Anda di halaman 1dari 10

Numerical calculations of

flow in a hydrocyclone
operating without an air core
Malcolm R. Davidson
CSIRO Division of Mineral Engineering, Lucas Heights Research Laboratories,
Lucas Heights, N.S.W., 2232, Australia
(Received March 1987; revised October 1987)

Steady flow in a hydrocyclone operating without an air core is modelled


by a finite difference solution of the Navier-Stokes equations, following
the approach of Pericleous and Rhodes, in which the shear stress due to
tangential motion is derived from the familiar Prandtl momentum transport
theory, applied to angular momentum. In this application, the Prandtl model
breaks down very near the axis of symmetry and a boundary condition,
corresponding to zero shear, must be imposed at a radius of about one
mixing length to ensure realistic flow predictions. Calculations are based
on a commonly used solution procedure for velocity and pressure which
uses the SIMPLE algorithm of Patankar and Spalding. Hybrid first-order
upwind/central differencing is used, and calculated flow velocities are obtained which agree with both published data and analytical predictions.
The corresponding transport equation for the dispersed (particle) phase
is solved similarly, and the predicted efficiency curve and distributions of
particle concentration are shown. Finally, predicted velocity and particle
distributions are compared with corresponding results based on quadratic
upstream differencing of the governing equations; it is concluded that
numerical diffusion has little effect on the predicted fluid or particle movement in this case.

Keywords: hydrocyclone, numerical model, swirling flow, numerical diffusion

Introduction
A hydrocyclone (Bradley ~, Svarovsky 2) is a particle
separation device, widely used in industry, in which a
particle-fluid mixture is injected tangentially, creating
a strongly swirling, recirculating flow. When the hydrocyclone is open to the atmosphere, an air core forms
along the centre line. Particles move relative to the
fluid according to a balance between centrifugal force
and viscous drag, with separation depending on particle size, density, and shape. For example, in classification according to particle size, large particles tend
to move to the cyclone periphery from where they pass
to the underflow while small particles are carried to
the centre and into the overflow.
Recently, Pericleous, and Rhodes 3 modelled the flow
and particle transport in a hydrocyclone using the com-

1988 Butterworth Publishers

mercially available fluid flow computer program


PHOENICS 4 to solve numerically the Navier-Stokes
equations, incorporating a modified Prandti mixing
length model of turbulent viscosity. An undiscussed
feature of their model is their equation for tangential
motion which is based on Prandtl's theory applied to
angular momentum. It implies, for constant density,
that angular momentum is transported like a passive
scalar and not according to the corresponding laminar
flow equation modified by turbulent viscosity. The experiment of KeisalP was simulated, and satisfactory
agreement between model and experiment was obtained for the tangential velocity distribution, pressure
drop, and flow split.
Despite the favourable comparisons reported above,
further testing of the Pericleous-Rhodes model is deAppl. Math. Modelling, 1988, Vol. 12, April

119

Numerical calculations of flow in a hydrocyclone: Malcolm R. Davidson

sirable to give confidence in its general validity as a


design tool, not only because of the uncertainty surrounding the applicability of individual turbulence
models in complex flows but also because of the possible importance of numerical diffusion in the firstorder upwind differencing used by PHOENICS. In this
paper we consider the related, but computationally
simpler, case of a hydrocyclone operating without an
air core (unlike the Kelsall case). This circumstance
occurs when the hydrocyclone forms part of a piping
network or in liquid-liquid separation (e.g., water-oil)
where the air core is actively suppressed.
We simulate the experiment of Knowles et al. 6 using
the model of Pericleous and Rhodes, and confirm that
a model equation for the tangential velocity, based on
the Prandtl theory, is appropriate. Simply using the
laminar flow equations, modified by turbulent viscosity, for the tangential as well as the other velocity
components leads to a predicted tangential velocity
distribution quite unlike that found experimentally, a
result consistent with the experience of others.
Calculations are based on methods embodied in the
computer programs TEACH (Gosman and Ideriah 7)
and PHOENICS, and are performed using the author's
own special-purpose program. In particular, velocity
and pressure are solved on a staggered mesh using the
SIMPLE algorithm of Patankar and Spalding. 8 As in
TEACH and PHOENICS, hybrid first-order upwind/central differencing is used for the fluid flow
equations and the transport equation for the dispersed
(particle) phase. Calculated vertical and tangential flow
velocities and pressure drops, for the given flow split,
are compared with the corresponding data of Knowles
et al. and (for vertical velocities) with the theoretical
predictions of Bloor and Ingham. 9 The predicted efficiency curve and distributions of particle concentration
are also shown. Finally, to assess the importance of
numerical diffusion in this application, the calculations
are compared with corresponding results based on the
more accurate quadratic upstream differencing of the
governing equations.

The model
The device used by Knowles et al. was a conventional
small-angle "cylinder-on-cone" hydrocyclone and is
shown schematically in F i g u r e 1 together with its relative dimensions. The diameter of the cylindrical section is denoted by D, its height by H, and the respective
diameters of the inflow, underflow, and overflow by
dl, d,, and do. The length of the vortex finder is denoted
by h, and the overall height of the device by L. The
familiar cylindrical coordinates (r, x, 0) are used with
the origin centred at the underflow.
To avoid solving a fully three-dimensional problem,
we assume that the flow is axisymmetric even though
this will clearly not be the case near the entry region
at the top. Instead of the fluid entering the cyclone
tangentially through a pipe, it is assumed to enter through
a circular strip of width d; on the cylindrical wall. The
radial velocity of the incoming fluid is chosen to be
120

