The Finite Element

Method for Flow and

Heat Transfer Analysis*

Evan Mitsoulis and John Vlachopoulos

Department of Chemical Engineering
McMaster University
Hamilton, Ontario, Canada

ABSTRACT
The finite element method (FEM) is discussed
and a specific formulation for flow problems is
outlined that can encompass non-Newtonian inelastic and viscoehtic fluids. A temperature for-

mulation is also considered t h t can be applied for

nonisothermal analyses of fluid flow. Some illustrative examples of the application of the method
in polymer processing are also presented.

INTRODUCTION

of control volumes (i.e., elements of the total

volume). In fact there exist several commercially
available computer packages for injection mold
design with such misleading labels. This is an
unfortunate situation and it is time that the differences between these methods are clearly understood by everyone in the field of polymer technology. The term finite element should be used to
describe numerical techniques based on variational
or weighted residual principles (Schechter, 1967;
Finlayson, 1972).
The FEM was originally developed for stress
analysis in complex airframe structures during the
1960s. Because of its diversity and flexibility as an
analysis tool, it has since been extended and applied to the broad field of continuum mechanics.

he finite element method (FEM) is a numerical analysis technique for obtaining approximate solutions to a wide variety of engineering problems (Huebner & Thornton, 1982). Like
the better known finite difference method (FDM),
the FEM is used to solve the appropriate differential equations that describe these problems. The
term finite element analysis is sometimes used
incorrectly to describe macroscopic mass and
energy balances performed for a specified number

*Financial assistance from the Natural Sciences and Engineering

Research Council of Canada is gratefully acknowledged.

ADVANCES IN POLYMER TECHNOLOGY

107

FINITE ELEMENT METHOD FOR FLOW AND HEA T TRANSFER ANALYSIS

Zienkiewicz and Cheung (1965) showed that the
method is applicable to all field problems that can
be cast into variational form. Some field problems
that have been solved with the FEM include electrostatic fields, magnetostatics, seepage, torsion,
irrotational flow, heat conduction, electric conduction, diffusion flow in porous media, etc. Furthermore, Oden (1969a and 1969b) gave a generalized interpretation of finite element models and
showed how finite element equations can be developed from well-established global energy balances
without resorting to classical variational principles. All these contributions served to expand
the use of FEM for virtually any problem that can
be described by differential equations.
Since the early 1970s, the FEM has been applied
for the solution of fluid mechanics problems and,
in particular, to slow viscous flows (small Reynolds number) that are usually encountered in the
processing of polymer melts (Palit & Fenner,
1972; Taylor & Hood, 1973; Tanner, 1973;
Nickell, Tanner, & Caswell, 1974). Recent attempts to use finite elements for the study of flow
of viscoelastic fluids are summarized and critically
examined by Crochet and Walters (1983) and
Mitsoulis, Vlachopoulos, and Mirza (1983a).
The main advantage of FEM over other numerical methods (notably, finite difference) is its ability to solve problems in irregular and complex
geometries with unusual boundary conditions
(Vlachopoulos, 1977). Once the general differential conservation equations have been cast in their
equivalent finite element formulation, a computer
program can be written and used for different
situations with only minor changes. These may include the geometry, boundary conditions and
material properties that are particular to each
problem.
The main disadvantage of the method is its
complicated formulation that requires a good
understanding of variational or weighted residual
principles and a good command of matrix algebra
and computer programming. However, the development of user-friendly computer packages can
help people unfamiliar with the intricacies of the
method, to apply a powerful tool to different engineering problems and obtain useful solutions.

HOW THE METHOD WORKS

In this section we will focus our attention on
fluid flow, although the analysis is quite general
108

and can be applied to other fields as well.

Fluids flow within specified solid boundaries
(e.g., process equipment) or flow freely in the atmosphere in unconstrained flows with free surfaces, The flow field can be described by the wellknown conservation equations of mass, momentum, and energy. For two-dimensional planar domains and creeping flow (Re << l), these equations
take the form

while at the boundary surface we have

Tx= u,,i

+ 7.,j

(44

T, =

+uYj

(4b)

q = q,i + q,j

(44

with
= -p&.

(I..

11

+ 7..

(5)

11

where

(I,,, a,,,
bx,b,
-

7,,, 7yy,7xy=

stress components

= body forces acting on a

plane area A (see Fig. 1)

TMT,

= applied surface tractions on

T
qx,qy

= temperature
= applied heat

VXYYY

i ,j

6ij

some part r, of the contour

r enclosing A

input per unit

area on some part rsof r
= velocity components
= direction cosines of the unit
outward normal vector n to
the boundary r at any point
= Kronecker delta.

