Académique Documents
Professionnel Documents
Culture Documents
b
viscosity of the binder system (Pa s)
0965-0393/03/040503+20$30.00 2003 IOP Publishing Ltd Printed in the UK 503
504 X Chen et al
r
relative viscosity of suspension
particle concentration
c
critical powder loading
m material constant
B material constant (Pa s)
T
b
material constant (K)
T temperature (K)
0
initial powder loading
R radius of the tube (mm)
Q ow rate (cm
3
s
1
)
u
i,j
axial velocity of the node i (at ow direction) at layer j
(at radial direction) (cms
1
)
N2 maximum interval numbers in radial direction
N grid numbers in ow direction
relaxation factor
t time (s)
w weight coefcient
z
pressure gradient in ow direction (Pa cm
1
)
w
particle concentration at the wall
w
shear rate at the wall (s
1
)
u= 0 (1)
p +
= 0. (2)
Considering a generalized Newtonian uid, the extra stress tensor
is given by
= (| |, T, )
. (3)
The apparent viscosity is a function of temperature T , particle concentration and magnitude
| | of the rate-of-strain tensor
}]
1/2
= [
1
2
{
2
rr
+
2
+
2
zz
+ 2
2
rz
}]
1/2
(4)
and the components of
=
u +(
u)
T
are given by
rr
= 2
u
r
r
,
= 2
u
r
r
,
zz
= 2
u
z
z
,
rz
=
zr
=
u
z
r
+
u
r
z
. (5)
For axisymmetric ows and cylindrical coordinates, equations (1) and (2) can be written as
u
r
r
+
u
r
r
+
u
z
z
= 0 (6)
p
r
+
rr
r
+
rr
r
+
zr
z
r
= 0 (7)
p
z
+
rz
r
+
rz
r
+
zz
z
= 0. (8)
It should be noted that with symmetry,
rz
=
zr
. In treating the particle suspension as a
single continuum, one must account for variation in viscosity with particle concentration. This
variation in viscosity of the concentrated suspension can be obtained as a simple correction to
the solvent viscosity. The functional relationship of can take several forms that adequately
model viscosity data for concentrated suspension uids (e.g. Krieger model) [18]. A viscosity
formulation of a concentrated suspension described by the Krieger model [18] is adopted here
=
b
_
1
c
_
m
. (9)
This rheological model describes the effect of the volume fraction of the powder and shear-
thinning on the ow behaviour of a concentrated suspension if the modied cross model is
employed to describe the viscosity of the binder:
b
=
0
1 + (
0
/
)
1n
(10)
with
0
= B exp
_
T
b
T
_
. (11)
Shear-induced particle migration modelling 507
The viscosity at zero shear rate,
0
, is used in this model. At high shear rate, equation (10)
can be modied into
b
= (
0
1n
0
)
n1
(12)
where
0
=
/
0
. Equation (12) is a power-law model which can describe the rheological
behaviour of the suspension. For pressure-driven tube ow, we have /z /r, u
r
u
z
,
u
z
= u
z
(r, z) and p = p(z). These assumptions, known as the lubrication approximation
theory [17], greatly simplify the conservation equations (7) and (8) to give
p
z
1
r
r
_
r
u
z
r
_
= 0. (13)
2.2. Diffusive-ux model
The shear-induced migration of particles is driven by concentration gradients and by shear rate
gradients [2, 3]. A phenomenological model based on these principles [8] in Poiseuille ow
yields the particle ux
J = a
2
K
c
u
z
r
_
K
2
a
2
u
z
r
. (14)
The dynamic change of particle concentration, , along the ow is governed by the particle
balance
D
Dt
= J. (15)
Following Phillips et al [8], the implementation of the diffusion equation in Poiseuille
ow can be written as
t
+
(u
z
)
z
=
_
a
2
K
c
u
z
r
_
+ K
2
a
2
u
z
r
_
. (16)
Shear-induced diffusion is assumed to be independent of interparticle interactions, which
include frictional and colloidal forces. These effects are incorporated in the phenomenological
constants K
c
and K
u
z
r
_
+ K
2
u
z
r
_
n = 0. (17)
The particle concentration is assumed to be uniform initially at the entrance of the tube, i.e.
