Sie sind auf Seite 1von 10

M. L. HAM1 and J. F. T.

PIITMAN

Department of Chemical Engineering


University College of Swansea
University of Wales
Swansea SA2 8PP, Wales, United Kingdom
Slow developed flow in an extruder channel has been set up
as a two-dimensional, variational problem using a helical coordinate system, thus avoiding the usual geometrical simplification. Continuity is enforced by an integral form of constraint,
and solutions for isothermal, Newtonian flow are obtained by a
finite element method for both shallow and deep, highly
curved channels. The performance ofthe solution procedure as
a function of Lagrangian multiplier is discussed. Convergence
to correct solutions is demonstrated for the shallow channel
case. Deep channel results are compared with analytic predictions, curvature corrected according to Booy. Further testing of
deep channel results will be made against experimental data.

INTRODUCTION
here has been a substantial amount of work on the
T a n a l y s i s and mathematical modelling of flow in screw
pumps (extruders); the characteristics of most approaches are summarized in Table 1 of Ref. (1).Models
which have most closely approached the real situation
have necessarily involved numerical solution of the governing differential equations. Pre-eminent in this category is the work of Griffith (2), Pearson, et al. (3)and
Fenner (4), who used finite difference methods and a
Cartesian coordinate system with the unwrapped channel approximation. The main features of the mathematical models used in these works have been summarized
(1)and it is not necessary to repeat them here. More
recently the work of Fenner and Palit (5-7) provides an
instance where the finite element method has been
used; the unwrapped channel approximation was,
however, retained. Solutions were obtained for twodimensional developed isothermal flow of power law
fluids. Non-isothermal solutions (5)were hampered by
the lack, at that time, of a true finite element analogue of
the finite difference technique of up-wind differencing,
which is required to deal with convective heat transfer at
higher Peclet numbers.
The present work follows on from Ref. (l),where a
variational analysis was set u p in helical coordinates,
and it differs from previous approaches in combining the
following three features:
The true geometry of the channel is taken into
consideration.
The finite element method is used.
Computed results are compared in detail with data
from carefully controlled experiments.
POLYMER ENGlNEERlNG AND SCIENCE, MARCH, 1980, Vol. 20, No. 5

ing
The case for the finite element method in mot
polymer processing has been made elsewhere (1,7, lox
and Pearson (11)has suggested that it might be used in
what we could describe as the ultimate approach to the
plasticating extruder problem. The case extends to all
situations where the geometry of the problem is one of
the parameters which we wish to vary, and where the
domain of the problem may not be a simple shape.
Although comparisons are hard to come by, it seems that
modern finite element techniques can be more efficient
in terms of computer usage than finite difference techniques. Certainly in terms of the sum of human-plusmachine effort, the ease with which a finite element
program can be applied to a whole range of problems
gives important benefits. A proposal to use finite elements in the present context was made in Ref. (1)and
the work of Pallit and Fenner represents the only example of its application to date in the extrusion problem.
Conclusive comparisons between numerical solutions
and experimental results are not abundant in this field.
The available data on extruder performance can be divided into two categories; those obtained under practical conditions and those from idealized experiments. In
the practical category, data are typically measured on
highly instrumented extruders pumping molten polymer (and probably also conveying and melting). Conditions are far removed from isothermal: flow is never fully
developed, and the fluid is highly non-Newtonian, with
considerable doubt existing as to the appropriateness of
any particular constitutive equation in this nonviscometric flow field. The possibility of wall slip exists
(12) and the flight may strip melt from the barrel (13).
Accurate data on physical properties and boundary con339

