You are on page 1of 18

Int. J. Miner. Process. 79 (2006) 235 252 www.elsevier.

com/locate/ijminpro

An extension of the particle-based approach to simulating the sedimentation of polydisperse suspensions


Monika Bargie a,, Elmer M. Tory b,1
a

Institute of Computer Science, AGH University of Science and Technology, al. Mickiewicza 30, 30-059 Krakow, Poland b Mathematics and Computer Science, Mount Allison University, Sackville, NB, Canada E4L 1G6 Received 1 September 2005; received in revised form 12 February 2006; accepted 7 March 2006 Available online 24 April 2006

Abstract We extend the particle-based approach to settling of polydisperse suspensions in which particles of at least one species move upward during all or part of the sedimentation process. The velocity of each downward-moving particle is governed by the solids concentration immediately below it; the velocity of each upward-moving particle by that immediately above it. This ensures that each particle remains in the rate-determining region in a small time-step. The robustness of the method is demonstrated by applying it to simulations in which the particles are initially distributed randomly. Despite the fluctuations that characterize these simulations, the essential features of the solids profile are maintained. The method was successfully applied to four hypothetical tridisperse suspensions. Simulations of two experiments produced settling curves that agreed closely with experimental values. 2006 Elsevier B.V. All rights reserved.
Keywords: simulation; sedimentation; polydisperse suspensions

1. Introduction Bargie et al. (2005) developed a particle-based approach to simulating the sedimentation of polydisperse suspensions and applied it to suspensions in which all particles have the same density and do not differ greatly in size. (Thus, all or almost all of the particles move downward.) In this method, the velocity of each particle is determined by the concentration in a thin layer (of thickness h) of suspension directly below that particle. The particle remains within its unique layer during the small time-step, ensuring that the appropriate
Corresponding author. Fax: +48 12 633 9406. E-mail addresses: mbargiel@uci.agh.edu.pl (M. Bargie), sherpa@nbnet.nb.ca (E.M. Tory). 1 Fax: +1 506 364 1183. 0301-7516/$ - see front matter 2006 Elsevier B.V. All rights reserved. doi:10.1016/j.minpro.2006.03.004

velocity-determining solids concentration is used. The flat surface at the bottom of the container is replaced by an artificial layer of packed particles in ( h, 0) that slows down the particles within h of the bottom. Particles settle into the packed bed with finite velocities and its height is calculated after each time-step. This method yields the concentration gradients that occur in some suspensions and accurately reproduces their propagation. Discontinuities are replaced by very sharp changes in concentration, which are actually more realistic. When suspensions contain particles that differ greatly in size and/or density, some particles may move upward, entering concentration layers above their initial positions. The aim of this paper is to adapt the method to handle such suspensions. The first consideration is to ensure that the suspension is stable.

236

M. Bargie, E.M. Tory / Int. J. Miner. Process. 79 (2006) 235252

Unlike suspensions of particles with the same density, which are always stable (Berres et al., 2003), suspensions of light and heavy particles are often unstable, especially at higher concentrations (Batchelor and Janse van Rensburg, 1986; Biesheuvel et al., 2001; Brger et al., 2002; Weiland et al., 1984). Unstable suspensions often settle very rapidly (Fessas and Weiland, 1981, 1984) and the one-dimensional equations developed for stable suspensions cannot be applied to them. However, suspensions of light and heavy particles are often stable at very low concentrations (Brger et al., 2002; Law et al., 1987). As they are displaced by larger or heavier particles, some particles may temporarily move upward in suspensions in which all particles are heavier than the fluid. Some of these suspensions are stable and some are not. Thus, there is considerable scope for the extension of the particle-based approach, but precautions are necessary to ensure that the suspensions are stable. 2. Stability of suspensions Let vk() be the settling velocity of the kth species in a polydisperse suspension, where = (1, , K)T is the vector of solids concentrations. Then the solids flux density is f() = (f1(), , fK())T, where fk() =kvk(). One-dimensional batch settling in a closed container is described by the system of conservation laws AF=At Af F=Az 0; 0VzVH; t z 0: 1 Abbreviating fk/i by fki, we write the Jacobian as 3 2 N f1K f11 6d d d 7 7 6 2 d d d 7 J F 6 7: 6 4d d d 5 fK 1 N fKK The system with initial concentration 0 is called hyperbolic if J(0) has K real eigenvalues, and strictly hyperbolic if these are pair-wise distinct (Brger et al., 2002). If some of the eigenvalues are complex, the suspension is unstable and its sedimentation must be described as a three-dimensional process (Brger et al., 2002). For two species, the instability criterion reduces to I2 F f11 f22 2 4f12 f21 < 0; 3

mined by evaluating a scalar discriminant function. The characteristic polynomial of J() is p3 k; F k3 RFk2 S Fk T F; 4

where the leading coefficient has been normalized to one and the functions R, S, and T are given (Brger et al., 2002) by R tr J F f11 f22 f33 ; 5 6 T det J F f11 f22 f33 f21 f13 f32 f31 f12 f23 f11 f23 f32 f22 f13 f31 f33 f12 f21 : 7 The equation p3(; ) = 0 has one real root and one pair of complexconjugate solutions if and only if the following inequality holds: I3 F S 3 =27 S 2 R2 =108 T 2 =4 R3 T =27 RST =6 > 0:

S f12 f21 f13 f31 f23 f32 f11 f22 f11 f33 f22 f33 ;

In this case, the first-order system of conservation laws is non-hyperbolic and the tridisperse suspension is unstable (Brger et al., 2002). Since hyperbolicity implies stability (Brger et al., 2002), we can also determine whether a given suspension is stable or unstable by finding the eigenvalues of J(0). 3. Model equations for polydisperse suspensions There are many empirical or semi-empirical equations for the one-dimensional settling velocity, v(), of a monodisperse suspension. Of these, the best known and most widely used is the RichardsonZaki equation (Richardson and Zaki, 1954) v/ ul 1 /n ; 0V/ < /max ; 9

where is the solids concentration, max is the concentration of the packed bed, and u is the Stokes velocity ul qs qf gd 2 =18gf : 10 Here s and f are the densities of the solid and fluid, respectively, g is the acceleration of gravity, d is the diameter of the sphere, and f is the dynamic viscosity of the fluid. Since spheres in our simulations cease settling when they hit the packed bed (Bargie et al.,

as derived by Batchelor and Janse van Rensburg (1986). For tridisperse suspensions, hyperbolicity can be deter-