In the finite difference method (FDM) we proceed by writing the above equations as difference
equations for an array of grid points. Therefore in
the FDM we have a pointwise approximation
to the governing equation. In the FEM we divide
the domain A in subdomains (finite elements) (see
VOL. 4, NO. 2

FINITE ELEMENT METHOD FOR FLOW AND HEAT TRANSFER ANALYSIS

Fig. 2), and in every one of them we seek the problem solution, thus having a piecewise approximation to the governing equations. Such discretization of the domain requires a different approach for solving the differential equations.
Since we are interested now in an area rather than
a point, the differential equations must be recast
in an approximate integral form which is the basis
for the FEM formulation. Several methods are
available for reducing the differential conservation equations to suitable integral equations.
These include (Huebner & Thornton, 1982):
the direct approach, based on the direct stiffness method of structural analysis
the variational approach, based on the calculus
of variations
the weighted residuals approach, based on Galerkins method
the energy balance approach, based on the balance of thermal andlor mechanical energy of a
system
The overwhelming majority of finite element
formulations are based on the variational approach, especially in structural mechanics where
the method of virtual work has been the most
popular (Martin & Carey, 1973). In fluid mechanics the Galerkin method is usually used
(Huebner & Thornton, 1982; Taylor & Hood,
1973; Tanner, 1973; Nickell, Tanner, & Caswell,
1974). Both methods result in the same form of
discretized equations. The primitive variables are
the velocities v, (or u) and vy (or v) and pressure p
(u-v-p formulation, Taylor & Hood, 1973).

VELOCITY-PRESSURE FORMULATION

The virtual work method when applied to the

continuity and momentum Eqs, (1) and (2) gives
the following integral equations in Cartesian index
notation (Martin & Carey, 1973; Mitsoulis, 1984):

where oijis the total stress tensor given by Eq. (9,

T~~is the extra stress tensor, Eij is the rate-of-strain
tensor, vi is the velocity vector, 6vi is the virtual
velocity vector, 6p is the virtual pressure, and A is
ADVANCES IN POLYMER TECHNOLOGY

FIGURE 2
Twc-dimensional domain
divided into triangular ele-

sy
I

the area of the domain. On the portion of the

boundary I where velocities are specified the
virtual velocities vanish. The term virtual
applies to arbitrary changes of the field variables.
We have made use of the definition of the rateof-strain tensor iii

which for two-dimensional problems reduces to

the following vector
109

FINITE ELEMENT METHOD FOR FLOW AND HEAT TRANSFER ANALYSIS

A constitutive equation relates the stress vector

la) to (8

(01

= [DIG1 + (rol

(10)

where [D] is a constitutive matrix and (ro)a stress

vector that accounts for all possible stresses present
in the fluid that cannot be included in the product
[D]{;]. For a generalized Newtonian fluid we have

where <> denotes a row vector, [I denotes a column vector and N, are the interpolation (shape)
functions. These are related to the natural
(area) coordinates Li for a triangle by the expressions
N, = Ll(2L, - l), N4= 4LlL2
N2 = L2(2L2- l),

N,

= 4L2L3

N3 = L3(2L3 - I),

N6 = 4L3Ll

NP = Li .

(13a)

( 13b)

The natural coordinates Li are defined by

Ai
L. = A

(14)

It follows that
thus

where 7 is the viscosity. For a viscoelastic fluid (ro)

may contain several terms such as extra rates-ofstrain, stresses andlor their derivatives, integrals,
etc.
Equations (6) and (7) must now be discretized
by an appropriate choice of approximation functions for the primitive variables u, v, p. Several
options are available (Huebner & Thornton,
1982). Here we consider a six-node triangular element (see Fig. 3). The behavior of the unknown
field variables over each element is approximated
by continuous functions expressed in terms of the
nodal values of the field variables and sometimes
the nodal values of their derivatives up to a certain
order. The functions defined over each finite element are called interpolation functions or
shape functions. Mathematical considerations
require that the interpolation functions for velocity should be higher by one order than the interpolation functions for pressure. Thus, a quadratic
variation is chosen for velocities u and v and a
linear variation for pressure p at the vertex nodes.
Thus, we have
110

L1+L,+L3=1.
(15)
Therefore, only two of the area coordinates L,, L,
and L, are independent.
The relation between area coordinates Li and
global coordinates x and y at any point P is given

Inverting

(17)
The above choice of interpolation for u and v dictates a linear interpolation for the derivatives of
velocities. The rate-of-strain vector {i) is then
approximated by
I

where [B] is the rate-of-strain matrix.

VOL. 4, NO. 2

FINITE ELEMENT METHOD FOR FLOW AND HEAT TRANSFER ANALYSIS

Substitution of the approximations (12) and
(18) back to the integral Eqs. (6) and (7) and also
taking into account the constitutive relation (10)
gives a final matrix equation of the form

FIGURE 3
A six-node triangular element for u-v-p
formulation.