M . L. Hami and J. F . T. Pittman


ditions may not be available. In short, the practical
situation may depart in so many respects from even the
most sophisticated of existing models that it is very
dimcult to pinpoint the reasons for the divergence of
experimental and computed results.
The only previous work where channel curvature is
taken into account (other than by approximate corrections) is the treatment by Zamodits, using helical coordinates (8). Isothermal, one-dimensional developed
flow was considered, making the wide channel assumption so that velocities depended only on radial position.
Results were obtained for power law fluids of various
indices and for various helix angles. These illustrated
interesting interactions between power law index and
channel curvature in determining throughput. The results, however, were not tested against experimental
data, nor was the work extended to the two-dimensional
problem including effects of the flight. It is true that
curvature effects in the metering section of a plastics
extruder are probably often small. However, the size of
the effects can at present be estimated only by using
an analysis, e. g., Ref. (9), which is itself approximate.
Deeper, more highly curved, channels occur in the
conveying and melting zones of plasticating extruders.
Flow in the melt pool should properly be modelled
taking into account channel curvature. It may be that a
flow model could be extended to deal with solids conveying (of powders, for example) by incorporating an appropriate constitutive equation and, possibly, wall slip. In
this case again, curvature may be more important. Finally, one could note that deep channel screws are used
other than in the plastics industry; in rubber and food
processing, for example. Given that there is no intrinsic
difficulty in modelling curvature realistically, it seems
worthwhile to do so, and avoid the restriction to shallow
channels inherent in most of the previous works. We
have used a helical coordinate system, though different
from that employed by Zamodits, which makes it possible to express steady developed flow in the helical channel as a two-dimensional problem.
In idealized experiments, on the other hand, conditions are chosen to match the main assumptions of the
theoretical model as closely as possible. In the present
work we adopt this approach, and now outline briefly
the underlying philosophy. Agreement between
idealized experiments and theory gives confidence in
the mathematical model at its present level of sophistication. Development proceeds by introducing elaborations into both experiments and model in parallel, and
comparisons can then indicate with more certainty what
phenomena need to be included in the model. The
experiments involved in this sort of approach will, at
least in the early stages, be far removed from the real
problem and may receive criticism on that ground, but
such criticism would be short-sighted.
The practical end-point to this kind of approach may
be set by various factors-available numerical techniques, computing costs, uncertainties over constitutive
equations, etc. At this point, but not before, the model
should be adjusted empirically in an attempt to account
for the remaining discrepancies. The predictive power
340

of the model will, though, be related to how far its


rigorous development has been carried. At this stage,
the model could in principle be incorporated into a
computer.-aided optimal design procedure such as that
described by Helmy and Parnaby (14). However, the
computing costs, using a sophisticated model, would be
very substantial unless the parameter range were already closely defined on the basis of experience. But
even before reaching this point, which is concerned with
quantitative prediction of machine performance, a good
model should be able to indicate important trends in the
relationshp between screw design and performance.
Numerical experiments can point the way to key physical experiments.
In the present work, we set out along the road
signposted above, and in this paper we recapitulate the
variational formulation of the flow and heat-transfer
problem in helical coordinates (I), describe a finite element procedure, and discuss its performance. Solutions
for isothermal Newtonian flow in both shallow and deep
channels are presented and compared with analytic results where applicable.
In subsequent work, experimental data will be compared with computations, and both computation and
experiment extended to purely viscous non-Newtonian
flow.

THE VARIATIONAL FORMULATION AND THE


HELICAL COORDINATE SYSTEM
In an earlier paper, a variational analysis of developed
flow in a helical screw channel was given (1).A functional
based on the General Evolution Criterion (15)was derived in vector-tensor form, and then expressed in a
helical coordinate system. The functional is applicable
to the incompressible, non-isothermal flow of a fluid
with viscosity dependent upon temperature and instantaneous rate of deformation (a generalized Newtonian
fluid). Here we give two separate functionals corresponding to the momentum and conservation equations,
since in practice these would be used iteratively in a
solution of the non-isothermal flow problem.

+ -u . V (P + p a ) + po
r

A:A
dV (1)

k
- (VT) 2

po

(%)TI dV

(2)

Subscripts zero indicate values corresponding to the


extremalizing velocity and temperature fields. The use
of these terms distinguishes the present approach from
the classical variational formulation, in which it is not
possible to allow for inertia terms (16) and a general
shear-dependent viscosity. Non-subscripted velocities
and temperatures are subject to variation. but subscripted terms are not. Bearing this in mind, it is easily
shown (17) that the Euler-Lagrange equations corresponding to extremalization of Y, with respect to inPOLYMER ENGINEERING AND SCIENCE, MARCH, 1980, Vol. 20, No. 5

