rules which govern state variable evolution. In the following, a new constitutive
model is not presented, but only a logical path is outlined.
In this perspective, the theoretical attempt presented below lays no claim to being exhaustive, but
rather seeks to give rise to scientific debate.
The interpretation outlined in Sections 4.2–4.4 makes use principally of two variables to describe
system evolution: the configurational entropy Sc and the kinetic energy Ec. Sc and Ec are assumed to
determine, respectively, how and whether the granular system evolves.
LOOSE SAND SPECIMEN COLLAPSES 109

Figure 17. (continued )

Thanks to some simplifying assumptions, in Sections 4.4.3 the approach proposed is shown to be
capable of justifying the kind of instability under consideration.
Naturally, according to the authors, the validity of these theoretical observations will have to be
confirmed by their application to constitutive relationships; nevertheless, in this paper such appli-
cations are not presented so that the approach may not lose its general applicability.

4.1. Stability definition


According to Liapunov’s definition of stability, if small consequences correspond to small dis-
turbances, the system is defined as stable. This definition may be mathematically converted into a
continuous dependence of solution on data. Therefore, in this sense, the collapses previously illu-
strated are the ultimate consequences of an unstable mechanical response. For instance, if we
compare the experimental mechanical responses corresponding to steps A and B of Figure 11, it
appears evident that a discontinuity takes place. The inputs are continuous, but the response is
discontinuous. On the contrary, when the material reaches the steady state, by means of strain-
controlled loadings, the mechanical response appears to be very different. In this case, the mechanical
behaviour is stable, because the dependency of the mechanical response on disturbances remains
continuous.
As anticipated in the introduction, the unstable phenomena considered will be theoretically
interpreted by taking into account the time dependency of the granular material’s mechanical
behaviour. This time dependency may be observed whether the mechanical response be stable or
unstable. The difference seems to consist in the evolution trend. From the analysis of the experi-
mental results shown above (Figure 9), it seems that, as the axial load is kept constant and if the
response is stable (Figure 9(a)), the axial strain rate continuously decreases whereas, if the response is
unstable (Figure 9(b)), the strain rate increases. Nevertheless, this difference is only apparent.
In reality, when the time period following a single instantaneous load increment is taken into
account, the axial velocity initially increases and subsequently continuously decreases. The experi-
mental results illustrated in Figure 9(a) do not show such behaviour. The initial acceleration takes
110 C. DI PRISCO AND S. IMPOSIMATO

place during the first minute and consequently is not recorded. The linear interpolation of experi-
mental data does not allow us to describe the actual initial trend (Figure 9(a)).
The initial system acceleration may take a few seconds when the mechanical response is stable, or
several minutes when instability occurs. The experimental results, obtained by Delage et al..,21
characterized by more accurate time recordings, confirm the initial axial acceleration, with reference
to all load increments. Therefore, with reference to the single axial-strain versus time trends, the
difference between stable and unstable mechanical responses is not qualitative but only quantitative.
The correct way to evaluate the instability occurrence is based on the analysis of the overall
mechanical behaviour during the load test.
In order to adapt Liapunov’s definition of stability to granular assemblies, it may be useful to
interpret the different grain microconfigurations as points of a set, and to relate them to the points
defined in a space in which the co-ordinates are the state variables. This link is meaningful if, and
only if, a representative volume continues to exist. This is true if no strain localization takes place.
Consequently, drained strain localization is excluded a priori, because the considered phenomenon
appears to be globally diffused and not localized.
With reference to granular continua, it may be useful to define as state variables, for instance, the
relative density (a scalar variable), and tensors (aij) describing the directional characteristics of the
material. With regard to these, many authors27,28 have recently introduced different tensors, which
describe directional properties of the microstructure, which may be used as state variables.
Between the points defining the single microconfigurations, and those to which they correspond,
defined in the space of state variables, there is no one to one correspondence (Figure 18). A single
point in the state variable space corresponds to an enormous number of different microconfigurations.
Having theoretically introduced a one to one relationship between the starting set and the points
belonging to the state variable space, and since it is possible to make this space topological, using any
mathematical definition of distance, it now becomes possible to apply Liapunov’s definition of
stability thermodynamically interpreted to granular assemblies:
‘A granular system is defined as stable, when the distance between the starting and the finishing
point, defined in the state variable space, continuously decreases in direct proportion to the decrease
in the size of the load disturbance, of whatever kind it may be.’

