Download as pdf or txt
Download as pdf or txt
You are on page 1of 10

COLLOIDS AND

Colloids and Surfaces


ELSEVIER A: Physicochemicaland EngineeringAspects 103 (1995) 1-10

SURFACES

Scale-up procedures and test methods in filtration: a test case on kaolin plant data
M. Eberl a, K.A. L a n d m a n a, p.j. Scales b.,
a Department of Mathematics, Advanced Mineral Products Research Centre, University of Melbourne, Parkville, Victoria, Australia 3052 b School of Chemistry, Advanced Mineral Products Research Centre, University of Melbourne, Parkville, Victoria, Australia 3052

Received 28 September 1994; accepted 18 February 1995

Abstract

The prediction of the filtration rate for a flocculated kaolin system in a plant filter is achieved using a compressional rheological model. This model requires two important rheological functions, namely the compressive yield stress and a hindered settling function, which relate to the compressibility and permeability of the filter cake. For our flocculated kaolin suspension, these functions are determined from laboratory experiments. The model results compare well with the data collected in an industrial scale plate and frame pressure filtration device. This allows predictions to be made from the model when the plant operations such as the applied pressure are changed.
Keywords: Compressive yield stress; Hindered settling factor; Kaolin suspension; Pressure filtration

1. Introduction

Solid-liquid separation processes, such as filtration, contiriuous thickening or batch settling operations, are important to the mineral and chemical industries. The modelling and optimisation of these processes is the focus of substantial solid-liquid separation research. The prediction of the filtration rate for flocculated systems in a plant filter is a difficult process, which has not been fully rationalised for simple laboratory systems. The aim of this paper is to apply the compressional rheological model developed by Landman et al. [ 1] to data obtained from a full scale filtration plant. The simplest models of the filtration process *Corresponding author. 0927-7757/95/$09.50 1995ElsevierScienceB.V. All rights reserved SSDI 0927-7757(95)03145-6

are the conventional engineering style models of Wakeman et al. [2], Shirato et al. [3], Terzaghi and Peck [4], and Sivaram and Swamee [5]. These models describe the flux of liquid through a uniform volume fraction of solids, and have been used extensively to model the effect of the filter resistance and bed resistance in filtration. The approach used here differs somewhat from many previous studies in that it utilises the non-linear pressure filtration model developed by Landman et al. [1]. This model requires two important rheological functions, namely the compressive yield stress and a hindered settling factor, which relate to the compressibility and permeability of the filter cake. For our flocculated kaolin suspension, these functions can be determined from laboratory experiments. Using these data, our phenomenological model was applied, and the results compared

M. Eberlet al./Colloids Surfaces A. Physicochem. Eng. Aspects 103 (1995) 1 10

with the data collected in an industrial scale plate and frame pressure filtration device, taking into account the pressure application profile and the measured filter cloth resistance. Therefore, given that the comparison between model and plant data is good, we are able to predict how a full scale filtration plant will operate under any pressure application strategy. Hence, our model allows for full scale-up.

Py(O)

suspension as individual flocs

suspension fully networked

1.1. Compressional rheological model


Connected aggregate structures of considerable size (flocs) can be formed in stable suspensions of particles when the electrostatic repulsion between particles is lowered by the addition of salts, or bridging forces are introduced by the addition of polymers. When the local volume fraction of flocs ~bexceeds a percolation or gel point, 0g, these flocs become interconnected to form a network with the properties of a solid structure. Compressive stresses on the suspension particles are transmitted throughout the network. The concept of solid stress, or particle pressure ps, is now introduced. If the solid stress is sufficiently small the network will resist compression. As the applied compressive force (e.g. a piston pressure) is increased, a point will be reached where the network will no longer be able to resist elastically, but will yield and irreversibly consolidate, owing to breaking or rearranging of bonds between particles and/or the formation of more contact points between particles. This consolidation process can be modelled using the concept of the network possessing a compressive yield stress Py(O). Py(O) can be expected to depend implicitly on the properties of the network, e.g. the strength of interparticle bonds and the shear/compression history of the system. In this model, it is assumed that Py(0) is an explicit function of q~ only. Py(O) can be expected to be an increasing function of 0 and will vanish below the volume fraction 0g at which the suspension is fully networked, as illustrated in Fig. 1. Buscall and White [-6] describe the dynamic compression of such a network as dO fO Ps < Py(O)