=
0
for 0 r R, at z = 0. (18)
3. Solution methodology
For a pressure-driven tube ow, when the no slip boundary condition is employed at the wall,
equation (13) can be integrated
=
r
2
(19)
508 X Chen et al
where
z
= p/z is the pressure gradient along the owdirection. Introducing the Krieger
model and the power-law model for polymeric binder, equations (9) and (12), respectively,
into equation (19), the pressure gradient
z
can be written as
z
=
Q
S
(20)
where ow conductivity S is expressed as
S =
1
z
_
z
2
0
1n
0
_
1/n
_
R
0
r
2+1/n
(1 /
c
)
m/n
d r. (21)
The viscosity can be obtained by
=
_
z
r
2
_
11/n
_
0
1n
0
_
1
c
_
m
_
1/n
(22)
and
u
z
=
z
_
R
r
r
2
dr (23)
or the volumetric ow rate Q can be calculated as a function of pressure gradient
z
as
Q = S
z
. (24)
When the modied cross-model was employed, a numerical method, namely the nite
difference method, was employed for solving equation (13). Introducing the Krieger model (9)
and the modied cross model (10) for polymeric binder into equation (13), respectively, the
velocity equation can be written as
du
z
dr
+
z
r
2
0
1n
0
_
1
c
_
m
_
_
du
z
dr
_
1n
+
1n
0
_
= 0. (25)
Divide the unit radius into N2 intervals, the axial velocity at the node i, u
i,j
(j = 0, 1, . . . , N2)
will be determined by the following N2 1 equations, using the backward difference scheme,
F
j
=
u
i,j1
u
i,j
r
+
z
r
2
0
1n
0
_
1
c
_
m
_
_
u
i,j1
u
i,j
r
_
1n
+
1n
0
_
= 0
j = 1, . . . , N2 (26)
which can be solved using the relaxed NewtonRaphson iteration
u
(m)
i,j
= u
(m1)
i,j
i
F
j
F
j
, m = 1, 2, . . . (27)
where is the relaxation factor or step length. F
j
=
1
r
+
z
r(1 n)
2
0
1n
0
r
_
1
c
_
m
_
u
i,j1
u
i,j
r
_
n
= 0. (28)
The relaxation factor should be chosen in such a way that the error norm u =
F
j
F
j
or the norm of the residual force F
j
decreases with iteration. In our calculation,
the initial guesses of velocity u
i=1,j
are obtained based on the power law.
Shear-induced particle migration modelling 509
j=0
j
j+1
j=N2
j-1
}
1
=
j j
r r r
2 + j
r
Figure 1. Discretization of a tube.
Employing the no slip boundary condition, i.e. u
i,N2
= 0, the last (i.e. the N2th) equation
of equation (26) becomes,
F
j=N2
=
u
i,N21
r
+
z
r
2
0
1n
0
_
1
c
_
m
_
_
u
i,N21
r
_
1n
+
1n
0
_
= 0. (29)
The derivative in equation (30) is given as
F
j=N2
=
1
r
+
z
r(1 n)
2
0
1n
0
_
1
c
_
m _
u
i,N21
r
_
n
= 0. (30)
The u
i,N22
can be solved by substituting the value of u
i,N21
into the (N2 1)th equation
in equation (26). In this way, all the velocity components in the radial direction of each
cross-section can be obtained one by one with the same NewtonRaphson iteration scheme
(equation (27)), with the function and derivative values calculated using equations (26)
and (28).
The nite difference numerical method with a semi-implicit (named CrankNicolson)
scheme based on control volume [22], which has attractive conservative properties, was
employed to solve the diffusive model equation (16), a partial differential equation involving
the development of particle concentrations with respect to both time and the radial coordinate.
The discretization scheme of diffusive equation (16) can be described as follows:
_
t +t
t
_
R
0
t
r dr dt +
_
t +t
t
_
R
0
(u
z
)
z
r dr dt
= a
2
_
t +t
t
_
R
0
_
r
_
rK
c
_
r
+
r
__
+
r
_
rK
2
m
r
__
dr dt .
(31)
To evaluate the terms in equation (31), an assumption is made about the variation of
concentration at time t or at time t + t for calculating the time integral or, alternatively, a
combination of concentrations at time t and t + t . We may rewrite the discretization forms
of equation (31) by means of a weighting parameter w between 0 and 1 with respect to time.