4.2. Configurational entropy


In order to describe how the system evolves, the configurational entropy29 Sc may be introduced.
The configurational entropy is assumed to be a function of the probability P of occurrence of a certain

Figure 18. Schematic correspondence between the set of microconfigurations and the hyper-space of the state variables
LOOSE SAND SPECIMEN COLLAPSES 111

state described by a point Ai in the state variable space (section 4.1):

Sc i ˆ S c … P i † …1†

The value of Pi is obtained by summing the pik over the set of k microconfigurations corresponding to
the same point Ai , where pik are the single probabilities associated with the microconfigurations k:

Pi ˆ
Pp
Ni
ik …2†
k ˆ1

where Ni is the number of microconfigurations belonging to the i set.


In the following section, a simplifying definition of p and P is proposed. This is necessary because
of the complexity of the system. The aim consists of highlighting the variables upon which Sc
depends, by disregarding a description of such a dependency.

4.2.1. Density of probability p. If we consider a fixed representative volume of sand, which is filled
with a great and variable number of grains, it is possible to affirm that the number of
microconfigurations that may be generated, depends only on the dimensions of the elemental
volume and on the geometrical characteristics of grains.
We can choose to define these different microstates geometrically, without any hypothesis as to
how the specimen is filled. In this case, the density of probability p, associated with each micro-
configuration, is the same. This ceases to be true when the gravity force is considered or the material
is assumed to be subjected to a certain state of stress. In fact, should this be the case the equilibrium
acts as a constraint and the density of probability p must be redefined.
Some microconfigurations miss any possibility of existing, because the microstructure must sustain
its own weight and the state of stress applied. Moreover, among the possible microconfigurations
some are characterized by high values of p, others by lower ones.
In order to define the value of p associated to a certain microconfiguration and to a particular
effective state or stress, i.e., to define the following dependency:
p ˆ p1 …s0ij † …3†

it is possible to introduce the method described below.


We may assume the density of probability p associated with a certain microconfiguation (Figure
19), defined by the measure of the surrounding domain of s0i *j (which is defined in the effective state
of stress), held by the maximum hyper-spherical domain within which the microconfiguration

Figure 19. Hyperspace of the effective state of stress: domain containing the states of effective stress which may be statically
sustained by the microconfiguration considered
112 C. DI PRISCO AND S. IMPOSIMATO

considered may exist (that is where the equilibrium condition is statically satisfied). The system may
or may not be balanced, depending on the geometry and on the mechanical properties of grains.

4.2.2. Probability P. Corresponding to a certain sand and to a given effective state of stress (s0i *),
j
in section 4.2 a definition of p has been introduced.
P has been defined as the sum of the p values associated with all the microconfigurations belonging
to the same set. Once the set of micro-configuration is defined, from section 4.2.1 the value of P may
be derived.
As we have introduced a relationship between the microconfigurations and the points belonging to
the state variable space, the sum previously defined is meaningful and a value of P may be associated
with each point defined in the state variable space.
As observed in section 4.2.1, if the granular assembly is considered only from a geometrical point
of view ( i.e., disregarding the mechanical aspects) the density of probability p is the same for each
microconfiguration. However, it is important to observe that the number of elements which form each
set of microconfigurations, is variable. In particular, among the sets associated to the points char-
acterized by the same value of relative density Dr, the sets associated with random micro-
configurational distributions of grains (isotropic microconfigurations) are larger.
Consequently, if we consider the points characterized by the same value of Dr, we observe that the
probability P connected to the points characterized by isotropic microstates is greater. This is no
longer true when the material is loaded and an effective state of stress is applied.

4.3. Micro-structural evolution


4.3.1. Kinetic Energy. If we consider load-controlled tests and, in particular, one load-increment
and the subsequent time period during which the state of stress is kept constant, we may observe the
following:
 If the current microstructure cannot statically exist under an increased state of stress (i.e., the
current microstructure cannot sustain the increased state of stress), the grains accelerate and a
system evolution takes place. As time passes during this process, each grain develops its own
momentum.
The introduction of the internal kinetic energy Ec allows us to produce a rough description of the
kinetic aspects of the dynamic evolution of the system. Ec is defined as the integral over the
representative volume of the overall kinetic energy of grains. This is strictly related to the specific
load disturbance and to the dissipated energy along contacts. In fact, it develops during microrear-
rangements and is dissipated by the interparticle frictional slidings. The internal kinetic energy
depends on the disturbance’s size and type but, in particular, on the vulnerability of the current
granular fabric.
It may be interesting to clarify the relationship between Ec and the strain rates, which are recorded
during experimental tests by interpreting the loose sand triaxial specimen as a unique macroelement.
This relationship is difficult to determine because, a priori, the two quantities cannot be assumed to
coincide but, if no turbulence takes place within the specimen, the link becomes explicit. In fact, if we
assume the displacement field to be homogeneous within the specimen, from the boundary condition
it is possible to derive an internal kinematic description of the continuum.
On the other hand, if any turbulence takes place, homogeneity within the specimen is lost and the
boundary conditions are not sufficient to derive the internal displacement field.
LOOSE SAND SPECIMEN COLLAPSES 113

