Você está na página 1de 18

Toorman, E. A. (1999). Geotechnique 49, No.

6, 709726

Sedimentation and self-weight consolidation: constitutive equations


and numerical modelling
E . A . TO O R M A N 
La solution de l'equation d'equilibrage des masses pour la sedimentation et la consolidation
dans les suspensions de sediments necessite des
equations de cloture pour les parametres suivants: diffusivite, permeabilite et contraintes effectives. Les donnees experimentales relatives
aux matieres non cohesives et cohesives sont
discutees et de nouveaux rapports semi-empiriques, representant une meilleure approximation
de la realite, sont proposes. Le caractere non
unique pour les sediments cohesifs est explique
sur le plan des differences dans l'histoire de
l'agregation. L'execution numerique de la theorie de l'unication pour la sedimentation et la
consolidation se heurte a de graves difcultes.
En particulier, les discontinuites necessitent l'adjonction d'une diffusivite articielle pour creer
l'amortissement necessaire pour eviter le depassement des oscillations induites. Une technique
speciale aux elements nis permet d'optimiser la
stabilite. Le modele est conrme avec plusieurs
ensembles de donnees et preuves an de prevoir
le comportement exact et l'evolution du prol
de densite pour les sediments non cohesifs et
cohesifs. Les differences restantes entre prols
de densites mesures et calcules pour le materiel
cohesif sont attribuees a l'insufsance des equations constitutives actuels.

The solution of the mass balance equation for


sedimentation and consolidation in sediment suspensions requires closure equations for the parameters diffusivity, permeability and effective
stress. Experimental data for non-cohesive and
cohesive materials are discussed. New semiempirical relations, which are a better approximation of reality, are proposed. The nonuniqueness for cohesive sediments is explained
in terms of differences in aggregation history.
The numerical implementation of the unifying
theory for sedimentation and consolidation faces
some serious problems. Particularly, the discontinuities require the addition of articial diffusivity to create the necessary damping to avoid
overshoot-induced oscillations. Improved stability is obtained using a special nite-element
technique. The model is validated with several
data sets and proves to predict the right behaviour and density prole evolution for both noncohesive and cohesive sediments. The remaining
differences between measured and computed
density proles for cohesive materials are attributed to the inadequacy of the currently used
constitutive equations.
KEYWORDS: consolidation; constitutive relations;
numerical modelling; sedimentation.

they are expressed as a function of sediment concentration (or void ratio) only. In Toorman (1996),
it has already been argued that this traditional
assumption cannot be valid for the effective stress.
This will be further investigated here and another
closure equation is proposed. Analysis of experimental data for cohesive sediments reveals a dependence of the permeability on the initial
conditions. This is investigated in the light of the
aggregation history.
In the literature one can nd many types of
numerical models for the prediction of sedimentation and/or consolidation. However, there is hardly
any which is applied over the total range of
concentrations (from zero to maximum compaction). This can be attributed to the problem of
solving the moving density discontinuities, such as

INTRODUCTION

A unifying theory for sedimentation and selfweight consolidation has been presented in a previous paper (Toorman, 1996). The solution of the
resulting sediment mass balance equation requires
closure relations for the suspension diffusivity, the
permeability (or the free ltration rate, redened
below) and the effective stress. Many types of
relatively simple empirical equations have been
proposed over the years (an overview is presented
by Alexis et al., 1993). They have in common that
Manuscript received 16 June 1997; revised manuscript
accepted 24 March 1999.
Discussion on this paper closes 30 June 2000; for further
details see p.ii.
 Katholieke Universiteit Leuven.

709

710

TOORMAN

the interface between the suspension and the consolidating bed. Furthermore, it is remarkable that
very few papers on the modelling of consolidation
show results of computed density proles and, even
more rarely, of a comparison with measured density proles. This clearly is the most valuable
validation of a model, as will be demonstrated
later. Most models are evaluated on their capacity
to match measured settlement curves. This can be
achieved relatively easily, even with the simplest
models. Of course, from an engineering point of
view, the latter test may be the most important, for
example, if one tries to minimize the storage
volume in dredged-material disposals. However,
there is much interest in the correct prediction of
density proles. For instance, in sediment transport
modelling one is interested in the correlation between the density of the bed and the erosion
resistance of the corresponding layer.
This paper investigates different approaches
(including their possibilities and limitations) to
numerical modelling of problems involving onedimensional sedimentation and consolidation. Special attention is given to different coordinate systems, different types of boundary conditions and
numerical stability. Results are presented of the
validation of the model with three sets of experimental data.
CONSTITUTIVE RELATIONS

It has been shown by Toorman (1996) that the


sedimentation and self-weight consolidation processes can be described by the general one-dimensional sediment mass balance equation:


@s @S
@
w0 @ 9
@s

w0 s
DB

@z @z
@t
s w @z
@z
(1)

where t is the time, z is the vertical Eulerian


coordinate, s is the solids volume fraction, S is
the sediment ux ( ws s , with ws being the
average settling rate of the particles), 9 is the
vertical effective stress (the adjective `vertical' is
dropped in the rest of the text), D B is the suspension diffusivity coefcient (which includes Brownian diffusion and other effects, such as differential
settling), and s and w are the unit weights of the
sediment and water, respectively. The parameter w0
has been dened previously (Toorman, 1996) as
w0
k
kr
ks
1e
s =w 1

(2)

where k is the permeability, k r is the reduced


permeability (e.g. Znidarcic, 1982) and e is the
void ratio. The parameter w0 was given the unfortunate and confusing name of `stress-free' ltration

rate. It equals the true settling rate, relative to the


mean suspension ow rate U vs s vw (1 s )
(where vs and vw are the average velocity of the
sediment and water, respectively), when the effective stress term and the diffusivity (representing
other particle interactions) are not present
(Toorman, 1996). It should not be confused with
the settlement rate of the surface of the consolidating soil. Even though this surface is stress-free (in
self-weight consolidation), as the (vertical) effective stress here is zero, the effective-stress gradient
is not. According to equation (1) the two rates are
different. It is proposed to rename the parameter
the `free ltration rate', which refers to the (often
ctitious) situation where the particles remain free
of interparticle contact interactions. It is equivalent
to the critical mean ow rate (U vw n, where n
is the porosity 1 s ) to uidize a sediment
layer in equilibrium (vs 0). When pore water is
forced through a homogeneous sample which is
subjected to gravity, uidization is reached when
the excess pore pressure equals the buoyant weight
(i.e. 9 0). Fluidization and sedimentation experiments indeed yield the same result (e.g.
Richardson & Zaki, 1954).
Equation (1) can be rewritten as (Toorman &
Huysentruyt, 1997)




@s
@
@ 9
@s
DB
w0 s 1

(3)
@z
@ 0
@t
@z
where 0 u0 is the buoyant stress, is the
total stress and u0 is the hydrostatic pressure. This
shows that (1 @ 9=@ 0 ) is the factor by which
the free ltration rate is reduced owing to particle
interactions in a consolidating network structure,
i.e. w0 @ 9=@ 0 is the rate of deceleration caused
by the soil matrix deformation.
In order to solve equation (3), closure relationships are required for the diffusivity coefcient and
for the two consolidation parameters, i.e. the effective stress and the free ltration rate (or the permeability).
Suspension diffusivity
The formulation of a closure relationship for the
suspension diffusivity coefcient D B is not
straightforward. Batchelor (1976) found theoretically that for dilute suspensions the diffusivity due
to Brownian motion increases linearly with the
concentration, according to D B D0 (1 1:45s ).
For higher concentrations, it is expected that D B
decreases again owing to hindrance by other particles.
The ratio between the diffusivity and the free
ltration rate can be estimated from experimental
data as follows. Consider a point where the net
sediment ux S is zero, e.g. at the bottom. Accord-

SEDIMENTATION AND SELF-WEIGHT CONSOLIDATION

ing to equation (1) applied to the suspension phase


(i.e. no effective stress), there is an equilibrium
between the upward diffusive ux D B @s =@z and
the downward settling ux w0 s when S 0.
Hence, the corresponding concentration gradient
(denoted Geq ), which can be determined from experimental density proles, equals




@s
w0 s
Geq
(4)
@z S0
D B S0
Solved for the diffusivity, this equation shows that
D B is proportional to the settling ux w0 s (with
1=Geq being the proportionality factor). This corresponds with the empirical ndings by Gosele &
Wambsgan (1983), who analysed bottom density
gradient data. But since D B cannot be zero in the
limit of s ! 0, because of Brownian diffusion,
and, according to the derivation of the diffusion
term (Toorman, 1996), D B is proportional to w0 ,
the following diffusivity relationship is proposed:
D B w0 (0 1 s )

(5)

which introduces two model parameters, 0 and 1 .


