J. Non-Newtonian Fluid Mech.

143 (2007) 133140

Inverse problem solution of squeeze flow for parameters

of generalized Newtonian fluid and wall slip
D.M. Kalyon , H.S. Tang 1
Stevens Institute of Technology, Castle Point St., Hoboken, NJ 07030, United States
Received 17 February 2007; accepted 21 February 2007

Abstract
The determination of the parameters describing the shear viscosity and wall slip behavior of complex fluids, including polymer melts, polymeric
suspensions, and polymeric gels, is a challenge. Here we use the squeeze flow of generalized Newtonian fluids with and without wall slip to
generate an inverse problem solution methodology for the estimation of the parameters of the shear viscosity and wall slip. Analytical as well as
the finite element method (FEM) based numerical solutions of the squeeze flow are used for the analysis and the determination of the parameters.
It is suggested that the analytical and numerical analyses can provide the basic where withal necessary for the solution of the inverse problem
of squeeze flows to characterize the shear viscosity and the wall slip parameters provided that the issues of uniqueness and stability are properly
addressed.
2007 Elsevier B.V. All rights reserved.
Keywords: Squeeze flow; Inverse problem; Viscoplasticity; Wall slip

1. Introduction
Many industrial materials including polymer melts, polymeric suspensions, and polymeric gels are complex fluids. Such
complex fluids present challenges for the characterization of
their rheological behavior [16]. This challenge stems from the
frequently encountered viscoplasticity and concomitant wall slip
behavior. Specialized techniques and multiple viscometers are
employed to simultaneously characterize the parameters of the
shear viscosity material function and wall slip versus the shear
stress relationship [13]. Generally, the procedure for the characterization of the shear viscosity and wall slip involves systematic
changes in the surface to volume ratio of the sample followed
by the analysis of the flow curves [15]. When capillary flow
is employed to generate flow rate versus pressure drop data, the
procedure for wall slip corrected shear viscosity determination
requires the use of multiple capillary dies involving systematically varied capillary lengths at constant diameter and different
capillary diameters at constant length over the diameter ratio (for
example, 12 capillaries were used in the study of the behavior

Corresponding author.
E-mail address: dkalyon@stevens.edu (D.M. Kalyon).
1 Present address: Pacific Northwest National Laboratory, 1100 Dexter Ave
N, Seattle, WA 98109, United States.
0377-0257/$ see front matter 2007 Elsevier B.V. All rights reserved.
doi:10.1016/j.jnnfm.2007.02.010

of concentrated suspensions by Yilmazer and Kalyon [1]). In

steady torsional flow between two parallel disks the procedure
requires the systematic change of the gap of the rheometer or
the imaging of the velocity distribution at the free surface of the
fluid [1,2].
Another approach applicable to the parameter estimation for
generalized Newtonian fluids subject to wall slip is to formulate
inverse problems. Inverse problem solutions have been used in
various applications including fluid mechanics and in rheological characterization [79]. However, the inverse problems can
be ill posed, that is, they can generate non-unique solutions and
they can be unstable [1012]. Moreover, as indicated by Wang
et al. [13], the greater is the number of unknown parameters to
be determined upon the solutions of the inverse problems, the
more difficult it becomes to obtain a unique solution.
The squeeze flow involves the compression of the fluid partially or completely filling the space between two parallel and
rigid circular disks, one or both of which are moving in the axial
direction at constant relative velocity, while the time-dependent
force is being measured, or under constant normal force while
the time-dependent relative velocity of the plate is measured
[14]. Useful approximate solutions to this unsteady state problem can be obtained by assuming that the speed of travel of
the disk is sufficiently slow so that the time derivatives can be
neglected, i.e., the quasi-steady state assumption [14,15]. A

134

D.M. Kalyon, H.S. Tang / J. Non-Newtonian Fluid Mech. 143 (2007) 133140

as a function of the second scalar invariant, II, of the rate of

deformation tensor. The squeeze flow is subject to the following
non-linear wall slip condition, i.e., slip velocity, Us , versus the
shear stress, rz :
Us = |rz |sb ,

Fig. 1. Schematic representation of the squeeze flow.