Fig. 1. The form of the compressive yield stress Py(~b)for a flocculatedsuspension. where dO/dt is the material derivative of the local volume fraction, i.e. the local rate of change of volume fraction as a particular network element is followed in time through the filtration process. Therefore, the network will consolidate as the local solid pressure is exceeded with the application of the external force. The other parameter appearing in Eq. (1) is the dynamic compressibility, ~c(0). If we accept the hypothesis that the hydrodynamic drainage of water from between the network structure is the rate-determining step in the consolidation process, calculations in Refs. [6] and [7] show that the dynamic compressibility is very large once Ps exceeds Py(O), for the collapse rapidly readjusts the network local volume fraction until the yield stress at that volume fraction exactly matches Ps. Hence, in this large dynamic compressibility limit, an exact knowledge of the parameter ~c(0) is not required and Eq. (1) can be replaced by ps(r,t) = Py(~b(r,t)) (2)

at any point r in the suspension at any time t. Py(O) and 0g for the flocculated kaolin suspension can be determined experimentally (Section 3.1).

1.2. Pressure filtration model


The pressure filter modelled here consists of a cylinder containing the suspension, compressed by

d~- ~K(0)(ps-P,(0)) p~>P,,(O)

(1)

M. Eberl et al./Colloids Surfaces A. Physicochem. Eng. Aspects 103 (1995) 1-10

a piston at one end and a porous membrane at the other end which only allows fluid to pass. Initially, the suspension is assumed to have a uniform volume fraction ~b o. There are two distinct cases as shown in Fig. 2.

filtration are based on the force balance of the hydrodynamic drag on the particles and pressure gradients [ 1,8]. For the one-dimensional filtration model, the solid phase satisfies
Ops

Case 1: The suspension is initially un-networked,


Here, filtration occurs in two stages as follows. Compact bed formation. Flocs are delivered and accumulate on the membrane by the fluid flux which exits through the membrane. In this compact bed, the local volume fraction ~b is non-uniform and exceeds ~g with Py(~b)> 0 up to a critical height I(t), where Py(~b)reaches zero and ~b= q~g(individual flocs above this point cannot transmit particle stress). Above l(t), individual flocs fall on the bed, and ~bremains at qto and Py(~b)= 0. l(t) will continue to increase until time tf when l( tr)= h( tO, where h(t) is the position of the piston. Compact bed consolidation. After time tr, the compact networked bed fills the whole filter and ~b will continue to increase until the filter cake cannot be compressed any further. Thereafter, q~is uniform everywhere and equal to ~b~.

t?z

dPy(~b) O~b dq~ c3z

)~ ~br() - (u-w) Vp 1 -

(3)

and the fluid phase is given by

c3pf 0z

)~ ~br(~b) (u-w) Vp 1-q~

(4)

Case 2: The suspension is initially networked,

o>~=
In this case, there is only one stage of filtration, namely the consolidation of the compact bed as described above. The underlying equations for modelling pressure
applied pressure Ap

~/////////////////////////z

suspension at initial volume fraction

Ap

h(t)

d)0
/ c o m p a c t bed

E l ........... ............
*>*g

We assume that gravitational forces and shear stresses are small compared to compressive forces in this paper [8-10]. Physically, Eq. (3) represents a force balance of the network stress gradient in the solid phase and the hydrodynamic drag on the network caused by the solid phase moving at a different downward velocity u to that of the local fluid w. The parameter 2 is a Stokes drag coefficient for an isolated solid particle of volume Vp (2= 6rCqap for a spherical particle of radius ap and fluid viscosity t/). (Here, there appears an 'additional' factor 1 - ~b in the denominator from the corresponding equation via Eq. (2) in Landman et al. [ 1], which comes from writing the fluid and solid equations down explicitly and rearranging them [8].) The function r(q~) is an important rheological parameter and may be thought of as the hindered settling factor, or a quantity which is inversely proportional to the permeability. It takes into account the hydrodynamic interactions between particles, which increase the drag on any given particle. This is a function which needs to be experimentally determined for the particular focculated suspension, just like Py(~b). The mass conservation equations for the solids and fluid are respectively

*>

a
~t

g
~

0(u)
Oz

Tl

(0 filter membrane

(5)
(6)

water f l u x - dh/dt

water flux
(b)

c?t (1 - qS)= ~zz [(1 - ~b)w]

(a)

Fig. 2. Schematic diagram of the filtrationprocess.