[KlnlXI, = [F),

or

where

TEMPERATURE FORMULATION
In the above, [K] is a symmetric stiffness matrix,
(XIis the vector of nodal unknowns and (FI a load
vector containing body forces, surface tractions,
forces due to extra stresses {701 and the boundary
conditions. The matrix integrations that appear in
Eqs. (20) are performed numerically. Equation
(19) is derived for the nfhelement. By assembling
the contributions from all the N elements and by
imposing the appropriate compatibility of the
nodal unknowns we derive the global stiffness
matrix and global load vector. The final set of
equations can be solved by a standard matrix solver (e.g., Choleski decomposition). For non-Newtonian problems, the nonlinear set of equations
must be solved by using some iterative technique
(e.g., Newton-Raphson). For a generalized Newtonian fluid, a direct substitution method is
recommended (Picard method) rather than the
Newton-Raphson method. The latter is superior
for fluids with a constant [D] matrix and a nonzero (701In Appendix A the reader may find an illustrative example of the application of the finite element method for the simple case of Poiseuille flow
between two flat plates. The differential equation
is cast in an integral form and the derivation of the
stiffness matrix and load vector is demonstrated
for a simple linear element.
ADVANCES IN POLYMER TECHNOLOGY

Using the virtual work method the differential

equation for conservation of energy (3) is written
in an integral form as

"1'

-7

+ pC,(GT)vi-

ax,

dA

(21)

where T is the temperature, 6T is the virtual temperature, k is the thermal conductivity, p is the
density, C, is the heat capacity, vi is the velocity
vector, Q is the rate of heat dissipated (Q = T:VT),
and q is the heat input per unit area on boundary
rq.Note that 6T = 0 on the part of the boundary
where T is specified.
The flow field can be discretized into three-node
triangular elements using linear interpolation
functions for temperature
T = NTT, + NTT,

+ NTT, = <NT>(T)

(22)

where <NT> is a row vector containing the shape

functions (linear functions of position) and [TI is a
column vector containing the nodal temperatures.
T o maintain grid and nodal compatibility between
the u-v-p and the T formulations and also improve
accuracy of the lower order of approximation for
temperatures, each six-node velocity-pressure tri111

FINITE ELEMENT METHOD FOR FLOW AND HEAT TRANSFER ANALYSIS

FIGURE 4
Subdivision of
a 6-node u-v-p
triangle into
four 3-node
linear Ttriangles.

angle can be subdivided into four temperature triangles (see Fig. 4).
The finite element equation for energy is obtained by substituting Eq. (22) into Eq. (21)
yielding for the nthelement (Huebner & Thornton,
1982)
([H,] + [C,IJ(T) + IQ,) = (0)

(23)

where [H,] is the thermal conductivity matrix, [C.]

is the heat capacity matrix, and IQn) is the heat
load vector. These are given by
dA

Qi= -SNiQdA
A

NiqdI.

(24a)

(2W

(Note that Cij is a nonsymmetric matrix.)

Following the usual assemblage procedure for
the contributions from all of the elements we obtain the matrix equation
[KTI~TJ= IFTI

(25)

where [KT]is the global heat stiffness matrix (nonsymmetric), IFT)is the global heat load vector containing boundary conditions and viscous dissipation, and (TI is the vector of the unknown nodal
temperatures. The linear system of equations can
112

then be solved by a standard matrix solver for

nonsymmetric matrices (e.g., Choleski decomposition).
For a complete nonisothermal analysis of flow
problems with temperature dependent properties,
the two formulations can be used in an intertwined
manner to update the flow and temperature fields.
Convergence of the overall systems of equations
cannot be guaranteed, in general. However, it has
been observed that in some well-defined polymer
processing problems, satisfactory convergence
was obtained after 2-3 iterations between the continuity-momentum (u-v-p) and the energy (T) formulations (changes in T less than 0.5C, Mitsoulis,
1984).
The major difficulty in the numerical solutions
of the energy equation arises from the presence of
the convective term. Past experience both with the
FDM and the FEM indicate that spatial oscillations of the solution appear when convection becomes important. The influence of convection in
heat transfer analysis is characterized by the Peclet
number defined by