The grid system as shown in gure 1 is constructed in the radial and ow directions. The grid
510 X Chen et al
points are shown in the radial direction fromthe centreline to the wall of the tube. The constant
axial and radial spacings are z and r, respectively. Let the subscripts i and j denote the
positions of the grid points, with i serving as an index in the axial (i.e. ow) direction and j in
the radial direction. The conservative nite difference approximation of the node i at layer j
and at time t is:
[wA(M
j+1
+ M
j
)]
t +t
i,j+1
+ [wA(M
j+1
+ 2M
j
+ M
j1
)
+2r
2
j (x + u
i,j,t
t )]
t +t
i,j
+ [wA(M
j1
+ M
j
)]
t +t
i,j1
= 2r
2
j
t
i1,j
[z u
i,j,t
(1 w)t ]
+2r
2
jt + A(1 w)(M
j1
+ M
j+1
2M
j
) + F
i,j
(32a)
where
A = zt a
2
(32b)
M
j+1
= (j + 1)
t
i,j+1
_
K
c
t
i,j+1
+
mK
s
(
t
i,j+1
)
2
c
t
i,j+1
_
(32c)
M
j
= j
t
i,j+1
_
K
c
t
i,j
+
mK
s
(
t
i,j
)
2
c
t
i,j
_
(32d)
M
j1
= (j 1)
t
i,j1
_
K
c
t
i,j1
+
mK
s
(
t
i,j1
)
2
c
t
i,j1
_
(32e)
F
i,j
=
K
c
zt
r
{[(j + 1)
t
i,j+1
+ j
t
i,j
]
t
i,j+1
+ [(j 1)
t
i,j1
+ j
t
i,j
]
t
i,j1
[(j + 1)
t
i,j+1
+ 2j
t
i,j
+ (j 1)
t
i,j1
]
t
i,j
} (32f)
if w =
1
2
, it is the CrankNicolson scheme. This means that the concentrations at time t
and t + t are equally weighted. All of our numerical computations were time dependent
and, whenever a steady solution was required, it was obtained as the time-asymptotic result of
transient solutions. This approach allowed us to decouple and linearize by means of successive
substitutions in a time step procedure. Figure 2 shows a ow chart that summarizes the main
steps in the computational procedure outlined as follows:
(a) Set the uniform particle concentration and temperature prole at the entrance. In
our investigation, isothermal and no viscous heating were assumed during the whole
computational procedure.
(b) Calculate iteratively ow conductivity (equation (21)), velocity and shear rate across the
tube according to equations (21), (23) and (19), respectively, by assuming a pressure
gradient until the specied ow rate at the entrance is reached.
(c) The velocity equation (equation (25)) was solved by using the relaxed NewtonRaphson
iteration in terms of equations (26), (27) and (28).
(d) The pressure gradient (
z
) was solved in an iterative manner based on the adjustment
of the values of particle concentration at the wall. The values of concentration of the
particles at the wall were determined implicitly by imposing the condition that the ow
bulk concentration (
) at any cross-section is equal to the initial concentration
0
, i.e.
0
=
_
R
0
u
z
r dr
_
R
0
u
z
r dr
. (33)
Shear-induced particle migration modelling 511
INPUT
Geometry, material, processing data
Loop on time step
Estimate flow conductivity S,
Pressure gradient, shear rate
and velocity
Modified Cross
viscosity model ?
Loop on velocity
Calculate
j
F (Equation 26)
and '
j
F (Equation 28)
Calculate
j i
u
,
(Equation 27)
'
) 1 (
,
) (
, j j i
m
j i
m
j i
F F u u =
j i
u
,
Newton-Raphson
type iteration
Solve diffusion equation
(Equation 32) using
Thomas algorithm
Update flow conductivity S,
Pressure gradient, shear rate
and velocity
Mass converged ?
Finished ?
start
End
Update
w
N
Y
N
Y
N
Y
N
Y
Power-law model
Converged?
Figure 2. Flow chart for the numerical procedure for pressure-driven tube ow.