Appl. Math. Modelling, 1988, Vol. 12, April

OVERFLOW

Vorl'ex

finder

Retotive Dimensions
d~ID =0 2B
do/D = 0 3t+
dJ D = 0163
hiD=Ok
HID=O 5
LID = 50

ct, = 5.65*

.____u #
r

-Nd. b-

Underflow
Figure I Schematic representation of the hydrocyclone of
Knowles et aL, 6 including relative dimensions

consistent with the specified volumetric feed rate, and


its tangential velocity is taken to be some fraction (3')
of the mean axial velocity in the inlet pipe (to allow
for viscous IossesJ).
Following Pericleous and Rhodes 3 and Pericleous et
al.,~ we assume that the motion of the fluid particle
mixture is steady, incompressible, and satisfies the Navier-Stokes equations with variable viscosity /zcrr (to
model the effect of turbulence) and density p (to account for particle concentration effects). Particles of
each given size are considered as a continuous phase
moving relative to the fluid mixture. The model equations governing axial, radial, and tangential velocities
(u, v, w, respectively), the pressure p, and particle
concentration c follow:
continuity:

--(pU)ax

l0

('pv) =

(I)

momentum:
O (push)+

1 a

7 g. (,pv6)

a( tXerr

Ox

+ r Or

(2)

Numerical calculations of flow in a hydrocyclone: Malcolm R. Davidson


for ~b = u , v , W (where W = rw is the circulation and
/gW is angular momentum), where

ap

S,, = - - Ox

(3a)

Op
S~ = - + / 9 - -

w 2

-/x~fr-g

Or

(3b)

r~

Sw = 0

(3c)

particle concentration:
_

1O

O (pUpC) + - - - (rpvpc)
Ox
r Or

c9
ac
= ~xx Fefr

( ;r)

1__0 rr~fr
+ r Or

(4)

where (up, vp) are the (x, r) components of particle


velocity, F~rr//9is the effective turbulent diffusion coefficient, and axial symmetry (o/ao = 0) is assumed. At
mesh points adjacent to a solid boundary, wall friction
must be included in equation (2). Additional terms in
the source equations (3a) and (3b) associated with a
varying/Z, rf (and which become zero when/zerr is constant) are found to have a negligible effect and are
omitted. For convenience we now restrict our attention
to low particle concentrations so that/9 can be taken
as uniform, and particle transport uncouples from the
underlying fluid flow which can now be determined
independently. Also we set F~rr = /~fr since they are
of the same order (see, e.g., Launder and Spalding II,
p. 28).
If/z~rr is replaced by the laminar viscosity /zt, the
equations for u and v (but not W) are identical to the
corresponding laminar flow equations. Pericleous and
Rhodes 3 do not discuss their model equation for W.
Away from solid boundaries, it appears to be that above,
which implies that angular momentum is transported
at interior points like a passive scalar since Sw = O,
unlike an equation based on the corresponding laminar
form for which Sw = -(2/r)(a/Or) (/X~rrW). This is
equivalent to assuming that the shear stress component
ZrO due to tangential motion is given by
~'m =

/X~fr

w)

(5a)

rather than the commonly used expression

:ow w)

(5b)

which is based on the corresponding laminar form. This


can be seen more clearly by writing the equation for
angular momentum in terms of components of the stress
tensor r:
I0 rvW)
x(puW) + 7 ~ ( /9
= r ( O (z0.,.) + Ir_~-r (r2v) )

(6)

Lilley and Chigier, 12 for example, give the relevant

equation for w, and a simple rearrangement yields


equation (6) for W. Setting ~ox = I.~eff(Ow/Ox) and 1"re
according to either (5a) or (5b) reduces equation (6) to
a form corresponding to equation (2) for tb = W, with
Sw = 0 or Sw = -(2/r)(O/Or)(I.~efrW), respectively.
Equation (5a) derives from Prandtl's momentum transport theory in which the angular momentum of fluid
particles displaced perpendicular to circular streamlines is assumed to remain constant (Wattendorf~3); the
derivation is analogous to that for unidirectional flow
(see, e.g., Davies, ~4 p. 14).
If the equation for W is based on equation (5b), the
momentum equations for all velocity components have
the same form as the corresponding laminar flow equations, modified by an isotropic turbulent viscosity. Such
isotropic turbulence closures become less valid in nonrecirculating swirling flow (e.g., Lilley and Chigier, ~2
Lilley I~) as the level of swirl increases, and are found
to be unsuitable for the strongly swirling recirculating
flows found in cyclone separators (Ayers et al.16). Indeed, it is our experience that such a model predicts
a tangential velocity which increases with radius across
the entire width of the cyclone, rather than increasing
only near the axis of symmetry before decreasing
towards the periphery, as is observed experimentally.
In view of this qualitative difference between theoretical and experimental behaviour ~, and the success
achieved by Pericleous and Rhodes using a model based
on equation (5a), there seems little point in considering
an approach based on equation (5b) with anisotropic
turbulent viscosity. A higher-order turbulence closure
has also been successfully used to model flow in gas
cyclones, 16"17but it is not considered here.
The Pericleous-Rhodes form for/zefr was based on
a generalisation of Prandtl's expression for/zerr, corresponding to unidirectional flow, in which the velocity
gradient was replaced by aw/ar - w/r. Here we use a
representation consistent with the Prandtl theory applied to angular momentum, on which the model Wequation is based, namely,
~rr = I~i + pA z

(7)

where ~i models the turbulence entering the hydrocyclone (since Ow/ar + w/r = (l/r)(OW/Or) is small near
the inlet where the axial velocity is low), and ,~ represents a mixing length. Here we have ignored the
(much smaller) underlying laminar viscosity/~t in equation (7). The same p,~fris used in each model momentum
equation. We set
la,i = O.04pwiAi