In the following, Ec will be interpreted as a useful tool in describing the onset of turbulence.
According to the authors, the instability takes place when the turbulence is not dumped, but itself
feeds.

4.3.2. State Variable evolution. In section 4.2 we have introduced the configurational entropy Sc
by statically describing the granular system. In fact, the definitions of p and P introduced above were
related only to static quantities and the system, as previously described, was assumed to be
motionless. On the other hand, in the present section the microstructural evolution is taken into
consideration and the configurational entropy itself will allow us to describe such an evolution. This
is possible because the variable Sc has been defined as a monotonic increasing function of the
probability P. Consequently, the coincidence between the most probable state and the state which
most frequently happens is translated in the following condition on Sc :
 In the state variable space, the granular system evolves towards the point characterized by the
current relative maximum value of Sc within the infinitesimal surroundings of the initial state. Sc
represents the variable which determines the way the granular system evolves, in the same way
as Ec determines the evolution itself. In fact, the system stops evolving when Ec goes down to
zero.

4.3.3. Further remarks. The evolution of a granular system is necessarily associated with a strain
increment. On the other hand, a strain increment is not necessarily associated with a state variable
evolution. For instance, we may obtain considerable strain increments, even though the image point
of the internal state of the material does not change. This is possible if the strain increments are
caused by the evolution of the fabric within the same set of microstructures. This happens when the
material reaches the so-called ‘steady state’.
This means that the system, defined in the state variable space, does not recognize any path which
allows an increase in Sc in the infinitesimal surroundings of the current state. On the contrary, up to
this condition, a state variable evolution takes place, but by increasing the stress level this evolution
takes place more slowly.
It is interesting to observe that, even if the strains associated with the steady state are considerable,
according to the thermodynamical definition introduced in section 4.1, the system remains stable. In
fact, according to the definition of steady state, the distance between the starting and the finishing
point, defined in the state variable space, associated with load disturbance, goes down to zero. Only
by analysing the phenomenon in the interior of the state variable space, is it possible to highlight a
clear separation between stable and unstable behaviour. In fact, by considering only the strain
variables, the mechanical behaviour of the material may be wrongly interpreted.

4.4. Instability analysis


In section 4.1 the occurrence of the instability has been described. In this paragraph the authors
propose a new approach that, thanks to the definition of the configurational entropy Sc previously
introduced, highlights the mechanical instabilities experimentally shown in section 3.
The new theoretical approach is based on the assumption that the density of probability p (defined
in section 4.2.1) depends on the kinetic variables describing the motion of each grain. In fact, during
the granular system evolution, the definition of p cannot remain static, but must become dynamic and
thus the evolution of the system becomes dependent on the time factor.
In section 4.3 the evolution of the micro-structure has been taken into account, nevertheless this
has been assumed to be quasi-static. This means that the kinetic energy is not nil, but is small enough
114 C. DI PRISCO AND S. IMPOSIMATO

not to influence the mechanical behaviour of the material at all (i.e., to be negligible in determining
Sc ). In fact, in the definition of p introduced in section 4.2.1, no kinematic variable appears.
The extension of this theoretical approach to real phenomena, implies disregarding any dynamic
effect on the mechanical behaviour of the material. Such an assumption is implicit in all the elas-
toplastic constitutive models but also in the elastoviscoplastic ones. In the latter the evolution of the
state variables takes place with time, but is not dependent on kinetic variables.