By adding Eqs. (5) and (6) and integrating, w may

M. Eberlet al./Colloids Surfaces' A." Physicoehem. Eng. Aspects 103 (1995) 1 10

be eliminated from Eqs. (3) and (4) as

dh
~bu + (1 - ~)w= - d t (7)

liquid was placed on top of the unformed filter cake, was used. Laboratory centrifuge experiments were conducted for a range of pressures less than 500 kPa.

the velocity of the piston. Eqs. (3), (5) and (7) are non-linear partial differential equations which form the basis of the rheological model for filtration. Given the rheological functions Py(~) and r(~b), and the correct boundary conditions, they can be solved numerically to give the volume fraction profiles in the filtration cylinder as a function of time along with the appropriate interface positions h(t) and

3. Results and discussion

3.1. Determining the compressive yield stress Py((~)


Suppose a filter press starts with material at a uniform volume fraction ~bo and height ho. A fixed pressure ao~ (or the value of the final pressure if ramped initially) is applied to give a final suspension bed height of h~. Then, by mass conservation, the volume fraction has a uniform value: ~oho

l(t).

2. Experimental section
The particles used in this study were an industrial kaolin sample with a density of 2.6 g cm -3. This was coagulated on a large scale and a sample taken for laboratory work. The feed to filtration was at an initial solid fraction of 36% w/w (q~= 0.18). Filtration data for the sample were collected on a plate and frame filter press where both the applied pressure and filtrate volume were monitored as a function of time. The pressure cycle consisted of a filter filling phase of approximately 6 min, a pressure ramping stage of approximately 10 min and then full pressure (~1500 kPa) for 60 min. The filter area was 1.76 m 2 and the final filter cake thickness was 1.55 cm. The average volume fraction as a function of time was calculated using a conservation of mass consideration. Laboratory tests to determine both the compressive yield stress Py(~b) and the hindered settling factor r(~b) of the suspension were performed on a Larox laboratory filter for pressures ranging from 500 to 1750 kPa. The filter area was 0.0232 m 2, and the filter cake thickness and final solid content were measured after each run. The reproducibility at each pressure was found to be excellent. The preferred method for determining r(~b) is the measurement of the steady state flux for a fully compressed filter cake. This method was made difficult owing to cake cracking. An alternative method, where the flux was determined when additional

h~
such that cr~ = Py(~b~) (8)

Therefore, applying different values of a~ and measuring the corresponding h~ allows the determination of ~b~ and hence the function Py(~b). This technique was used with a laboratory pressure filter at pressures above 500 kPa, which provides values of Py(~b) for values of ~b much greater than the initial volume fraction. For lower pressures, the batch centrifuge technique of Buscall and White [6] was used to estimate Py(~b). In this method, a sample of suspension, with an initial volume fraction ~b and height o ho, is spun at a particular speed until an equilibrium height heq is reached. The speed is increased and the exercise is repeated. The data comprise a curve of heq versus g, the acceleration at the bottom of the tube. The equations governing the network can then be shown to yield a parametrisation in g of the volume fraction and the solid stress at the bottom of the tube, which in turn determines Py(~). This method has been used successfully at lower pressures [10-13]. The two sets of Py(~b)data from the laboratory filter press and centrifuge were combined. A power law and an exponential curve were fitted to this

M. Eberl et al./Colloids Surfaces A: Physicochem. Eng. Aspects 103 (1995) 1 10

combined data:

600

Pr~

k[(~b/~bg)"- 1 ]