(8)

where the subscript i denotes an inlet value. Equation


(8) is based on/zz = pA~w' (Launder and Spalding, jl p.
25) with turbulence velocity w' = 0.04w,. for axial flow
in a pipe (Davies, j4 p. 28). Following Pericleous and
Rhodes, we set A; = 0.1dg and A = /3d, where d = D
in the cyclone and d = d,, in the vortex finder with the
fraction /3 to be adjusted to give " b e s t " agreement
between the calculated and experimental velocity dis-

Appl. Math. Modelling, 1988, Vol. 12, April

121

Numerical calculations of flow in a hydrocyclone: Malcolm R. Davidson


tributions. No attempt is made here to account for the
differential effect of swirl on mixing length, m'm Stability considerations indicate that turbulent mixing is
inhibited or enhanced, depending on whether the angular momentum increases or decreases with radius.
In hydrocyclones, the observed tangential velocities
are such that the angular momentum increases with
radius. 2
The velocity, relative to the fluid, of particles (assumed to be spherical) with density pp > p and diameter
dp can be expressed as 3
Ure,

. Up.

U.

.
(;(~

Vr~l = Vp -- V =

I)g~DD) 1/2
-- 1

rCo ]

(9a)

(9b)

where g is the gravitational acceleration, C o is the drag


coefficient, and it is assumed that the centrifugal or
gravitational force on a particle (modified by bouyancy) is in equilibrium with the drag force. The drag
coefficient/particle Reynolds number relationship (see,
e.g., Svarovsky2) is approximated by
C o = 24/R%

= 0.42

Rep-< 57 (laminar, i.e., Stokes law)


Rep -< 57 (turbulent)
(10)

where
Rep = p(UZrel + v~el)Z/2dp//xt
is the particle Reynolds number. To ensure continuity
of the CD function, the Re, ranges of equation (10) are
changed slightly from those used by Pericleous and
Rhodes.

and cell volume which is occupied by fluid. The continuity equation is treated in the same way.
As discussed earlier, the radial and tangential velocities of fluid entering the computational domain are
based on the assumption of axial symmetry and take
values consistent with the known volumetric feed rate
(Q) (Here Q = 28.4 L/m, and we ignore viscous losses
at entry by setting 3' = 1). At the underflow and overflow planes W is taken equal to its upstream value to
approximate (assumed) zero-gradient conditions there.
In the experiment of Knowles et a/., 80% of the incoming flow passes to the overflow; we choose uniform
axial velocities at the underflow and overflow consistent with this flow split. The alternative of setting
the underflow and overflow pressures according to ap/ar
= p w 2 / r and predicting the flow split, as was done by
Pericleous and Rhodes, is not considered here, since
the absence of an air core results in an uncertain interpretation of the given pressure data at the exits. For
a rotating liquid core, radial (and axial) pressure gradients are much larger than for an air core, and the
specification by Knowles e t a / . of a single pressure
value at each exit becomes ambiguous.
Flow normal to the boundary of the computational
domain is set to zero, of course, except at those parts
defining the inlet, underflow, and overflow. Following
Pericleous and Rhodes, wall friction is applied at mesh
points adjacent to the walls of the vortex finder and
the roof the hydrocyclone (i.e., solid boundaries defined by coordinate lines) according to the familiar logarithmic wall law; and at points adjacent to the cone
boundary by adding a momentum sink to equation (2)
given by

2~1/2--~C~
rm

F4j = fp(u 2 + V 2 + w p

Numerical considerations
We solve equations (1), (2), and (independently) (4)
numerically on the nonuniform mesh shown in F i g u r e
2. Note that the radial mesh spacing shown adjacent
to the axis is further subdivided four times in an effort
to resolve the rapid change in tangential velocity from
zero on the axis of symmetry to the peak values observed by Knowles et al. 6 immediately adjacent. (This
matter will be discussed later in more detail.) Variables
W, p, and c are defined at the centre of each cell, and
normal velocities at cell walls are defined at their midpoints. Following the PHOENICS approach, the discretised momentum equations of variables whose control volumes are intersected by the cone boundary are
adjusted to account for the fraction of the cell boundary

(11)

where f is the friction factor, A is the cone surface


area intersecting the control volume V, and ~b = u,v, W.
We choose f = 0.006 corresponding to a value for
smooth pipes (Davies, ~4p. 33) when the Reynolds number has the same order (104) as that in the Knowles
hydrocyclone.
The region near the axis of symmetry presents some
difficulty. Near the axis the tangential velocity distribution is expected to have the form corresponding to
solid-body rotation (w = r and W ~ r 2) since w = 0 on
r = 0. Such behaviour is typically observed in hydrocyclones operating with an air core. However, when
the air core is absent, this region of solid-body rotation
appears to be extremely close to the axis since it is not
evident in the data of Knowles et al. 6 Despite mesh

AXIS OF SYMMETRY
Figure 2

Nonuniform computational mesh used, but with the radial mesh spacing shown adjacent to the axis of symmetry subdivided
an additional four times. Following Pericleous and Rhodes) the computational domain has been extended along the vortex finder