M. Bargie, E.M. Tory / Int. J. Miner. Process. 79 (2006) 235252

237

2005), we can safely eliminate the upper bound in Eq. (9) and its generalizations. The appropriate generalization of the Richardson Zaki equation to K species is the MasliyahLockett Bassoon (MLB) equation (Masliyah, 1979; Lockett and Bassoon, 1979; Brger et al., 2002) vk F l1 /n2 " dk qk qF # dj /j qj qF ; 11 where j is the concentration of the jth species, qF qf 1 / q1 /1 q2 /2 N qK /K ; 12
2 2 dk dk =d1 ;

Eq. (18) clearly reduces to Eq. (9) when 1 = 2 = = K = s. According to the MLB equation, the solids flux of the kth species in a polydisperse suspension with K species is fk F /k vk F l1 /n2 fkF; where " fkF /k dk q k r d F r d F
K X j 1 K X j1

19

K X j1

dj q j /j 20

# dj /j :

13 14

When all spheres have the same diameter, Eq. (19) simplifies to " # K X n 2 fk F /k l1 / q 1 gk 2 / /j gj :
j 1

2 l gd1 =18gf ul1 =q1 qf ;

21 Then n 2fk; fki l1 /n3 1 /fki where " # K X fk F /k q 1 gk 2 / /j gj


j1

k is the density and dk is the diameter of spheres of the kth-largest species, and u1 is the Stokes velocity of the largest species. Note that u1 < 0 when 1 f > 0. Eq. (11) can also be written (Brger et al., 2002) as " vk F l1 /n2 dk q k r d F
K X j 1

22

23

# dj /j q j r d F ; 15

=( 1, 2, , K). When all where k = k f and spheres have the same diameter (k = 1, k = 1, , K), Eq. (15) simplifies to " # K X n 2 v k F l 1 / q k 2 / /j q j ;
j1

and q fki 1 dki gk g d F/ 2 /k g d F gi / 2:

24

since j1 /j / and r d F j1 q j /j . In this case, we denote the heaviest species as 1 and define k = q 1 : gk q Then Eq. (16) can be written as " vk F l1 /n2 q 1 gk 2 /
K X j 1

PK

PK

16

Substitution of Eq. (22) in Eqs. (5)(7) reveals that [(1 )n 3]6 is a common factor in Eq. (8). Thus, we () (n 2)fk() for fki() in can substitute (1 )fki 6 Eqs. (5)(7) and use them in Eq. (8). Since 1 is also a common factor, we can use a dimensionless stability index.
Table 1 Properties of suspensions in Examples 1 to 4 Example 1 2 1000 400 200 0.20 0.15 0.15 3 1000 800 200 0.20 0.20 0.10 4 1000 400 100 0.20 0.15 0.05 kg/m3 kg/m3 kg/m3 1 2 3 1 2 3 Units

17 # /j gj : 18

1000 500 300 0.20 0.15 0.05

238

M. Bargie, E.M. Tory / Int. J. Miner. Process. 79 (2006) 235252

4. Suspensions of light and heavy particles The system consists of an ensemble of N particles of K species distributed randomly in a cylinder of height H and cross-sectional area N A 25 K P H /0 = V k k
3 where Vk = dk /6 is the volume of a sphere of species k. The value of A is meaningless since we assume one-

dimensional motion, but it is necessary to set A so that the heights and concentrations in the simulations match those in the experiments. The number of spheres of the kth species is Nk t AH /0 k =Vk b: 26

k 1

Some dilute bidisperse suspensions with one species lighter than the fluid and one heavier are stable. Thus, we can use Eq. (15) with K = 2 to calculate the one-dimensional velocity of each species. We adapt the method of
0s 0s 0s

100 0s

height [mm]

50

0 100 400 s 400 s 400 s 400 s

height [mm]

50

0 100 700 s 700 s 700 s 0.0 0.5 700 s 1.0

height [mm]

50

0 100 900 s 900 s 900 s 900 s

height [mm]

50

0 100 987 s 987 s 987 s 987 s

height [mm]

50

0 0 0.5
1

0.5
2

0.5
3

0.5

Fig. 1. Solids profiles for Example 1. Each point represents the solids concentration in a height of 0.25 mm.

M. Bargie, E.M. Tory / Int. J. Miner. Process. 79 (2006) 235252

239

Bargie et al. (2005) to simulate such experiments. Retaining the artificial packed bed in ( h, 0), we introduce an additional artificial layer in (H, H + h) and distinguish between upward- and downward-moving species. There are two cases to consider. In the first, the suspension completely fills the container, which is closed at both ends. Then, the light particles (species 2) settle upward onto the packed bed at the top in exactly the same way that the heavy particles settle downward onto the packed bed at the bottom. However, the simulation can also handle the accumulation of packed spheres at a free surface and their protrusion beyond it. In this case, the weight of the particles accumulated at the top is Nt2V2 and thus the volume of displaced fluid is NtVf = Nt2V2/f. The total volume of the region above the free surface is NtVa = Nt(V2 Vf)/2max. Thus, the height of the packed region above the free surface is Ha = NtVa/A. The algorithm of Bargie et al. (2005) calculates the height of the packed bed as settling proceeds. Thus, the bottom of the bed of packed particles at the top is known at all times whether there is a flat plate or a free surface there. Each particle i (at height zi(tn) and velocity vi(tn) in the nth time-step) has its own unique bin (of height h) that contains all the particles j for which  i z tn h; zi tn if vi tn1 V 0; j z a 27 zi tn ; zi tn h if vi tn1 > 0: The settling velocity of each particle is calculated from Eq. (15). It is well known that particles closely approaching a packed bed are slowed down. Bargie et al. (2005) imitated this effect on downward-moving particles by including particles from the bed in the thin velocitycontrolling layer below the particle. However, when particles differ substantially in size or density, it is necessary to distinguish between the effects of sedimenting and settled particles. Sedimenting particles displace fluid that may carry smaller or lighter particles upward. Thus, we need to ensure that particles in the packed bed slow down incoming particles but do not cause them to move upward. We achieve this by including settled particles only in the appropriate term in the foregoing equations. For example, Eq. (18) becomes " vk F l1 /n2 q 1 gk 2 /
K X j1