(9)
Py~ = k( 10 " ~ - 10"+Q
with various values of k and n. The data and the curve fitting are shown in Fig. 3. The correlation coefficient was ~0.98 for the power law fit with n = 7 and ~0.99 for the exponential fit with n = 8.1. Both forms will be discussed in Section 3.4. The Pv(~)) versus volume fraction graphs indicate that for this system ~g=0.2; further comments on this appear in Section 3.4.
E

500

Stage 1 Intermediate S t a g ~

. o*

400

300

~.S.~*mO~ ==r'" Pressure(kPa)


~ ' ~ =~ 500 750 I000
o

200

100

1250 1500 1750


I I

u
i 1 O+

0
)

20

30

40

50

60

3.2. Determining the hindered settling factor r(qb)


Various flux measurements, discussed below, determine the function r(~). The experimental setup for these measurements is same as in Fig. 2(a), but with an extra layer of water added between the piston and the suspension/filter bed. The filtrate volume data and corresponding t/V versus V plots for the laboratory filter are shown in Figs. 4(a) and 4(b). This filtration can be divided into the two stages of Case 1 in Section 1.2 plus an intermediate stage, which provides a check for the calculations.

(a)
120000

Time

(minutes)

Pressure (kPa)
e
500

m=a
[]
[]

100000

750
1000 []

a e

80000


[]

1250 1500
1750
n

=m e []
B

11 m i
m

e 11
4,

~memm

m[] []

60000
[] *

...." ..m mm

[]

t ~ .o
o B

_%== =

~g

40000

20000 0.000

0.002

0.004

0.006

0.008

0.010

0.012

(b)
2O0O ~" .....
- -

(m)

Power Exponential

law

fit fit

Fig. 4. (a) Filtrate volume versus time t obtained from the laboratory filter press for filtration stages 1 and 2. (b) The t/V versus V curves for stage 1 and the intermediate stage, where V is the volume of fluid expressed per unit area of membrane.

"o
1000

Filter press

Stage 1: Filtration stage


The data can be analysed using the model of Wakeman et al. [ 2 ] , which considers the total pressure drop in the fluid phase to be the sum of the pressure drop across the growing compact bed and the pressure drop across the membrane:

==

~o E
0

CentrifuJ
h i , i i

0.2

0.3

0.4

0.5

Volume Fraction (~)


Fig. 3. Experimental data and fitted power law, and an exponential curve for P~,(q~) versus q~ for flocculated kaolin.

Ap = tl~cV~ +tlR ~

dV

dV

(10)

where V is the volume of fluid expressed per unit area of membrane, Ap(t) is the applied pressure, R

M. Eberl et al./Colloids SurJaces A." Physicochem. Eng. Aspects 103 (1995) 1-10
2.4

is a membrane hydraulic resistance, q is the viscosity of the fluid and c is the effective solid mass concentration in the filter feed. The quantity ~ in this theory is the compact bed hydraulic resistance per unit mass of bed and is assumed constant as the bed grows. This quantity should be a strong function of the volume fraction, which we know varies across the bed. However, there exists an effective volume fraction which can be used in this calculation [14]. When Ap is a constant, as for these laboratory experiments, integrating Eq. (10) leads to
t

......
2.2

'E
~n rL ~ 2.0

Power law fit Exponential fit

/&

1.8

1.6

~E

1.4

y
I 0.36 I 0.37 I 0.38 I 0.39 0.40

10 0.35

- (c~crl/2Ap)V+ (rIR/Ap) = m~ V+ b

( 11 )

Volume Fraction ( O m )

Hence, q~c and qR can be obtained from the slope ml and intercept b of a t/V versus V graph. This process occurs for times up to t = tf. These straight line graphs are shown in Fig. 4(b).