Note that equation (4) no longer holds when
effective stresses develop. As the diffusivity then is
negligibly small, its actual value does not need to
be known. The diffusion term is thus dropped in
the consolidation stage.
Experimental determination of the consolidation
parameters
The accurate determination of the consolidation
parameters for a soft sediment with high water
content is difcult. Classical geotechnical techniques cannot be applied. Instead, the values are
obtained from self-weight consolidation tests in
settling columns. All information is computed from
measured pore water pressure proles, density proles and the settling curve (the watersediment
interface level as a function of time) (Berlamont et
al., 1993a; Sills, 1997). Pore water pressures can
be measured with different types of piezometers
(open standpipes or electronic transducers). Densities are non-destructively measured with gamma or
X-ray transmission probes.
The error on the density measurements is typically of the order of 1% (on the bulk density of
the watersediment mixture, i.e. 10% on the
sediment concentration) for gamma densimeters.
The measurements have a spatial resolution of the
order of 510 mm. Sills (1995) claims an accuracy
for bulk density of 0:2% for X-ray densimetry,
with a spatial resolution of 1 mm. The accuracy of
the piezometers used is typically 10 Pa. However,
in practice, the error on the measured pore pressures is of the order of 50 Pa. A possible cause of
the deviation from the theoretical smallest error

711

may be the malfunctioning of the lters in the


pressure ports, either by trapped air or by clogging
with sediment. Another possible cause is drift of
the electronic pressure transducers or the assumption of the same capillary pressure correction for
different standpipe piezometers.
At the beginning of consolidation, when the soil
structure is weak, effective stresses are small. Near
the surface they remain small because of the small
loading by the weight of the sediment above.
Under these circumstances the effective stress is
the result of the subtraction of two entities of equal
order of magnitude. As long as 9 is below
100200 Pa, the relative error on the effective
stress can be enormous. This is a very serious
problem. It implies that measurements in the top
10 cm of a sediment bed are unreliable (Toorman
& Huysentruyt, 1997). There is a great need for
more accurate measurement techniques.
Permeability
Experimental determination. The permeability k
is a measure of the inverse of the ow resistance
of the pore water ow in a saturated porous
medium subjected to a pressure gradient. Experimental values for k are obtained from application
of the generalized Darcy (or DarcyGersevanov)
law to batch sedimentation, resulting in
(1 s )

vw vs ws
1 @u

i
k
k
w @z

(6)

with vs ws (z axis positive upwards). The


equality ws (1 s )(vw vs ) is only valid in the
case of an impervious bottom and follows from the
continuity equation (Toorman, 1996).
The excess pore pressure u at a certain depth
z below the water surface is calculated as the
difference between the measured pore water pressure uw and the computed hydrostatic pressure
u0 w z. The excess pore pressure gradient or
hydraulic gradient i is the non-dimensional local
slope of the u prole. The settling rate ws can be
estimated from the difference between subsequent
density proles in two ways:
 


@z
1
@
ws

@ t constant
(s w )s @ t constant z
(7)
Notice that `constant ' is equivalent to `constant
material coordinate'.
The accuracy of the experimental determination
of the parameters i and ws is very low. It depends
on the accuracy of the density probe and piezometers, the number of measurement points (i.e. the
number of pressure ports in the settling column
set-up and the frequency of density prole record-

712

TOORMAN

ing, respectively) from which the slope is computed, as well as the approximation method (using
some sort of interpolation) used to compute the
slope.
The methodology described above only applies
to a saturated soil matrix in which effective stresses have developed. At lower concentrations, in the
suspension phase, one has to use the relationship
between permeability and free ltration rate, dened by equation (2). As long as the diffusivity is
negligible, ws w0 and one can apply an analytical method based on the theory of Kynch (1952) to
estimate ws as a function of the concentration (e.g.
Toorman, 1992). In the case of an initially homogeneous suspension, one nds the settling velocity
and corresponding concentration at the interface
between the sediment and water h is the interface
level as
ws

dh h

dt
t

s 0

h0
h ws (t t0 )

(8)

where t is the time, 0 is the initial sediment


concentration and the subscript 0 refers to the
initial conditions.
When diffusion becomes important, i.e. for cohesive sediments which form a compressible bed,
Kynch's method is no longer valid (Auzerais et al.,
1988). Later improvements of Kynch's theory by
Tiller (1981), Fitch (1983) and Font (1988) for
cohesive sediments, to account for the non-linear
growth of the bed, only apply to the sedimentation
zone and require extra data (i.e. the suspension
bed interface level as a function of time), which
are often not available. The effect of diffusivity
can simply be eliminated by measuring the initial
surface settling rate of a uniform suspension at
different initial concentrations (e.g. Shannon et al.,
1963).
It is generally assumed that, for a given soil
type and uid viscosity, the permeability only
varies with the density (or void ratio). Many different empirical relationships have been proposed.
Alexis et al. (1993) present an overview. These
relationships are generally based on measurements
over a limited range of concentrations.
However, there are two physical constraints on
the value of the permeability, which often are not
taken into account by these relationships. When
the solids volume fraction becomes zero, the ow
resistance is zero, i.e. the permeability becomes
innite. This can be seen from equation (2), bearing in mind that w0 becomes equal to the settling
velocity of a single particle when there are no
other particles in the uid, i.e. the limiting value of
the settling rate for s ! 0 is the Stokes fall
velocity. For the other extreme value, where there

are no more pores, i.e. at s 1, the permeability


should become zero. This also follows from equation (6) applied to a saturated porous layer in
equilibrium (vs 0), subjected to a pressure gradient (i . 0) by which water is forced to ow
through the static soil matrix. The condition
s 1 will most likely never occur, as the maximum compaction volume fraction max is generally
smaller than 1 owing to the incompressibility of
the primary particles.
Settling rate and permeability of non-cohesive,
rigid particles. The settling rate of rigid, natural,
non-cohesive particles (such as sand) is relatively
high owing to the particle size. The constitutive
relationship of the settling rate as a function of the
sediment volume fraction is based on experimental
data obtained by different methods (mainly sedimentation and uidization tests) for ne, closely
sized (mostly spherical) particles. The results for
the settling rate, non-dimensionalized relative to the
Stokes fall velocity wSt, as a function of solids
concentration are nearly the same for all these data.
A selection of data is shown in Fig. 1. Barnea &
Mizrahi (1973) present a discussion of the majority
of the data published up to that time. A dependence
on the particle Reynolds number Rep has been
identied as a major factor explaining the variations
(up to 20%) between different sets of data.
Particularly in the dilute concentration range
(s , 0:05), it is observed that the settling rate
decreases more rapidly with increasing concentration when Rep is very small. In the dilute limit
(s ! 0) the ratio ws =wSt can be approximated by
a relationship of the form 1 asn , where n varies
from 1=3 for Rep  1 (the creeping-ow range, i.e.
where inertia is negligible) to 1 for large Rep and in
shear ow. The value of n can be related to the
spatial arrangement of the particles (Davis &
Acrivos, 1985).
Several empirical, semi-empirical and theoretical
relationships have been proposed over the years.
Intercomparisons between various relationships and
data have been presented by Oliver (1961), Barnea
& Mizrahi (1973) and Garside & Al-Dibouni
(1977). The majority of these laws yield a zero
particle velocity at s 1. But in sedimentation
tests the settling rate becomes zero at a smaller
concentration max , corresponding to the maximum-compaction arrangement of the particles. On
the other hand, as pointed out previously, a bed
composed of rigid particles at its maximum compaction still has interconnecting pores. Consequently, the permeability is not zero at the
concentration max at which the settling rate is
zero. Furthermore, at low concentrations the settling rate and the free ltration rate are equal.
Comparison of data from sedimentation and uidization experiments for non-cohesive particles (e.g.

