12

Disk-Stack Modeling

In this chapter, the separation performance of a disk-stack centrifuge

is modeled. A schematic of the dropping-bottom intermittent discharge
disk centrifuge is shown in Fig. 12.1. The model discussed in this chap-
ter is also applicable to the nozzle disk and the manual discharge disk.
This model is validated with results from a pilot test using a disk-stack
centrifuge. As shown in Chapter 13, Performance Projection of Centrifuges
in Bioseparation, the disk model can be used to simulate performance of
disk-stack centrifuge of different sizes operating at various feed rates and
rotational speeds applicable to various bioprocess and biotech applications.
A suspension of particles is continuously fed to the disk centrifuge.
Under centrifugal acceleration G, heavier solids settle to the upper sur-
face (Fig. 12.3) of the annular disk channel formed between two disk
surfaces, displacing lighter liquid phase toward the small radius. The
collected sediment slides down the upper surface of the disk channel
under the longitudinal component of the G-force until it leaves the disk
stack and is collected at the solids holding annular space of the bowl.

12.1 Disk Model

In the simple model [1] presented herein, the following assumptions are
made:

1. The solid concentration is relatively dilute, such that the presence of

solid does not affect the flow field.
2. The aspect ratio of the channel (longitudinal length to the channel
height or spacer height between adjacent disks) is very large and the
details of the entrance and exit geometry respectively at both ends
can be ignored. For modeling purpose, the triangular region at the
inlet of the disk channel can be eliminated; similarly the triangular
region at the exit of the disk channel can be added (Fig. 12.2), so
that the channel has inlet area and exit area exactly perpendicular
to the longitudinal direction of the channel. By virtue of the large
aspect ratio and the strong rotational field at high rotation speed, the

Centrifugal Separations in Biotechnology. DOI: https://doi.org/10.1016/B978-0-08-102634-2.00012-X

© 2020 Elsevier Ltd. All rights reserved. 261
262 Centrifugal Separations in Biotechnology

Ri
Ro
D

Figure 12.1 Schematic of a disk-stack centrifuge.

Flow

Figure 12.2 Entrance and exit channel region (annulus) being ignored.

flow develops quickly (i.e., for an entrance distance equivalent to

several channel spacing) to a steady velocity profile.
3. The flow is uniformly distributed across all the ‘n’ channels. This
assumption is very weak, as shown in Fig. 9.9. In any case, only a
single channel is considered in the model.
4. The flow field in each channel (Fig. 12.3) is independent of particle
sedimentation; however, the reverse holds in that the sedimentation
of a particle depends on the velocity profile in the channel. This
statement holds for dilute suspension and where particles are rela-
tively small compared to the channel spacing.
5. The velocity profile in the suspension is assumed to be relatively con-
stant across the entire channel within a given channel, despite there are
Ekman layers and Ekman flow adjacent to the upper or lower surfaces
 Disk-Stack Modeling 263

h Ri

Upper surface

Particle θ Ω

G 1
2
3 L

Lower surface

y z
yc
Ro

Figure 12.3 Velocity profiles and trajectory of particles in the disk

channel.

of the disk surfaces and also a thin clear liquid flowing on the upper
surface of the disk channel analogous to the lamella settler [2].

The flow field in the channel is assumed to be dominated by viscous

effect. For a rotation speed of, say, 5000 rpm or 524 rad/s, kinematic vis-
cosity 10 times that of water at 0.1 cm2/s, the combined viscous Ekman
layers, respectively, at the upper and lower disk surfaces is of order of 2
(ν/Ω)1/2 5 0.28 mm. (The angle effect has been ignored.) Typical spacing
between adjacent disks can be as small as 0.25 mm and as large as 1 mm 1 .
For high performance, disk spacing is small, therefore the assumption of vis-
cous flow in the channel is not unwarranted. Consequently, we do not con-
sider the geostrophic flow (no viscous effect) in the channel unless the disk
spacing is large, over 1 mm.
Referring to Fig. 12.4, a particle in the channel is subject to convec-
tion from the main flow along z-direction, as well as relative motion due
to sedimentation under G.
Uniform distribution in Fig. 12.3 implies that the flow rate to each
channel is identical and only a single channel needs to be modeled. We
separately consider the continuum phase (which is the suspending liquid)
and the particles (the dispersed phase).
264 Centrifugal Separations in Biotechnology

U
Particle

θ vs cosθ
vs
vs sinθ

Figure 12.4 Particle velocity in the channel.

12.1.1 Continuum Phase

The continuity and momentum balances (without considering the settling
particles) imply the following.