122

Appl. Math. Modelling, 1988, Vol. 12, April

Numerical calculations of flow in a hydrocyclone: Malcolm R. Davidson


refinement near the axis, the expected behaviour there
could not be achieved without some additional constraint on the solution for W other than that of implied
symmetry. Indeed, the assumption (that fluid particles
in turbulent motion retain their angular momentum
during radial translation) underlying the model equation for W breaks down near the axis of symmetry,
since perturbed as well as mean values of angular momentum are zero on r = 0 (otherwise the corresponding
tangential velocities are infinite), whereas at a radius
corresponding to the mixing length they are in general
nonzero. However, in contrast with the present work,
Pericleous and Rhodes 3 do not report any difficulty
near the axis of symmetry or the need to apply special
conditions there. Presumably this is associated with
the presence of an air core in which air with low angular
momentum enters the hydrocyclone along the axis at
the underflow and overflow. Such counter-current flow
at the underflow and overflow planes is not possible
in our model owing to the assumption of uniform velocity profiles at each exit; however there is no evidence in the data of Knowles et al. 6 of downwards flow
along the axis of symmetry counter to the upwards
moving overflow stream, which can occur under some
conditions (Dabir and Petty2). Furthermore, setting
the pressure at the underflow (to allow for the possibility of both inflow and outflow there), while retaining
a uniform overflow velocity profile, results in a negligible change in the predicted flow pattern.
In this paper we assume that the rate of strain
OWor Wr- rl (0~rW" 2W)
is zero (consistent with solid-body rotation) near the
axis. In particular, we set
OW
Or

2W
= 0
r

a t r = rc

(12)

where rc is a small nonzero radius. The width of cells


nearest the axis is set equal to re; equation (12) is then
computationally equivalent, when linear interpolation
is used, to setting W = 0 at nodes adjacent to the axis
(i.e., at r -- r,.). (A tangential velocity which falls to
zero before reaching the axis has been predicted for
air cyclones using a higher-order turbulence model, t6.J7)
We expect r,. to have the same order of magnitude as
the mixing length, since, roughly speaking, fluid particles in the region r < r,. can cross the axis of symmetry, whereas those elsewhere cannot. As was indicated earlier, the cells in Figure 2 immediately adjacent
to the axis are subdivided to give four computational
points within the experimentally unresolved region; in
that case 2r,/D = 0.014. Doubling r,. (i.e., a near-axial
subdivision by two rather than four) is found to have
a negligible effect on the calculated flow velocities except for a reduction of about 8% in the peak tangential
velocity, and a consequent reduction of about 20% in
the pressure drop between the inlet and the axis of
symmetry. The large variation in the pressure drop
occurs because the peak w value lies very near the axis

and Op/Or ~ pw2/r. An alternative, more fundamental


(e.g., second moment) turbulence model, which does
not require the modelling parameter r,. may give better
results near the axis; however, the predicted pressure
drop will still be sensitive to errors in the rapidly varying w profile there.
The methods on which our calculations are based
are embodied in the computer programs TEACH and
PHOENICS and are described in detail by Patankar. 2~
Briefly, the governing flow equations are integrated
over control volumes, and the resulting difference
equations are solved for velocity and pressure using
the SIMPLE algorithm of Patankar and Spalding. 8 The
discretised momentum equations are first solved for a
given pressure field, which is then adjusted to ensure
mass balance within cells. The process is repeated until
residual errors in the momentum and continuity equations are sufficiently small. False transient techniques
are used here, incorporating recommendations of
Raithby and Schneider. 22 In particular, we set their
time step factor E = 0.01 and choose the pressure
relaxation parameter according to their equation (30).
In this paper best agreement with the experimental
velocity and pressure drop data of Knowles et al. 6 is
obtained by choosing the mixing length fraction/3 =
0.005. Corresponding comparisons with those data are
discussed in the next section. Increasing/3 increases
the mixing length and predicts a flatter tangential velocity distribution across the hydrocyclone with lower
peak values, because of an increased flux of angular
momentum across the inner core boundary at r = rc.
Decreasing/3 has the converse effect.

Comparison with theory and experiment


Calculated profiles of tangential and vertical velocity
are compared with the corresponding experimental data
of Knowles et al. 6 in Figures (3a) and (3b). On theoretical grounds the tangential velocity must initially
increase from zero away from the axis, and our axial
boundary condition is chosen to ensure that such behavior is predicted, even though the data does not
resolve this near-axial region. Across most of the hydrocyclone the tangential velocity decreases, with predicted and experimental values in satisfactory agreement. Further, the general form of the predicted profile
(increasing away from the axis before decreasing towards
the periphery) is consistent with that observed in hydrocyclones operating with and without an air core
(Kelsall 5 Dabir and Petty 2) and in air cyclones (Ayers
et al. 16).
Knowles et al. obtained vertical velocity profiles
from data averaged within the field of view, showing
the expected downward flow along the hydrocyclone
wall and upward flow along the axis. These data are
plotted in Figure (3b) and lie satisfactorily close to our
predicted curves. The exception is the profile immediately below the vortex finder; we do not plot the
view-averaged data there, since Knowles et al. use
other data to obtain the velocity profile. This profile
is unlike the calculated shape, showing a striking un-

Appl. Math. Modelling, 1988, Vol. 12, April

123

Numerical calculations of flow in a hydrocyclone: Malcolm R. Davidson

3.0

-9s

1.o

10
00

......
00

0t~
2riD
(o)

08

-1.0L

~
00

0t~
2riD
(bl

08

75

Figure 3 Calculated profiles (solid lines) of tangential (a) and


vertical (b) velocity compared with corresponding experimental
data of Knowles et aL 6 at different heights. In (b), comparison
is also made with theoretical vertical velocities (dashed lines)
derived from the analysis of Bloor and Ingham s

dulation near the axis (interpreted in terms of a shortcircuit flow direct from inlet to overflow). Indeed, we
see from both the tangential and vertical velocity profiles in Figure 3 that the model loses accuracy of description near the vortex finder. This is not surprising
since our assumption of axial symmetry is not valid in
the inlet region.
In Figures (3b) and (4) the vertical velocity profiles
are compared with theoretical predictions derived from
the analysis of Bloor and Ingham. 9 These authors assumed that the flow in the (r, x) plane is inviscid with
a prescribed vorticity distribution of fairly general form.
The approximate expression
u = Br-'n(3a,. - 5r/z)