Fig. 5. Experimental data and fitted power law, and exponential curve for (2/Vp)r(~b) versus ~b.

an

compact bed can be found as 3 1 ~b,,= 4 ~ + 4~bo (13)

Intermediate stage: Constant flux stage


As shown in Fig. 4(a), the flux is now constant as the supernatant filtrate is pushed through the completely formed bed. Let this flux be dV/dt= m2. For consistency, m2 should be equal to the flux at the end of stage 1, namely dV/dt evaluated at t=tr. Experimentally, this point was determined as the start of the linear portion of the data in Fig. 4(a). Using Eqs. (10) and (11), this flux is given by

where ~b is the initial volume fraction and ~b~ is o the final volume fraction satisfying ~r~ =Py(~b~). Comparing the Wakeman et al. model to the rheological model, we have obtained the required relationship between r(t~) and the slope ml of the t/V versus V plots [14,15], namely

2
~pp r(~bm)-

2Apml(C~m/OO-1)(1 -~bm) 2
~b., (14)

dV dt

1 at 2ml V( tf ) + b

t=tf

(12)

Agreement between the two methodologies was found to be reasonable considering the wide pressure span. Stage 2 is the bed consolidation stage and is not required in the calculation of r(~b). Landman and co-workers [14,15] showed how the compressional rheological model and the engineering approach of the Shirato-Wakeman model can be reconciled. The filtration parameters extracted from the experiments, i.e. ml and b, can be related to the more fundamental theological parameters, namely r(~b). A mean value approximation to the volume fraction in the non-uniform

If the filter cake is stable at the end of Stage 2 in filtration so that liquid can just flow through the cake without disturbance, then a more direct measurement of (2/Vp)r(~)) can be made. (2/Vp)r(Oo~) can then be obtained from plots of Ap/(dV/dt) versus final volume fraction q~o~. As discussed earlier, such experiments could not be carried out for this system. The corresponding (2/Vp)r(~)) data obtained using the mean value method are shown in Fig. 5. A power law and an exponential function were fitted to these data, as 2 Vp r(~b)= k(1 -~b)" Vp r(~b)--k x 10"* (15)

M. Eberl et al./Colloids Surfaces A. Physicochem. Eng. Aspects 103 (1995) 1-10

with various values of k and n. Since only large pressure values were used in the laboratory experiments, data are available over a very limited range of volume fractions ~b; hence, the curves must be extrapolated over the full range ~b < ~b< ~boowhen o implementing the model. A power law and an exponential fit to the data are shown, with n = 10.4 for the power law fit and n--7.25 for the exponential fit. The correlation coefficient for both fits was approximately 0.86. To check the consistency and physical plausibility of the extrapolated value k of (2/Vp) obtained from Eq. (15) with q~= 0 and r(~b)--*1, an equivalent spherical particle radius was determined to be ~0.18 gm. Calculations based on typical kaolin particles used in this study gave an equivalent radius of ~ 0.2 gm. The agreement is pleasing.

made of the average volume fraction q~av, by a conservation of mass argument:


1

c~ = ~ ) ,v

f[ ~~bdz = -ho~o h( t )

(18)

3.3. Determining the membrane resistance


An estimate of the membrane resistance R of the plant filter press can be obtained from initial filtration data in Fig. 4(b). Using Eq. (11), r/R is related to the intercept b of the t/V versus V plot by

Apb = fIR

(16)

where Ap is the constant applied pressure. For the plant filter measurements, the pressure is ramped. To correct for this, we introduce an effective time z to replace the time t in Eq. (11) with z = f~ Ap(t) dt
O'oo

The compressional rheological model of Landman et al. [ 1 ] was implemented with power law representations for both Py(~b) and r(q~) with a ramped applied pressure for 10 min, after which the pressure was kept constant at 1500kPa, as shown in Fig. 6. The model calculates the volume fraction as a function of position at each time increment as well as the filter height h(t). This Figure shows the experimental plant measurements and the results of the model. The agreement between the two is excellent. Some comments on the sensitivity of the model are now made. The underlying partial differential equations of the model depend on the derivative of Pr(~b) with respect to volume fraction, as seen in Eq. (3), as well as Py(~b) and r(~b). We found that the predicted filtration behaviour is not very dependent on the type of fit used for both Py(O) and r(0), as long as the derivative of Pr(~b) for each type of fit (i.e. power law or exponential in this case) was similar. It should also be noted that determining Py(~b) requires accurate measurements over the entire range of volume fractions. Pr(~) curves based only on the high pressure compression data give higher
0,5

(17)
0.4

where a~ is the maximum value of the applied pressure. The plant membrane resistance was calculated to be small compared to the resistance of the filter bed (their ratio is ~0.03). The membrane resistance has been included in the calculations, although the difference between these results and those for zero membrane resistance is less than 1%.