practice, we adjusted Eq. (15) and used it in all simulations. We follow Bargie et al. (2005) by replacing the bottom of the container by an artificial layer of packed spheres. The modified equations, such as Eq. (28), imply that we need to specify only the packing density of this layer, not its composition. Consequently, for all particles within h of the packed bed (at the bottom or at the top), we assume / 1 a/max a/ ; where  i z B b =h a Bt zi =h 29

for bottom packed bed; for top packed bed;

30

Bb and Bt are the levels of the bottom and top packed + + bed, respectively, and + = 1 + + K is the concentration of active particles in the region between zi and the packed bed. When particle i is about to settle onto the packed bed, the number of active particles in its bin + becomes too small to calculate k with reasonable accuracy. To overcome this problem, we find for every species, k, a particle, l, closest to the packed bed, with
(a)
100 Example1 80

height [mm]

60 40 20 0
packed bed species 1 species 2 species 3

500

1000

time [s] (b)


0.05 Example1 0.0 -0.05 -0.1 -0.15
species 1, tridisperse species 2, tridisperse species 2, bidisperse species 3, tridisperse species 3, bidisperse species 3, monodisperse

# / j gj ; 28

velocity [mm/s]

500

1000

time [s]
Fig. 2. (a) Height vs. time plots for the three interfaces and the packed bed for Example 1. (b) Mean velocities of each species in the three zones.

where includes particles in the packed bed that are in the controlling layer, while + and j+ exclude them. In

240

M. Bargie, E.M. Tory / Int. J. Miner. Process. 79 (2006) 235252

+ still reasonable value of concentration, k , and velocity, l vk. Then all particles of species k closer to the packed l bed than particle l are given velocity vk .

5. Suspensions of heavy particles of different densities There are many stable bidisperse suspensions in which the lighter species moves upward in the lower part of the suspension. Once the heavier particles have settled out of
100 0s 0s

the upper region, it becomes a monodisperse region and the lighter particles move downward. In tridisperse suspensions, one or two species may initially move upward. Our algorithm modifies that of Bargie et al. (2005) by flagging each particle as upward- or downwardmoving. As indicated in Eq. (27), the velocity of each upward-moving particle is governed by the concentration in the region immediately above it; the velocity of each downward-moving particle by the concentration immediately below it. At some point, each upward-moving
0s 0s

height [mm]

50

0 100 500 s 500 s 500 s 500 s

height [mm]

50

0 100 1000 s 1000 s 1000 s 0.0 0.5 1000 s 1.0

height [mm]

50

0 100 1500 s 1500 s 1500 s 1500 s

height [mm]

50

0 100 1931 s 50 1931 s 1931 s 1931 s

height [mm]

0 0 0.5
1

0.5
2

0.5
3

0.5

Fig. 3. Solids profiles for Example 2.

M. Bargie, E.M. Tory / Int. J. Miner. Process. 79 (2006) 235252

241

particle reaches a region where has a value that makes that particle's velocity negative. Thereafter, the particle moves downward and its velocity is governed by the concentration immediately below it. Eq. (28) implies that particle velocities are positive when the expression in square brackets is negative. For tridisperse suspensions in which d1 = d2 = d3, we can capture the direction of movement by letting F gk 2 //1 /2 g2 /3 g3 : vk 31

7. Numerical examples Tridisperse suspensions challenge numerical methods. The method used by Al-Naaf and Selim (1989) takes the Smith effect (Bargie et al., 2005) into account, but makes no allowance for the propagation of concentration gradients upward from the packed bed (Bargie et al., 2005). We used Eq. (31) to design four hypothetical cases in which spheres of at least one species move upward. In every case, u1 = 10 3 m/s. The densities and concentrations of all species in these examples are given in Table 1. In these examples, max = 0.64. All the spheres have the same size, so settling velocities can be calculated from Eq. (28). Using Eq. (8), we determined that I3(0) < 0, so these suspensions are stable. Example 1. Fig. 1 shows the solids profiles at various times. The initial profiles reflect the random initial placement of particles. Each point represents the concentration in a 0.25 mm interval of height. At t = 400 s, the packed bed consists primarily of species 1, but species 2 is present at roughly its initial concentration

Particles that move upward may be overtaken by the rising packed bed (Bargie et al., 2005). This is especially important at the high concentrations necessary to ensure that at least one species moves upward. 6. The algorithm The algorithm used in this paper has been described in Bargie et al. (2005). The modifications that were made to handle upward-moving particles are given in the Appendix. Almost all remarks considering time requirements of the algorithm presented in Bargie et al. (2005) are still valid. However, running both programs with the same input data, we found that the new approach caused the program to run about 2025% slower (depending on the number of particles). This slow-down is caused by two factors: (1) Since we consider particles moving in both directions (downward and upward), it was necessary to split the loop for concentration/ velocity calculation into two parts. Accordingly, we move through all the particles twice (instead of once as in Bargie et al., 2005). The first loop, searching the particles from top to bottom, calculates the velocities of those moving downward while the second one (going from the bottom up) deals with particles moving upward. (2) We introduced new methods of calculating the concentration and velocity for particles close to the packed bed. This too caused additional splitting of the velocity loop, since we consider separately particles lying further than h from the packed bed and those within h. However, we believe this factor is much less prominent in the slow-down of the algorithm. The memory requirements of the both programs are identical (22 bytes per particle). In all the examples and experiments described below we used 220 particles.