where V is a characteristic velocity and H is a characteristic length. The Peclet number represents the
ratio of heat transferred by convection to the heat
transferred by conduction. For polymer melt processing the Peclet number will be high (in the order
of 102-105).
Oscillations are exhibited at moderate to high
Pe and are due to different mathematical characteristics of the first and second order operators
(i.e., viaT/axi and VT) involved in the energy
equation.
To suppress the undesirable oscillations, either
the grid density must be increased wherever rapid
temperature changes are expected or upwind finite
elements should be used (Mitsoulis, 1984; Stolle,
1982; Christie, Griffiths, Mitchell, & Zienkiewicz,
1976; Heinrich, Huyakorn, Zienkiewicz, &
Mitchell, 1977; Hughes, 1979). The method of
upwinding has also been used in finite differences (Roache, 1976). In the FEM method, upwinding consists of numerically integrating Eq.
(24b) at specially selected upwinding integration
points A, B, C instead of the usual GaussQuadrature integration points, a, b, c as illustrated in Figure 5 . The optimum distance of points
A, 3, C depends on the local Peclet number. The
method can be seen as a weight factor approach
VOL. 4. NO. 2

FINITE ELEMENT METHOD FOR FLOW AND HEAT TRANSFER ANALYSIS

usually employed for stabilizing a solution process. It should be noted that although upwinding
suppresses oscillations, it also decreases accuracy.
A remedy is, of course, a more refined grid at the
expense of computations. A more detailed study
of upwinding for triangular elements is given in
Mitsoulis, 1984.

FIGURE 5
Upwinding
scheme for triangular elements.

STREAM FUNCTION FORMULATION

In fluid mechanics the stream function $ and

the vorticity w are of great importance. For twodimensional flows these are defined by

It can easily be shown that

Poisson equation
v2*

+ and w satisfy the

= -w .

(29)

The above equation is valid for any two-dimensional flow of any fluid. Its usefulness lies in the
fact that enables us to visualize the flow field by
constructing streamlines from the stream function

*.

The same finite element formulation used for

temperatures (linear 3-node triangles) can also be
used for the stream function 4, by realizing that
the energy equation with no convection satisfies a
Poisson equation of the form
kV2T= -Q

(30)

where Q = ?:VV is the viscous dissipation term.

Comparison with Eq. (29) shows the equivalence
for k = 1, T = $, and Q = w . A solution of a flow
field (u-v-p formulation) gives w , and the stream
function can, therefore, be obtained a posteriori
from the FEM with the appropriate boundary
conditions.
FREE SURFACE FLOWS
Flows with free surfaces in polymer processing
are encountered whenever a fluid enters or exits
from a processing device (e.g., calendering) or
flows freely in the atmosphere after being processed in a confined region (exit flow from dies).
ADVANCES IN POLYMER TECHNOLOGY

These problems are particularly difficult to solve

since not only the flow characteristics (velocities,
pressure, temperature, etc.) are unknown but also
the location of the free surface that forms part of
the boundary is unknown. The only condition that
makes this class of problems solvable is the fact
that no fluid can flow through a free surface,
which is a streamline. Hence, at the free surface
- n-v=O
(3 1)

where is the unit normal vector to the surface

and the velocity vector.
The FEM is very attractive for free surface
problems because of the ability of the elements to
take the shape of the freely flowing fluid. The general procedure is to assume at first a location for
the free surface (usually an extension of the solid
boundary line that the fluid last encountered) and
solve the system of conservation equations for
these boundaries. Once the velocity field has been
found, a streamline can be constructed at the free
surface based on the velocities v, and v, found on
that surface and the equation that defines a
streamline, i.e.

Integration of Eq. (32) can be carried out (usually

numerically by applying Simpsons rule) to locate
a new surface h(x) which is given by

h(x) = h,

+ 0 LVvxd x

(33)

where h, is the initially assumed location of the

free surface and I its length in the x-direction.
113

FINITE ELEMENT METHOD FOR FLOW AND HEAT TRANSFER ANALYSIS

With the new boundary known, the system of
conservation equations is solved again in the new
domain. Recent velocities are used to compute a
new surface and the process is repeated until no
change in the location of the free surface is observed. Three to five iterations are usually sufficient for convergence if the initial free surface is
judiciously chosen.
The whole process can be seen as a series of
solutions for different problems, although we consider only one material and one process. Such calculations are very difficult and time consuming,
even with todays computers, and present many
problems because of computer memory requirements and other complexities imposed by the finite
element method.

expressed in terms of Deborah number (De),

Weissenberg number (Ws) or the stress ratio (SR),
.which is also referred to as recoverable shear.
These dimensionless quantities can be shown to be
identical under certain conditions. Here, we define
Deborah number

Calculations with viscoelastic fluid models are

performed iteratively by starting from inelastic
= 0).
constitutive relations (Le., Eq. (10) with kO)
The main difficulty encountered is the lack of convergence when the elastic effects are comparable
to the viscous effects. The elasticity level is usually

45O-tapered
entry flow of a
power-law fluid
through a slit
die.