0.3

/
measured data
0 i 20 I 40 t 60

& 1500

1000 w

500

2
> 0.2

...... simulation using power law fit simulation using exponential fit
0.1 80

3.4. Determining the average volume fraction


In the plant pressure filter, h(t) is determined by the quantity of fluid lost. Calculations were then

Time

(minutes)

Fig. 6. The measured average volume fraction for the plant filtration data, the applied pressure as a function of time and the model results for the average volume fraction.

M. Eberlet al./Colloids Surfaces A: Physicockem. Eng. Aspects 103 (1995) 1-10

values for n, resulting in differences in the predicted volume fractions. A further model sensitivity is the dependence of the membrane resistance on ~b. A constant membrane resistance has been assumed, although it is likely to depend on ~b. For the pressure filter described here, the membrane resistance was calculated to be small compared to the filter cake resistance ( ~ 3% of the resistance of the final filter cake). Thus, it had a negligible effect on the predicted average volume fraction. Despite the model sensitivities, perhaps the greatest discrepancy is the apparent ~bgas predicted from the Py(~b) data and the measured ~g for this kaolin sample. The Py(~b) versus volume fraction data indicate ~g=0.2, SO that for this system ~b < ~bg. However, as determined from sedimentao tion experiments, ~bg is approximately 0.1. Hence, for the plant filtration measurements with ~bo= 0.193, the choice of ~b < ~bg or ~b > ~bg is problemo o atic. We have implemented the model for both these cases. For ~0 > ~bg, the model gives a larger more uniform q~ throughout the filter bed and a faster initial increase in the average volume fraction. However, for the kaolin system, this initial difference was small. For the large pressures used in this filtration process, the volume fraction quickly moves to values much higher than ~b and o, therefore choosing ~b > ~bg rather than ~b < ~bg has o o little effect. If we need to predict the operation of the plant at low or moderate pressures, then a more accurate value of ~bg would be required.

pressure ramp and a fast pressure ramp for both ~b < ~bg and ~bo > ~b [ 1]. A number of trends are o s observed. The larger the applied pressure, the faster the filtration process. The greatest non-uniformity in the filter cake will be observed for the case when the maximum pressure is applied from the beginning of the filtration cycle. Ramping of the applied pressure will produce a more uniform solid profile across the filter bed, but filtration will be slower, as shown in Fig. 7. The filter resistance produces more uniform solid densities across the filter bed and increases the filtration time, as illustrated in Fig. 8. These trends could be utilised to the benefit of a filtration plant.

4. Conclusions
The data presented demonstrate the applicability of the filtration model of Landman et al. [-1 ] to the prediction of plant filtration behaviour from laboratory data. The analysis shows that the essential physics of filtration is being predicted by the model. The model trends can be used to enhance filtration performance. In particular," the model makes various predictions, as follows. The dependence of the filter cake thickness on filtration time. The dependence of filtration time on the application of the pressure cycle. The relationship of the permeability and the compressibility to the evolving filter cake, and the non-linear dependence of these two parameters on the volume fraction. This work gives confidence to the application of the fundamental rheological model to aid in the design and efficiency of pressure filtration devices, and to the adjustment of coagulant dosage and pH effects in changing the theological behaviour of the feed slurry [15].

3.5. Changing plant operations


The good agreement between the plant data and the model confirms the applicability of the model and demonstrates that laboratory data are scalable to plant filtration processes. We can make some general comments about q~av. It is possible to find asymptotic solutions for the initial stage of pressure filtration. For the case of ~bo < ~g, h(t) and hence q~,v vary linearly with t for non-zero membrane resistance and vary with ?/2 for zero membrane resistance. The filtration model predicts the average volume fraction ~b~vas a function of time in dimensionless units for the case of constant pressure, a slow

Acknowledgements
The data presented in this study were obtained in an industrial collaboration. We wish to acknow-

M. Eberl et aL/Colloids Surfaces .4: Physicochem. Eng. Aspects 103 (1995) 1-10
0.18 0.18 tO

tO o t~ ii
E >

f"

m tl

-i

//
.....