Finite Element Solutions for Flow in a Single Screw Extruder, Including Curvature Effects
compressible velocity fields are indeed the momentum
conservation equations for a generalized Newtonian
fluid. /I.,, is the local viscosity value, and for nonisothermal flow provides coupling with the energy functional. The energy functional is a modification of that
derived previously (1)and is more convenient in that the
field variable is in the denominators. It is identical to the
form used by Palit (5), and again it is easy to show (17)
that the Euler Lagrange equations give the correct
energy-conservation relationship. The functional does
not include provision for stored elastic energy, consistent with the assumption of a purely viscous flow.
The helical coordinate system which has been used is
as follows:
(1)Radius, r , measured from the screw axis.
(2) Helical distance, s, measured along the intersection of a cylinder of radius r and a helical reference
surface fixed relative to the barrel, coaxial with it, and
with the same pitch and sense as the screw.
(3) Axial distance, t, measured from the reference
helix.
The coordinates are dimensionless, having been
scaled by the barrel internal radius R. The helical reference surface is thus the coordinate plane t = 0. s = O is a
plane passing through the axis and fixed relative to the
barrel. The system is non-orthogonal and is illustrated in
Fig. 1 . It differs from that previously used by Zamodits
(8) for the extruder problem, and from that proposed by
Tung and Lawrence (18) for the treatment of flow in
static mixers, etc. The latter states that the coordinate
transformations given in our previous paper (1)do not
give the coordinate curves we describe, that our vector
component transformations do not follow the rules for
contravariant components, and that our continuity equation is incorrect. We are therefore obliged to go into
some detail to clear up this misunderstanding. In their
paper, Tung and Lawrence review some standard material on general curvilinear coordinate systems, and go on
to derive equations of motion and continuity using the
general laws for contravariant component transformations. Their remarks about our work show that they have
assumed that we are also using contravariant components. In fact, this is not the case; our system is mixed.
Ours base vector is tangential to the intersection of the r
= constant and t = constant surfaces, and is thus a
natural base normalized to unit magnitude; likewise the
t-direction base, which is tangential to the intersection
of r = constant and s = constant. The r components,
however, are not referred to a unit vector tangential to
the intersection ofs = constant andt = constant. We use
instead a base orthogonal to the s and t bases just mentioned. It is thus a reciprocal rather than a natural base
(see Fig. 1).An attempt to derive our component transformations or continuity equation assuming a system of
natural base vectors, as Tung and Lawrence appear to
have done, will inevitably fail. In fact, we avoided their
approach and derived our results by the simple and
direct method which is given in the Appendix. We
denote dimensional velocity components by u:, u i , u l ,
and scale these by 2 rrRN to obtain dimensionless components u T ,us,ut.
POLYMER ENGINEERING AND SCIENCE, MARCH, 1980, Vol. 20, No. 5

axial
pitch

r, s, t.

Fig. 1 . The helical coordiiiaie sysieni. The screw is shown at the


instant during its reuoluiiori when the reference helix coiiicides
with the center of t h e j i g h t .

Referring back now to E q 1 and 2, we note that the


domain of the volume integrations is fixed relative to the
screw and is taken as one pitch length of the channel
bounded by: (a) the inner surface of the barrel; (b) the
surfaces of screw and flight; (c) the helical surface running through the center of the flight clearance, which
coincides with t = 0 for an instant on every revolution of
the screw; and (d)axial surfaces cutting across the channel at the beginning and end of the pitch, which coincide
with s = 0, once every revolution.
The geometry of the screw is (partially)defined by the
following dimensionless constants
radius of screw shaft
k, = internal barrel radius, R
(3)

kz = radius of the flight tip/R


k3 = flight width measured parallel
k4 = screw pitch measured parallel

to the axis/R
to the axid2rrR

In the present work we consider developed flow, that


is, there is no dependence of velocity (or temperature)
34 1

M . L. Hami and J. F. T . Pittman


upon s. We also make the usual assumption of negligible
inertia forces, and deal here with isothermal flow. Thus
Y,, E9 1 , less the first term of the integrand, is to be
expressed in the helical system, and since there is no
dependence on s, an integration with respect tos can be
carried out directly. The result (1) is (dropping the subscript on Y):

u, = us = ut = 0

k { D } [L,]{Dq)drdt (4)
P*

The domain of integration is now the axial channel


cross-section (see Fig. 2 ) . The dimensionless groups are
as follows:
F = 47T"R 3 N 2 p*', a characteristic rate of viscous heat
generation, N is screw speed, P* is a reference viscosity
(constant).
N p , = 4 R2Pc/p* U , Poiseuille number, P G is the axial
pressure gradient, U is a characteristic flow velocity
magnitude.
NTh = U I ~ T R N ,Thompson number. poIp* is the
ratio of local to reference viscosity.
The row matrix { D } is (1 x 9) and contains the three
dimensionless velocity components and their spatial derivations. Row matrix { M } , (1 X 9), involves only
geometrical factors together with r, s andt, as does the (9
x 9) matrix [ L , ] . Details of these matrices and the
manipulations leading to Eq. 3 are given in Ref. (17).
Note that throughout we indicate a row matrix by, for
example, { D } , and the corresponding column by (03.
The treatment of the term involving pressure inEqs 1
and 4 requires comment. For incompressible flows generally, pressure can be removed from the variational
problem, since