.on0
I

8.000

7.000

6.000

5.000

4.000

3.000

2.000

---

-D

Weissenberg number Ws = A-

-w

A d - -

V
H

S, = W

(36)

27,

where (NJWis the first normal stress difference

and 7wthe shear stress, both evaluated at the wall.
Finite element calculations with viscoelastic
fluid models require very large computer times
and most of the results published in the literature
are of little practical significance. Recently,
Mitsoulis (1984) proposed a short-cut method for
viscoelastic fluid flow calculations. This method is
applicable to problems with a dominant flow

=
c

-=

*
-

----- --

-__)

_.)-

#
*

(35)

where V is a characteristic velocity and H a characteristic length, and

-__)

(34)

where A is a characteristic fluid time and 8 a characteristic process time,

Stress ratio
VISCOELASTIC FLUIDS

De = -

PS : xXy=5000

FINITE ELEMENT METHOD FOR FLOW AND HEAT TRANSFER ANALYSIS

direction (Le., flows through long dies). Mitsoulis
uses constitutive models that are valid in viscometric flow with experimentally measured shear
and elongational viscosities and normal stress coefficients. Successful simulations include the prediction of vortex growth in die entry flows and extrudate swell.

SOME FEM APPLICATIONS

IN POLYMER PROCESSING

The preceding analysis has been successfully

employed in a FEM computer program called
MACVIP (Mitsoulis, Vlachopoulos, & Mirza,
1983b) that was developed for the solution of twodimensional and axisymmetric heat and flow
problems of non-Newtonian inelastic and viscoelastic fluids in polymer processing. Problems that
have been solved include the slider bearing problem, driven cavity flows, tapered entry flow in a
die, entry flows in sudden contractions, exit flows
from slit and capillary dies, extrudate swell, and a
full two-dimensional analysis of calendering that
includes the determination of the melt bank free
surface. Some of the results are illustrated in
Figures 6-12.
Figure 6 presents the flow field for a

45"-tapered entry flow of a polystyrene melt

through a slit die and Figure 7 the streamline pattern for entry flow in a 1O:l sudden contraction.
Such analysis can be used for die design. Figure 8
shows the final finite element grid for the exit flow
problem from a slit die and the determination of
the extrudate swell for a polystyrene melt. Figures
9 and 10 show the velocity vectors and streamlines. Figures 11 and 12 show the streamlines and
isotherms in the melt bank for rigid PVC calendering. These figures are shown here merely to illustrate the type of results obtained by our finite element program. Detailed information on the
problems examined can be found elsewhere
(Mitsoulis, Vlachopoulos, & Mirza, 1984; and two
forthcoming articles).
From the foregoing discussion it is apparent
that in the application of the FEM, large matrices
are handled and very large numbers of arithmetic
operations are performed. Consequently, the computer memory and time requirements are large,
especially for nonlinear andlor free surface problems. Some typical CPU times are shown in Table
I, for McMaster's CDC 64001CYBER 170 (available memory 377 K). It should be pointed out that
the solution of the flow field in geometries such as
sudden contractions and sudden expansions is particularly difficult because of stress singularities
(mathematical discontinuities) at the entrance cor~

FIGURE 7
Streamline pattern for entry

POLYSTYR E N E : D e =5.47 (t:507~-')

1.0

-.8

-.I

planar sudden
fiowina10:l
contraction of
a polystyrene
melt.

.8
-6

.4

.2
.O

-.Z
-.I
-a6

-.8

-1.0

-2.0

-1.6

-1.2

.O

.4

.8

1.2

1.6

2 .o

X-COORDINRTE* 1 CM I
ADVANCES IN POLYMER TECHNOLOGY

115

FINITE ELEMENT METHOD FOR FLOW AND HEAT TRANSFER ANAL YSIS
FIGURE 8
Final finite element grid for
the exit flow
from a slit die
and the determination of extrudate swell
for a polystyrene melt.

3 .O
2 .O

1.0
I
0.0

-1.0
-2 .O

-3 .O

-2.0

-1.0

0.0

1.0

2.0

3.0

4.0

5.0

I
6.0

7.0

8.0

OISTFlNCE I X/H
FIGURE 9
Velocity vectors for a polystyrene melt
exiting from a
slit die.

3 00

PS
De=2.7

2 .O

'162.7

3 .O

4 -0

1.0

: 0.0
-1.0

SLIT

-2 a 0

-3 m O

116

-3 .O

-2 .O

-1.0

DIE
0 00

1.0
DISTFINCEP X/H

2 00

5 .O

VOL. 4, NO.2

FINITE ELEMENT METHOD FOR FLOW AND HEAT TRANSFER ANALYSIS

3 .O
2 .o

0.0

1.0

2.0

FIGURE 10
Streamline pattern for a polystyrene melt
exiting from a
slit die.