1. Continuity:
The flow rate per unit channel is equal to the overall feed rate Q divided
evenly among n channels in n 1 1 disks, thus
Q 5 2πnRUh (12.1)

where R is the radius, U(z) is the local longitudinal velocity in the

z-direction, and y is the transverse coordinate measured from the
upper surface of the disk channel, and h is the channel height, that
is, distance between adjacent disks (Fig. 12.3). This equation holds
from the feed entrance R 5 Ro along increasing z to the disk channel
exit at the small radius Ri
2. z-momentum:

Despite that we have assumed U being constant, independent of the trans-

verse coordinate y. To satisfy continuity or mass conservation, Eq. (12.1),
the longitudinal velocity has to increase with reducing radius.


Q=n 1
UðzÞ 5 (12.2)
2πh R

12.1.2 Dispersed Phase

The dispersed phase is sufficiently dilute such that the presence of parti-
cles does not affect the flow field, on contrary the rate of deposition and
removal of them depend on the main flow. The relative velocity of settling
particle in relation to the liquid (in relation to the surrounding liquid under
steady-state) with negligible acceleration and deceleration is given by Stokes’
law as discussed in Chapter 2, Principles of Centrifugal Sedimentation.
 Disk-Stack Modeling 265

Given x being the particle size (equivalent spherical diameter), Stokes’ law
in a centrifugal field states
 
ρs 2 ρL Gx2
vs 5 (12.3)
18μ
The above holds under 0 # x ,N, 0 # y # h, 0 # z # L. The fre-
quency distribution f(x) of particle residing in a given size range is the
derivative of the cumulative undersize distribution F(x)
dFðxÞ
f ðxÞ 5 (12.4a)
dx
ðx
FðxÞ 5 f ðxÞdx (12.4b)
0

The frequency size distribution f(x) or cumulative size distribution F(x)

is assumed to be known from measurement.
The trajectory can be determined for a given particle located at trans-
verse location y at the inlet (i.e., z 5 0) in a flow field, as illustrated in
Fig. 12.3. This approach is to determine whether the particle trajectory
intercepts the upper collecting surface of the channel prior to traveling
through the channel length ending at z 5 L. Consider a particle, with size
x starting at z 5 0 and taking on trajectory 1, which gets settled as it tra-
vels along the disk channel. If the particle had been positioned differently
at the inlet and had taken trajectory 3, it would have escaped from being
settled. With reference to Fig. 12.3, trajectory 2, with particle at channel
inlet z 5 0 and y 5 yc, is the critical trajectory wherein particle with size x
barely gets captured as it travels across the disk channel (Fig. 12.4).
Therefore the fraction captured in the sediment Zs for particle size x is
h 2 yc
Zs ðxÞ 5 (12.5)
h
where h is the channel height as set by the spacer between adjacent disks.
Assuming particles of all sizes are initially distributed uniformly across the
inlet of the channel from the limiting trajectory, the solids recovery or size
capture can be determined. Another important point is to determine the cut
size, that is the particle positioned at the inlet lower corner of the channel
z 5 0 and y 5 0 (Fig. 12.3) that undertakes the critical trajectory such that it
ends up at y 5 h at z 5 L. Particles larger than the cut size would be fully
settled in the channel; vice versa, only a fraction would be settled for parti-
cles smaller than the cut size. With the above information, the fraction cap-
tured for particles of all sizes can be schematically represented in Fig. 12.5.
266 Centrifugal Separations in Biotechnology

1
Zs
0.8
Ze
0.6
Zs , Z e

0.4

0.2

0
0 0.5 1 1.5 2
x/xc

Figure 12.5 Capture fraction for different particle size x.

As stated, the particle settling velocity is Stokes’ velocity relative to the

moving liquid under the influence of the G-acceleration (Fig. 12.4). There
are two velocity components that should be considered: one component is
perpendicular to the underside of the disk surface in which a particle is set-
tling toward with a velocity vscosθ; another is the longitudinal component,
vssinθ, which is in the negative direction of the streamwise direction. This
second component should be superimposed with the streamwise flow with
velocity U(R), that is U 2 vssinθ. Let us consider a small incremental path
(dz, dy) along the particle trajectory. Thus it follows that
dy dz
dt 5 5
vs cosθ U 2 vs sinθ
or
dz U 2 vs sinθ
5 (12.6)
dy vs cosθ
From Fig. 12.3 by virtue of the geometry
R 5 Ro 2 zsinθ (12.7)

It follows that dR 5 2sinθdz and substituting this in Eq. (12.6),

1 dR b
2 aRsinθ
2 5 R
(12.8)
sinθ dy aRcosθ
where


1 Δρ
a5 ðΩxηÞ2 (12.9a)
18 μ
 Disk-Stack Modeling 267

Q=n
b5 (12.9b)
2πh
In Eq. (12.9a), η is added to account for the acceleration efficiency.
Note that the coefficient a is derived from Stokes’ law, whereas coeffi-
cient b is from the continuity Eq. (12.2). Integrating Eq. (12.5) along the
critical trajectory where particle initially at y 5 yc just barely gets cap-
tured. The two limits have to be R 5 Ro to R 5 Ri, and y 5 yc and y 5 h,
respectively. Also for all practical purposes the component of the
Stokes’ velocity is assumed to be much smaller than the streamwise
velocity component, thus aR2sinθ ,, b/R.
ð Ri ðh
aR2 cosθ dR
2 5 dy (12.9c)
Ro b 2 aR sinθ sinθ
2
yc