(a)
100 Example 2 80

height [mm]

60 40 20 0
packed bed species 1 species 2 species 3

500

1000

1500

2000

time [s]

(b)
0.05 Example 2

velocity [mm/s]

0.0

-0.05

species1, tridisperse species 2, tridisperse species 2, bidisperse species 3, tridisperse species 3, bidisperse species 3, monodisperse

-0.1

500

1000

1500

2000

time [s]
Fig. 4. (a) Height vs. time plots for the three interfaces and the packed bed for Example 2. (b) Mean velocities of each species in the three zones.

242

M. Bargie, E.M. Tory / Int. J. Miner. Process. 79 (2006) 235252

and many particles of species 3 are also trapped by the rising bed. Above the packed bed, there is a discontinuity, then a concentration gradient (mostly in the value of 1), and a second discontinuity above the gradient. The uppermost region of the suspension contains only particles of species 3 at a concentration much higher than the initial value. Just below this region is a bidisperse suspension of constant concentration in which 2 is much greater than its initial value. The subsequent profiles show the discontinuities and gradients after species 1 has completely settled.
100

Fig. 2(a) shows the positions of the packed bed and the three interfaces that delineate the tri-, bi-, and monodisperse regions. The changing rates of fall reflect the upward propagation of concentration gradients. The simplicity of this figure masks the complicated movement of the individual species. Fig. 2(b) shows the average velocities of the active particles of each species in the various regions of the suspensions. In the tridisperse suspension, the average velocity of spheres of species 1 becomes less negative as more of them are in regions of higher concentrations. Spheres of species 2

height [mm]

0s 50

0s

0s

0s

0 100 500 s 50 500 s 500 s 500 s

height [mm]

0 100 0.0 800 s 50 800 s 800 s 0.5 800 s 1.0

height [mm]

0 100 1000 s 50 1000 s 1000 s 1000 s

height [mm]

0 100 1370 s 50 1370 s 1370 s 1370 s

height [mm]

0 0 0.5
1

1 0

0.5
2

1 0

0.5
3

1 0

0.5

Fig. 5. Solids profiles for Example 3.

M. Bargie, E.M. Tory / Int. J. Miner. Process. 79 (2006) 235252

243

initially move downward, but their average velocity gradually increases with time and becomes positive when the concentration of species 1 has increased substantially. Species 3 moves upward throughout. In the bidisperse region, species 2 and 3 initially move downward at roughly constant rates, reflecting the essentially constant values of 2 and 3 at t = 400 s. Later, the average velocity of spheres of species 2 becomes less negative while spheres of species 3 move upward. These changes reflect the increased concentrations in the gradients at t = 700 s. In the monodisperse region, spheres of species 3 settle with an approximately constant velocity, reflecting the roughly constant concentration in that region at t = 400 s. This velocity becomes less negative in the concentration gradients at t = 700 s and t = 900 s. Example 2. Spheres of species 2 and 3 initially move upward (Fig. 4(b)), but Fig. 3 shows that some of them are trapped by the rising bed. Since velocities of species 2 are very small, its concentration in the packed bed is still approximately 0.15. The much smaller proportion of species 3 in the packed bed reflects the higher velocity that allows more of them to escape (Fig. 4(b)). The sharp change in 1 at t = 500 s approximates a discontinuity. There are many short-range fluctuations in the region of constant concentration, but 1 0.2 on average. Despite the scatter, species 3 has also maintained its concentration, except near the upper interface, where the Smith effect is notable. The concentration of species 2 varies greatly. One branch consists of most of the points with values slightly greater than 0.15; the other has fewer points with values much less than 0.15. Fluctuations are much less in the region that contains only species 2 and 3. Subsequent profiles show the further sedimentation of these species. When settling is complete, the ternary packing shows wide fluctuations in concentration with height, but the mean solids fraction there is exactly 0.64. The concentrations in the upper regions are much less variable. Fig. 4(a) shows the positions of the three interfaces with time. Aside from the first few seconds when the zones are being established, the tri-bidisperse interface falls with a constant velocity. Because there is a sharp change in 1 rather than a discontinuity, there is a sharp change in velocity in the last few seconds. The other two interfaces fall with constant velocities until concentration gradients propagate to them. As in the previous example, the simplicity of the settling curves belies the complexity of the particle velocities in the suspension. Fig. 4(b) shows the average velocities of each species. The mean velocity of

species 1 is almost constant until all of its spheres have settled out. The mean velocities of species 2 and 3 are also roughly constant in the tridisperse suspension, with both species moving upward. After species 1 has settled out, the mean velocity of species 2 in the bidisperse suspension increases (decreases in absolute value), rather slowly at first and then more quickly. The mean velocity of species 3 remains positive and roughly constant. In the monodisperse suspension, the mean velocity of species 3 is roughly constant until about 1200 s and then increases slowly. The fluctuating velocities at very small times reflect the small number of spheres in the bidisperse and monodisperse regions. Example 3. Fig. 5 shows the solids profiles at various times. At t = 500 s, the concentration of the packed bed varies only slightly, despite considerable variation in the concentrations of the different species. Above the bed are a discontinuity and a concentration gradient, followed by a region of (roughly) constant concentration. Fluctuations in are less than those in k. As first species 1 (t = 800 s) and then species 2 (t = 1000 s) settle

(a)
100 Example 3 80

height [mm]

60 40 20 0
packed bed species 1 species 2 species 3

500

1000

1500

time [s]

(b)
0.05 Example 3

velocity [mm/s]

0.0

-0.05

species 1, tridisperse species 2, tridisperse species 2, bidisperse species 3, tridisperse species 3, bidisperse species 3, monodisperse

-0.1

500

1000

1500

time [s]
Fig. 6. (a) Height vs. time plots for the three interfaces and the packed bed for Example 3. (b) Mean velocities of each species in the three zones.