4.4.1. A simplified dynamic definition of p. When a load-controlled test is considered, as the load-
increment may be assumed to be impulsive, an accurate dynamic description of the system response
must be outlined. From a probabilistic point of view, this entails a dynamic description of the density
of probability p. In the following, the authors will assume that the dynamic description of the system
may be simplified and that the value of p is a function only of the effective state of stress (section
4.2.1), of the micro-inertia and of the kinetic energy Ec (section 4.3.1). This implies that the momenta
of grains are disregarded.
The direct dependency of p on Ec may be physically interpreted by means of a very simple
hydraulic analogy. As happens in fluid flows, the increase in Ec may cause turbulent motions in the
continuum and these are assumed to influence the mechanical behaviour of the material.
Then, we may write:
p ˆ p1 …s0i j ; micro-inertiae†p2 …Ec † …4†
where p1 is the extension of p defined in equation (3) (section 4.2.1) to the dynamic case, while p2
will be described below.
By means of equation (4), we assume that the density of probability p is a function of two
independent factors:
 the equilibrium condition (dynamically adapted),
 the possible onset of turbulence.
In the first term, the micro-inertiae allow the granular system to sustain dynamically some effective
states of stress excluded in static conditions. More simply, thanks to micro-inertiae in dynamic
conditions, the number of possible micro-configurations changes and the value of p1 (defined in
section 4.2.1) changes, too.
Instead, the second term p2 of equation (4) may be interpreted as a modulating function, which
takes into account the disorder induced by turbulence. p2 may be assumed to be a constant function of
unit value, when Ec is sufficiently small, in all other cases p2 is a weight function. This is char-
acterized by larger values corresponding to disordered microstructures and smaller values corre-
sponding to ordered microstructures.

4.4.2. Drained triaxial strain controlled tests. As the authors in the previous sections underlined
the possible importance of the Ec variable in the description of the granular system evolution, it may
be interesting to discuss the possible strain-rate dependence of the material’s mechanical behaviour.
Some experimental results, obtained by performing strain controlled standard triaxial compression
tests, are illustrated in Figure 20. The three curves are relative to the same type of test on specimens
characterized by the same initial relative density but these tests were performed by imposing different
strain rates.
The experimental curves show a clear dependency of the mechanical response on the strain rate.
Naturally, this dependency can neither be reproduced by means of a traditional elasto-plastic con-
stitutive model, nor by means of an elasto-viscoplastic one (for instance Perjna30). By using an elasto-
viscoplastic constitutive model and by increasing the strain rate, an initial more rigid behaviour
LOOSE SAND SPECIMEN COLLAPSES 115

Figure 20. Strain controlled standard triaxial tests on saturated very loose Hostun sand specimen, performed at different strain
rates: (a) stress–strain behviour; (b) olumetric curve.

would be obtained. Moreover, the value of the q limit would not be influenced by the value of strain
rate.
Such experimental results may be theoretically interpreted only by a global discussion of the
influence of the strain rate on the microstructural evolution.
First of all, we must assume that the kinetic energy Ec is constant, during constant strain rate tests.
This is possible because the granular system capability of dissipating energy allows us to exclude that
a quantity of energy is stored as grains’ vibrational energy. Therefore, we may assume that, also with
116 C. DI PRISCO AND S. IMPOSIMATO

reference to high values of strain rate, a monotonic dependence of the external value of Ec on the
internal one exists. Moreover during strain controlled loadings, characterized by a constant axial
strain rate, we can disregard the values of the micro-inertial terms.
In Section 4.4.1 we have observed that, by changing Ec , p2 and p change too. In particular, by
increasing Ec , the probability of p associated with more disordered microstructures increases. Con-
sequently, by increasing the strain rate (i.e., by increasing the Ec ), it derives a different path in the
state variable space and a steady state characterized by a less ordered microstructure.
Generally, the system has two different possibilities to satisfy the equilibrium condition: to
improve the contact distribution along the direction of maximum stress (i.e., by inducing the
directional rearrangement of the microstructure-induced anisotropy) or by increasing the global
number of contacts (by increasing the relative density). The increase in Ec is assumed to partially
‘inhibit’ the induced anisotropy. Consequently, a denser microstructure is obtained.
In fact, in Figure 20(b), the experimental curves show a compaction influenced by the strain rate. In
particular, by increasing the strain rate, the compaction increases. Instead, by analysing the axial
stress-strain curves of Figure 20(a) and Figure 21, we may observe that, by increasing the strain rate,
the maximum q value decreases. This seems to suggest us the following observation:
 the increase in relative density is not large enough to balance the decrease in contact surfaces
along the direction of maximum stress.
These experimental results seem to imply a curious softening behaviour of the material linked to
the strain rate and not envisaged, by performing a constant strain rate test.
Such a strain rate dependency is analogous to that analysed by Rice and Ruina31 with reference to
the stability of fault slip.
From equation (4) of section 4.4.1. we may conclude that the increase in Ec causes a change in the
path followed by the granular system. This causes two different effects:

Figure 21. Strain–controlled tests: asymptotic values of q corresponding to different axial strain-rates
LOOSE SAND SPECIMEN COLLAPSES 117

 the ultimate state of the system is denser


 the maximum q value is lower.
Both these effects may be justified, by assuming an ultimate state, described in the state variable
space which is less ordered and a slightly denser.