.,.,/
constant pressure
......... slow ramp fast ramp

E
O >

Increasing R

>

>

0.00
(a)

(a)
0.8

0.00 0,0

0.0

Time
0.19

0.19[
/

Time
=0

/,"
a

.."""'"
/ Z
gs *

O m J. U.

,,
,.s /

G)

E
m O
Q~

03

?
> constant pressure : slow ramp fast r a m p 0.10

0.10 0.12

0.00
Time

0.25

0.00

(b)

Time

Fig. 8. The calculated average volume fractions as a function of time (in scaled units) for different membrane resistances (a)
(b) 3

~o < G, (b) ~o > G-

constant pressure

a.

Tfast raV

ledge and thank the contributions of our industrial collaborators for data measurement and for the freedom to publish our findings. We wish to thank Lee White for his advice, and Matthew Green and Deanne Labbett for the centrifuge work. The financial support of the Advanced Mineral
Fig. 7. The calculated average volume fraction as a function of time {in scaled units) for different pressure application rates. (a)

Q.

s l o w ramp

o.o

Time

1.5

o <G, tb) o > G

10

M. Eberl et al./Colloids Surfaces A: Physicochem. Eng. Aspects 103 (1995) 1-10


concentrated suspensions I: The theory of sedimentation, J. Chem. Soc., Faraday Trans. 1, 83 (1987) 873. I. Howells, K.A. Landman, A. Panjkov, C. Sirakoff and L.R. White, Time-dependent batch settling of flocculated suspensions, Appl. Math Modelling, 14 (1990) 77. K.A. Landman and W.B. Russel, Filtration at large pressures for strongly flocculated suspensions, Phys. Fluids A, 5 (1993) 550. R. Buscall, P.D.A. Mills and G.E. Yates, Colloids Surfaces, 18 (1986) 341. R. Buscall, I.J. McGowan, P.D.A. Mills, R.F. Stewart, D. Sutton, L.R. White and G.E. Yates, J. Non-Newtonian Fluid Mech, 24 (1987) 183. N.J. De Guingand, Master of Eng. Thesis, University of Melbourne, 1986. M.D. Green and D.V. Boger, The compressive yield stress of zirconia suspensions, Proc. Australia Soc. Rheol. Vlth National Conference on Rheology, Clayton, Australia, 10-12 June 1992, pp. 33-36. M.D. Green, N.J. De Guingand, Q.D. Nguyen and D.V. Boger, The shear and compression rheology of bauxite residue - - an overview, Proc. International Bauxite Tailings Workshop, Perth, Australia, November 1992, pp. 116-125. K.A. Landman, L.R. White and M. Eberl, Pressure filtration of flocculated suspensions, AIChE J., in press. K.A. Landman and L.R. White, Solid/liquid separation of flocculated systems, Adv. Colloid Interface Sci., 51 (1994) 175.

Products Research Centre, a Special Research Centre of the Australian Research Council, our industrial collaborators, and the International Fine Particle Research Institute, Inc., are gratefully acknowledged.

[7]

[8]

[9] [10]

References
[1] K.A. Landman, C. Sirakoff and L.R. White, Dewatering of flocculated suspensions by pressure filtration, Phys. Fluids A, 3 (1991) 1495. [2] R.J. Wakeman, M.N. Sabri and E.S. Tarleton, Factors affecting the formation and properties of wet compacts, Powder Technol., 65 (1991) 283. [3] M. Shirato, T. Murase and M. Iwata, in R.J. Wakeman (Ed.), Progress in Filtration and Separation: 4, Elsevier, Amsterdam, 1986. [4] K. Terzaghi and P.B. Peck, Soil Mechanics in Engineering Practice, Wiley, New York, 1948. [5] B. Sivaram and P.K. Swamee, A computational method for consolidation coefficient, J. Jpn. Soc. Soil Mech. Found. Eng., 17 (1977) 48. [6] R. Buscall and L.R. White, On the Consolidation of [11] [12]

[13]

[14] [15]

You might also like