number of studies have been made on the squeeze flow for generalized Newtonian fluids [1624]. Nevertheless, although its
literature is rich, the squeeze flow has not been fully exploited
to facilitate the identification of parameters of constitutive equations and wall slip. In this paper, the isothermal squeeze flow of
generalized Newtonian fluids subject to wall slip is analyzed
using analytical and FEM-based numerical methods to elucidate parameter determination capabilities of squeeze flow for
the characterization of shear viscosity and wall slip.
2. Squeeze ow: background
A schematic representation of the squeeze flow is shown in
Fig. 1. An incompressible fluid is placed in between the two
circular disks with a radius of R. Although different experimental configurations are possible, including the use of a constant
normal force, we have selected to work here with our current
experimental configuration involving the top disk moving down
at a constant speed of V, and the bottom plate being stationary.
The time-dependent gap between the plates is designated as h,
and the total force acting on the top plate is f.
Industrially relevant polymer melts and polymeric suspensions and gels generally exhibit relatively high shear viscosity
values to give rise to creeping flow conditions upon being subjected to squeeze flow. Under the resulting typical relatively
low Reynolds number conditions the inertial terms in the equations of motion can be neglected. The governing equations for
axisymmetric flow of incompressible fluids under quasi-steady
state and isothermal conditions become:
1 rvr
vz
+
= 0,
r r
z

(1)

rz

p
1 rrr
+

+
= 0,
r r
z
r
r

(2)

zz
p
1 rrz
+
+
= 0.
r r
z
z

(3)

Here vr and vz are the velocity components in the radial and axial
directions, i.e., r and z, respectively, p is the pressure, and rr ,
zz and rz are components of the deviatoric stress tensor. If
the rate of movement of the plate, V, is sufficiently slow to allow
viscoelastic effects to be considered to be negligible, generalized
Newtonian fluid constitutive behavior is applicable, i.e., =
(II), and  are the stress and rate of deformation tensors,
respectively. (II) is the shear viscosity material function varying

(4)

where the Navier slip coefficient, , and the slip exponent, sb,
are parameters of the wall slip behavior [5]. The slip coefficients for the top and bottom disks, i.e., t and b , respectively,
can be different due to differences in roughness or the materials
of construction of the two plates. For 1D flow the shear stress
for the HerschelBulkley type viscoplastic constitutive equation becomes rz = y m|dvr /dz|n1 (dvr /dz) for |rz |y
( sign is used for negative shear stress, rz ) and the shear rate
(dvr /dz) = 0 for | rz | < y (Herschel and Bulkley [25]). Thus,
for the HerschelBulkley fluid subject to wall slip there are five
parameters (m, n, y , , sb) that need to be determined for representing the shear viscosity along with the wall slip condition.
For concentrated suspensions, sb can be estimated from the shear
viscosity material function of the binder and the properties of
the particles [5] and thus the determination of only the other
four parameters will be elucidated in this study. The solution of
the squeeze flow problem for the HerschelBulkley fluid also
provides the solutions for Newtonian fluid with n = 1 and y = 0;
Bingham fluid with n = 1; power-law behavior with y = 0; all
subject to either no slip or slip at the wall.
The normal force acting on the top plate, f, is
 R
f =
2(p + zz )|z=h r dr
0


=



dp

dr



r + 2zz r 
2

dr.

(5)

z=h

The problem (Eqs. (1)(5)) can be solved either numerically or

analytically to determine the normal force, f, acting on the moving plate [1523]. A numerical analysis of the time-dependent
squeeze flow of viscoplastic fluids [24] is also available. The
analytical solutions, which generally rely on the lubrication
assumption for viscoplastic fluids, are subject to the problem
associated with the extensional deformation of the fluid in
between the two plates, as dictated by the continuity equation
(Eq. (1)). One possible solution is the definition of a pseudoplug that can be weakly yielded, as demonstrated for the thin
film flow of viscoplastic fluids over inclined planes [26,27].
For the Newtonian fluid (n = 1 and y = 0) subject to Naviers
wall slip condition with constant slip coefficient, , the pressure gradient (dp/dr), is derived from Eqs. (1)(4) for the
lubrication flow regime (h  R) as

dp
6mVr
= 3
.
dr
h + 6mh2

(6)

Compared with p, zz is assumed to be negligible and the normal

force becomes:
f

3mR4 V
.
2(h3 + 6mh2 )

(7)

D.M. Kalyon, H.S. Tang / J. Non-Newtonian Fluid Mech. 143 (2007) 133140

Under conditions for which the slip effect is relatively small,

i.e.,  h/6m, and upon ignoring high order terms the normal
force, f, is given as
9R4 V 2
3R4 V
m

m .
(8)
2h3
h4
For the no slip condition, i.e., = 0, Eq. (8) becomes the Stefan
equation [28].
For a power-law fluid ( y , = 0) without wall slip, the lubrication flow analysis provides:


2n + 1 n n n
dp
2m

= 1+2n
V r ,
(9)
dr
h
n

f =

derivatives of objective functions are evaluated using central

differences:
Jm+m,n,y , Jmm,n,y ,
J

,
(14)
m
2m
Jm,n+n,y , Jm,nn,y ,
J

,
(15)
n
2n
Jm,n,y +y , Jm,n,y y ,
J

,
(16)
y
2y
Jm,n,y ,+ Jm,n,y ,
J

2

(17)

4. Inverse problem solutions

and
f =

135

2mRn+3
(n + 3)h2n+1

2n + 1
n

n

V n,

(10)

i.e., the Scott equation [29].

3. Inverse problem for parameter estimation
Given a set of experimental data for the forces 0 < f1e <
e exerted on the top plate to drive the flow at
f2e < < fM
gaps h1 > h2 > > hM > 0, the least square objective function
becomes:

M 
fi 2
1
e
1 e ,
(11)
J(F, F )
M
fi
i=1

Fe

e }, F = {f , f , . . ., f }, and the latwhere

= {f1e , f2e , . . . , fM
1 2
M
ter represents the total normal forces on the plates given by an
analytical or a numerical solution.
Let m, n, y , and be real numbers and = {m, n, y , }. The
parameter estimation can be formulated as the following inverse
problem:

find {} = {|J(F, F e ) = min J(F, F e )},


subject to fi =
0


2(p + zz )i z=h r dr,

(12)

4.1. Newtonian uid with wall slip

For simple cases, for which analytical solutions of the squeeze
flow are available (Eqs. (8) and (10)), the generation of the
inverse problem solution methodology is straightforward. For
example, consider the squeeze flow of a Newtonian fluid with
wall slip under lubrication flow conditions. The plate force, f,
given by Eq. (8) can be arranged as
fi = Ai m Bi m2 ,

1 i M,

(18)

where Ai = 3R4 V/2hi 3 > 0, Bi = 9R4 V/hi 4 > 0. Letting

= m and = m2 , the inverse problem can be set up by employing the objective function:

M 
A i Bi 2
1
J(F, F e ) =
1
(19)
M
fie
i=1

to find (m, ) that minimizes J on plane. Since on plane,

J C1 (R2 ), the minimum of J is found from:


M 
A i Bi
Ai
2
J
1

,
(20)
=
0=

M
fie
fie
i=1

i = 1, 2, . . . , M.
(13)

 
M 
J
A i Bi
Bi
2
0=
1
=
e

M
fi
fie

(21)

i=1

From Eqs. (20) and (21), one has

In case of the Newtonian fluid with wall slip, only two parameters
M
m and are in effect, and they are estimated using the force fi
e 2 M A /f e M A B /f e2 M B /f e
i=1 (Bi /fi )
i=1 i i
i=1 i i i
i=1 i i
=
given by Eq. (8). In case of the power-law fluid without wall

2 ,
M

M
e 2 M (B /f e )2
e2
slip, m and n are determined using the force fi given by Eq.
i=1 (Ai /fi )
i=1 i i
i=1 (Ai Bi /fi )
(10). For the HerschelBulkley fluid with wall slip, all the four
(22)
parameters m, n, y , and need to be identified, and the forces
fi are computed using FEM.

2
M
e )2 M A /f e M (A /f e )2 M A B /f e2 M B /f e
(A
B
/f
i
i
i
i=1
i=1 i i
i=1 i i
i=1 i i i
i=1 i i

=
(23)

2  .
M
M
M
2
2
2 M
2
e
e
e
e
i=1 Ai Bi /fi
i=1 (Ai /fi )
i=1 (Bi /fi )
i=1 (Ai Bi /fi )
Associated with FEM-based solution procedure, the minimization of Eq. (12) can be carried out using various methods
including the steepest descent method, the conjugate gradient method or combinations of multiple methods [30]. The

It is seen that under the condition of

2

M
 M  
M 

 A i Bi
A i 2  Bi 2

= 0,
2
fie
fie
fie
i=1

i=1

i=1

(24)

136

D.M. Kalyon, H.S. Tang / J. Non-Newtonian Fluid Mech. 143 (2007) 133140

(, ), given by Eqs. (22) and (23), is the unique minimizer in

R2 for the objective function defined by Eq. (19), and the minimizer determines the parameters m = , = / 2 . It is required
that = 0, which is guaranteed under the condition:
 M
M 
M
M