713

SEDIMENTATION AND SELF-WEIGHT CONSOLIDATION

Relative settling rate

0.1

Equation (9)
(Richardson & Zaki, 1954)

0.01
Equation (11)
Equation (12)

Equation (10)
(Barnea & Mizrahi, 1973)
(Brinkman, 1947)
0.001
0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1.0

Solids volume fraction

Fig. 1. Settling rate, relative to the representative Stokes velocity, as a


function of volumetric concentration for non-cohesive (spherical)
particles. Comparison between some experimental data (m, creepingow data from Oliver (1961); h, Gurel (1951) and series XII from
Shannon et al. (1963) and various semi-empirical relationships (lines)

Richardson & Zaki, 1954; Barnea & Mizrahi,


1973) indeed shows that the free ltration rate and
the settling rate are equal over the whole range of
concentrations considered (s , 0:5). This leads to
the conclusion that at a certain concentration, close
to the maximum packing fraction, the free ltration
rate and the settling rate can no longer be equal.
According to equation (3), the difference should be
attributed to the particle interaction in the matrix
which is being formed and the corresponding effective stresses.
Consequently, the popular relationship of
Richardson & Zaki (1954) should not be used for
the settling rate, but can be used for the free
ltration rate (or the permeability):
w0
(1 s ) a
(9)
wSt
where wSt is the average Stokes fall velocity
(computed for a particle with the mean size d 50 ).
The settling rate becomes zero at the maximum
concentration max , which is the maximum packing
fraction (max , 1; e.g. for ne sand it has a value
of 0:64). A best t to the data in Fig. 1 is found
for a 4:9.
For Rep  1 the following semiempirical equation has been proposed by Barnea & Mizrahi
(1973):
w0
(1 s )2

1=3
wSt (1 s ) exp[5s =3(1 s )]

(10)

Curves obtained from these two equations are also

shown in Fig. 1. If equation (9) or (10) were used


for the settling rate, it would predict a solid bed
without pores as the nal situation, which is impossible. Few of the proposed relationships full
the condition of a zero settling rate value at the
maximum compaction volume fraction max and
can be used to represent the settling rate. One of
the exceptions is the theoretical equation of
Brinkman (1947), even though it was aimed at
estimating the permeability. It generally underpredicts the settling rate outside the dilute region. The
fourth-order polynomial proposed by Shannon et
al. (1963) unnecessarily introduces too many empirical parameters. The following simple empirical
function can be proposed as an alternative to
describe the settling rate:


ws
s
(11)
exp(s =1 ) 1
wSt
max
with max 0:642 and 1 0:27. Fig. 1 shows
that this equation gives a good t over the total
range of concentrations. But, if equation (9) is
chosen to describe the free ltration rate, it is
better to propose an equation for the settling rate
based on this equation, because at low concentrations ws should equal w0 . The following equation
can be proposed:
ws w0 [1 (es =max 1 ) b ]

(12)

A value for the exponent b of 20 gives a good t


(Fig. 1). Higher values may be possible too, but

714

TOORMAN

the best-tting value cannot be determined, because


of the lack of experimental data for validation in
the concentration range near max.
The sediment ux S ws s as a function of s
is presented in Fig. 2. This ux curve can be used
for the analysis of the presence and magnitude of
density discontinuities in a settling column with a
homogeneous initial concentration (Kynch, 1952;
Fitch, 1983; Auzerais et al., 1988). The presence
of a discontinuity is found graphically by drawing
a straight line from the point corresponding to the
initial condition to the point of maximum compression. When this line lies below the ux curve a
discontinuity exists. When the line intersects the
ux curve the concentration below the discontinuity is lower than max and is dened by the contact
point of a new line which is tangential to the ux
curve at the other end (Fig. 2). In reality much
turbulent energy is required to obtain a uniform
suspension over the total water column, introducing
diffusion, which is not considered in the theoretical
analysis of discontinuities. But, when diffusion is
negligible (e.g. when s . 0:15 in the case of
Shannon et al., 1963), the sedimentation of rigid
particles can be solved using the analytical method
of Kynch (1952).
Permeability of cohesive, deformable particles.
From the data analysis of many settling-column
experiments on clay and estuarine muds, carried out
at the Hydraulics Laboratory of the Katholieke
Universiteit Leuven, it was concluded that a negative exponential relationship between k and the

excess density re ( bulk density r minus uid


density rw ) provides the best simple t within the
range of measured concentrations (Berlamont et al.,
1993b; Torfs et al., 1996). Results for kaolin are
presented in Fig. 3. A simple analytical function
which ts the semilogarithmic empirical data and
satises the limiting conditions is:


s
ks
1 w0 wSt exp(s =1 )(1 s )
w
(13)
where 1 is an empirical parameter and wSt is the
representative Stokes fall velocity. In Fig. 3 experimental data for the permeability for kaolin are
tted with this equation. These experimental values
have been obtained from settling-column data
using equation (7) for four different constant values
of at different times (Huysentruyt, 1995). The
data are presented in terms of excess density re,
rather than void ratio, because re is proportional to
the solids volume fraction and it is directly obtained from measured densities without requiring
the knowledge of the dry density (Toorman, 1996).
Figure 3 shows that the experimental data for
the permeability of cohesive sediments can also be
tted using the RichardsonZaki relationship,
equation (9), but the value of the power exponent
a must be taken as about one order of magnitude
higher than for non-cohesive particles. However,
equation (9) (and (13) likewise) yields unsatisfactory results in numerical simulations, i.e. it is
observed that the consolidation is generally slower

0.08
0.07

Relative settling flux

0.06
0.05
0.04
Equation (9)
(Richardson & Zaki, 1954)

0.03
0.02

Equation (12)
Equation
(11)

(Brinkman, 1947)
0.01

Equation (10)
(Barnea & Mizrahi, 1973)

0
0

0.1

0.2

0 .3

0.4

0.5

0.6

0.7

0.8

0.9

1.0

Solids volume fraction

Fig. 2. Settling ux, relative to the representative Stokes velocity, as a


function of volumetric concentration for non-cohesive (spherical)
particles, corresponding to Fig. 1 (same date and legend). The arrow
indicates the theoretical discontinuity path from an initial concentration
s 0:15 to the corresponding bed surface concentration (s  0:45)

715

SEDIMENTATION AND SELF-WEIGHT CONSOLIDATION

0.1
0.01

Permeability: m/s

0.001

w0

0.0001
1025
1026
1027
1028
0

50

100

150

200

250

300

350

400

Excess density: kg/m3

Fig. 3. Experimental data for permeability as a function of excess density


re for Scheldt river mud (m) and for China clay obtained from settlingcolumn tests with different initial densities (h, 1040 kg=m3 ; j,
1095 kg=m3 ), tted by semiempirical relationships and their corresponding free ltration rates (full line, RichardsonZaki relationship, equation
(8); dashed line, equation (11) with max 1; dotted line, possible true
correlation)

than predicted by equation (9) (Fig. 4), which


means that the permeability is overpredicted.
Another method to estimate permeabilities is to
start from the settling rates obtained by application
of the theory of Kynch (1952). There is a traditional misconception that this method is only ap-

plicable to sedimentation and not to consolidation.


However, bearing in mind that the settling rate
ws w0 (1 @ 0 =@ 0 ), when diffusion is neglected, and assuming that this whole expression is a
function of concentration alone (Kynch's hypothesis), Kynch's method can be applied to a consoli-

2.0
1.8
1.6

Height: m

1.4
1.2
1.0
0.8
0.6
0.4
0.2
0
0

1000

10000

100000

1000000 10000000 100000000

Time: s

Fig. 4. Comparison between experimental () and computed (lines) settling


curves for China clay (settling-column test 32 from Huysentruyt (1995); initial
density 1095 kg=m3 ): nite-element solution without (dashed line) and with
(full line) taking into account the additional decrease of permeability