512 X Chen et al
To accelerate the convergence of the solution, the initial values of particle concentration
at the wall can be determined by the evaluation of the no-ux condition (equation (17)) in
the steady state. The resulting rst order non-linear equation with respect to the particle
concentration (
w
) at the wall is obtained [8] as:
1
w
(
c
)
mK
/K
c
_
1
w
c
_
mK
/K
c
(
c
)
w
= 0. (34)
Here
w
and
w
are the values of and at the tube wall. The interval halving method
was employed for solving this nonlinear equation. The control accuracy dened as the
nal solution accuracy of the interval halving method [21] for the solution is 0.0001. The
error in the estimate of the solution must be less than |2
n
c
|, where n is the number of
iterations performed.
(e) Equation (32) is a tridiagonal-matrix, which was solved by the Thomas algorithm [22]
at each time level. The procedure is repeated to progress the solution for a further time
step. Subsequently, the viscosity, shear rate, velocity distribution and ow conductivity
at every nite difference grid are calculated.
In the above calculations, values of at the wall, velocity in each cross-section and of
the driving pressure gradient P/z, are solved iteratively until the boundary conditions and
imposed volumetric ow rate could be satised.
4. Results and discussion
To measure shear-induced particle migration, particle distribution was quantied by conducting
experiments using viscous nickel spherical particles having a sieved mean size of 58 m
suspended in a non-Newtonian solvent composed of ethylene vinyl acetate (EVA460) resin
manufactured by Dupont. A 40% volume fraction of particles was used in this investigation.
The viscous suspension was initially mixed using the Haake Rheomix600 torque mixer.
Ascanningelectronmicroscope (SEMJEOL-5600) was usedtocapture images of the feedstock
with specially designed capillary split dies. Image analyser software (Image ProPlus) was used
to obtain particle concentration at the intermediate radial annulus [19].
The material data employed for numerical simulation and experimental measurement
are tabulated in tables 1 and 2. The rheological constants B, T
b
in equation (11) and
,
n in equation (10) of the EVA460 binder system were determined by data tting of viscosity
measurement shown in gure 3(a) using the modied cross-model and power-law model,
respectively. For EVA460, the uid exhibits a Newtonian plateau value at low shear rate and
shear-thinning at high shear rate. As shown in gure 3(a), the modied cross-model can t
the experimental data well both at low and high shear rates. The power-law model can t the
experimental data well only at a high shear rate. The effect of powder loading on the relative
viscosity of the suspension, dened as the ratio of viscosity of suspension to that of carrier
uid [23] was plotted in gure 3(b). With an increase in particle loading, the relative viscosity
of suspension will increase.
The value of
c
in the Krieger model (equation (9)) could depend on the uniformity of the
particle size, the effective microstructure of the packedconguration, andeventype of ow. For
uniform diameter spheres, the value of
c
can vary over a range from 0.52 to 0.74. Although
the value of
c
could be adjusted, it is normally assumed to be a constant for a particular
powderbinder system. For this investigation,
c
was assumed to be 0.68 for a pressure-driven
tube ow [20]. For the determination of the material constant min equation (9), it is important
to carry out experiments under conditions of minimal or no particle migration. As it was shown
Shear-induced particle migration modelling 513
Table 1. Physical and rheological parameters of binder and powder system [20].
Materials EVA460 Nickel powder
Constants B in T
b
in
in n in a in m in
c
in
eqn (10) eqn (10) eqn (10) eqn (10) eqn (14) eqn (9) eqn (9)
(Pa s) (K) (Pa) (m)
Values of 2.530 4576.76 128 773.86 0.385 29 0.82 0.68
constants
Table 2. Geometry and testing conditions.
Size
Die type
Diameter
(mm)
D
Length
(mm)
L
Flow rate
(m
3
s
1
)
Q
Temperature
(
C)
T
Circular split die 3 40
9.82 10
7
180
Circular standard die 1 30
[12] that a parallel-plate viscometer introduced little migration, m was determined to be 0.82
using this testing method.
As particle migration is in the radial direction, the size of the grid must be ne in the
radial direction to ensure sufcient numerical accuracy. Figure 4 shows the relative error plot
of particle concentration near the wall dened in table 3 versus grid densities in the radial
direction. The relative errors were independent of any further renement of grid density for
both initial particle concentrations 30% and 40% if the grid was ner than 80 40 (80 in
the radial and 40 in the ow direction, respectively). Thus, grid 80 40 was used for all
subsequent calculations.