244

M. Bargie, E.M. Tory / Int. J. Miner. Process. 79 (2006) 235252

out of the upper regions, increases in the remaining species reflect the Smith effect. The final packed bed (t = 1370 s) varies little in total concentration, but substantially in its composition. Fig. 6(a) shows the positions of the three interfaces with time. The tri-bidisperse interface falls at an essentially constant rate until it is just above the packed bed. The change near the end reflects the gradient at t = 500 s. The rate of fall of the bi-monodisperse interface is also constant for a long time, but the period
100 0s 0s

of changing rates is slightly longer (reflecting the gradient at t = 800 s). The upper interface falls at a constant rate for more than 1000 s. This rate then changes slowly and finally abruptly, as predicted by the profile at t = 1000 s. Fig. 6(b) shows the mean velocities of the three species. Those in the tridisperse suspension change only slightly until species 1 has almost completely settled out. After the first few seconds, the velocities of both species in the bidisperse suspension are roughly constant until almost 700 s. Species 2
0s 0s

height [mm]

50

0 100 400 s 400 s 400 s 400 s

height [mm]

50

0 100 700 s 700 s 700 s 0.0 0.5 700 s 1.0

height [mm]

50

0 100 1200 s 1200 s 1200 s 1200 s

height [mm]

50

0 100 1657 s 1657 s 1657 s 1657 s

height [mm]

50

0 0 0.5
1

0.5
2

0.5
3

0.5

Fig. 7. Solids profiles for Example 4.

M. Bargie, E.M. Tory / Int. J. Miner. Process. 79 (2006) 235252

245

initially moves downward and species 3 initially moves upward. Species 3 continues to move upward when only species 2 and 3 are present. In the monodisperse suspension, a constant velocity is followed by a gradual change. Example 4. Fig. 7 shows the solids profiles. At t = 400 s, the packed bed consists mostly of species 1, but many spheres of species 2 (which move slowly upward, as shown in Fig. 8(b)) are trapped by the rising bed. Since species 3 moves more quickly upward, very few of its spheres are trapped. Above the bed, there is a discontinuity followed by a concentration gradient in 1. This corresponds with an increase in 2 and 3. The bidisperse region demonstrates the Smith effect, which is especially notable for species 2. Finally, the monodisperse region at the top shows a great increase in 3. At t = 700 s, the top of the packed bed consists almost entirely of species 2 with a small concentration of species 3. Above the bed is a discontinuity in species 2 followed by a concentration gradient. These correspond to an increase in 3 from the Smith effect. At t = 1200 s, a discontinuity and a concentration gradient

in 3 lie above the packed bed, which varies substantially in its composition. Fig. 8(a) shows the settling curves. The concentration gradients produce the gradual changes in velocity and the discontinuities in concentration produce the abrupt changes to zero. Fig. 8(b) shows the mean velocities of the three species. Species 2 and 3 move upward initially and species 3 continues to move up when species 1 has settled out. The mean for species 3 increases as a greater proportion are in the concentration gradient. 8. Comparison with experimental data Experiment 1. Law et al. (1987) studied a suspension of almost equal-sized spheres of polymethyl methacrylate (1 = 1186 kg/m3, d1 = 2.41 10 4 m) and polystyrene (2 = 1050 kg/m3, d2 = 2.37 10 4 m) in an aqueous solution of sodium chloride (f = 1120 kg/m3). Computational (Brger et al., 2002) and experimental (Law et al., 1987) evidence indicates that the suspension with 1 = 2 = 0.08 is stable. Accordingly, we used Eq. (15) with K = 2 to simulate this experiment. Law et al. (1987) found that 1max = 0.56, 2max = 0.55, and the RichardsonZaki exponent is n = 5.39. Taking wall effects into account, they calculated u1 = 0.001241 m/s and u2 = 0.001348 m/s. Without wall effects, u1 = 0.001380 m/s and u2 = 0.001500 m/s. In our simulation, the latter values provided a better fit of the experimental data. Fig. 9 shows the solids profiles for Experiment 1. At t = 100 s, the middle section still contains both species. Except for the concentrations of the packed beds, these profiles are similar to those in Fig. 8 of Brger et al. (2000). By t = 200 s, the two species have separated. The simulated settling curves, shown in Fig. 10, agree very closely with the experimental values in Fig. 5 of Law et al. (1987) and also with the calculated values in Fig. 9 of Brger et al. (2000), who used a shock-capturing numerical scheme to simulate this experimental study. Experiment 2. Al-Naaf and Selim (1989) compared the predictions of various sedimentation equations with experimental results for polydisperse suspensions. The data, including some from a supplementary document (footnote on page 263 of their paper), for one of the experiments with a tridisperse suspension (mixture 7 of their Table 3) are as follows: 1 = 0.191, 2 = 0.099, 3 = 0.097, d1 = 3.26 10 4 m, d2=1.63 10 4 m, d3 = 8.1 10 5 m, 1 = 4300 kg/m3, 2 = 2970 kg/m3, 3 = 2500 kg/m3, f = 1113 kg/m3, f = 0.0174 kg/m s. According to Table D.9 of the supplementary document, the experimental values of the velocities of the lower,

(a)

100 Example 4 80

height [mm]

60 40 20 0
packed bed species 1 species 2 species 3

500

1000

1500

time [s]

(b)

0.1 Example 4

velocity [mm/s]

0.05 0.0 -0.05 -0.1 -0.15


species 1, tridisperse species 2, tridisperse species 2, bidisperse species 3, tridisperse species 3, bidisperse species 3, monodisperse

500

1000

1500

time [s]
Fig. 8. (a) Height vs. time plots for the three interfaces and the packed bed for Example 4. (b) Mean velocities of each species in the three zones.

246

M. Bargie, E.M. Tory / Int. J. Miner. Process. 79 (2006) 235252

300 0s 0s 100 s 100 s

height [mm]

200

100

0 0 300 200 s 200 s 296 s 296 s 0.5


1

1 0

0.5
2

1 0

0.5
1

1 0

0.5
2

height [mm]

200

100

0 0 0.5
1

1 0

0.5
2

1 0

0.5
1

1 0

0.5
2

Fig. 9. Solids profiles for Experiment 1.

middle, and upper interfaces were 0.1085 cm/s, 0.0313 cm/s, and 0.0098 cm/s, respectively. Substitution of the relevant values in Eq. (10) yields u1 =
300

200

100

Experiment 1 bottom packed bed top packed bed species 1 species 2

100

200

300

time [s]
Fig. 10. Simulation of Experiment 1 (Law et al., 1987). When the two species have separated, the slopes of the two lines denoting their boundaries change abruptly.