(13)

for the vertical velocity was derived for hydrocyclones


with small cone angle 2a,. and negligible underflow
rate, where z is the height above the cone vertex (differing from x, which is the height above the underflow
plane) and B is a constant depending on the volumetric
flow rate Q. In this paper the underflow rate is not
negligible; nevertheless, provided streamlines intersecting the underflow lie close to the hydrocyclone
wall, we still expect equation (13) to apply when Q is
replaced by the overflow rate (0.8Q here)., This is
demonstrated in Figures (3b) and (4), in which the
numerical and analytical predictions agree well except
in a region near the axis (equation (13) implies an infinite velocity on the axis). This agreement with the
analysis of Bloor and Ingham confirms that turbulent
mixing plays little role in the transfer of axial or radial
momentum in the interior of the hydrocyclone (in contrast to its importance in the transport of angular momentum) except, perhaps, near the axis. Indeed, numerical predictions of u at the axial positions shown

124 Appl. Math. Modelling, 1988, Vol. 12, April

in Figure 3b vary little with a fourfold change in mixing


length, except near the axis immediately below the
vortex finder. Similar observations apply deeper within
the hydrocyclone; the tt variation with mixing length
being greatest in the upflowing fluid near the axis, but
with little variation on the axis itself.
Calculated values of/zerr are greatest adjacent to the
axis of symmetry with values of 2lzerr/pwiD falling rapidly from a central value at r = r,. of about 2.2 x 10 -z
to about 0.3 x 10 -2 at 2r/D = 0.2 and thereafter decreasing slowly towards the inlet value of 0.2 x 10 -2
(of course, very close to the wall it falls rapidly towards
the laminar value). However, the turbulent viscosity
should decrease as r---~ 0 in the region of solid-body
rotation (r < r,.), since from stability arguments we
expect radial mixing there to be suppressed relative to
that in the main body of the flow where angular momentum increases less rapidly with radius. Furthermore, radial stabilisation has been observed at the core
of vortex tubes (Escudier et al.23). In our calculations,
we do not choose a reduced value of ~cfr on the axis;
however, this will have little effect on the results, since
the region r < r,. lies outside the calculation domain of
w and we have already shown that predicted values of
u in this region are not sensitive to the choice of ~crr
for r,. small. In cases for which r,. is not small, correct
prediction of the flow in the core region is expected to
require proper modelling of the turbulent mixing there.
An assumption of Bioor and Ingham's model is that
the tangential velocity (w) is independent of height x.
Although inspection of Figure 3(a) would suggest that
this is true of the numerical results, Figure 5 reveals
that, for 2x/D < 7.0, the predicted tangential velocity
decreases towards the hydrocyclone wall more rapidly
as the underflow is approached. A fall in w, as the wall
is approached along lines of constant radius, occurs
because of wall friction, and indeed the effect is less
if the friction factor f is reduced. Such behaviour close
to the wall has been reported by Kelsall. 5 The degree
of axial nonuniformity in w which occurs in the present
case is in contrast to the almost height-independent w
distribution in the Kelsall hydrocyclone, and is prob'

'

0B
04.

' ~

~00

-08-

b-=~
,

O0

03

06

09

2rid
Figure 4 Profiles of calculated vertical velocity (solid lines)
compared with corresponding predictions (dashed lines) based
on the analysis of Bloor and Ingham 9 at a range of heights including those for which data are unavailable

Numerical calculations of flow in a hydrocyclone: Malcolm R. Davidson


30 1
!
2,5

agreement is acceptable in view of the uncertainty surrounding the experimental p , and po values. We find
little axial variation in the pressure, with the radial
pressure distribution at positions below the vortex finder
lying close to that shown in Figure 6. The greatest
variation in this direction occurs along the axis itself,
with most of the pressure drop between underflow and
overflow occurring in the vortex finder.

I
-

20
15
1.0
0.5

Other results

O0
00

03

06

09

Streamlines defined by

2rid

Figure 5 Profiles of c a l c u l a t e d t a n g e n t i a l velocity at a r a n g e of

u -

h e i g h t s i n c l u d i n g t h o s e for w h i c h d a t a a r e u n a v a i l a b l e

200
15 0

50
I
03

I
06

I
09

2riD

Figure 6 Radial pressure variation across the h y d r o c y c l o n e

ably the consequence of the smaller cone angle (less


spin-up, due to conservation of angular momentum, to
offset the effect of wall friction) and greater height of
the Knowles hydrocyclone, together with the lower
feed velocity (thicker wall boundary layer). The predicted x dependence of w is expected to contribute to
the discrepancy between numerical and analytical predictions of u (Figure 4) away from the axis of symmetry.
As discussed earlier, and we can see from Figure
6, radial pressure gradients in a region about the axis
are high in the absence of an air core, and it is not
clear to which radius the underflow and overflow pressure values given by Knowles et al. should be assigned.
If we assume that they are the pressure values on the
axis of symmetry, then the dimensionless pressure drops
(p, - p,,)/pw~ and (pi - po)/pw~ have calculated values
of 5.7 and 25.2, respectively, compared with corresponding experimental values of 5.5 and 20.1. This

0.0
00

10

I a,
I'ax

v . . . .