4.4.3. Instability occurrence. By simplifying the approach outlined in section 4.4.2, the system
evolution may be described schematically by means of two different variables:
 Dr
y
where Dr is the relative density (a state variable) and y is not a state variable, but a scalar quantity
introduced for simplicity which may be explained as follows.
By introducing in the effective stress space a state variable tensor aij (section 4.1) which describes
the anisotropy of the material, we may define the scalar quantity y as follows:
y ˆ y…s0hk ; aij † …5†

y may be interpreted as the eccentricity of s0ij with reference to the current aij .
Therefore, each microstructure evolution may be schematically classified as in Figure 22. Point I
represents the current state of the material and each arrow defines the direction of possible evolution.
Type 1 is characterized by an increase in y and a decrease in Dr; type 2 by an increase in y and Dr;
and type 3 by a decrease in y and an increase in Dr.
During a standard triaxial compression test on loose sand specimens, when the mechanical
behaviour is stable, the evolution type 2 is followed by the system. On the other hand, according to
the authors, when the instability occurs, evolution type 3 takes place.
In order to highlight when and why this kind of evolution occurs and why it is unstable, it is
necessary to describe the system evolution during the time period following the load increment which
causes the unstable mechanical response.
We assume that the system initially is motionless. Subsequently, at the time instant t0‡ successive
to the load-increment, the micro-inertial terms may be not negligible, but Ec may be assumed to be
nil. From equation (4) it derives that the current path is characterized by an increase in Dr and in y
(Path number 2). Then, Ec increases and the values of p associated with more disordered micro-
structures increase too. After a certain period of time, because of the disturbance caused by the Ec
developing, among the values of the density of probability p associated with the microconfigurations
belonging to the infinitesimal surroundings of the current microconfiguration, the maximum corre-
sponds to a less ordered internal fabric (path number 3).

Figure 22. System evolution: stable and unstable paths.


118 C. DI PRISCO AND S. IMPOSIMATO

If the micro-inertial terms are nil, within the quadrant 3, a line may be drawn separating the
possible and the impossible states (section 4.4.1). When these terms are not negligible, this limit
rotates clockwise. According to the authors, during the time period following the load increment
mentioned previously, point J, possibly thanks to the micro-inertial terms, becomes the most prob-
able.
When evolution type 3 is chosen by the system, even if the density increases, the increase in
density may be not enough to replace the contacts lost along the direction of the maximum stress
caused by the negative anisotropic evolution.
In this case, because of the decrease in the number of contacts along the direction of maximum
stress, both the micro-inertial term and the Ec may increase. The system continues to accelerate and
the point image in the state variable space continuously changes.
According to the definition given in section 4.1, this behaviour may be defined as unstable, because
corresponding to a small load disturbance, a great distance between the starting and the finishing
points (defined in the state variable space) is obtained. The phenomenon is self-feeding, because the
path causes an acceleration and an increase in Ec automatically. The collapses may occur slowly or
quickly, but the mechanism does not change.
During the collapse, when the microstructure becomes isotropic enough, the increase in relative
density (i.e., the increase in number of contacts) allows the system to decelerate. Ec decreases and,
consequently, the phenomenon stops. In this manner, we may assume that the instabilities experi-
mentally shown are caused by the dependency of the mechanical behaviour of loose sand on the
kinetic energy.

4.4.4. Further remarks. In section 4.4.2 the increase in the relative density Dr associated with the
instability phenomenon was analysed. As observed experimentally (section 3.3), if a loose sand
specimen is saturated, a rapid increase in pore pressure takes place.
The presence of water during collapse causes a decrease in the effective mean pressure and
consequently an increase in the effective stress level. This combines with the dynamic factor
described above in determining the unstable mechanical behaviour of the granular system.
When the liquefaction phenomenon is considered with reference to boundary value problems, the
propagation of the instability must be taken into account. Therefore, when the propagation of this
type of instability in saturated soils is analysed, it is important to consider both the mechanical factors
previously mentioned: the dynamic effect and the subsequent rapid increase in pore pressure.