Bi 2  A i  A i B i  B i

= 0.
(25)
2
fie
fie
fie
fie
i=1

i=1

i=1

i=1

then Eq. (31) will have a solution if G has a zero point. Notice
that i < j, 0 < fie /fje < 1, and hi /hj > 1, it can be shown that

when n > nupper

4.2. Power-law uid without wall slip

Starting with the Scott equation, Eq. (10), and considering
that R, V = const, the objective function becomes:

2
M
(2n+1)

mh
g(n)
1
i
1
,
(26)
J(F, F e ) =
M
fie
i=1

where
2Rn+3
g(n) =
(n + 3)

2n + 1
n

n

V n.

(27)

At minimum points:


M
(2n+1)
(2n+1)
J 2 
mhi
hi
g(n)
g(n)
0= =
1

,
m M
fie
fie
i=1

(28)

0=

M 
2

J
=
n
M

i=1

mh(2n+1)
g(n)
i
fie

mh(2n+1)
(2 ln(hi ))g(n)
i

fie


hi 2n+1
> 0,
hj


e e
1 ln(fj /fi )min

1 .
2
ln(h1 / hM )

G > 0, qij 1
when n < nlower

fie
fje

fie
fje

dg(n)/dn

(29)
which yield:
M (2n+1) e
hi
/fi
.
m = M i=1(4n+2)
2
g(n)/fie
i=1 hi

(30)

If n exists uniquely, then m will be uniquely determined and the

minimizer (m, n) is unique. From Eqs. (28) and (29), and upon
rearrangement the following is obtained:

(34)

Therefore, G must have at least one zero point n (nlower , nupper )

for Eq. (31) to have a solution. A simple but helpful way to check
the existence and uniqueness of the solution for Eq. (31) is to plot
out G on the Gn plane and examine how many times curve G
crosses the n-axis. The number of times it crosses the n-axis is the
number of the roots of Eq. (31). When G has multiple solutions,
one needs to eliminate unrealistic solutions. For instance, those
with n > 1 and n < 1 must be excluded for shear-thinning and
thickening materials, respectively. For the case of M = 2, i.e.,
when only two experimental data are used, n exists uniquely
and the minimizer becomes:
m=

(33)

h1 (2n+1) f1e f2e + h2 (2n+1) f1e f2e

2

mh(2n+1)
i


hi 2n+1
< 0,
hj


e /f e )
1
ln(fM
1

1 ,
2 ln(hi / hj )min

G < 0, qij 1

(4n+2) e2
f2

+ h2

n = nlower = nupper

1
=
2

(h1

(4n+2) e2
f1 )g(n)


ln(f2e /f1e )
1 .
ln(h1 / h2 )

(35)

(36)

As an example, let us consider one set of experimental squeeze

flow data (force versus gap) collected by Zhang et al. [16],
reproduced in Fig. 2. The numerical solution of Eqs. (30)
and (31) gives m = 48,700 Pa s0.2 , n = 0.2. The forces determined by the estimated parameters are plotted in Fig. 2.
Additional numerical test runs show that, in the range of
10,000 < m < 500,000 and 0.05 < n < 2, the estimate is unique.

(4n+2) (2n+1)
hj
(ln(hi ) ln(hj ))
2 e
e
fi f j
i=1,j=2,i<j
M1,M


hi

fe
1 ie
fj

hi
hj

2n+1 

=0

(31)

which provides the value of n. The question of whether n has a

solution depends on whether Eq. (31) has a solution. Let
G=

(4n+2) (2n+1)
hj
(ln(hi ) ln(hj ))
2 e
e
fi f j
i=1,j=2,i<j
M1,M


hi

fe
1 ie
fj

hi
hj

2n+1 

(32)

Fig. 2. Force vs. gap for a power-law fluid obtained from experiment [16] and
computation. R = 0.0254 m and V = 2.18 104 m/s.