PS
De=2.7

1.0

-2

.o

+
lJ0
:
I

-3.0
-2.0

-1.0

3.0
4.0
5.0
DISTFlNCE I X/ H

6.0

7.0

8.0

FIGURE 11
Streamline pattern in the melt
bank for PVC
calendering
(from Mitsoulis, Vlachopoulos, Mirza,
Calendering
Analysis).

1.0
.I

.C
.4

.z
.O

-.2

8u

-.4

ROLL 3

-.c

R i g i d PVC

-.I

-1.0
-1.2

FIGURE 12
Isotherms in
the melt bank
for PVC calendering (from
Mitsoulis,
Vlachopoulos,
Mirza, Calendering Analysis).

1.0

.I

.c
I

.g
c

P
..

.z
.o
-.I
.I

-.c
-.I
-1.0

-1.2
-5.7

-5.5-5.3-5.1-~.Y-~.7-~.5-~.1-~.1-1.Y-1.7-~.5-1.1-1.1-2.-2.7-2.1-2.1-2.1-1.Y-1.7-1.5-1.~-1.1

X-COOROIWTEI

ADVANCES IN POLYMER TECHNOLOGY

-.

-.7

-.5

-.I

-.I

.I

.I

.5

CM

117

FINITE ELEMENT METHOD FOR FLOW AND HEAT TRANSFER ANALYSIS

for banded nonsymmetric systems are inappropriate for such large numbers of nodal variables.
The frontal method of solution (Irons, 1970;
Hood, 1976) with diagonal pivoting is then applied
which leads to a drastic reduction of core storage.
Nevertheless, while memory requirements can
thus be overcome, computer run time remains
very high.

ner to the die and at the die exit. A large number

of elements must be concentrated in these areas to
better capture the irregular behavior. The inclusion of viscoelasticity through a standard constitutive equation requires numerical differentiation of
velocities and velocity gradients which may introduce additional errors. T o avoid this, some researchers, notably M. J. Crochet of Louvain la
Neuve, Belgium, and R. A. Brown and R. C.
Armstrong of MIT, use the stresses as primary
variables along with velocities and pressure. This
amounts to six unknowns per node for the twodimensional (u, v, p, 7,,, 7yy, 7J and seven unknowns per node for the axisymmetric problem
(u, w p, T,,, T ~ T,, 7& The total number of nodal
variables can therefore become very large and
sometimes prohibitive for engineering calculations. For example, Crochets results for the extrudate swell of a viscoelastic model fluid called
Oldroyd-B were obtained with a grid consisting of
75 quadrilateral elements, 357 nodes, and a total
number of 1889 variables for a slit die and 2246
variables for a capillary die. The total running
time was reported to be 50 CPU hours on a
PRIME 750 computer. Brown and Armstrong
have used up to 7000 variables with commensurate
time and memory requirements. Standard solvers

CONCLUDING REMARKS
While there are still many problems to be
ironed out, the finite element method is a very
powerful tool for solving flow and heat transfer
problems. It appears to be the only rational
method of solution for problems involving complicated geometrical boundaries and complex
interactions of flow and temperature fields. The
development of user-friendly computer packages
will make the method accessible to persons that
have a limited background in variational calculus
or matrix algebra. The wide use of such a
powerful tool will enhance our understanding of
polymer flow, mixing and melting and may spearhead new dramatic developments in polymer processing.

TABLE I
Typical Requirements for Problems Solved by the FEM (MACVIP Program) Using a CDC 64OO/CYBER 170
Problem

Fluid

Tapered Entry Flow

Inelastic
Power-Law
Isothermal

No

64

153

317

260

70

0.15

Inelastic
Power-Law
Isothermal

No

200

459

881

350

200

0.4

Viscoelastic
Isothermal

No

200

459

88 1

350

200

0.3.

Exit flowExtrudate Swell

Viscoelastic
Isothermal

Yes

184

423

858

360

225

0.5*

Calendering of Rigid PVC

Inelastic
Power-Law
Non-isothermal
Slip at wall

Yes

118

28 1

588

345

120

2.5

Sudden Entry Flow

Free
Surface

Number
of
Elements

Number
of
Nodes

Number
of
Unknowns

Computer
Memory
Required
(k)

Run
Time per
Iteration
(5)

Overall
Run
Time
(hr)

*For each elasticity level (Le., for each Deborah number).