4.1. Predicted and experimental particle concentrations
Figure 5 shows the comparisons of simulation and experimental results for particle
concentrations 30% and 40% in the radial direction at the exit section. It was experimentally
observed that with a decrease in concentration at the wall, there was an increase in concentration
at the region next to the wall. The experimental observations were obtained at ve layers in
the radial direction of the sample with the centre of the layers at reduced radius r/R = 0.1,
0.3, 0.5, 0.7 and 0.9, respectively.
The constant empirical parameters K
c
and K
= 0.65 (K
c
/K
= 0.49)
were obtained if the power-law model was employed and the values K
c
= 0.33, K
= 0.65
(K
c
/K
= 0.51) were obtained if the modied cross model was employed. The agreements
between predictions and experimental observations were good. The maximum relative error
with particle concentrations 30% and 40% is less than 15% and 10%, respectively. The results
of numerical predictions with the power-law and modied cross model and experimental
observations are summarized in table 3.
Both experimental and numerical results showed a near plug-like concentration prole
for a non-Newtonian concentrated suspension during pressure-driven tube ow. The shape is
514 X Chen et al
Log shear rate (1/s)
L
o
g
v
i
s
c
o
s
i
t
y
(
P
a
.
s
)
5
4
3
2
1
-4 -2 0 2 4 6
(a)
(b)
Figure 3. (a) Rheological behaviour of EVA420 binder at 180
C: () experimental;
() modied cross model; (- - - -) power-law model. (b) Effect of powder loading on relative
viscosity of the suspension. Experimental conditions: T = 180
C, a = 29 m, = 5 s
1
.
() Krieger model (
c
= 0.68, m = 0.82); ( ) experimental data.
)
Figure 4. Relative error plot of particle concentration near the wall at the exit section versus grid
size in a radial direction. Conditions: T = 180
C, a/R = 0.02, L = 40 mm, R = 1.5 mm,
K
c
/K
= 0.49 Q = 9.82 10
7
m
3
s
1
. Grid numbers in ow direction: 40; particle
concentration: ( ) 30%; () 40%.
S
h
e
a
r
-
i
n
d
u
c
e
d
p
a
r
t
i
c
l
e
m
i
g
r
a
t
i
o
n
m
o
d
e
l
l
i
n
g
5
1
5
Table 3. Comparisons of particle concentration between numerical predictions and experimental observations with initial particle concentration of 40%.
Experimental
data
Numerical
prediction Absolute error Relative error (%)
Concentration
bulk Model
Constant
K
c
/K
exp
max
(%)
exp
min
(%)
num
max
(%)
num
min
(%)
num
max
exp
max
(%)
num
min
exp
min
(%)
num
max
exp
max
num
min
exp
min
= 30% Power-law 0.492 31.33 24.77 31.29 27.38 0.04 2.61 0.1 8.7
Modied cross 0.510 31.20 27.89 0.13 2.12 0.325 7.1
= 40% Power-law 0.492 42.62 34.17 42.52 35.13 0.10 0.96 0.25 2.40
Modied cross 0.510 42.42 35.15 0.20 0.98 0.50 2.45
Note:
num
max
and
num
min
are not the maximum and minimum values respectively predicted. They were the values at the corresponding
exp
max
and
exp
min
, respectively.
516 X Chen et al
25
29
33
37
41
45
0 0.25 0.5 0.75 1
Reduced radius (r/R)
C
o
n
c
e
n
t
r
a
t
i
o
n
(
%
)
Centerline of the tube Wall of the tube
19
23
27
31
35
0 0.2 0.4 0.6 0.8 1.0
Reduced radius (r/R)
C
o
n
c
e
n
t
r
a
t
i
o
n
(
%
)
Wall of the tube Centerline of the tube
(a)
(b)
Figure 5. (a) Concentration prole at exit section (z = 1.0) for
= 40%. Experimental
conditions: T = 180
C, a/R = 0.02, L = 40 mm, R = 1.5 mm, Q = 9.82 10
7
m
3
s
1
.