5. CONCLUSIONS
Sudden and unexpected collapses, taking place during experimental drained load controlled triaxial
tests on loose Hostun sand specimens, were shown.
From an experimental point of view, it was underlined that these instabilities are not necessarily
linked to the presence of water within the sand specimens. In fact, these collapses may occur whether
saturated or dry loose sand specimens are tested.
The instability considered appears to be strictly linked to the load-control, i.e., it may be inter-
preted as an unstable creep phenomenon.
In order to justify the experimental results shown above from a theoretical point of view, the
authors tried to highlight the factors influencing the mechanical behaviour of the material.
The mechanical response was associated with the microstructural evolution of the granular system
and, according to the authors, such an evolution may be described theoretically by means of the
configurational entropy Sc and the kinetic energy Ec .
LOOSE SAND SPECIMEN COLLAPSES 119

Sc was interpreted as the variable determining the way in which the granular system evolves at the
macrolevel, while Ec determines whether this evolution takes place. The configurational entropy
associated with a certain point in the state variable space was assumed to be dependent on the
effective state of stress and of the current kinetic energy Ec .
In fact, the density of probability p at the microlevel, introduced in order to define the config-
urational entropy Sc at macrolevel, was assumed to be a function of two independent probabilistic
constraints:
 the equilibrium condition (dynamically adapted)
 the energetic content.
The simplifying assumptions introduced allowed the authors to outline the dynamic aspects of the
unstable phenomenon, by paying particular attention to the influence of the energetic content of the
system on its own evolution.
In the light of these considerations, the incapability of reproducing the instability considered,
shown by the elastoplastic and elastoviscoplastic constitutive models commonly implemented, may
be justified.
In fact, these constitutive relationships assume the process of fabric evolution to be quasi-static.
This appears to be unrealistic with reference to the phenomenon considered and, in this paper, the
authors have tried to introduce a new way of defining the rules which govern such an evolution. The
whole theoretical interpretation has been developed by analysing laboratory experimental test results.
Any extension of this approach, in order to interpret boundary value problems (spontaneous lique-
factions, flow slides), should be coupled to a more accurate analysis of the phenomenon propagation,
both in dry and saturated continua.

ACKNOWLEDGEMENTS

This research was conducted within the framework of Project 2, localization phenomena in geo-
mechanics, of the A.L.E.R.T. Geomaterials Programme, funded by the E.U. (Human Capital and
Mobility). Financial support from Italian C.N.R. and M.U.R.S.T. is also gratefully acknowledged.
Moreover, the authors would like to acknowledge Prof. R. Nova for his helpful and precious
support, Prof. I. Vardoulakis for his hints, Ing. F. Calvetti for pleasant discussions and Dr. S. Losasso
for his careful linguistic observations.

REFERENCES
1. H. K. S. Begemann, H. L. Koning and J. Linderberg, ‘Critical density of sand’, Proc. 9th. Int. Conf. Soil Mech., Tokyo,
Japan, 1977.
2. J. D. Eckersley, ‘Instrumented Laboratory flowslides’, Géotechnique, 40 (3), 489–502 (1990).
3. S. Sasitharan, P. K. Robertson, D. C. Sego and N. R. Morgenstern, ‘Collapse behaviour of sand’, Can. Geotech. J., 30,
569–577 (1993).
4. S. Sasitharan, P. K. Robertson, D. C. Sego and N. R. Morgenstern, ‘State boundary surface for very loose sand and its
pratical implication’, Can. Geotech. J., 31, 321–334 (1994).
5. P. Skopek, N. R. Morgenstern, P. K. Robertson and D. C. Sego, ‘Collapse of dry sand’, Can. Geotech. J., 31, 1008–1014
(1994).
6. C. di Prisco and S. Imposimato, ‘Time dependent mechnical behaviour of loose sand’, Mechanics of Cohesive-Frictional
Materials and Structures, 1, 45–73 (1996).
7. R. F. Scott, ‘Solidification and consolidation of a liquefied sand column’, Soils and Foundations, 26 (4), 23–31 (1986).
8. D. Kolymbas, ‘Localised consolidation’, in Localisation and Bifurcation Theory for Soils and Rocks, eds. R. Chambon, J.
Desrues and I. Vardoulakis, Balkema, pp. 83–124, 1994.
9. V. A. Osinov and G. Gudehus, ‘Plane shear waves and loss of stability in saturated granular body’ Mechanics of Cohesive-
frictional Materials and Structures, 1, 25–44 (1996).
10. K. Terzaghi, ‘Varieties of submarines slope failures’, Norwegian Geotechnical Institute, 25, 1–16 (1957).
120 C. DI PRISCO AND S. IMPOSIMATO