118

VOL. 4, NO. 2

FINITE ELEMENT METHOD FOR FLOW AND HEA T TRANSFER ANALYSIS

REFERENCES
Christie, D., Griffiths, D. F.. Mitchell, A. R., and Zienkiewicz. 0.C.
1976. Finite Element Methods for Second Order Differential Equations with Significant First Derivatives, Int. 1. Numer. Methods Eng.
10: 1389.
Crochet, M. J. and Walters, K. 1983. Numerical Methods in Nan-Newtonian Fluid Mechanics, Ann. Rev. Fluid Mefh. IS: 241.
Finlayson, B. A. 1972. The Method of Weighted Reciduals and Variational Principles, New York: Academic Press.
Heinrich, J. C., Huyakorn, P. S., Zienkiewicz, 0. C., and Mitchell,
A. R. 1977. An Upwind Finite Element Scheme for Two-Dimensional
Convective Transport Equation. Int. J. Numer. Methods Eng. 11:
131.
Hood, P. 1976. Frontal Solution Program for Unsymmetric Matrices,
Int. J . Numer. Methods Eng. I 0 329.
Huebner, K. H., and Thornton, E. A. 1982. The Finite Element Method
f o r Engineers, Second Ed., New York: Wiley.
Hughes, T. J. R. 1979. A Simple Scheme for Developing Upwind Finite
Elements, Int. 1.Numer. Methods Eng. 1 2 1359.
Irons, D. M. 1970. A Frontal Solution Program for Finite Element
Analysis, Int. J. Numer. Methods Eng. 2: 5 .
Martin, H. C. and Carey, G. F. 1973. Introduction to Finite Element
Analysis, New York: McGraw-Hill.
Mitsoulis, E. 1984. Finite Element Analysis of Two-Dimensional PolymerMelt Flows, Ph.D. Thesis, Dept. Chem. Eng., McMaster Univ.,
Hamilton. Ontario. Canada.
Mitsoulis, E.. Vlachopoulos, J., and Mirza, F. A. 1983a. Finite Element
Analysis of Two-Dimensional Polymer Melt Flows, Po/ym. Proc.
Eng. I : 281.
Mitsoulis, E., Vlachopoulos, J., and Mirza, F. A. 1983b. MACVIPA Finite Element Program f o r Creeping Viscoelatic Flows, Internal
Report. Faculty of Engineering, McMaster Univ., Hamilton, Ontario,
Canada.

From variational calculus (see for example

Schechter, 1967) it is well known that this problem
is equivalent to finding the function u(y) that minimizes the functional

APPENDIX A

Poiseuille Flow by FEM

In this Appendix we illustrate the application of
the FEM for the case of Poiseuille flow of a Newtonian fluid between two flat plates. Our objective
is to demonstrate how a variational principle is
used and how the stiffness matrix and the load
vector are generated.
For the analysis we consider a simple linear element, shown in Figure A1 and designate ends 1
and 2 as nodes.
The governing differential equation is

-d_p dx

Mitsoulis, E., Vlachopoulos, J., and Mirza, F. A. 1984. Numerical

Simulation of Entry and Exit Flows in Slit Dies, to appear in Polym.
Eng. Sci. 24.
Mitsoulis, E., Vlachopoulos. J., and Mirza, F. A. A Numerical Study of
the Effect of Normal Stresses and Elongational Viscosity on Entry
Vortex Growth and Extrudate Swell, to appear in Polym. Eng. Sci.
Mitsoulis. E., Vlachopoulos, J., and Mirza, F. A. Calendering Analysis
without the Lubrication Approximation, to appear in Polym. Eng.
Sci.
Nickell, R. E.. Tanner, R. I., and Caswell, B. 1974. The Solution of Viscous Incompressible Jet and Free-Surface Flows Using Finite Element
Methods. J . Fluid Mech. 45: 189.
Oden, J. T. 1969a. A General Theory of Finite Elements. I: Topological
Considerations. Int. 1. Numer. Methods Eng. I : 2.
Oden, J. T. 1969b. A General Theory of Finite Elements. 11: Applications, Int. J. Numer. Mefhods Eng. I : 3.
Palit, K. and Fenner, R. T. 1972. Finite Element Analysis of TwoDimensional Slow Nan-Newtonian Flows, AIChE f 8 6.
Roache, P. J. 1976. Computational Fluid Dynamics, Albuquerque, NM:
Hermosa Publishers.
Schechter, R. S. 1967. The Variational Method in Engineering, New
York: McGraw-Hill.
Stolle, D. F. E. 1982. Finite Element Modelling of Creep and Instability of Large Ice Masses, Ph.D. Thesis, Dept. Civil Eng., McMaster
Univ., Hamilton, Ontario. Canada.
Tanner. R. I . 1973. Die-Swell Reconsidered: Some Numerical Solutions
using a Finite Element Program, Appl. Polym. Symp. 20: 201.
Taylor, C. and Hood, P. 1973. A Numerical Solution of the NavierStokes Equations Using the Finite Element Technique, Computers
Fluids I : I .
Vlachopoulos, J. 1977. Should You Use Finite Difference or Finite Element Methods. S.P.E. 35th ANTEC, Montreal, Tech. Papers, Val.
23, 519.
Zienkiewicz, 0. C.. and Cheung, Y. K. 1%5. Finite Elements in the
Solution of Field Problems, Engineer 220.