D.M. Kalyon, H.S. Tang / J. Non-Newtonian Fluid Mech. 143 (2007) 133140

137

Zhang et al. [16] report the results of their rheological characterization as: m = 44,900 Pa s0.25 and n = 0.25, which are close
to the estimates of these parameters given above.
4.3. HerschelBulkley uid with wall slip
The inverse problem solution of the isothermal squeeze flow
for the HerschelBulkley fluid subject to wall slip requires a
numerical solution. Here, the FEM approach given by Lawal and
Kalyon [20] is followed to solve the governing equations, and a
combination of the steepest descent and the conjugate gradient
methods [30] is employed to carry out the minimization required
for the solution of the inverse problem. Starting from an initial
guess, the minimization begins with the steepest descent method
and, after certain number of steps in searching, switches to the
conjugate gradient method. The FEM code was validated using
experimental data, and the effectiveness of the minimization
program has been tested (see Tang and Kalyon [31]).
The validity of the inverse problem solution methodology
was probed directly by employing the FEM solution for both
the linear (sb = 1) and non-linear (sb = 1) wall slip conditions.
Here typical results for the linear slip condition are provided.
For validation a given set of four parameters {m, n, y , }
and a set of plate gaps {hi } were employed to generate a
set of forces {fie } using FEM, which were used as the set of
experimental forces. For example, a set of {hi } and {fie } were
obtained for V = 2.1 105 m/s and R = 0.0286 m using the set
of four parameters {m0 , n0 , y0 , 0 } = {5066 Pa s0.45 , 1978 Pa,
2.14 107 m s1 Pa1 }. This set of parameters is referred as
to P-I. P-I is in the range of material parameters discussed in
[20]. The four parameters {m, n, y , } are made dimensionless
by
m
y
, n = n, y =
,
m0
[m0 (V/R)n ]
 m  V n
0

.
=
V
R
m =

Then, P-I becomes {m*, n*, y , *} = {1, 0.45, 10, 2}. Obviously, P-I is a solution with J = 0 to the inverse problem. In
the following, numerical tests are reported to demonstrate the
limitations of the conventional inverse problem solution. If the
inverse problem solution methodology functions flawlessly than
the parameters identified should coincide with those of the set
P-I, which was used to generate the gap versus force data.
Let us first set the initial point to start the search for the minima under conventional search conditions (Eqs. (12) and (13) to:
{m*, n*, y , *} = {0.07, 0.73, 2.1, 3.0}. These initially selected
parameters are not close to the exact solution P-I. The inverse
problem solution which uses this set of initial parameters deviates from the exact solution, i.e., the search converges to a local
minimum with J = 6.7 105 and yields a significantly different solution {m*, n*, y , *} = {0.85, 0.34, 4.1, 4.3}, referred
as to P-II. As shown in Fig. 3, however, the forces given by P-I
and P-II are very close to each other, in spite of the two sets of
parameters being very different from each other. From Eq. (5)
it is known that the forces are determined by the distribution of

Fig. 3. Example for ill-posedness of inverse problem for the HerschelBulkley

fluid subject to wall slip. Forces are generated by two different sets of parameters
P-I {m*, n*, y , *} = {1, 0.45, 10, 2} and P-II {m*, n*, y , *} = {0.85, 0.34,
4.1, 4.3}.

(p + zz ) on the plates. Fig. 4 shows the corresponding p + zz distributions for the two sets of parameters. It is seen that, although
the parameters contained in sets P-I and P-II differ a lot, their
corresponding distributions of p + zz at different gaps are very
close to each other.
These typical calculations, reported in Figs. 3 and 4, suggest
that the inverse problem solution procedure for searching the values of the four parameters associated with the HerschelBulkley
fluid subject to wall slip, may not be stable/unique and thus ill
posed (two distinctly different sets of parameters give rise to a
similar group of normal force on the plate). This lack of uniqueness of the inverse problem solution for the parameters of shear
viscosity and wall slip was indeed found to be true for many
other computations, including those with different ranges for
{hi }, {fie }, {m*, n*, y , *} and different FEM mesh sizes when
all four parameters describing the shear viscosity and the wall
slip behavior were sought simultaneously.
Thus, this demonstrated lack of uniqueness renders it difficult to concomitantly identify all the four parameters {m, n, y ,
} using the conventional procedure. Two different approaches
were sought. In the first approach a smaller number of parameters
were sought assuming that some of the parameters can be deter-

Fig. 4. Distributions of (p + zz ) at the wall for different gaps. Circles-parameter

set P-I {m*, n*, y , *} = {1, 0.45, 10, 2} and dashesparameter set P-II {m*,
n*, y , *} = {0.85, 0.34, 4.1, 4.3}.