1.0605 10 2 m/s. The supplementary document states that settling rates calculated with the MLB equation exceeded the experimental values by 22.5%, 14.7%, and 10.2% in the lower, middle, and upper regions, respectively. Since the greatest difference occurs at the highest concentration, and the lowest at the lowest, it is clear that a larger value of n would produce a much better fit. This conclusion is supported by the fact that the larger value of n (Garside and AlDibouni, 1977) produced a better fit than the smaller (due to Richardson and Zaki, 1954). For these data, n = 5.48 produced the best fit. Using Matlab Symbolic Math toolbox, we calculated the eigenvalues of the Jacobian J(0) and confirmed that this tridisperse suspension is stable. Fig. 11 shows the solids profiles at various times. In the initial profile, the scatter is large for species 1 and very small for species 3. Each sphere of species 1 has eight times the volume of species 2 and slightly less than twice the concentration. Consequently, the number of spheres of species 2 in the simulation is roughly four times the number of species 1. Similarly, the number of spheres of species 3 is roughly eight times the number of

height [mm]

M. Bargie, E.M. Tory / Int. J. Miner. Process. 79 (2006) 235252

247

100 0s 0s 0s 0s

height [mm]

50

0 100 50 s 50 s 50 s 50 s

height [mm]

50

0 100 200 s 200 s 200 s 0.0 0.5 200 s 1.0

height [mm]

50

0 100 400 s 400 s 400 s 400 s

height [mm]

50

0 100 600 s 600 s 600 s 600 s

height [mm]

50

0 0 0.5
1

0.5
2

0.5
3

0.5

Fig. 11. Solids profiles for Experiment 2 (Al-Naaf and Selim, 1989).

species 2. Since the initial heights of the particles are distributed randomly over H, increasing the number of particles decreases the fluctuations in concentrations. At t = 50 s, the concentrations of species 2 and 3 in the packed bed are less than the initial values because particles of these species move upward. Above the packed bed is a discontinuity in 1, followed by a concentration gradient to the initial concentration. The decrease in 1 corresponds to increases in 2 and 3 to roughly their initial values. Variability in 2 and 3

often increases when species 2 and 3 move upward. Note the Smith effect in the upper regions. By t = 200 s, all the particles of species 1 and most of species 2 have settled. There is a discontinuity in 2 and above that is a concentration gradient. There is a gradient in 3 and above that is a region with the constant concentration attained at t = 50 s. By t = 400 s, species 1 and 2 have completely settled. The top of the packed bed consists entirely of spheres of species 3. Above the bed is a discontinuity, followed by a concentration gradient. By

248

M. Bargie, E.M. Tory / Int. J. Miner. Process. 79 (2006) 235252

t = 600 s, only the discontinuity and the lower part of the gradient remain. By t = 717.94 s the spheres in this gradient have settled into the packed bed (not shown). Note that the profiles for t 50 s are smoother than the initial profiles. Fig. 12(a) shows the settling paths of the three interfaces and the rise of the packed bed. All species show an abrupt change to zero velocity, but the period of changing rates is very short for species 1, moderately long for species 2 and very long for species 3. The long period is caused by the concentration gradient shown in Fig. 11 ( t = 400 s). The abrupt change to zero corresponds to the discontinuity at the top of the packed bed. Fig. 12(b) shows the mean settling velocities of the three species. In the tridisperse suspension, species 2 and 3 move upward. In the bidisperse suspension, species 3 is almost stationary, moving very slightly upward. The mean velocity of species 2 changes gradually as proportionally more of its spheres are in a concentration gradient. The remnant of this gradient is seen in Fig. 11 (t = 200 s). In the monodisperse

suspension, the slow change in the mean velocity of species 3 results from the concentration gradients shown in Fig. 11 (t = 400 s and t = 600 s). 9. Conclusions and discussion By using a rate-controlling region above upwardmoving particles, we have extended the particle-based method to light and heavy species and to concentrated suspensions in which at least one species initially moves upward. We have also improved the approximation of the effect of the packed bed by distinguishing between the effects of active and settled particles in the ratedetermining zone. The approximation used by Bargie et al. (2005) is good enough for particles that differ only moderately in size and have the same density, but cannot be used for spheres that differ in density or greatly in size because it can prevent smaller or lighter spheres from hitting the packed bed. Using the concentration of the region immediately above an upward-moving particle ensures that it remains in that region in the small time-step. Unlike the use of the concentration in the region immediately below downward-moving particles, it does not smooth out fluctuations. Despite these fluctuations, the essential features of the solids profile are maintained. As soon as all velocities are downward, the profiles are smoothed. We could have used initially uniform distributions (Bargie et al., 2005), which would have reduced the variability, but we chose to demonstrate the robustness of the method by carrying out simulations that more closely represent small-scale experiments. Every mixing of a suspension produces different initial positions for each particle. Nevertheless, interface velocities are very reproducible if the number of particles is large. We carried out two simulations of the Al-Naaf-Selim experiment. The fluctuations in the profiles were different, but the profiles were the same in all essential features. No differences could be discerned in either the paths of the interfaces, the rise of the packed bed, or the mean velocities. Settling conditions can also affect the concentration of the packed bed (Bezrukov et al., 2001). The final solids concentration can vary from 0.55 (Law et al., 1987) to 0.64 for uniform spheres, depending on the conditions of sedimentation. Very slow sedimentation of uniform spheres (Verhoeven, 1963) yields values close to the simulated value of 0.58 (Bezrukov et al., 2001). Final concentrations also depend on the size and shape of the container relative to the size of the spheres. For spheres settling in cylinders, max increases with decreasing sphere-cylinder diameter ratio. When the species have different diameters, max varies with

(a)

100 Experiment 2 80

height [mm]

60 40 20 0
packed bed species 1 species 2 species 3

200

400

600

800

time [s]

(b)
0.2 Experiment 2 0.0 -0.2 -0.4 -0.6 -0.8 -1.0 0 200 400
species 1, tridisperse species 2, tridisperse species 2 ,bidisperse species 3, tridisperse species 3, bidisperse species 3, monodisperse

velocity [mm/s]

600

800

time [s]
Fig. 12. (a) Height vs. time plots for the three interfaces and the packed bed for Experiment 2. The slopes of the extended straight lines are experimental values. They coincide with the straight-line portion of the simulated settling curves. (b) Mean velocities of each species in each of the three zones.