(14)

are shown in Figure 7, where the stream function , on


the axis of symmetry is taken to be zero. In equation
(14) velocities normal to cell faces intersected by the
cone boundary (u only in this case)are multiplied by
the face fraction occupied by fluid; by analogy with a
porous medium, the velocities used in equation (14)
are Darcy velocities, not pore velocities. The slight
wriggle in the streamlines close to the cone boundary
results from imperfect modelling of this nonrectangular
boundary by the rectangular grid. The flow patterns
exhibit the expected deeply penetrating reversing flow
characteristic of hydrocyclones, with fluid flowing down
along the outer wall and up along the axis of symmetry.
Streamlines which intersect the underflow lie very close
to the cone boundary and are not shown. As in the
Pericleous-Rhodes simulation of the Kelsall experiment, the flow pattern contains a region of recirculation (closed streamlines) lying between the downflowing and upflowing streams.
A peculiarity of the predicted flow pattern is the
absence of any short-circuit flow from the inlet, across
the roof of the hydrocyclone, to the vortex finder.
Physically, such a flow occurs in the top boundary
layer because the tangential velocity there is smaller
than that of the fluid below. This results in an outward
centrifugal force on the boundary layer fluid which is
less than the opposing force of the radial pressure field,
and so inwards radial flow occurs. Short-circuit flow
does not occur in this simulation because we do not
resolve the top boundary layer, relying instead on the
logarithmic wall law to determine the near-wall velocity. Furthermore, inspection of the predicted velocity
data reveals little axial variation in w in the inlet region
as the top boundary is approached. Calculated values
of both u and v are small near the outer wall of the
vortex finder; the detail of the predicted flow pattern
there (streamlines rising then reversing down along the

25.0

0.0.

I a~
r ar

20

30

40

50

&0

70

8.0

90

100

2x/D

Figure 7 Calculated streamlines in the hydrocyclone, w h e r e ~ * = ~/(0.25D2wi) is the d i m e n s i o n l e s s stream function

Appl. Math. Modelling, 1988, Vol. 12, April

125

Numerical calculations of flow in a hydrocyclone: Malcolm R. Davidson


08 I

F S"

"

,-.1

0A. ~

'

00--

'

'

CI[I:10~

0.S
I

0.8
O0

0BL.

0t~
00

'

'

'

'

~
1.0

20

30

40

50

.....

_ _ . _ - - 1 ~ ~ ~

- -

60

_ ~
70

80

90

100

2x/D

Figure 8 Contours of constant particle concentration in the hydrocyclone for particles having diameters (a) 10 microns, (b) 20 microns,
(c) 30 microns, assuming the feed concentration is low. The solid and dashed curves are based on the HYBRID and QUICK differencing
schemes, respectively

10

vortex finder) is therefore of doubtful significance due


to loss of precision, and is consequently not shown in

Figure 7.
Curves of constant particle concentration (c/cl) are
plotted in Figure 8 for particles with diameters (d,) 10,
20, and 30 microns, where c; is the feed concentration
of particles of given size, and c is assumed to be sufficiently small to have a negligible effect on the fluid
density. Attention is focused on the solid lines in Figure 8; the significance of the dashed curves will be
discussed later. We have set c = c; at upstream inlet
nodal points just outside the flow domain, and the particle velocities relative to the fluid (Urel,/.)rel) are taken
to be zero normal to solid boundaries. Furthermore,
we have set vr~ = 0 on the boundary at the inlet, since
we expect the actual geometry of the inlet region to
prevent any backwards drift of particles.
Because the radial drift velocity Vr, is positive for
particles which are denser than water, the particle concentration in Figure 8 is greater near the periphery of
the hydrocyclone than it is near the axis. Also, Vre,
depends on the centrifugal acceleration w2/r, whereas
the (negative) vertical drift velocity u~e,depends on the
gravitational aceleration g; u~~is thus about three times
smaller than Vr~near the inlet radius but about 150
times smaller near the axis, where it consequently has
a negligible effect on the predicted particle concentration. The very small x variation in concentration near
the axis is a reflection of similarly small variations in
w (Figure 5), and hence /-)rel, there.
As dp increases, Vr~,also increases, and we can see
from Figure 8 that the particle concentration at a given
position on the hydrocyclone wall rises (it also falls on
the axis); hence the fraction of particles passing through
the underflow increases, as expected. The corresponding efficiency curve reflecting this behaviour is shown
in Figure 9. Unfortunately, no efficiency data are available for comparison in Figure 9. At very low dp values
there is little buildup of particles along the wall and

126

Appl. Math. Modelling, 1988, Vol. 12, April

~06
J
0k

z 0z

13E 0.0
U-

10
20
30
t,0
PARTICLE DIAHETER (HICRONS)
Figure 9 Fraction of particles entering the hydrocyclone which
appear at the u n d e r f l o w as a function of particle size (efficiency
curve)

c ~ c,. everywhere, so that the particle split between


underflow and overflow is approximately equal to the
flow split. As dp and the concentration along the wall
increase, the concentration contours necessarily become more closely spaced.
T h e effect o f n u m e r i c a l d i f f u s i o n

Our calculations thus far have been based on a hybrid


differencing scheme (HYBRID), widely used to model
highly convective flows, in which first-order upwind
differencing is applied to stabilise the calculation, except when the cell Reynolds number is less than 2, in
which case central differencing is used. However, the
accuracy of first-order upwind differencing is limited,
and artificial numerical diffusion errors can be important. Such errors may be significant in the presence of
source terms (Leonard24), and when the streamlines
are skewed relative to the axes of the numerical grid
(de Vahl Davis and Mallinson25), as is the case here.
Despite the agreement with experimental data reported
here, the possibility remains that, in the absence of