716

TOORMAN

dating sediment layer as well. Data analysis by


Toorman & Huysentruyt (1997) demonstrates that
a corresponding constitutive relationship between
effective stress gradient and concentration is quite
acceptable. The reason why the theory of Kynch is
not strictly valid in consolidation is the fact that
diffusive effects are not negligible for cohesive
sediments (Auzerais et al., 1988). However, comparison of bulk densities at the interface computed
using this analytical method with measured values
generally shows an underprediction of the surface
density up to only a maximum of 2%, which
indicates that the diffusivity is small over a large
range of concentrations. Therefore, the curve of
settling rate versus concentration estimated with
Kynch's theory is representative of the true trend
and contains valuable information.
As can be seen in Fig. 5 (curve e), the settling
rate as a function of concentration, estimated with
the theory of Kynch (1952), always shows a typical
drop to lower values over a certain intermediate
concentration range. This cannot be attributed to
the effective stress because that affects the settling
rate only towards the highest concentrations, i.e. it
controls the downward bend for high densities, as
can be concluded from the data analysis by
Toorman & Huysentruyt (1997). Close examination
of the permeability data (Fig. 3) does not exclude
the possibility of the same trend of a drop at
intermediate densities as in the settling rate, since
over a certain range the majority of the data points

lie below the curve dened by equation (9). The


true variation cannot be distinguished with certainty as it is concealed by the relatively large
experimental errors. Recent, more accurate data,
obtained for Scheldt river mud (Fig. 3), seem to
conrm the trend of the permeability as a function
of concentration, indicated by the dotted line in
Fig. 3. Fig. 4 shows that the drop in permeability,
as suggested by the application of Kynch's theory,
is a necessity to obtain a correct prediction of the
settling curve.
The most likely physical explanation for the
trend in these data is related to structural changes
within the particles. One should bear in mind that
the `particles' in cohesive sediment suspensions are
actually aggregates (or ocs) of the primary clay
particles, the size, structure, density and strength of
which depend on the hydrodynamic conditions
(shear forces), the local concentration (collisions)
and bio- and physico-chemical factors. Grain size
analysis of the kaolin of Fig. 3 indicated that 75%
of the particles were smaller 2 m. Extrapolation
of the settling rate in Fig. 1 indicates that the
equivalent diameter of the kaolin aggregate is of
the order of 10 m. The true size is likely to be
even larger as the aggregates are often open-spaced
and contain a signicant amount of immobilized
pore water. One can easily imagine that these large,
open-structured, weak ocs break up into smaller
ones owing to interaction with neighbouring particles and shear by displaced water. Immobilized

0.1
0.01

Test
a
b
c
d
e

Settling rate: m/s

0.001
0.0001

w0

Initial density: kg/m3


1004
1019
1025
1098
1095

Initial height: m
0.25
0.25
0.25
0.25
2.0

25

10
10

26

1027
1028
1029
e

10210
0

50

100

150

200

250

300

350

400

Excess density: kg/m3

Fig. 5. Experimental data for settling rates as a function of excess density


re for China clay, obtained from settling-column tests with different initial
conditions. Settling rates computed from experimental settling curves with
Kynch's sedimentation theory. Dashed lines, possible permeability relations
from Fig. 3. Notice that the `drop' in permeability and in settling rate
(both for data set `e') occur at the same excess density

SEDIMENTATION AND SELF-WEIGHT CONSOLIDATION

pore water is released during this process. The


same happens under the increasing load of particles under which a oc is buried during the
deposition process; oc pore water can be squeezed
out. The resulting smaller, stronger, more compact
aggregates form a more compact bed with lower
permeability. Consequently, the permeability of a
cohesive sediment bed is not a unique function of
concentration, but also depends on the occulation
and bed formation history and the depth of burial
(i.e. the load or the effective stress). This thixotropic behaviour is typical of cohesive sediments.
Consider rst the original large aggregate particles in a suspension and assume they are incompressible. This situation can be treated in the same
way as a non-cohesive sediment, where the maximum compaction volumetric concentration max
would be very low. Contrary to non-cohesive sediment particles, the aggregate particles in cohesive
sediment suspensions are deformable. Hence, the
settling rate on a xed bottom of a cohesive
particle, which is deformed under the load of other
particles, is not zero, as the centre of gravity
moves downward during the deformation. Consequently, the maximum compaction concentration
increases and the settling-rate reduction, related to
incompressible compaction, is decreased by the
contribution of the particle deformation rate. This
can be translated mathematically by making the
maximum compaction concentration a function of
parameters still to be determined (probably effective stress and aggregate structure). Also, the parameter wSt is expected to vary, depending on the
oc density and equivalent diameter. This suggests
that a structural model may have to be added,
similarly to the treatment of the ow behaviour of
thixotropic dense suspensions (Toorman, 1997).
Another observation, for which an explanation
should be sought in the thixotropic properties, is
that application of Kynch's theory to settling-curve
data yields very different relationships between
settling rate and concentration for different tests on
the same material (Fig. 5). Tests with different
initial concentrations show that the settling rate at
a certain concentration increases with increasing
initial concentration. Flocculation is a time-dependent process; for each condition of concentration
and shear stress an equilibrium oc size exists.
The initial mixtures were obtained by mechanical
mixing, i.e. in an environment with high shear
rates. Consequently, the initial ocs can be expected to be compact and of high density and
strength. In the settling column, shear rates are
negligibly small and the concentration is the dominant controlling parameter. The initial ocs have
entered another environment, in which their structure is no longer in equilibrium. For each data set,
the rapid decrease of the settling rate with increasing concentration near the initial value suggests

717

that aggregation takes place. According to Stokes'


law, the settling rate can only decrease if the
aggregate density decreases faster than the square
of the aggregate size, i.e. open-structured aggregates must be formed. For instance, in Fig. 5, the
initial concentration for data set `c' is approximately the same as for the last data point from `a'.
The difference in oc structure at this concentration leads to a completely different history and bed
structure.
Other results show that in a short column the
settling rate at a certain concentration is lower than
in a long column (Fig. 5). The effect of the initial
height can be understood by considering a bottom
layer in the long column of the same thickness as
the total height of the sediment bed in the short
column. Initially the concentration is constant over
the total height and equal in both columns; hence,
the settling rate initially is everywhere the same,
except at the non-deforming bottom, where the
particles are stacked. Here a bed is being formed
and the bed interface rises. This will be identical
in both columns until the bedsuspension interface
reaches the height of the watersuspension interface in the short column. In the short column the
bed will consolidate without any change of the
load at the waterbed interface. At the same height
as this interface in the short column, the load will
increase in the long column as the bed continues
to grow. The consolidation of the bottom layer will
differ from here on from that in the short column.
Consider now a layer of equal density and equal
structure at the same height in both columns. In
the short column this layer will be subject to a
lower load than in the long column. The chance
that the aggregate structure will yield in the long
column, thereby releasing immobilized pore water,
is larger. Particularly, for the bottom aggregate
particles, it can then be understood why the settling rate goes to zero at much lower densities in
the short column compared to the long column.
It should be stressed that thixotropic effects are
expected to play an important role mainly in selfweight consolidation of cohesive sediments, following sedimentation, where load increments are large
compared to the maximum load and where, consequently, large variations in soil structure occur
within a relatively short time. This conclusion is in
accordance with that of Imai (1981), who postulates that `the difference in the fabric may become
less remarkable with increasing stress, and the
fabric may be equalized.'
Effective stress
The (vertical) effective stress 9 is, by denition, the difference between the total stress ,
obtained by integration of the density prole from
the level considered up to the surface, and the

718

TOORMAN

measured pore water pressure uw . Physically, the


effective stress on a certain layer corresponds to
the effective load carried by this layer, i.e. the
fraction of the total weight of all the particles
above this layer which is self-supported. Traditionally, the effective stress is assumed to be a function
of concentration alone. This is the result of geotechnical tests on uniform samples under high
loading conditions (e.g. oedometer). In self-weight
consolidation it is observed that the effective stress
increases gradually with depth from zero at the
sediment surface, independent of its non-constant
(for increasing time) concentration, to a maximum
value at the bottom. Furthermore, the effective
stress in self-weight consolidation is subject to the
constraint that it can never be larger than the
buoyant weight of the sediment above (Toorman &
Huysentruyt, 1997). This limiting condition is
reached when the sediment layer is in equilibrium,
i.e. when primary consolidation has stopped.
For sediment beds consisting of non-cohesive
monodisperse particles, uniform equilibrium density proles are obtained (see below). The effective
stress then equals the buoyant stress and is a linear
function of depth in this case. Experimental data
indicate that there is no noticeable diffusive effect
in this case when the initial volumetric concentration is above 0:15 (Shannon et al., 1963). The
settling rate ws w0 [s 1=(s w )@ 9=@z] is
found to be a function of concentration alone and
the theory of Kynch (1952) can be applied. Hence,
according to equation (1), not the effective stress,
but its gradient must then be a function of concentration alone.
For a cohesive sediment, however, diffusive effects are clearly present (e.g. Auzerais et al.,
1988). This implies that for these materials there
should be a second contribution to the effectivestress term which represents the resistance to the
deformation of the compressible bed structure.
These two considerations form the physical
background required to write the effective-stress
term as the combination of a settling-reduction ux
and a diffusive ux (Toorman & Huysentruyt,
1997), such that the total sediment ux S during
consolidation becomes
w0 @ 9
s w @z