138

D.M. Kalyon, H.S. Tang / J. Non-Newtonian Fluid Mech. 143 (2007) 133140

mined independently from other experiments or theoretically. In

the second approach the solution domain was subdivided into
multiple subdomains and initial points pertaining to each subdomain were taken to arrive at the global minimum, which was
considered to represent the solution of the problem for all four
parameters.
Let us first start with the estimation of a smaller number
of parameters, a process that may be applicable if some of
the remaining parameters are determined using other methods.
This generally requires that a combination of multiple flows
is utilized such as capillary and rectangular slit flows (e.g.,
Tang and Kalyon [31] for the determination of the additional
parameters). Two typical examples for FEM-based parameter
estimation seeking to identify only two parameters are discussed
next.
In the first example we again use hi and fi e given in Fig. 3.
However, we let {m*, n*} = {1, 0.45}, and start to search for
{y , } from four different initial guesses which are far away
from their exact values: {y , } = {0.1, 0.02}, {y , } = {100,
200}, {y , } = {0.1, 200}, and {y , } = {100, 0.02}. The
search ends up at the exact solution {y , } = {10, 2}, regardless of which initial guess is used. Solutions were generated over
a broad range of {y , } values, all giving rise to unique solutions to suggest that the solution of the inverse problem becomes
unique over a large range of initial values of {y , }.
In the second example, we employ experimental squeeze flow
data of a poly(dimethyl siloxane), PDMS, melt (SE 30 available
from GE Silicones Company) at 25 C as shown in Fig. 5. Here,
experimental data collected with three different plate velocities,
V, are used. This case is solved for the power-law fluid subject to
non-linear wall slip condition, i.e., Eq. (4), thus four parameters
represent the shear viscosity and wall slip. However, from the
capillary flow experiments, it was determined that two of the four
parameters are = 3.52 1015 m s1 Pa2.165 , and sb = 2.165
[31] and m and n values are thus the only remaining unknowns.
Starting from four different sets of initial values, i.e., {m,
n} = {5000, 0.1}, {m, n} = {50,000, 0.6}, {m, n} = {50,000, 0.1}
and {m, n} = {5000, 0.6}, the minimization gives a unique solution to identify the parameters m = 16,500 Pa s0.39 and n = 0.39

Fig. 5. Experimental and computed forces for pure PDMS. R = 0.0286 m and
V = 2.1 104 , 4.2 104 and 8.4 104 m/s.

with J = 8.2 103 . The forces computed from the identified

{m, n} are also given in Fig. 5. The estimates of these two parameters also agree with the additional experimental findings; these
two parameters were estimated upon wall slip corrections using
the Mooney procedure [3,4] from capillary rheometry data as
m = 11,946 Pa s0.45 and n = 0.45.
Extensive numerical experimentation suggested that when
the number of parameters sought is equal to or greater than
three, reasonable estimates could only be made if the initial
guesses for parameters sought are relatively close to the true
values. However, how would one know where the true solution lies, prior to the characterization of the fluid, so that the
initial guesses are made to approach the true values of parameters? What is suggested here is the division of the parameter
space spanned by {m, n, y , } into subdomains and starting
the minimization with multiple initial conditions belonging to
each subdomain. In conjunction with this method the objective
function arising from each subdomain can be compared with the
others and a global minimum can be determined. To illustrate,
let us suppose the smallest possible values for m, n, y , and are,
respectively, mmin , nmin , ymin , and min and the largest possible
values are, respectively, mmax , nmax , ymax , and max . We divide
the whole possible domain of parameters [mmin , mmax ] [nmin ,
nmax ] [y min , y max ] [min , max ] into multiple subdomains
(with the total number of subdomains given as Lm Ln Ly L ) as
m = mmin + (I 1)m,
n = nmin + (J 1)n,
y = y min + (K 1)y ,
= min + (M 1),

1 I Lm + 1,
1 J Ln + 1,
1 K Ly + 1,
1 M L + 1,

where m = (mmax mmin )/Lm , n = [nmax nmin ]/Ln , y =

(ymax ymin )/Ly ,  = [max min ]/L , and Lm , Ln , Ly ,
and L are the numbers of subdomains in directions of m, n,
y , and , respectively. We start the minimization from each {I,
J, K, M), and upon obtaining all of the solutions pertaining to different sets of initial conditions the global minimum is obtained
and the corresponding set of parameters obtained is considered
to represent the solution. Provided that the division is sufficiently
fine and the objective function is continuous and smooth, it is
anticipated that the procedure will find the global minimum. In
the following example to this approach four parameters of a suspension of PDMS were sought starting from the experimental
squeeze flow versus time data.
A suspension consisting of a poly(dimethyl siloxane), PDMS,
fluid (GE Silicones SE 30) incorporated with 40% by volume of
hollow glass spheres (Potters Industry, mean size of 12 micron)
was subjected to squeeze flow. The experimental normal force
versus time data for this suspension are shown in Fig. 6. The
radius of the disks, R, was 0.0286 m and the plate velocity, V,
was varied as V = 2.1 104 , 4.2 104 and 8.4 104 m/s.
The value of the slip exponent, sb, could be a priori specified
as sb = 2.165 on the basis of the characterized shear viscosity of
the PDMS binder [5].
The entire parameter domain was subdivided to one hundred parameter subdomains, each of which contained a different