1 u. +
V(P

p a ) dV =

gradient, P G . No other information on the pressure field


is needed. In the present problem, the duct has a less
simple shape than that illustrated in Fig. 3, but the same
ideas apply and Eq. 4 consequently involves only the
axial pressure gradient, rather than pressure as a field
variable.
The boundary conditions used with Eq 4 are as follows:
(a) No slip at the barrel:

V u(P + p a ) dV

For the case where velocity boundary values are given


over the whole of the surface, S, we see that the term
involving pressure is not subject to variation and can be
dropped from the functional. Alternatively, the situation for any developed flow can be illustrated as in Fig. 3 .
The velocity fields over surfaces S, and Sz are identical.
The pressure difference, PJ, between any pair of corresponding points is independent of position on the
surfaces. Thus the contribution to the integral from
surfaces S , and S2 depends, for a given duct length,
only on the velocity field and the downstream pressure

of the decelopedjlow prvhleni i s a n uxial


cross-section of the screu: channel shown in hold outline.

(b) No slip at the surface of the screw, including the


flight:
ur= 0
us = -

ut =

d k4' + 1-2

k4

These components correspond to a solid body rotation


about the axis with speed N , and in the positive direction.
(c) Velocities in the flight clearanc'e are obtained from
a solution for flow in the narrow annulus between infinite, co-axial cylinders. Details of this are available
elsewhere (17).Velocities are given as afunction ofr and
depend on fluid viscosity, screw speed, flight clearance,
and axial pressure drop across the flight. Calculation of
the latter requires some assumption about pressure distribution in the channel, in addition to a value for overall
axial pressure gradient. We have assumed that isobaric
surfaces in the channel are helical and orthogonal to the
flight. This is known not to be true, but the approximation introduced is unimportant, as will be discussed in a
subsequent paper. It is important to recall that this
assumption affects only the pressure-driven leakage
flow, which is generally negligible. In all other respects,
as mentioned earlier, the solution is independent of'the
pressure distribution over the channel cross-section.
Numerical solutions for the velocity components ur(r,
t), us(r,t) and ut(r, t) are found using the finite element
method to extremalize E q 4 , subject to the boundary
conditions and the incompressibility condition. Flow
rate through the channel is found by integrating the
circumferential component of the resulting velocity field
over the axial channel cross-section, having first subtracted a term to allow for the fact that the domain of
integration rotates with the screw (17). The result is:

Q~~ =

!,I,2 T ~ N3 [

drdt

(6)

Fig.2 . The domuin

342

Fig. 3. Illustration vf a simple, developed duct flow.


POLYMER ENGINEERING AND SCIENCE, MARCH, 1980, Vol. 20, No. 5

Finite Element Solutions f o r Flow in a Single Screw Extruder, Including Curvature Effects
To get the throughput of the machine, one has to correct
this channel flow rate by the amount of leakage over the
flight. Expressions are given elsewhere (17)in terms of
the helical system for the drag and pressure-driven
components of the leakage flow. However, since the
'clearance is generally very small, leakage can be
adequately calculated for the isothermal Newtonian case
using the analysis by Mohr and Mallouk (19) which
neglects curvature. This is also true for the simplest form
of non-Newtonian isothermal flow.
Results will be presented later in terms ofdimensionless flow rate and axial pressure gradient, as used by
Fenner (4).These were originally defined in terms of a
Cartesian representation and are interpreted here as
follows:

rIQ=

Q
2mRNWH

COS'

&,

(7)

where W is channel width measured parallel to the axis,


and H is channel depth. &, is the helix angle at the barrel
internal radius.

where ;i = jZ 7 is a characteristic shear stress obtained


from

Y=

2mRN cos
H

~$b

and jZ is viscosity at the shear rate 7.