M. Bargie, E.M. Tory / Int. J. Miner. Process. 79 (2006) 235252

249

composition (Bezrukov et al., 2001). Concentrations can exceed 0.7 for binary or ternary packings, depending on the proportions of the species and the ratio of sphere diameters. No information was available regarding the concentration of the packed bed in the Al-Naaf-Selim experiment, so we used max = 0.64. In cases where the concentration of the bed is known or can be predicted, it should be possible to use an iteration in which the increase in height calculated for a short time-step is adjusted to take into account the composition of the spheres deposited in that time-step. The simulation automatically adjusts to a change in the concentration of the packed bed. This concentration affects the paths of the interfaces. For example, the long changing-rate period for species 3 would be shortened and the final break in the settling curve would be more pronounced if the concentration at the top of the packed bed were less. Although our method can use vk() calculated from any equation, we believe that the MLB equation has the proper structure. Its merits have been discussed extensively by Brger et al. (2002) and Berres et al. (2003). We contend that the constants of the equation are not universal. In particular, n varies from case to case and should be used as a best-fit parameter. For example, n can vary from 4.0 (based on data from Verhoeven (1963)) to 6.55 (Batchelor, 1972). The value depends on the distribution of positions of particles. Batchelor's result is applicable to colloidal suspensions in which Brownian motion keeps the distribution uniform (AlNaafa and Selim, 1992). Sometimes the Richardson Zaki equation fits experimental values very closely, but the Stokes velocity obtained by extrapolation does not agree with the measured or calculated value. In the two experiments simulated here, we used the actual values of the Stokes velocity (uncorrected) and adjusted n to fit the experimental settling velocities. List of symbols f vector of solid flux densities fk flux density of kth species fki derivative of fk with respect to i h thickness of velocity-determining layer H initial height of suspension Ha height of the packed region above the free surface I2 two-species stability index defined by Eq. (3) I3 three-species stability index defined by Eq. (8) J Jacobian defined by Eq. (2) K number of species N total number of spheres Nk number of spheres of kth species

Nt p3 R S t T u u 1 vk vk Vk z

number of buoyant spheres in the packed bed at the top of the suspension characteristic polynomial of J for three species function defined by Eq. (5) function defined by Eq. (6) time function defined by Eq. (7) Stokes velocity Stokes velocity of the largest species (or densest species if all diameters are equal) velocity of particles of kth species velocity defined by Eq. (31) volume of a sphere of the kth species height

Greek symbols k k / 1 row vector of density ratios (1, 2, , K) k ratio defined by Eq. (13) ki Kronecker delta eigenvalue of Jacobian matrix J k density of kth species s density of solid particles in a monodisperse suspension f density of fluid k k f row vector of densities ( 1, 2, , K) vector of solids concentrations (1, 2, , K)T total solids concentration of sedimenting particles + concentration of active particles in the region between the test particle and the packed bed k concentration of kth species + k concentration of active particles of kth species Acknowledgements We thank the referees for carefully checking the manuscript and suggesting improvements in the presentation of our results. Appendix A. Algorithms for sedimentation of polydisperse suspensions A.1. Calculation of concentrations and velocities At any time t > 0, the system consists of the bottom packed bed (of Nb particles), a column of Na active particles, and a top packed bed (of Nt buoyant particles). Obviously, N = Nb + Na + Nt and Nt = 0 if k > 0, k = 1, , K. The particles are kept in an array sorted according to

250

M. Bargie, E.M. Tory / Int. J. Miner. Process. 79 (2006) 235252

their coordinates. Thus, it is convenient to locate all the particles lying in the ith bin. For particles moving downward we start from the topmost active particle, M = N Nt, and travel down until we find the first particle j with height zj < zM h. Now we have established the bin for particle M. Then we take the next particle M 1. Its bin consists of all particles from the previous bin (except for M) and possibly a few particles below j. So now we have to travel only from particle j down. The procedure for upward-moving particles is symmetric. We start from the lowermost active particle, L = Nb + 1 and travel up to find all particles from its bin and subsequently bins for particles L + 1, L + 2, and so on. In the diagrams below, ki refers to the species to which particle i belongs, and ck is the number of particles of species k in the current bin. // Find velocities for particles moving downward (v < 0) // Particles more than h from the packed bed jp = N Nt for k = 1 to K do ck = 0 for i = N Nt downto Nb + 1 while zi > Bb + h do if vi < 0 then for j = jp downto Nb + 1 while zj > zi h do ckj = ckj + 1 jp = j =0 for k = 1 to K do k = ckVk/(Ah) = + k end do calculate vi from Eq. (15) end if cki = cki 1// disregard particle i in (i 1)th bin end do // Particles less than h from the packed bed e = false for i downto Nb + 1 while not e do if vi < 0 then = (zi Bb)/h for j = jp downto Nb + 1 while zj > zi h do ckj = ckj + 1 jp = j + = 0 for k = 1 to K do + k = ckVk/(Ah) + + = + + k end do if + < max = max(1 ) + +

calculate vi from Eq. (15) vk i = v i else e = true v i = vk i end if end if cki = cki 1//disregard particle i in (i 1)th bin end do // Particles very close to the packed bed for i downto Nb + 1 do if vi < 0 then vi = vki // Find velocities for particles moving upward (v > 0) // Particles more than h from the packed bed jp = Nb + 1 for k = 1 to K do ck = 0 for i = Nb + 1 to N Nt while zi < Bt h do if vi > 0 then for j = jp to N Nt while zj < zi + h do ck j = ck j + 1 jp = j =0 for k = 1 to K do k = ckVk/(Ah) = + k end do calculate vi from Eq. (15) end if cki = cki 1 // disregard particle i in (i 1)th bin end do // Particles less than h from the packed bed e = false for i to N Nt while not e do if vi > 0 then = (Bt zi)/h for j = jp to N Nt while zj < zi + h do ck j = ck j + 1 jp = j + = 0 for k = 1 to K do + k =ckVk/(Ah) + + = + + k end do if + < max = max(1 )++ calculate vi from Eq. (15) vk i = v i else e = true v i = vk i