Assuming vs/U ,, 1, with substitution of a and b from Eqs. (12.9a)

and (12.9b), respectively, after rearranging we get


a cotθ  3  π ΔR ðΩηÞ2 hcotθ  3 
h 2 yc 5 Ro 2 R3i 5   Ro 2 R3i x2 (12.10)
b 3 27 μ Q=n

Assuming uniform particle concentration across the entire channel,

the fractional capture of particle with size x from Eq. (12.5) becomes


2
h 2 yc π ΔR ðΩηÞ2 cotθ  3  x
Zs 5 5   Ro 2 R3i x2 5 (12.11)
h 27 μ Q=n xc

where xc is defined as
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi

   ffi
27 μ Q=n 1
xc 5   (12.12)
π ΔR η2 Ω2 cotθ R3o 2 R3i

From Eq. (12.11) the dimensionless Leung number, abbreviated as

Le number, that governs the settling behavior of the disk stack can be
defined as
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
!
u

u 3Q μ tanθ
t    
n ρs 2 ρL Ro 2 R3i
3

Le 5 (12.13)
Ωxo η
where n is the number of disks, Ro is the radius of the outer disk and Ri
is that of the inner disk, θ is the angle subtended between the disk
268 Centrifugal Separations in Biotechnology

surface and the vertical axis, μ is the suspension viscosity, ρs is the solid
density, ρL is the liquid density, Ω is the angular velocity, and η is the
acceleration efficiency. The cut size is thus given by
xc 3
5 pffiffiffi Le (12.14)
xo π

In Eq. 12.13, n is the number of disk channels. Given the number of

disks should be n +1, but as n is usually very large, we refer n also as
the number of disks, which is a good approximation. The rest of the
governing equations are identical to the ones (i.e., solid recovery etc.)
that have already been presented in Chapter 11, Visualization and
Modeling of Flow and Separation in Tubular Centrifuge, given that the
Zs function is identical with the tubular model.

12.2 Model Validation

It would be useful to validate the model with some experimental results [3].
A suspension of cell culture contains viable mammalian cells 1020 μm,
nonviable cells 510 μm, and cell debris less than 5 μm. (In other applica-
tions after passing through the homogenizer, cell debris is usually much
smaller, less than 0.5 μm.) The size distribution in terms of number of parti-
cles is shown in Fig. 12.6. The suspension was separated in a pilot disk-stack
centrifuge with a bowl diameter 190 mm. The cell debris and nonviable cells
are removed in the centrate and the viable cells are captured in the concen-
trate. Fig. 12.7 shows the centrate solids by volume leaving the 190-mm disk
centrifuge plotted against the feed rate to the centrifuge. (The solids %v/v of
the centrifuge centrate was determined from spindown at 14,000 g for

1800
Number of particles

1500 Non
Cell Viable cells Feed
debris viable
1200
cells
900

600

300

0
0 5 10 15 20 25
x (μm)
Figure 12.6 Particle size distribution of feed suspension [3]. Reproduced
with permission from the American Filtration and Separation Society.
 Disk-Stack Modeling 269

190-mm bowl dia. 10,000g μ=1.9 cp, Δ ρ/ρ =0.01

0.10

%Centrate solids, v/v*

0.09
0.08 Prediction Qcrit =10.5 L/m
0.07 Tests x c =11.4 μm
0.06
0.05
0.04
0.03
0.02
0.01
0.00
0 5 10 15
Q (L/m)
*Samples spun at 14,000g @ 10 min

Figure 12.7 Centrate solids from a 190-mm disk centrifuge. Solid

curve is prediction from the present model and test data are from
Reference [3].

190-mm bowl dia. 10,000g μ=1.9 cp, Δρ/ρ =0.01

100
%Solids recovery

Prediction
99
Tests
98
>99% recovery of
solids, @ Q<7 L/m
97
>99.5% @Q<4 L/m
96

95
0 5 10 15
Q (L/m)

Figure 12.8 Solid recovery versus feed rate. Solid curve is prediction
from the present model and test data are derived from Fig. 12.7.