S ws s w0 s

@ 9
w0 s 1
@ 0

w0 s (1 E) De

@s
@z

(14)

where E(s ) is the effective-stress settling-rate reduction factor, which is related to the interparticle
contact; De (s ) is the effective-stress diffusivity,
which is related to the deformation of the weak

structure in occulated soils. The parameter De


corresponds to an inverse compressibility coefcient. The last equality of equation (14) suggests
that this diffusivity is proportional to w0 .
This new formulation complicates the problem
of the determination of a closure equation for the
effective stress, because now two relations must be
found. Theoretically the necessary information can
be obtained from the local equilibrium condition
(S 0), which now becomes




@s
w0 s (1 E)
Geq (s )

De
@z S0
S0
(15)
This equation applies not only to the bottom, but
also to the equilibrium bed density prole, reached
at the end of primary consolidation when all the
excess pore pressures have dissipated. By rearranging the total sediment ux balance (equation (14)),
the value of E can be estimated for any point of a
density prole, provided that ws , w0 (or, equivalently, the permeability k, according to equation
(2)) and Geq are known, from


ws
1 @s 1
E(s ) 1
1
(16)
Geq (s ) @z
w0
It is a disadvantage that calibration of this equation
requires a consolidation test which is continued
until all pore pressures are dissipated, which for
some materials, such as estuarine dredged material,
can take several months. This would mean that
predictions of self-weight consolidation of such
materials could only be made on the basis of tests
that lasted as long as the eld situation. The test
duration may be reduced by accelerated consolidation in a centrifugal set-up.
On the other hand, when the parameters w0 (or
k), E and De as a function of concentration are
known, integration of the equilibrium condition
(equation (15)) allows the calculation of the equilibrium density prole (and, consequently, the nal
bed height) without solving for the transient behaviour (equation (1)), provided that the total mass
per unit area is known.
Surface densication
Creep is often dened as the densication at a
constant effective stress (e.g. Sills, 1995). The
increase of the density at the soil/water interface,
as observed in the experimental proles shown
here for both rigid particles (Fig. 6) and compressible aggregate particles (Figs 7 and 8), fulls this
condition, because the effective stress at the bed
surface is always zero. The case of rigid spherical
particles (Fig. 6) suggests that what actually happens is that the accumulation rate of depositing

719

SEDIMENTATION AND SELF-WEIGHT CONSOLIDATION

0.35

Height: m

Time: s
0.3

10

0.25

40

0.2

70

0.15

100

0.1

130

150

170

200

0.05
0
0

100

200

300

400

500

600

700

800

900

1000

Excess density: kg/m3

Fig. 6. Computed density proles for spherical glass beads. Dashed line,
analytical solution (Kynch's sedimentation theory); full line, niteelement solution. Initial height H 0 0:312 m; initial solids concentration 0 0:15; D 104 S; diffusivity limit Dlim 1013 @e =@z
2.0
1h
1.8
6h

1.6

Height: m

1.4
1.2
1.0

24 h
67:25 h

0.8

211 h

0.6
812 h

0.4
0.2
0
0

50

100

150

200

250

300

350

400

Excess density: kg/m3

Fig. 7. Comparison between experimental (dashed lines) and computed


(full lines) density prole evolution for a China clay suspension.
Experimental data from Huysentruyt (1995). Initial height H 0
1:998 m; initial density r0 1095 kg=m3 (0 0:06)

particles is faster than the soil matrix arrangement


rate of the particles. This means that this apparent
`creep' phenomenon is the net effect of two processes with different timescales. As long as the
compaction is not completed, the surface particles
are balanced between gravity and the drag forces
of pore water expelled upwards from the consolidating bed below.
Equation (3) shows that densication occurs as

long as @ 9=@ 0 , 1 (assuming D B 0). This is


the major reason for surface densication. Again,
the same apparent problem has arisen here as for
the interpretation of the free ltration rate, outlined
at the beginning. Even though the effective stress
at the surface is constant, the effective-stress gradient is not. The effective-stress gradient will increase until @ 9=@ 0 1, which is the local
equilibrium condition. It must be concluded that

720

TOORMAN

0.7

0
0:12 h

0.6

0:20 h

Height: m

0.5

0:36 h
1:45 h
2:33 h

0.4

12:06 h

0.3

104 h
0.2

334 h

0.1
0
0

50

100

150

200

250

300

Excess density: kg/m3

Fig. 8. Comparison between experimental (dashed lines) and computed


(full lines) density prole evolution for a natural mud suspension.
Experimental data from test RB09 of Bowden (1988) (redrawn after
Bowden (1988)). Initial height H0 0:676 m; initial density r0
1060 kg=m3

surface densication is not a pure creep phenomenon. In the case of non-cohesive particles no
creep is involved at all. A new denition of creep
is required. True creep, as observed in certain soils
after excess pore pressures have dissipated, should
be related to biochemically induced structural
changes, probably driven by non-equilibrium gradients of the corresponding properties. It is very
likely that during primary consolidation true creep
already takes place (Sills, 1995). Hence, for cohesive sediments part of the surface densication
may be attributed to creep.
NUMERICAL IMPLEMENTATION

Coordinates and remeshing


The mass balance equation, equation (1), has
been expressed in a xed Eulerian reference frame
because it is solved over the non-deforming domain of the total water column including the consolidating bed. This is particularly of interest when
this model is implemented in a three-dimensional
numerical model for sediment transport, where the
water column remains part of the computational
domain. In geotechnical models, material or advective coordinates are generally employed and the
computational domain is restricted to the deforming saturated sediment layer (e.g. Schiffman et al.,
1985). The interrelationship between the different
coordinate frames has been summarized by
Toorman (1996).
Software has been developed for the solution of
the mass balance equation in an Eulerian coordi-

nate frame in which the nodes are not necessarily


xed (resulting in a special type of advective
coordinate). In order to account for the nodal
displacement, the sediment balance equation has to
be modied, corresponding to a mixed Euler
Lagrange approach (e.g. Huerta & Liu, 1988), to
@s
@s
c
@t
@z


@
wo @ 9
@s

w0 s
DB
@z
s w @z
@z

(17)

where c is the nodal mesh velocity, which is


approximated as the difference between the new
and the old coordinate, divided by the time step of
the numerical scheme. The nodal displacement
must be imposed. There are several options.
The advantage of material coordinates is that
the mass distribution over the internodal layers
remains constant. This corresponds to computing
the new nodal coordinate (at each iteration step)
such that the total buoyant mass of the sediment
above this point remains constant:
0,i 0,i1
z i z i1
(18)
(s w )(s,i s,i1 )=2
where the index i refers to the node number. The
constant value of the buoyant weight 0 at each
point is dened by the initial mesh and density
prole.
The model has been tested with different parameter sets. In principle, the mixed EulerLagrange
approach does not require as many nodes as for a

SEDIMENTATION AND SELF-WEIGHT CONSOLIDATION

xed grid (c 0), but the denition of the initial


mesh is not straightforward, because there should
be enough renement around density steps (i.e.
large density gradients such as the interface).
Furthermore, it is found that the convergence rate
is slowed down by the remeshing. A nal disadvantage of the mixed method is that the computational domain must be restricted to the sediment
layer. Solution on a xed, uniform, ne grid produces identical results with fewer convergence
problems. Therefore, the latter is preferred, particularly for one-dimensional problems, even when
more nodes are required.
Notice that when c is taken equal to the sediment velocity vs ws, a formulation is obtained
which is fully equivalent to the one of Gibson et al.
(1967) (see Toorman, 1996). Equation (1) should
then no longer be solved in the Eulerian frame,
because it would be dangerous to remesh with
z ws t, since ws is not constant but decreases
during t and this could lead to `overtaking' of
nodes and subsequent mesh distortion. Instead, one
must work in the reduced coordinate system and
with the void ratio as the independent variable to
reduce the non-linearity (see Toorman, 1996). This
method, however, is only applicable if the computational domain is restricted to the sediment.
Numerical solution method and stability
In the present work, the equation was solved by
the nite-element method, using linear interpolation
functions and a rst-order implicit time-stepping
scheme. Several iterations were carried out at each
time step because the equation is non-linear owing
to the dependence of the permeability and the
effective-stress gradient on the concentration and/or
its gradient. The time step t was increased at
each time interval by multiplication with a factor
1:1 in order to compensate for the slow-down of
the consolidation process, i.e. to speed up the
computation. However, this factor was automatically adjusted during the computation depending
on the convergence behaviour, i.e. when the convergence criterion was not fullled within 20 iterations, the time step was temporarily reduced again.
The high density gradient areas caused serious
slow-down of the convergence rate, which is controlled by the minimal value of the diffusivity
occurring.
The numerical model solves equation (1) in
terms of the excess density re ( r rw ) in order
to allow direct comparison with the measured
density proles without needing knowledge of the
value of the intrinsic density rs of the solids. In
order to get the computation of an initially homogeneous mixture started, a small concentration
gradient was created in the top and bottom elements, by subtracting and adding 1% of the initial

721

uniform value in the top and bottom nodes, respectively.