11. G. Castro, ‘Liquefaction of sand’, Harvard Soil Mechanics Series, vol. 81, Cambridge, MA, 1969.
12. S. L. Kramer and H. B. Seed, ‘Initiation of soil liquefaction under static loading conditions’, Geotech. Eng., 114 (4), 412–
430 (1988).
13. P. V. Lade, ‘Static instability and liquefaction of loose fine sandy slopes’, J. Eng. Mechanics, 118 (1), 51–71 (1992).
14. F. Darve, ‘Liquefaction phenomenon: modelling, stability and uniqueness’, in Verifications of Numerical Procedures for
the Analysis of Soil Liquefaction Problems, eds. Arulanandan and Scott, Balkema, pp. 1305–1319, 1994.
15. F. Darve, E. Flavigny and M. Meghachou, ‘Constitutive modelling and instabilities of soil behaviour’, Computers and
Geotechnics, 17, 203–224 (1995).
16. J. M. Konrad, ‘Undrained response of loosely compacted sands during monotonic and cyclic compression tests’,
Géotechnique, 43 (1), 69–89 (1993).
17. K. Ishihara, ‘Liquefaction and flow failure during earthquakes’, Géotechnique, 43 (3), 351–415 (1993).
18. C. di Prisco, R. Matiotti and R. Nova, ‘Theoretical investigation of the undrained stability of shallow submerged slopes’,
Géotechnique, 45, 479–496 (1995).
19. A. Andresen and L. Bjerrum, ‘Slides in subaqueous slopes in loose sand and silt’, Norwegian Geotech. Institute, 81, 1–9
(1968).
20. J. Lindenberg and H. L. Koning, ‘Critical density of sand’, Géotechnique, 31, N (2), 231–245 (1981).
21. P. Delage, X. Noc, G. Saidy and E. de Laure, ‘Aspects volumiques du fluages des sables’, in 25 Colloque du Groupe de
Rheologie, Grenoble, pp. 147–158, 1990.
22. S. Murayama, K. Michihiro and T. Sakagami, ‘Creep characteristics of sands’, Soils and Foundations, 24 (2), 1–15 (1984).
23. P. V. Lade, ‘Creep effects on static and cyclic instability of granular soils’, J. Geotech. Eng., 120 (2), 404–419 (1994).
24. R. Matiotti, C. di Prisco and R. Nova, ‘Experimental observations on static liquefaction of loose sand’, in Earthquake
Geotechnical Engineering, Ed. Ishihara, Balkema, pp. 817–822, 1995.
25. A. M. Liapunov, ‘General problem of motus stability’, French tranlation, Ann. Fac. Sci. Univ. Toulouse, 9, (1892).
26. L. Bjerrum, S. Kringstad and O. Kummeneje, ‘The shear strength of a fine sand’, Proc.V I.C.S.M.F.E., vol. 1, Paris, France,
pp. 29–37, 1961.
27. J. Christoffersen, M. M. Mehrabadi and S. Nemat-Nasser, ‘A micromechanical description of granular material behaviour’,
J. of Applied Mechanics, 48, 339–344 (1981).
28. S. C. Cowin, ‘Microstructural continuum models for granular materials’, Proc. of the U.S Japan Seminar on Continuum
mechanical and Statistical Approaches in the Mechanics of Granular Materials, eds. Cowin and Satake, Tokyo, Japan, pp.
162–170, 1978.
29. K. C. Valanis, J. F. Peters and J. Gill, ‘Configurational entropy, non–associativity and uniqueness in granular media’, Acta
Mechanica, 100, 79–93 (1993).
30. P. Perzyna, ‘The constitutive equations for rate sensitive plastic materials’, Quart. Appl. Math., 20, 321–332 (1963).
31. J. R. Rice and A. L. Ruina, ‘Stability of steady frictional slipping’, J. of Applied Mechanics, 50, 343–349 (1983).

You might also like