10 minutes in a spintube containing a centrate sample.) The solid curve is the

model prediction and the data are results from the experiment [3]. The two
agrees extremely well with each other. It is clear that the centrate solids
increase with increasing feed rate. When the feed rate reaches 10 L/min the
centrate solids shoots up dramatically. For example, the cut size xc is deter-
mined to be at 11.4 μm, and this is within the size range of the mammalian
cells (Fig. 12.6). When the centrifuge is operating at this high rate, some of
the particles that were intended to be captured also escape in the centrate. In
this case the feed rate should be maintained below 67 L/min to avoid loss
of product in the centrate.
Another measure is to examine the solids recovery as a function of
feed rate, which is depicted in Fig. 12.8. As shown, the solids recovery
270 Centrifugal Separations in Biotechnology

drops gradually initially at low rate and when the feed rate reaches
10 L/min, the solid recovery drops precipitously, even with a small increase
in feed rate. Note that the model prediction compares very well with the
test data.

12.3 Complications
One major complication is that the feed may not distribute uniformly to
all the channels. Fig. 9.9 shows a computation fluid dynamics simulation
of flow into a stack of plates under the Earth’s gravity, wherein a compli-
cated flow pattern occurs near the entrance of the channels. This is per-
ceived to happen also with disk-stack centrifuge. The nonuniformity is
accounted for by the efficiency η in the Le number. Low efficiency
implies larger Le number which is undesirable.
Another complication is that the separated clarified liquid stays on the
upper side of the disk and flows rapidly up the channel by buoyancy
force. Because it is lighter in density compared with the surrounding sus-
pension, the velocity of the clarified layer is very high and can entrain
particles from the suspension adjacent to it. In fact there could be instabil-
ity between the two liquid layers [4]. This is simulated under 1 g using an
inclined plate settler, as shown in Fig. 12.9. Presumably similar phenom-
ena also occur for high-speed centrifugation.

Suspension
Upper disk
wall

Axis

1g
Sediment

Clarified
Feed
liquid
Lower disk
Thin clarified wall
layer

Figure 12.9 Flow of thin clarified layer and thick suspension layer in a
simulated disk channel under the Earth’s gravitational acceleration.
 Disk-Stack Modeling 271

One last complication is that the flow might not have been fully
accelerated when it reaches the entrance of the disk stack in which the
feed experiences lower centrifugal acceleration G 5 v2/R, which may be
less than that of the solid-body acceleration G 5 Ω 2R. The efficiency
factor accounts to some extent for this effect and indeed comes up with
a Le number which is much larger than it should have been if the feed
were 100% accelerated.

12.4 Summary
A simple trajectory model on particle deposition in a disk-stack centri-
fuge has been presented. A viscous flow is assumed in the channel formed
between adjacent disks. The flow is independent on the details of the
velocity profile. Among the typical process properties of the solids and
liquid, a necessary input to the model is the feed size distribution. The
model has been validated against an experiment on classification cells and
debris using a disk-stack centrifuge. The model presented can be used to
complement testing or predict results based on conditions for which test
results are not available. It can also be used to project performance for a
different size machine, such as scale-up/scale-down. This serves as a basis
for simulation for which feed particle size measurement is available
(Chapter 15, Flocculation with Decanter Centrifuges), or for which analyt-
ical forms are assumed on the feed size distribution with support from
testing (Chapter 16, Case Studies of Monotonic and Unimodal Size
Distribution Models, and Chapter 17, Classifying Bimodal Particle Size
Distribution and Case Study of Inclusion Body Classification).

References
[1] W. Leung, “Simulating Centrifugal Recovery of Protein in Biopharmaceuti-
cal Production using AT-SPIN Simulator”, presented at the American
Filtration and Separation Society Annual Conference, 2005, Atlanta, GA.
[2] W.W.F. Leung, R. Probstein, Lamella and tube settlers - Part 1 Model
and operation, Ind. Eng. Chem. Process Des. Dev 22 (1983) 5867.
[3] M. Rohr, H. Tebbe, Clarification of mammalian cells by stacked disk
centrifugation, in: presented at the American Filtration and Separation
Society Annual Conference, 2002, Galveston, TX.
[4] W. Leung, Lamella and tube settlers - Part 2 Flow stability, Ind. Eng.
Chem. Process Des. Dev. 22 (1983) 6873.
272 Centrifugal Separations in Biotechnology

Problems
(12.1) Why are the streamlines shown in Fig. 12.3 curved? Can the par-
ticles influence the fluid streamlines?
(12.2) Why can one ignore vssinθ in comparison with the throughflow
velocity U? (Hint: Make an estimate on their orders of magnitude.)
(12.3) Show that Eq. (12.12) is identical to Eqs. (9.4a), (9.5), and (9.6).
What is the rationale for Eqs. (9.4a), (9.5), and (9.6). [Hint:
Compare Eq. (9.4a) with Eq. (9.9)]
(12.4) For larger particles ( . 100 μm) in which inertial effect of the
particle becomes important, would the trajectory model still work?
Why and why not?