Results: ( ) experimental; () modiedcross model (K
c
/K
= 0.51); (- - - -) power-lawmodel
(K
c
/K
= 0.51); (- - - -) power-law
model (K
c
/K
= 0.49).
quite different fromthe cusp-like prole of concentration predicted at the centreline of the tube
where the local shear rate vanished for Newtonian suspension uids (n = 1) [8]. The maximum
concentration occurred at about r/R = 0.85. This is because the concentration distribution
prole in the cross-section has not yet attained its nal steady-state prole and thus is transient-
effect related. Also, this is due to the effect of shear-thinning of non-Newtonian uid. The
effect of shear-thinning increases the gradient of concentration and shear deformation rate and
amplied particle migration near the wall. As shown in gures 5(a) and (b), the numerical
model could capture this feature.
S
h
e
a
r
-
i
n
d
u
c
e
d
p
a
r
t
i
c
l
e
m
i
g
r
a
t
i
o
n
m
o
d
e
l
l
i
n
g
5
1
7
Table 4. Comparisons of pressure between numerical predictions and experimental observations with initial particle concentration of 40%.
Numerical prediction Relative error (
r
) (%)
With migration With migration
Without
migration
P
0
(MPa)
Power-law
model
K
c
/K
= 0.492
P
1
(MPa)
Modied
cross model
K
c
/K
= 0.510
P
2
(MPa)
Experimental
data
P
3
(MPa)
Without
migration
r
=
P
0
P
3
P
3
Power-law
model
r
=
P
1
P
3
P
3
Modied
cross model
r
=
P
2
P
3
P
3
Volumetric
ow rate
(cm
3
s
1
)
(D = 1 mm,
L = 3 cm)
(
)
30%
(
)
40%
(
)
30%
(
)
40%
(
)
30%
(
)
40%
(
)
30%
(
)
40%
(
)
30%
(
)
40%
(
)
30%
(
)
40%
(
)
30%
(
)
40%
0.009 82 10.36 15.84 9.68 13.45 9.00 12.80 8.01 11.77 29.34 34.58 20.85 14.27 13.36 8.75
0.019 6 13.52 20.15 12.63 17.48 11.70 16.71 11.87 15.80 13.9 27.53 6.4 10.63 1.43 5.76
0.049 19.23 29.41 17.98 24.87 16.72 23.77 17.61 22.68 9.2 29.67 2.1 9.63 5.05 4.81
0.098 2 25.4 38.43 23.50 32.50 22.85 31.06 23.86 29.77 6.45 29.09 1.5 9.18 1.51 4.30
0.196 32.81 50.15 30.66 42.41 30.50 40.53 30.70 41.63 13.40 20.47 0.13 1.86 0.65 2.64
0.49 46.68 71.37 43.63 60.46 43.57 57.68 43.90 60.41 6.33 18.15 0.62 0.08 0.62 4.50
518 X Chen et al
0
15
30
45
60
75
0 0.1 0.2 0.3 0.4 0.5
0.5
Volumetric flow rate Q (cm
3
/s)
P
r
e
s
s
u
r
e
P
(
M
P
a
)
0
10
20
30
40
50
0 0.1 0.2 0.3 0.4
Volume flow rate (Q cm
3
/s)
P
r
e
s
s
u
r
e
(
M
P
a
)
(a)
(b)
Figure 6. (a) Comparison of pressure drop versus ow rate between predicted and experimental
data for
= 40%. Experimental conditions: T = 180
C, a/R = 0.02, L = 30 mm, R = 0.5 mm.
Results: () experimental; ( ) modied cross model (K
c
/K
= 0.51); (- -- -) power-law
model (K
c
/K
= 0.49); (
= 0.49);
(
(
%
)
Centerline of the tube Wall of the tube
Figure 7. Effect of non-Newtonian index n on steady-state proles of particle concentration
in a radial direction for
= 65%. Conditions: T = 180
C, a/R = 0.2, K
c
/K
= 0.51,
Q = 9.82 10
7
m
3
s
1
. Results: () n = 0.385; ( ) n = 0.6; () n = 0.8;
(
) n = 1.
maximum relative error occurred at low volumetric ow rate (Q = 9.82 10
3
cm
3
s
1
).