Numerical calculations of flow in a hydrocyclone: Malcolm R. Davidson


numerical diffusion, some other choice of mixing length
would be appropriate. Furthermore, the validity of extrapolated model behaviour must be in doubt unless
the effects of false diffusion are minimal.
An alternative differencing scheme (called QUICK),
proposed by Leonard 24and involving an upstream shifted
quadratic interpolation at each control surface, has been
shown to be more accurate in general than the conventional HYBRID scheme for a variety of flows
(Leschziner, 26 Han et a1.,27 Leschziner and Rodi, 28 Patel and Markatos29), but it converges less readily.
We assess the importance of numerical diffusion in
the present application by solving the governing equations for flow and particle transport using both
HYBRID and QUICK. The exception is the angular
momentum equation for which HYBRID is used
throughout; when QUICK is used for this equation as
well, convergence is unacceptably slow. To isolate the
effect of numerical diffusion on the solution of the
particle transport equation, we compare corresponding
HYBRID and QUICK results based on the same underlying (HYBRID) flow calculation. In using QUICK,
we applied a source decomposition of the corresponding difference formulae tested by Han et al. 27 (their
equation (8)).
The dashed contour lines of particle concentration
in Figure 8 are based on QUICK differencing. We see
that numerical diffusion has little effect on the predicted concentration pattern except in a region along
the wall of the hydrocyclone where it results in concentrations which are too low by about 1-7%. This
occurs because the downward flow along the cone wall
necessarily cuts the x-axis at an angle, whereas the
upward flow in the centre is almost parallel to the axis.
Near the cone wall, the false diffusivity is calculated
(see e.g. Leschziner 26) to be of similar order to the
corresponding turbulent diffusivity. Nevertheless, the
effect o f numerical diffusion there is small. Furthermore, the integrated effect of false diffusion on the
efficiency curve in Figure 9 is found to be negligible.
Numerical diffusion is found to have little effect on
the predicted vertical velocity distribution. This is expected since diffusion of axial momentum has already
been shown to have a negligible role away from the
axis of symmetry, and hence false diffusion of similar
order will also be unimportant (in this consideration
only the region along the cone wall is relevant). As
discussed earlier, angular momentum decreases down
along the hydrocyclone wall; numerical diffusion is
thus expected to result in tangential velocities there
which are too high. However, as for the particle concentration, the error is expected to be small.

be imposed at a radius of similar magnitude to the


mixing length to ensure realistic flow predictions. Velocity and pressure are calculated on a staggered mesh
using the SIMPLE algorithm of Patankar and Spalding;8 also hybrid first-order upwind/central differencing and false transient techniques are used, incorporating recommendations of Raithby and Schneider. 22
Calculated vertical and tangential flow velocities and
pressure drops are shown to fit the corresponding data
of Knowles et al. for a suitable choice of mixing length.
Further, calculated vertical velocities are found to agree
with theoretical predictions based on the analysis of
Bloor and Ingham, 9 except in a region near the axis.
Comparison of results based on both hybrid and quadratic upwind differencing is used to assess the importance of numerical diffusion. False diffusion is found
to have little effect on the predicted particle concentration, except along the wall of the hydrocyclone where
it results in errors of about 1-7%; also the effect on
the predicted efficiency curve, a n d on the calculated
flow in the (r, x) plane, is negligible. A similarly small
effect on the transport of angular momentum seems
likely.

Conclusion

S,,,S,,,Sw

Notation
A
B
C
Ci

Co
D

di, do, d.
d~
E

f
g
H
h
L
P
pi, po, p,,

Q
(r, x)
r{.
Rep

The model of Pericleous and Rhodes 3 is used to describe the flow and particle transport in a hydrocycione
operating without an air core, and the experiment of
Knowles et al. 6 is simulated. In this work the model
breaks down to the axis of symmetry, and a special
boundary condition (corresponding to zero shear) must

ll.. U, 14'
lip, Up
/Ircl, /-)rel

surface area of cone boundary intersecting


a control volume
constant in equation (13)
particle concentration
feed concentration of particles
drag coefficient
diameter of cylindrical section of hydrocyclone
diameters of inflow, overflow, and underflow, respectively
particle diameter
time-step factor defined in Ref. 22
friction factor
gravitational acceleration
height of cylindrical section of hydrocyclone
length of vortex finder
overall height of hydrocyclone
pressure
pressure at inflow, overflow, and underflow, respectively
volumetric feed rate
cylindrical coordinates as in Figure I
width of computational cells nearest the
axis
particle Reynolds number
momentum sources defined in equation (3)
axial, radial, and tangential fluid velocities, respectively
axial and radial components of particle velocity, respectively
axial and radial velocities, respectively, of
particles relative to the fluid (drift velocity)

Appl. Math. Modelling, 1988, Vol. 12, April

127

Numerical calculations of flow in a hydrocyclone: Malcolm R. Davidson

control volume of the computational grid


= r w , where p W is angular momentum
turbulent velocity fluctuation in the inlet
pipe
height above the cone vertex

V
W
W~

Greek symbols
0l
half-angle of the hydrocyclone
mixing length fraction
13
reduction factor of tangential velocity at
7
the inlet
turbulent exchange coefficient for particle
Feet
transport
cylindrical polar angle
0
A
mixing length
turbulent viscosity in the hydrocyclone and
/Zeff, /zi
at the inlet, respectively
laminar viscosity
]./,I

,b

References
1
2
3
4

5
6
7

8
9
10

12
13
14
15
16
17
18
19

U, t3, o r W

density of fluid and particles, respectively


turbulent shear stress
stream function

P,P.