THE FINITE ELEMENT FORMULATION


Briefly, the options in a finite element approach to
incompressible flow are: the pressure-velocity formulation, the stream function formulation, or a constrained
extremalization.
The first of these, the momentum and continuity equations, are treated simultaneously using the Galerkin
method (20, 21). The nodal variables are the velocity
components and pressure. As already indicated, we do
not wish to increase the unknowns of the problem by
introducing pressure as a field variable, so t h e
pressure-velocity approach has not been followed.
The use of a stream function raises to two the order of
derivatives in the functional. An early example was the
work ofAtkinson, et al. (22),who used a non-conforming
cubic triangular element. The need forCl continuity, or
a proven non-conforming element, probably necessitates the use of higher order shape functions. Palit and
Fenner (6, 7 )used the stream function formulation with
a non-conforming quadratic triangular element, which
gave good results, but it has not been subjected to the
patch test (23-25)nor has its behavior as a function of
mesh design been discussed.
Continuity can be enforced by applying a constraint to
the extremalization of, e.g., E q 1 , in either differential
or integral form. Incorporation of the differential constraint V . u = 0 into the functional introduces a
position-dependent Lagrangian multiplier, which can
be identified as pressure (26). In the present work, we
POLYMER ENGINEERING AND SCIENCE, MARCH, 1980, Vol. 20, No. 5

have chosen to use the integral form (27)introduced into


the functional of E q 1 with a constant multiplier,

YM = Y M + a/v(V.U),dV

(9)

YM is now to be extremalized with respect to all velocity

fields satisfying the boundary conditions. After addition


of the constraint term to 4, expressing it in helical coordinates and integrating with respect to s, we get

+ 2 P* z {D} [L,] {D?)

drdt

(10)

where the (9 x 9 ) matrix [L,] is sparse and is a function of


r only. The advantage of the integral constraint approach
is that the nodal variables remain simply the three velocity components, keeping down the size of the problem
and allowing the use of simple elements. Against this
must be balanced the need to arrive at an appropriate
value of the multiplier, a.
The multiplier can be identified as (28)

a=

variational derivative of YM
variational derivative

/ (V -

(11)

u)'dV

Since the constraint is itself a minimum condition, the


denominator goes to zero as the constraint is satisfied.
The numerator, however, remains finite at the constrained extremum, so analytically V * u + 0 as a + m.
An element contribution to the constraint term, however, involves the product of a with the discretized
version of V . u , and so may in principle remain finite in
the limit. ThTs product can in fact be identified as
pressure. In practice, due to approximations in the
numerical solution, V * u will not go identically to zero,
and the constraint termwill eventually dominate as a
becomes very large (c.f., Ref. (29)).We may, however,
expect to find a range of a values for which both continuity and momentum conservation are adequately
satisfied.
These remarks apply independently of the particular
element chosen. The choice of element needs to be
made keeping in mind the problem of overconstraint
(25)which can occur when continuity is enforced. Various ways of avoiding overconstraint have been proposed, an important one being the use of reduced integration with higher order elements (25). With linear
triangular elements in a Cartesian system, it has been
shown (30) that overconstraint can be avoided if the
mesh is constructed from quadrilaterals with crossed
diagonals. This is also true for the present helical coordinate system.
In any numerical solution, one has to strike a balance
between the amount of algebra and programming, and
the work to be done by the computer. The algebra
involved with the helical system, which has been rather
glossed over in the previous section, provides an incen343

M . L. Hami and 1.F . T. Pittman


tive to choose the simplest possible element, and accept
that a large number of nodes may be needed for a good
solution.
We have therefore chosen linear triangles arranged as
indicated above.
Expressing the functional of E q 10 in terms of the
nodal variables within element e , we obtain

having used {or}


= {u,} [H,] [GI, where {u,} is the row
of nine nodal variables in element e , [He] involves only
the element nodal coordinates, and [GI is a function of r
and t , { M } is a row involving geometrical constants and r
(17). Integration is now over element e. Differentiating
with respect to the nodal variables:

Assembly of the element contributions gives the extremalizing conditions, to be solved for the totality of
unknowns {u}:

As indicated, t h e integrals are governed by t h e


geometry of the mesh, and are independent of fluid
viscosity and operating conditions. They are accordingly
evaluated in a separate program, once only for a given
mesh, and the results stored for use in solutions with
N T h and a. As a consequence of
various values of po,,IpO,
the helical coordinate system, the integrands contain
terms such as 1ldk4' r z and cannot be integrated
exactly by low-order Gaussian quadrature as is usual in
FE implementations. W e have used a Rhoinberg
scheme of an order sufficient to give effectively exact
results, as demonstrated by numerical experiments. A
second program assembles and solves E q 14 exploiting
the symmetric, banded, positive, definite character of
the coefficient matrix, so that only Mb(b + 1)coefficients
are held in high-speed store at any moment (b is semiband width). A mesh-generator program was also written. This very substantially reduces the labor ofproducing specifications for fine meshes, and automatically
orders nodes and elements correctly.