This is because the power-law model did not describe the characteristic rheological
behaviour of the binder system at low shear rate. Generally, the power-law model over-
estimates the value of viscosity at low shear rate, as shown in gure 3(a). When the
modied cross model was employed, the range of the relative error of pressure predicted
decreased to less than 10% over the whole range of ow rate investigated for particle
concentration 40%.
Better prediction in pressure from the power-law model than that from the modied cross
model was obtained at high volumetric owrate due to the use of a different K
c
/K
, which was
obtained by using a data tting technique in terms of the distribution of particle concentration
measured. The power-law model seems to tend to predict less particle migration and thus
higher pressure. However, the simulated results in particle concentrations of both 30% and
40%indicated that the modied cross model is better than the power-lawmodel in concentrated
suspension ow for a wide range of shear rates.
4.3. Fully developed steady ow
For a fully developed, steady-state, pressure-driven uid tube owusing the power-lawmodel,
the analytical solution to equation (16) can be obtained as
r
R
=
_
_
n
_
c
w
c
_
mn((K
/K
c
)(1/n))
. (35)
This analytical solution for a power-law uid is different from that given by the original
model [8] for the Newtonian uid in that the non-Newtonian index n has an effect on
concentration prole. Figure 7 shows a comparison of predicted concentration proles for
particle concentration 65% with different non-Newtonian indexes n under fully developed
ow. When n = 1, i.e. Newtonian uid, the cusp-like concentration prole was predicted
as expected as in the model of Phillips et al [8]. When the index n decreases (n < 1), the
520 X Chen et al
0
0.02
0.04
0.06
0.08
0.1
0.12
0 0.2 0.4 0.6 0. 1
1
8
8
Reduced length (z=l/L,L=40mm)
Entrance of the tube Exit of the tube
0
0.05
0.1
0.15
0.2
0.25
0.3
0 0.2 0.4 0.6 0.
Reduced length (z=l/L,L=40mm) Entrance of the tube Exit of the tube
(b)
(a)
Figure 8. (a) Predicted
at various a/R ratios for bulk concentration
= 60%. Experimental
conditions: T = 180
C, L = 40 mm, K
c
/K
= 0.51, Q = 9.82 10
7
m
3
s
1
. Predictions at
(a/R) ratio: () 0.08; () 0.04; () 0.02; ( ) 0.01; (
= 0.51, Q = 9.82 10
7
m
3
s
1
. Predictions at (a/R) ratio: () 0.08;
() 0.04; () 0.02; ( ) 0.01; (
) 0.005.
cusp-like prole at the centre of the tube attens gradually. It is observed that the effect of
non-Newtonian index n is to relax the cusp-like appearance in the concentration prole at the
centre of the tube. This was due to the effect of shear-thinning of non-Newtonian rheological
behaviour. That is, the effect of shear-thinning was to decrease the viscosity and increase the
shear rate at the wall, which amplied particle migration.
4.4. Effects of the ratio of particle size to tube radius, r/R
Figure 8 shows the deviation
of particle concentration versus reduced length (z = l/L) at
the volumetric ow rate Q = 9.82 10
1
cm
3
s
1
with initial particle concentrations of 60%
and 65% for various values of a/R.
was dened as (
w
)/
w
at the wall from the bulk concentration
. With increasing values of
a/R, the deviation
increases due to particle migration. The deviation
is relatively small
Shear-induced particle migration modelling 521
(5%) if the value of a/R is below 0.005. This means that if a/R is small, there will be little
particle migration.
Figure 8 also shows that the deviation
decreases for increasing initial values of particle
concentration at a specied value of a/R. With a lower initial particle concentration of 60%
as shown in gure 8(a), the deviation of particle concentration is around 25% for the value of
a/R = 0.08 at the exit section of the tube.
is about 10% for an initial particle concentration
of 65% as shown in gure 8(b). However, if the ratio of particle size to characteristic size
of the part exceeds a certain value, there will still be particle migration occurring during the
process even if the particle concentration is close to the critical particle concentration.
5. Conclusion
The complex behaviour of a suspension containing monosize sphere particles in a generalized
Newtonian binder system was simulated using an extended diffusive-ux model. This model
couples a generalized Newtonian stress/strain relationship with a shear-induced migration
model of the suspended particles in which the local effective viscosity is dependent on the
local volume fraction of the particles and shear deformation rate. The shear-induced particle
migration model could capture the essential features in pressure-driven suspension tube ow.