The grid generator allowed the construction of a
block-structured grid. Each block could automatically be rened towards its top and bottom. The
sediment layer could be subdivided into sublayers
such that the contact surfaces of the sublayers
corresponded to the expected levels of high concentration gradients. For reasons explained above,
a uniform grid was used for the results presented
here. The program discarded elements at the top
where the concentration was below a user-specied
threshold value (here 1020 kg=m3 ), in order to
reduce the computation.
The mudwater interface in reality corresponds
to a discontinuity in the density prole, which
cannot be obtained numerically. Instead there is a
rapid density increase over the elements around the
interface location, which can be stabilized by the
diffusivity. Physical grounds for the introduction of
a diffusion term in the suspension phase have been
presented by Toorman (1996). When the diffusivity
is too small, as is the case for the physical
diffusivity in the supernatant water, the high density gradients cause overshoot and subsequent oscillations in the density prole. A minimum value
of the diffusivity is required to damp these oscillations. Since the required damping is proportional
to the mesh size, the element length should be
small enough that the effect of the added numerical
diffusion is negligible.
Additional stability was obtained by implementing a simplied streamline-upwind PetrovGalerkin
scheme. This method adds a second term, proportional to the spatial derivative of the shape function, to the nite-element weight function (Brooks
& Hughes, 1982). The proportionality factor (the
advective velocity times the so-called intrinsic
time) is a function of the Peclet number (Pe, the
element Reynolds number). A practical problem is
that the high degree of non-linearity of the local
advective velocity tends to destabilize the solution
procedure. A simple method to circumvent the
computation of Pe and the global advective velocity is to assume that Pe is very high over the total
column. The proportionality factor then reduces to
the element length. Upwinding techniques generally produce numerical diffusion. Its relative magnitude has not been quantied in this case, but the
model results seem to suggest that it is unimportant. The high-Peclet-number approximation seems
to work very well as long as the element size is
small enough. This method allows much steeper
interfaces to be simulated.
The determination of the exact location of the
interface in the numerically simulated density proles may at rst sight seem a bit difcult since the
simulated surface is never perfectly horizontal (no
discontinuity). The most accurate denition of the

722

TOORMAN

interface location was obtained by taking the coordinate corresponding to the rst maximum in the
density gradient, starting from the top of the
column.
Boundary conditions
In theory two types of boundary conditions can
be imposed for the surface as well as for the
bottom. The rst type is a known sediment ux,
i.e. a natural or Neumann condition. This condition
is the best one to use at the bottom, where the
sediment ux is zero. However, careful consideration should be given to the physical meaning of
this condition. The settling rate traditionally is
supposed to be a decreasing function of concentration and becomes zero only at the maximum
compaction concentration. Hence, the downward
(gravitational) ux can never be zero at lower
concentrations. Nevertheless, there is no ux
through the bottom. Hence, it must be balanced
exactly by the upward diffusive ux.
At the surface, a zero sediment ux condition
allows the simulation of the formation of the interface in a natural way, again balancing the diffusive
ux. This condition should be applied when using
a xed Eulerian grid. In can also be applied when
using material coordinates. In both cases the sharp
density gradients at the sedimentwater interface
may lead to numerical problems (overshoot and
oscillation). A non-zero sediment ux should be
applied when sediment enters the domain from the
surface boundary, e.g. in the case of mud-dumping
in a conned disposal site.
The second type of boundary condition is a
known sediment concentration, i.e. an essential or
Dirichlet condition. This boundary condition is
often used for the surface in a soil-mechanical
approach in a reduced coordinate frame. The use
of a Dirichlet condition at the top avoids the
overshoot problem. However, as densication occurs at the surface, this boundary condition
should not be used. For the same reason, this
condition generally cannot be used at the bottom.
The only physically correct boundary condition is
a zero sediment ux condition at both ends.
However, if one imposes the condition that the
bottom concentration is the maximum value, then
a true discontinuity must be considered. This is
difcult to handle numerically. Lee & Sills
(1981) apply this condition in their simplied
analytical solution for consolidation on a pervious
bottom.
In the case of bottom drainage, the uid ux
through the bottom has to be imposed as well,
which can be done through the additional term
which then occurs in the mass balance equation
(equation (27) of Toorman, 1996). This case will
not be considered here.

MODEL VALIDATION

Sedimentation of non-cohesive sediment


The proper behaviour of the model can be
demonstrated by the simulation of a nearly monodisperse non-cohesive suspension. A data set of
well-documented settling-column experiments by
Shannon et al. (1963, 1964) with spherical glass
beads (normally distributed particle size, mean diameter 66:49 m, standard deviation 6:70 m,
effective mean sphere diameter d e 66:94 m,
solids density rs 2450 kg=m3 ) was used. The
settling rate as a function of volumetric concentration was obtained experimentally from the initial
constant-rate settlement and is presented in Fig. 1.
The maximum bed concentration max 0:642 is a
little higher than the theoretical concentration for
randomly packed spheres ( 0:61) but smaller than
the maximum packing fraction ( 0:74).
The analysis by Shannon et al. (1963, 1964)
suggests that the diffusive ux for this material is
negligible when the initial volume fraction is at
least 0:15, as in the case selected here. Consequently, the experiments can be simulated analytically using the theory of Kynch (1952). For lower
initial concentrations diffusive effects have been
observed, which can be attributed to turbulence in
the suspending uid, generated during the mixing
to homogenize the suspension.
The experimental data for the settling rate as a
function of the concentration can be tted by just
one curve of the form of equation (11) or (12). As
shown above (see equation (14)), in the absence of
diffusivity (De 0), the settling rate consists of a
contribution w0 and a second term w0 E, which
represents the reduction due to interparticle contact. This effective-stress contribution is found as
the difference w w0 , where w0 is assumed to be
given by equation (9). Comparison of equations
(12) and (14) gives the following expression:
E (e=max 1 ) b

(19)

Therefore, equation (12) is preferred over equation


(11), as it has a more sound physical basis. The
`doubly concave' sediment ux curve, claimed by
Shannon et al. (1963), can be obtained for very
large values of the parameter b. Note that the use
of the RichardsonZaki relationship, equation (9),
would lead to a serious overprediction of the bed
density, since it implies that settling only stops
when the maximum compaction volumetric fraction
equals 1.
Even though diffusion is negligible, it is necessary to add some for numerical-stability purposes.
The articial diffusivity was chosen to be proportional to the sediment ux, with a limiting minimal
value of the order of 108 m2 =s. Even better
results, particularly at the sediment-water interface,
are obtained by dening the limiting value to be

SEDIMENTATION AND SELF-WEIGHT CONSOLIDATION

proportional to the concentration gradient. The


typical evolution of density proles is shown in
Fig. 6. Comparison of the density proles computed using the analytical method based on the
theory of Kynch (1952) with the numerical solution
shows the effect of the added diffusivity. Thanks to
the upwind stabilization method the model allows
the simulation of very at interfaces, which was
impossible in previous simulations (e.g. Toorman
& Berlamont, 1993).
Sedimentation of cohesive sediment
Two data sets of laboratory sedimentation and
self-weight consolidation column tests have been
used to validate the model for cohesive sediment
suspensions. The rst test was carried out in
Oxford on a natural mud from the River Parrett
estuary, designated Combwich mud (Bowden,
1988); the second was on a kaolinite suspension
(China clay from ECC Int. Ltd, St Austell, UK;
73% , 2 m), carried out in our laboratory
(Huysentruyt, 1995). The density proles of these
two sediments show a very similar evolution (compare Figs 7 and 8). The intermediate `step' in the
density prole, which corresponds to the initially
growing consolidating layer, occurs only under
specic initial conditions (see Auzerais et al.
(1988), for a discussion).
The empirical relationship for the diffusivity De
was estimated from the equilibrium density prole.
Only for the kaolin test was this prole available.
A satisfactory t of this prole is given by the
relationship
s
1 1 ln(1 z= H 1 )
(20)
max
where H 1 is the nal bed thickness and an
empirical constant which controls the slope steepness of the prole. By applying equation (15), one
nds
De w0 (1 E)

s
H 1 exp[(1 s =max )]
max
(21)