M. Bargie, E.M. Tory / Int. J. Miner. Process. 79 (2006) 235252

251

end if end if cki = cki 1 // disregard particle i in (i 1)th bin end do // particles very close to the packed bed for i to N Nt do if vi > 0 then vi = vki A.2. Settling of the particles The coordinates of the spheres are updated according to the equation z i z i vi D t ; i Nb 1; N ; N Nt : 32

The particle array is sorted in every time-step to facilitate easy calculation of concentrations. However, it would be inefficient to sort the whole array, since the settled particles do not move and hence their order does not change. On the other hand, it is not sufficient to sort only the active particles since, near the settled-active boundary (Bb at the bottom and Bt at the top), the settled and active particles could have intermixed as a result of the motion process. We set t such that the active particles cannot travel further than h in one time-step. Therefore, below Bb h and above Bt + h there are only settled particles that do not need re-sorting. Hence, we sort only the part of the particle array between positions nb and nt where nb maxNb 1 t/max Ah=VKb b; 1; nt minN Nt t/max Ah=VKt b; N ; 33 34

proceed from particle nt and go downward until we encounter a particle of height zi > Bt. The main settling loop is is given below for i = nb to N Nt while zi Bb do if si = 1 then // settle only active particles Nb = Nb + 1 si = 0 // mark particle as settled Bb = Bb + Vi /(Amax) // raise packed bed end if end do for i = nt downto Nb + 1 while zi Bt do if si = 1 then // settle only active particles Nt = Nt + 1 si = 0 // mark particle as settled Bt = Bt V i /(Amax) // raise packed bed end if end do

References
Al-Naaf, M.A., Selim, M.S., 1989. Sedimentation of polydisperse concentrated suspensions. Can. J. Chem. Eng. 67, 253264. Al-Naafa, M.A., Selim, M.S., 1992. Sedimentation of monodisperse and bidisperse hard-sphere colloidal suspensions. AIChE J. 38, 16181630. Bargie, M., Ford, R.A., Tory, E.M., 2005. Simulation of sedimentation of polydisperse suspensions: a particle-based approach. AIChE J. 51, 24572468. Batchelor, G.K., 1972. Sedimentation in a dilute dispersion of spheres. J. Fluid Mech. 52, 245268. Batchelor, G.K., Janse van Rensburg, R.W., 1986. Structure formation in bidisperse sedimentation. J. Fluid Mech. 119, 379407. Berres, S., Brger, R., Karlsen, K.H., Tory, E.M., 2003. Strongly degenerate parabolic-hyperbolic systems modeling polydisperse sedimentation with compression. SIAM J. Appl. Math. 64, 4180. Bezrukov, A., Stoyan, D., Bargie, M., 2001. Spatial statistics for simulated packings of spheres. Image Anal. Stereol. 20, 203206. Biesheuvel, P.M., Verweij, H., Breedveld, V., 2001. Evaluation of instability criterion for bidisperse sedimentation. AIChE J. 47, 4552. Brger, R., Concha, F., Fjelde, K.-K., Karlsen, K.H., 2000. Numerical simulation of the settling of polydisperse suspensions of spheres. Powder Technol. 113, 3054. Brger, R., Karlsen, K.H., Tory, E.M., Wendland, W.L., 2002. Model equations and instability regions for the sedimentation of equal spheres. Z. Angew. Math. Mech. 82, 699722. Fessas, Y.P., Weiland, R.H., 1981. Convective solids settling induced by a buoyant phase. AIChE J. 27, 588592. Fessas, Y.P., Weiland, R.H., 1984. The settling of suspensions promoted by rigid buoyant spheres. Int. J. Multiph. Flow 10, 485507. Garside, J., Al-Dibouni, M.R., 1977. Velocityvoidage relationships for fluidization and sedimentation in solidliquid systems. IEC Proc. Des. Dev. 16, 206214. Law, H.-S., Masliyah, J.H., MacTaggart, R.S., Nandakumar, K., 1987. Gravity separation of bidisperse suspensions: light and heavy particle species. Chem. Eng. Sci. 42, 15271538.

where Kb and Kt denote the smallest heavy and buoyant species, respectively. In the sorting process we lose information about the state of the particle (settled, or active), so we give the particle structure an additional field that contains information about its status. The value of this field is as follows:  s
i

0 settled particle 1 active particle:

35

As the particles settle into the packed bed, we do not know the index of the lowest active particle. Once again we use estimates nb and nt given by Eqs. (33) and (34). Then, for the bottom packed bed, we start from particle nb and travel upward until the first particle for which zi > Bb. For the top packed bed we

252

M. Bargie, E.M. Tory / Int. J. Miner. Process. 79 (2006) 235252 Verhoeven, J., 1963. Prediction of batch settling behaviour. B. Chem. Eng. thesis, McMaster University. Weiland, R.H., Fessas, Y.P., Ramarao, B.V., 1984. On instabilities arising during sedimentation of two-component mixtures of solids. J. Fluid Mech. 142, 383389.

Lockett, M.J., Bassoon, K.S., 1979. Sedimentation of binary particle mixtures. Powder Technol. 24, 17. Masliyah, J.H., 1979. Hindered settling in a multi-species particle system. Chem. Eng. Sci. 34, 11661168. Richardson, J.F., Zaki, W.N., 1954. Sedimentation and fluidization: Part 1. Trans. Inst. Chem. Eng. (London) 32, 3553.

You might also like