The model, with the material constants determined, was demonstrated to have good agreement
with experimental results. The results indicated that particle migration could be a signicant
phenomenon. If it is ignored, signicant error could exist in the particle concentration
distribution and pressure predicted.
References
[1] German R M 1991 Powder Injection Molding (Princeton, NJ: Metal Powder Industries Federation)
[2] Leighton D and Acrivos A 1987a The shear-induced self-diffusion in concentrated suspensions J. Fluid Mech.
181 41539
[3] Leighton D and Acrivos A 1987b Measurement of shear-induced self-diffusion in concentrated suspensions of
spheres J. Fluid Mech. 177 10931
[4] Abbot J R, TetlowN, GrahamAL, Altobell S A, Fukushima E, Mondy LAand Stephens TA1991 Experimental
observations of particle migration in concentrated suspensions: Couette ow J. Rheol. 35 77395
[5] Gadala-Maria F and Acrivos A 1980 Shear-induced structure in a concentrated suspension of solid spheres
J. Rheol. 24 799811
[6] Karnis A, Goldsmith H L and Mason S G 1966 The kinetics of owing dispersions: I. Concentrated suspensions
of rigid particles J. Colloid Interface Sci. 22 53153
[7] Hookham P A 1986 Concentration and velocity measurements in suspensions owing through a rectangular
channel PhD Thesis California Institute and Technology
[8] Phillips R J, Armstrong R C, Brown R A, Graham A L and Abbot J R 1992 A constitutive equation for
concentrated suspensions that accounts for shear-induced particle migration Phys. Fluids A 4 3040
[9] Hampton RE, Mammoli AA, GrahamALand TetlowN1997 Migration of particles undergoing pressure-driven
ow in a circular conduit J. Rheol. 41 62140
[10] Koh C, Leal L G and Hookham P A 1994 An experimental investigation of concentrated suspension ow in a
rectangular channel J. Fluid Mech. 256 132
[11] Averbakh A, Shauly A, Nir A and Semiat R 1997 Slow viscous ows of highly concentrated suspensions
part II. Particle migration, velocity and concentration proles in rectangular ducts Int. J. Multiphase Flow 23
61329
[12] Chow A W, Sinton S W and Iwamiya J H 1994 Shear-induced particle migration in Couette and parallel plate
viscosmeters: NMR imaging and stress measurements Phys. Fluids A 6 256175
[13] Phan-Thien N 1994 A brief review of suspension mechanics 7th Nat. Conf. on Rheology (Brisbane: Australian
Society of Rheology) pp 710
[14] Nott P R and Brady J F 1994 Pressure-driven ow of suspensions: simulation and theory J. Fluid Mech. 275
15799
522 X Chen et al
[15] Fang Z and Phan-Thien N 1999 A particle suspension model: an unstructured nite-volume implementation
J. Non-Newtonian Fluid Mech. 80 13553
[16] Allende M and Kalyon D M 2000 Assessment of particle-migration effects in pressure-driven viscometric ows
J. Rheol. 44 7990
[17] Wang HP and Lee HS 1989 Numerical techniques for free and moving boundary problems Computer Modeling
for Polymer Processing: Fundamentals ed C L Tucker III (Hanser)
[18] Krieger I M 1972 Rheology of monodisperse lattice Adv. Colloid Interface Sci. 3 11136
[19] Lam Y C, Chen X, Tan K W, Ma J, Tam K C and Yu S C M 2002 Experimental measurements and simulation of
particle migration in a pressure-driven tube ow: applications to ow in runner J. Injection Molding Technol.
6 4557
[20] Tan K W, Chen X, Lam Y C, Ma J and Tam K C 2003 Experimental investigation of shear-induced particle
migration in steady-state isothermic extrusion J. Soc. Rheol. Japan accepted
[21] Carnahan B, Luther H A and James O W 1969 Applied Numerical Methods (New York: Wiley)
[22] Patanker S V 1980 Numerical Heat Transfer and Fluid Flow (New York: McGraw-Hill)
[23] Christopher W M 1994 Rheology: Principles, Measurements, and Applications (Weinheim: VCH)