The same relationship, with different parameter


values, was used for the simulation of the Oxford
experiment. This assumption seems to be justied
by the fact that the same shape of the equilibrium
density prole reappears in other published data
for cohesive sediments (e.g. Imai, 1981; Been &
Sills, 1981; Sills, 1997). A similar empirical prole, described by
n (n0 n1 )ez n1

(22)

(where the index 0 refers to the surface and 1 to


the asymptotic value at innite depth, and z is the
depth), ts many sets of eld data for sea-beds and

723

mudats and is widely used as a steady-state


porosity equation in the study of near-surface
biochemical processes (e.g. Rabouille & Gaillard,
1991). A theoretical justication for the latter
equation has recently been proposed by Boudreau
(1999).
The value of De , given by equation (21) turns
out to be very small over a wide range of concentrations. Therefore, in both cases, the free ltration
rate w0 was assumed to equal the settling rate
obtained from applying Kynch's (1952) theory to
the settling curve. For the Oxford data this was
compared with settling rates computed for each
point of the experimental density proles using the
method of Galvin & Waters (1985), which is also
based on Kynch's theory. The two results are very
similar (Fig. 9). In order to account for the drop in
the permeability, the parameter wSt was multiplied
with an empirical correction factor of the form
1 =[1 exp(s =2 )], which was tted to
the data (2 , and are empirical constants: is
the fraction by which the apparent Stokes velocity
wSt decreases; it can be shown that 2 ln determines the concentration where the transition occurs
and 2 =4 the slope of the transition). Generally,
the parameters of this correction factor depend on
the occulation history.
As a result of the extremely low diffusivity for
concentrations well below the maximum, a temporary density peak occurs at the bottom, which
slowly disappears as the bottom density approaches
the maximum value. On the contrary, density peaks
formed by segregated material often remain, as is
known from experiments and which can be numerically simulated by considering two sediment fractions (Toorman & Berlamont, 1993). Hence,
interpretation of the bottom peak requires care.
The time origin for the Oxford data had to be
shifted by 95 min to the point where the constantrate settling started in order to skip an initial
period where only very little settlement of the
surface could be observed. This stabilization period
may be explained by the formation of continuous
drainage paths (Michaels & Bolger, 1962) and
possibly changes in aggregate structure due to the
new environment with smaller shear stresses than
in the mixing tank in which the initial homogeneous suspension was prepared. Imai (1980) calls
this the occulation stage.
Comparison of computed and measured density
proles (Figs 7 and 8) shows that the model
qualitatively predicts the typical features of the
density proles. The deviations between model and
measurement are generally smaller than 25 kg=m3 ,
which is about twice as much as the theoretical
measurement error. The major differences are the
wider and smoother density step and a less steep
density gradient in the upper part of the consolidating bed for the simulation results. In fact, the

724

TOORMAN

0.001
0.0001

Settling rate: m/s

1025
1026
1027
1028
1029
10210
10211
0

50

100

150

200

250

300

Excess density: kg/m3

Fig. 9. Settling rate as a function of excess density for Bowden's (1988)


experiment RB09 on Combwich mud. Symbols, Kynch's (1952) theory
applied to settling curve; dashed line, Galvin & Waters (1985) method;
full line, curve t

measured density in the top layer seems to be


nearly uniform, particularly in the Oxford experiments. A sensitivity analysis shows that the currently used constitutive equations cannot account
for this constant-density zone. It could only be
explained if the permeability were to decrease with
depth (or, equivalently, with load), which implies a
variation in aggregate structure at the same density.
A check of the mass balance for the Oxford data
shows that the total mass for each density prole is
not exactly equal. Comparison of the areas of the
measured and simulated density proles in Fig. 8
shows that the largest discrepancy, an excess mass
of 2%, is found for the experimental proles at
12:06 h and 104 h. This is still within the error
bounds of the accumulated measurement error obtained by integration, as the theoretical accuracy is
0:2% per data point for the X-ray system (Sills,
1997). Taking into account the relatively large
measurement errors and the absence of such a
near-zero-gradient top layer in many other experimental data, one cannot be certain about the observed trend in the presented data and no denite
conclusion can be drawn about the quality of the
model simulation.
DISCUSSION AND CONCLUSIONS

The prediction of sedimentation and consolidation requires the knowledge of closure relationships
for the free ltration rate (or permeability), effective stress and diffusivity. The traditional empirical
constitutive equations do not full all the physical
requirements. The lack of accurate measurement

techniques has contributed a great deal to the


problem of dening generally valid closure equations. New techniques and data-processing methods
should be developed
Nevertheless, on the basis of the shortcomings
from a theoretical point of view, new semiempirical
relationships have been proposed. The well-known
RichardsonZaki relationship turns out to be a
good approximation for the permeability of nondeformable particles. The effective-stress term in
the mass balance equation has been split into two
contributions. The rst represents the support of
particles and the second, which seems only to be
present for cohesive soils, the resistance of the
deforming occulated network structure. The performance of these new closure equations in the
numerical model is very promising, since they
allow very realistic simulations of settling-column
experiments. Experimental data of better quality are
required to verify or improve these relationships.
It has been concluded that the permeability for
cohesive sediments is not a unique function of the
concentration. Differences in the results obtained
for different initial concentrations or suspension
heights can be explained in terms of the history of
the aggregate structure. A structural-kinetics type
of equation, such as used in thixotropic shear ow
(Toorman, 1996) or occulation in turbulent ow
(Winterwerp, 1998) should be added. Further study
is required.
It is important to realize that the simulations
presented here for cohesive sediments should not
be considered as predictions, because the closure
relationship for w0 was obtained from application

SEDIMENTATION AND SELF-WEIGHT CONSOLIDATION

of Kynch's theory to the experimental settling


curves corresponding to each of the two cases. As
long as the constitutive equation for permeability
is not generalized to include structural changes
(thixotropy), accurate true predictions will not be
possible.
It has been shown that surface densication does
not need to be explained as creep, with which it
has been confused as a result of an unfortunate
denition of creep. The model results show that
surface densication is indeed predicted, even with
Kynch's theory.
The sedimentation equation can be solved in a
xed Eulerian coordinate frame with the computational domain including the supernatant clear
water. The only correct boundary conditions for
the top and bottom of the computational domain
are natural boundary conditions, i.e. a known sediment ux. The simulation of the high density
gradients requires special numerical techniques to
stabilize the solution. This was done here by using
a simplied form of the streamline-upwind Petrov
Galerkin nite-element method. The simplication
consists of a high-Peclet-number approximation for
each element. From a mathematical point of view,
the numerical model performs very well: sharp
interfaces and stability are obtained and the general
features of the evolution of the density proles are
reproduced. Even for non-cohesive sediments with
negligible diffusion, and thus very steep density
gradients, the results are quite good. The major
differences between model and experiment can be
attributed to an inadequate description of the constitutive relationships and, more importantly, the
lack of quality data for validation. High-quality
data sets from sedimentation and self-weight consolidation tests are thus strongly desired.
Another problem which should be addressed is
that of polydispersity. Most natural sediments are
mixtures of different types and sizes of particles.
At concentrations below the gel point, segregation
of the coarser particles occurs. The prediction of
the relative distribution of the different fractions
with time over the depth of a deposit requires the
solution of the mass balance for each fraction.
Toorman & Berlamont (1993) presented preliminary simulation results for mixtures of sand and
cohesive sediment, including the case of a laminated bed. Additional problems arise owing to the
interaction (e.g. mutual hindrance) of the fractions
(Toorman & Berlamont, 1993). The solution for
multiple particle systems is still in an infant stage
and will be subject of future research.
ACKNOWLEDGEMENTS

The author holds the position of postdoctoral


researcher of the Fund for Scientic Research,
Flanders. Some of this work was undertaken as