We shall ignore the fact that this problem has an

exact solution and proceed to find an approximate solution. We approximate the velocity u(y)
by the following quadratic polynomial
FIGURE A1
Single linear
element for
flow between

d2u

dyz

where dpldx = p is assumed uniform across the

element length L. The nodal values of u(y) are
u(0) = u,

(A21

u(L) = u2

(A31

and are called nodal unknowns or degrees of

freedom. We want to find the function of u(y)
that satisfies Eq. (Al).
ADVANCES IN POLYMER TECHNOLOGY

LL

0
.-

1
119

FINITE ELEMENT METHOD FOR FLOW AND HEAT TRANSFER ANALYSIS

u = a + by + cy2.

(AS)

Since we have three unknowns-a, b, c-and two

degrees of freedom-u, and u2-we also need to
introduce another equation. This is done by differentiating Eq. (AS)

Q = -p
2

4
~ j . ; -u;
3L

4
4
+ -u:
+ -L+;
+
3L
3

+ 2f1u2 - 2f,u, - 2 3 L - -j-UIU2

8
8

- -j-r1u2 + +,ul]

Differentiation yields
We find now a , b and c in terms of the nodal unknowns
(A74
(A7b)
UI

- j.1L)/L2 ( A ~ c )

where we have introduced an extra degree of freedom f l , the unknown velocity gradient. Equations
(AS) and (A6) are then written in terms of the
nodal unknowns as

u = u I + j.y + 1/Lz(uZ- uI - j.,L)y2

+ = du/dy =

+ 21LZ(u2- uI - j.,L)y

(A8)

ar,

+ 2u, - 2u, - 4Lf,

(A9)

Noting that in Eq. (A4) one of the terms is associated with viscosity while the other with the pressure gradient we may write
J=Q+W

++8L +

---,[2Lf1
a Q - 21

Equation (A17) can be written in a matrix form as

(A101

where

and
or in shorthandnotation

To minimize the functional we must have

or

where xi stands for u,, u2 and

Substitution of Eq. (AS) in (A1 1) yields
Q = l ~ LP ~ [2 ~ ~ + - j~. l L( ) ~
U] z~d ~-. (A151
- ~ ~
After the appropriate mathematical manipulations
we obtain
120

(Fi) = [KIIxi) -

The matrix [K] is a symmetric matrix and is

usually called by historical default the stiffness
matrix, due to its original derivation in structural
mechanics. The vector (xi) is the vector of unknown nodal variables for which we seek the solution.
Similarly, we can substitute Eq. (A8) into Eq.
(A12) t o obtain

FINITE ELEMENT METHOD FOR FLOW AND HEAT TRANSFER ANALYSIS

After the appropriate mathematical manipulations
we obtain
1

W=pL

For this simple case we can proceed and solve

explicitly Eq. (A24). A further simplification is
obtained if we consider the kinematic boundary
condition at the wall, i.e.,
uz = 0

By differentiating we have

This eliminates the second row and column in Eq.

(A24) which now becomes

2
-pL
3
1
-p L
3

-Fi

Z
1E

1
-pL2
6

or

(A251

[:
-

=-plLi\
and finally

pLz
2p

=--=---

L2 dp
2p dx

i ,= o

The vector IFi) is then called load vector, again

from its original structural mechanics derivation.
Thus we have

(stiffness matrix)

(A27a)
(A27b)

The above derivation has been obtained by

using a single linear element with two nodes 1 and
2. It is obvious that more elements can be put together in various ways. In all cases a stiffness matrix and a load vector must be derived for each element. A standard procedure for assembling all the
elements together can then be applied that takes
care of nodal compatibility. The kinematic boundary conditions and/ or external forces must also
be employed. The final assemblage leads to a
global stiffness matrix and a global load
vector that can be solved by a standard matrix
solver.
It is interesting to note that for this simple problem the exact solution of the differential equation
(All is

(load vector)

and for the midplane of the channel of width 2L

we have

or in shorthand notation
[KlIxiJ = (FJ.

(A24b)

Equation (A24) is the standard finite element

form that all formulations lead to (i.e.y a matrix
equation that includes a stiffness matrix and a
load vector).

ADVANCES IN POLYMER TECHNOLOGY

u
I

L2 dp
2p dx

=---

(A29a)

f, = o

(A29b)

(i.e., the two solutions are identical).

121