D.M. Kalyon, H.S. Tang / J. Non-Newtonian Fluid Mech. 143 (2007) 133140

139

apparent upon considering that the squeeze flow test takes only
a few minutes versus the weeks of work generally involved in
collecting conventional data, for example, capillary flow using
multiple capillaries and multiple apparent shear rates for each
capillary.
Acknowledgment
We are grateful to Ms. E. Birinci of SIT for the squeeze flow
data used as the examples.
References

Fig. 6. Experimental and computed forces for PDMS incorporated with

40 vol.% of glass spheres. R = 0.0286 m and V = 2.1 104 , 4.2 104 and
8.4 104 m/s.

set of initial parameters. The values of the objective function,

J defined by Eq. (11), were determined for all one hundred
sets of initial conditions from which the global minimum could
be determined. This procedure gave rise to the four parameters of the suspension represented as a HerschelBulkley fluid
subject to non-linear wall slip as: m = 42,500 Pa s0.50 , n = 0.50,
y = 177 Pa, = 1.55 1014 m s1 Pa2.165 . The forces given
by the estimate are also plotted in Fig. 6. At rz = 102 kPa these
parameters define the slip velocity, Us , as Us = 0.0012 m/s. The
utilization of capillary flow, and corrections of the data following
the procedures given in [4] provided m = 36,000, n = 0.49 and
Us = 0.0017 m/s at rz = 102 kPa, suggesting acceptable agreement with the capillary flow results. Therefore, these typical
results suggest that adequate estimates can indeed be obtained
from the inverse problem solution of the squeeze flow using
FEM if the solution domain is broken into multiple parameter
subdomains and systematically varied sets of initial conditions
to coincide with all subdomains are used to allow the global
minimum and hence the solution of the inverse problem for four
parameters to be obtained at reasonable accuracy.
5. Concluding remarks
The inverse problem solutions of squeeze flow outlined here
are promising approaches for the estimation of parameters of the
shear viscosity of generalized Newtonian fluids with or without
wall slip under quasi-steady state conditions and in the absence
of viscoelastic effects (when plate motion is relatively slow).
The uniqueness of the solution can be guaranteed only when two
parameters are sought. Seeking more than two parameters to represent the shear viscosity and wall slip behavior is complicated
on the basis of non-uniqueness and instability problems. However, if the parameter space is divided into multiple subdomains
so that initial conditions would coincide with each subdomain,
the reliability of the search method increases and reasonable estimates for obtaining up to four parameters can be obtained. The
advantage of the squeeze flow in conjunction with inverse problem solution for the determination of the parameters becomes