725

part of the MAST-2 G8 Coastal Morphodynamics


Programme, funded partly by the Commission of
the European Communities, Directorate General
for Science, Research and Development, under
contract MAS2 CT92-0027. The experiments on
kaolin were carried out by my former colleagues
Heidi Huysentruyt and Hilde Torfs.
NOTATION
D diffusion coefcient
De compression resistance diffusivity
e void ratio
Geq equilibrium concentration gradient
g gravity constant
i hydraulic gradient
k permeability
k r reduced permeability k=(1 e)
n porosity
S solids ux
t time
u excess pore pressure
u0 hydrostatic pressure
uw uid pressure
w0 free ltration rate
ws settling rate
z vertical Eulerian coordinate
unit weight
s sediment unit weight
w water unit weight
E effective-stress settling-rate reduction factor
r suspension density
re excess density r rw
rs sediment intrinsic density
rw water density
total stress
0 buoyant stress u0
9 effective stress
s solids volume fraction

Subscripts

s sediment
w water

REFERENCES
Alexis, A., Thomas, P. & Gallois, S. (1993). Consolidation of cohesive sediments, Final Report for the
MAST-1 G6M Programme. St Nazaire: Civil Engineering Department, IUT (in French).
Auzerais, F. M., Jackson, R. & Russel, W. B. (1988). The
resolution of shocks and the effect of compressible
sediments in transient settling. J. Fluid Mech. 195,
437462.
Barnea, E. & Mizrahi, J. (1973). A generalized approach
to the uid dynamics of particulate systems. Part I:
general correlation for uidization and sedimentation
in solid multiparticle systems. Chem. Engng J. 5,
171189.
Batchelor, G. K. (1976). Brownian diffusion of particles
with hydrodynamic interaction. J. Fluid Mech. 74,
No. 1, 129.
Been, K. & Sills, G. C. (1981). Self-weight consolidation

726

TOORMAN

of soft soils: an experimental and theoretical study.


Geotechnique 31, 519535.
Berlamont, J., Ockenden, M., Toorman, E. & Winterwerp,
J. (1993a). The characterisation of cohesive sediment
properties. Coast. Engng 21, 105128.
Berlamont, J., Van den Bosch, L. & Toorman E. (1993b).
Effective stresses and permeability in consolidating
mud. Proc. 23rd Int. Conf. on Coast. Engng, Venice
3, 29622975.
Boudreau, B. P. (1999). The porosity equation for steadystate early diagenesis. Am. J. Sci., in press.
Bowden, R. K. (1988). Compression behaviour and shear
strength characteristics of a natural silty clay, sedimented in the laboratory. PhD thesis, University of
Oxford.
Brinkman, (1947). A calculation of the viscous force
exerted by a owing uid on a dense swarm of
particles. Appl. Sci. Res., A1: 2734.
Brooks, A. N. & Hughes, T. J. R. (1982). Streamline
upwind/PetrovGalerkin formulations for convection
dominated ows with particular emphasis on the
incompressible NavierStokes equations. Comp. Meth.
Appl. Mech. Engng 32, Nos. 13, 199259.
Davis, R. H. & Acrivos, A. (1985). Sedimentation of
noncolloidal particles at low Reynolds numbers. Ann.
Rev. Fluid Mech. 17, 91118.
Fitch, B. (1983). Kynch theory and compression zones.
AIChE J. 29, No. 6, 940947.
Font, R. (1988). Compression zone effect in batch sedimentation. AIChE J. 34, No. 2, 229238.
Galvin, K. P. & Waters, A. G. (1985). A novel approach
for determining the solids ux curve for continuous
thickening. In Flocculation, sedimentation and consolidation (eds B. M. Moudgil and P. Somasundaran),
pp. 363372. Engineering Foundation, New York.
Garside, J. & Al-Dibouni, R. (1977). Velocityvoidage
relationships for uidization and sedimentation in
solidliquid systems. Ind. Engng Chem., Process Des.
Dev. 16, No. 2, 206214.
Gibson, R. E., England, G. L. & Hussey, M. J. L. (1967).
The theory of one-dimensional consolidation of saturated claysI. Geotechnique 17, 261273.
Gosele, W. & Wambsgan, R. (1983). Thickening process
in gravity thickeners. Ger. Chem. Engng 6, 351355.
Gurel, S. (1951). PhD thesis, University of Birmingham
(Title unknown. cf Oliver 1961).
Huerta, A. & Liu, W. K. (1988). Viscous ow with large
free surface motion. Comput. Meth. Appl. Mech.
Engng 69, 277324.
Huysentruyt, H. (1995). Consolidation of mud. Settling
column tests, report for MAST2 G8M. Leuven: Hydraulics Laboratory, Katholieke Universiteit Leuven.
Imai, G. (1980). Settling behavior of clay suspensions.
Soils Found. 20, 6177.
Imai, G. (1981). Experimental studies on sedimentation
mechanism and sediment formation of clay materials.
Soils Found. 21, 720.
Kynch, G. J. (1952). A theory of sedimentation. Trans.
Faraday Soc. 48, 166176.
Lee, K. & Sills, G. C. (1981). The consolidation of a soil
stratum, including self-weight effects and large strains.
Int. J. Numer. Anal. Methods Geomech. 5, 405428.
Michaels, A. S. & Bolger, J. C. (1962). Settling rates and
sediment volumes of occulated kaolin suspensions.

I&EC Fundamentals 1, No. 1, 2433.


Oliver, D. R. (1961). The sedimentation in a suspension
of closely-sized spherical particles. Chem. Engng Sci.
15, 230242.
Rabouille, C. & Gaillard, J.-F. (1991). A coupled model
representing the deep-sea organic carbon mineralization and oxygen consumption of surcial sediments.
J. Geophy. Res. 96, 27612776.
Richardson, J. F. & Zaki, W. N. (1954). Sedimentation
and uidisation. Part I. Trans. Inst. Chem. Engrs. 2,
3553.
Shannon, P. T., Stroupe, E. & Torey, E. M. (1963). Batch
and continuous thickening. Basic theory. Solids ux
for rigid spheres. I&EC Fundamentals 2, No. 3,
203211.
Shannon, P. T., Dehaas, R. B., Stroupe, E. & Torey, E. M.
(1964). Batch and continuous thickening. Prediction
of batch settling behaviour from initial rate data with
results from rigid spheres. I&EC Fundamentals 3,
No. 3, 250260.
Schiffman, R. L., Pane, V. & Sunara, V. (1985). Sedimentation and consolidation. In Flocculation, sedimentation and consolidation (eds B. M. Moudgil and
P. Somasundaran), pp. 57121. Engineering Foundation, New York.
Sills, G. C. (1995). Time dependent processes in soil
consolidation. Proceedings of the International Symposium on Compression and Consolidation in Clayey
Soils (eds H. Yoshikuni and O. Kusakabe), pp. 875
890. Rotterdam: Balkema.
Sills, G. C. (1997). Consolidation of cohesive sediments
in settling columns. In Cohesive sediments (eds
N. Burt, R. Parker and J. Watts), pp. 107120.
Chichester: Wiley.
Tiller, F. M. (1981). Revision of Kynch sedimentation
theory. AIChE J. 27, No. 5, 823829.
Toorman, E. A. (1992). Modelling of uid mud ow and
consolidation. PhD thesis, Katholieke Universiteit
Leuven.
Toorman, E. A. (1996). Sedimentation and self-weight
consolidation: general unifying theory. Geotechnique
46, No. 1, 103113.
Toorman, E. A. (1997). Modelling the thixotropic behaviour of dense cohesive sediment suspensions. Rheologica Acta 36, 5665.
Toorman, E. A. & Berlamont, J. E. (1993). Settling and
consolidation of mixtures of cohesive and non-cohesive sediments. In Advances in hydro-science and
engineering (ed. S. Y. Wang), pp. 606613, University of Mississippi.
Toorman, E. A. & Huysentruyt, H. (1997). Towards a
new constitutive equation for effective stresses in
consolidating mud. In Cohesive sediments (eds
N. Burt, R. Parker and J. Watts), pp. 121132.
Chichester: Wiley.
Torfs, H., Mitchener, H., Huysentruyt, H. & Toorman, E.
(1996). Settling and consolidation of mud/sand mixtures. Coast. Engng 29, No. 12, 2746.
Winterwerp, J. C. (1998). A simple model for turbulence
induced occulation of cohesive sediments. J. Hydraul. Res. 36, No. 3, 309326.
Znidarcic D. (1982). Laboratory determination of consolidation properties of cohesive soil. PhD thesis, University of Colorado, Boulder.

Você também pode gostar