I1

Bradley, D. The Hydl'ocyc/one. Pergamon, London, 1965


Svarovsky, L. Hydrocyclones. Holt, Rinehart, and Winston,
London, 1984
Pericleous, K. A. and Rhodes, N. The hydrocyclone classifier--a numerical approach. Int. J. Miner. Process. 1986, 17,
23-43
Rosten, H. I., Spalding, D. B., and Tatchell, D. G. PHOENICS: a general-purpose program for fluid-flow, heat-transfer
and chemical-reaction processes. Proceedings of the Third International Conference on Engineering Software, 1983, pp.
639-655.
Kelsall, D. F. A study of the motion of solid particles in a
hydraulic cyclone. Trans. lnstn. Chem. Engrs. 1952, 30, 87-104
Knowles, S. R., Woods, D. R., and Feuerstein, i. A. The
velocity distribution within a hydrocyclone operating without
an air core. Can. J Chem. Eng, 1973, 51,263-271
Gosman, A. D. and Ideriah, F. J. K. TEACH-T: a general
computer program for two-dimensional, turbulent, recirculating flows. Lecture Notes, Dept. Mech. Eng., Imperial College,
London, 1976
Patankar, S. V. and Spalding, D. B. A calculation procedure
for heat, mass and momentum transfer in three-dimensional
parabolic flows. Int. J. Heat Mass Transfer 1972, 15, 1787-1806
Bloor, M. I. G. and Ingham, D. B. Theoretical aspects of hydrocyclone flow. In Progress in Filtration and Separation, vol.
3, ed. R. J. Wakeman. Elsevier, New York, 1983
Pericleous, K. A., Rhodes, N., and Cutting, G, W. A mathematical model for predicting the flow field in a hydrocyclone

128 Appl. Math. Modelling, 1988, Vol. 12, April

20

21

22
23
24
25
26
27

28

29

classifier. Proceedings of the Second International Conference


on Hydrocyclones, Bath, U.K., 1984, pp. 27-40
Launder, B. E. and Spalding, D. B. Mathematical Models of
Tarbulence. Academic Press, London, 1972
Lilley, D. G. and Chigier, N. A. Nonisotropic turbulent stress
distribution in swirling flows from mean value distributions.
Int. J. Heat Mass Transfer 1971, 14, 573-585
Wattendorf, F. L. A study of the effect of curvature on fully
developed turbulent flow. Proc. Roy. Soc. 1935, A148, 565-598
Davies, J. T. Turbulence Phenomena. Academic Press, London, 1972
Lilley, D. G. Prediction of inert turbulent swirl flows.
AIAA J. 1973, 11(7), 955-960
Ayers, W. H., Boysan, F., Swithenbank, J., and Ewan,
B. C. R. Theoretical modelling of cyclone performance. Filtration and Separation Jan/Feb 1985, 39-43
Boysan, F., Ayers, W. H., and Swithenbank, J. A fundamental
mathematical modelling approach to cyclone design. Trans.
IChemE. 1982, 60, 222-230
Bradshaw, P. Effects of streamline curvature on turbulent flow.
AGARDograph No. 169, 1973
Koosinlin, M. L., Launder, B. E., and Sharma, B. I. Prediction
of momentum, heat and mass transfer in swirling, turbulent
boundary layers. J. Heat Transfer, Trans. ASME 1974, 98,
204-209
Dabir, B. and Petty, C. A. Laser doppler anemometry measurements of tangential and axial velocities in a hydrocyclone
operating without an air core. Proceedings of the Second International Conference on Hydrocyclones, Bath, U.K., 1984,
pp. 15-2(?
Patankar, S. V. Numerical heat transfer and fluid flow. In
Cornpatationa/ Methods in Mechanics and Thermal Sciences,
eds. W. J. Minkowycz and E. M. Sparrow. McGraw-Hill, London, 1980
Raithby, G. D. and Schneider, G. E. Numerical solution of
problems in incompressible fluid flow: treatment of the velocity-pressure coupling. Namerica/Heat Transfer 1979, 2, 417-440
Escudier, M. P., Bornstein, J., and Zehnder, N. Observations
and LDA measurements of confined turbulent vortex flow. J.
Fhdd Mech. 1980, 98, 49-63
Leonard, B. P. A stable and accurate convective modelling
procedure based on quadratic upstream interpolation. Compat.
Methods Appl. Mech. Eng. 1979, 19, 59-98
de Vahl Davis, G. and Mallinson, G. D. An evaluation of upwind and central difference approximations by a study of recirculating flow. Computers and Fhdds 1976, 4, 29-43
Leschziner, M. A. Practical evaluation of three finite difference
schemes for the computation of steady-state recirculating flows.
Compat. Methods App/. Mech. Eng. 1980, 23, 293-312
Han, T., Humphrey, J. A. C., and Launder, B. E. A comparison of hybrid and quadratic upstream differencing in high Reynolds number elliptic flows. Comput. Methods App/. Mech.
Eng., 1981, 29, 81-95
Leschziner, M. A. and Rodi, W. Calculation of annular and
twin parallel jets using various discretization schemes and turbulence-model variations. J. Fhdds Eng., Trans. ASME 1981,
103, 352-360
Patel, M. K. and Markatos, N. C. An evaluation of eight discretization schemes for two-dimensional convection-diffusion
equations. Int. J. Namer. Fhdds 1986, 6, 129-154

Anda mungkin juga menyukai