[1] U. Yilmazer, D.M. Kalyon, Slip effects in capillary and parallel disk
torsional flows of highly filled suspensions, J. Rheol. 33 (1989) 1197
1212.
[2] B. Aral, D.M. Kalyon, Effects of temperature and surface roughness on
time-dependent development of wall slip in steady torsional flow of concentrated suspensions, J. Rheol. 38 (1994) 957972.
[3] M. Mooney, Explicit formulas for slip and fluidity, J. Rheol. 2 (1931)
210222.
[4] D.M. Kalyon, P. Yaras, B. Aral, U. Yilmazer, Rheological behavior of a
concentrated suspension: a solid rocket fuel simulant, J. Rheol. 37 (1993)
3553.
[5] D.M. Kalyon, Apparent slip and viscoplasticity of concentrated suspensions, J. Rheol. 49 (2005) 621640.
[6] B. Aral, D.M. Kalyon, Viscoelastic material functions of noncolloidal suspensions with spherical particles, J. Rheol. 41 (1997) 599620.
[7] D.S. Schnur, N. Zabaras, Finite element solution of two-dimensional
inverse elastic problems using spatial smoothing, Int. J. Numer. Meth. Eng.
30 (1990) 5775.
[8] Y.L. Yeow, J.W. Taylor, Obtaining the shear rate profile of steady laminar
tube flow of Newtonian and non-Newtonian fluids from nuclear magnetic
resonance imaging and laser Doppler velocimetry data, J. Rheol. 46 (2002)
351365.
[9] Y.L. Yeow, Y.K. Leong, Shear rate and wall slip velocity functions of
polyvinyl chloride melts based on slit die viscometry data, Polym. Eng.
Sci. 44 (2004) 153162.
[10] V. Druskin, On the uniqueness of inverse problems from incomplete boundary data, SIAM J. Appl. Math. 58 (1998) 15911603.
[11] I. Knowles, Uniqueness for an elliptic inverse problem, SIAM J. Appl.
Math. 59 (1999) 13561370.
[12] Y.L. Keung, J. Zou, An efficient linear solver for nonlinear parameter
identification problems, SIAM J. Sci. Comput. 22 (2000) 15111526.
[13] W. Wang, H. Ishikawa, H. Yuki, An inverse method for determining material
properties of a multi-layer medium by boundary element method, Int. J.
Solids Struct. 38 (2001) 89078920.
[14] B. Bird, R. Armstrong, O. Hassager, Dynamics of Polymeric Liquids, vol.
1, 2nd ed., Wiley, New York, 1987, Chapter 4.
[15] G. Covey, B. Stanmore, Use of the parallel-plate plastometer for the characterization of viscous fluids with a yield stress, J. Non-Newton. Fluid Mech.
8 (1981) 249260.
[16] W. Zhang, N. Silvi, J. Vlachopoulos, Modeling and experiments of squeezing flow of polymer melts, Int. Polym. Proc. 10 (1995) 155164.
[17] M.J. Adams, I. Aydin, B.J. Briscoe, S.K. Sinha, A finite element analysis of
the squeeze flow of an elasto-viscoplastic paste material, J. Non-Newton.
Fluid Mech. 71 (1997) 4157.
[18] A. Lawal, D.M. Kalyon, Squeeze flow of viscoplastic fluids subject to wall
slip, Polym. Eng. Sci. 38 (1998) 17931804.
[19] H.M. Laun, M. Rady, O. Hassager, Analytical solutions for squeeze flow
with partial wall slip, J. Non-Newton. Fluid Mech. 81 (1999) 115.
[20] A. Lawal, D.M. Kalyon, Compressive squeeze flow of generalized Newtonian fluids with apparent wall slip, Int. Polym. Proc. 15 (2000) 6371.
[21] D.N. Smyrnaios, J.A. Tsamopoulos, Squeeze flow of Bingham plastics, J.
Non-Newton. Fluid Mech. 100 (2001) 165189.

140

D.M. Kalyon, H.S. Tang / J. Non-Newtonian Fluid Mech. 143 (2007) 133140

[22] N. Roussel, C. Lanos, Plastic fluid flow parameters identification using a

simple squeezing test, Appl. Rheol. 13 (2003) 132141.
[23] J. Engmann, C. Servais, A. Burbidge, Squeeze flow theory and applications to rheometry: A review, J. Non-Newton. Fluid Mech. 132 (2005)
127.
[24] G. Karapetsas, J. Tsamopoulos, Transient squeeze flow of viscoplastic
materials, J. Non-Newton. Fluid Mech. 133 (2006) 3556.
[25] W.H. Herschel, R. Bulkley, Measurement of consistency as applied to
rubber-benzene solutions, Am. Soc. Test. Proc. 26 (1926) 621633.
[26] N.J. Balmforth, R.V. Craster, A consistent thin-layer theory for Bingham
plastics, J. Non-Newton. Fluid Mech. 84 (1999) 6581.

[27] N.J. Balmforth, R.V. Craster, R. Sassi, Shallow viscoplastic flow on an

inclined plane, J. Fluid Mech. 470 (2002) 129.
[28] J. Stefan, K. Sitzungber, Versuche ber die scheinbare adhesion, Akad. Wiss.
Math. Natur. Wien 69 (1874) 713735.
[29] J.R. Scott, Theory and application of parallel-plate plastometer, Trans. Int.
Rubber Ind. 7 (1931) 169175.
[30] H. Press, S.A. Teukolsky, W.T. Vetterling, B.P. Flannery, Numerical
Recipes in Fortran, Cambridge University Press, Cambridge, 1992.
[31] H.S. Tang, D.M. Kalyon, Estimation of the parameters of HerschelBulkley
fluid under wall slip using a combination of capillary and squeeze flow
viscometers, Rheol. Acta 43 (2004) 8088.