RESULTS AND DISCUSSION


Dependence of the Solution on Lagrangian Multiplier
Value
In the limiting case of an infinitely wide shallow extruder channel, pointwise continuity is satisfied automatically. Thus it is not surprising to find, in the present
work, that solutions for a shallow channel, (depth/width
HMI = 0.05, depthhadius H/R = 0.02), show little dependence on the value of ar//.~*. Using 256 elements, flow
rates at N = 60 RPM are constant within a fraction nfa
percent for values of d / . ~from
* 103 to greater than lo6.
Similar behavior is exhibited by point velocity components, although velocities close to the flights are more
sensitive. A better test of the formulation is provided by
solutions for a deep, highly curved channel (see Table 1
for dimensions). Here, the component equations of motion in the helical system cannot be separated as they can
be in the simple analytic theory just mentioned. Thus,
although us does not occur in the continuity equation

u
au,
L+
r
dr

au,
- - 0,

at

solutions for the helical component, together with u,


and ut, would be expected to show dependence upon
d p * . This is confirmed, and for a uniform 256 element
*
mesh, we find further that the region of d / . ~independence is much reduced. It expands, however, when
finer meshes are used. Results for a uniform 1024 element mesh are shown in Figs. 4 a n d 5 for velocities at the
channel center, for three fractional depths, y (a fractional depth y = 1corresponds to barrel surface, y = 0, to
the screw root). An example of flow-rate variation with
d p * for the highly curved channel is shown in Fig. 6 ,
indicating, by comparison with Figs. 4 and5, the greater
sensitivity to d p * of velocities close to the flight. A finer
mesh would further expand the region of independence
and provide better velocity values nearer the flight if
required. Subsequent solutions were all carried out in
the region of dp* independence.
Values of (V . g)' were integrated over the volume of
every element, and made dimensionless using a characteristic shear rate as follows:
J(V.u)2dV
=

element vol

x channel volume over


one pitch

(16)

344

(15)

The sums over the whole mesh for the shallow (256
Table 1. Deep Screw Dimensions
Barrel internal radius, R, rnrn
19
rnrn
Barrel axial length, rnrn
1000
rnrn
Axial channel width, W, rnrn
109
rnrn
Channel depth, H, rnrn
9.6 rnrn
Axial pitch, rnrn
121
rnm
11.9 rnrn
Axial flight width, rnrn
Flight clearance, rnrn
0.18 rnm
Helix angle at radius R, 4,,, deg 45.2
HIW
0.087
HiR
0.506

POLYMER ENGINEERING AND SCIENCE. MARCH, 1980, Vol. 20, No. 5

Finite Element Solutions f o r Flow in a Single Screw Extruder, Including Curvature Effects

a concentration of finer elements in this region might be


beneficial.
The results in Figs. 4 , 5 and 6 are for drag flow (I&, =
0). For pressure flow (II, = 0), u , = h t l d t = 0, so
continuity isautomatically satisfied. Since the equations
of motion are linear (we have neglected inertia), any
intermediate operating condition can be obtained by
superposition of solutions, and so we should expect alp*
dependence in all cases to be the same as for drag flow.
This is confirmed by numerical experiments.

1.4
1.2
1.0
us

.5

.6

:9

1c3

1 o5

1I

1 c6

1 0

u poise-

Fig. 4 . Variation ofu,switha at the channel center. DragfIow, II,


= 0 . Mesh size 1024 elements. Curve 1 : Y = 0.25, Curve 2 : Y =
0.50, and Curve 3 : Y = 0.75.
1.6
:1.

1.2
1.0
I

.E
.6
4

.2

102

103

t@

105

a poise-

Id

106

F i g . 5 . Variution of u( with a a t the channel center. Drugjlou;, II,,


= 0 . Mesh size 1024 elements. Curve 1 : Y = 0.25, Curve 2 : Y =
0.50, and Curce S : Y = 0.75.

16

Q
in3513

Present solutions for the wide, shallow channel were


compared with the analytic results, e.g., Ref. (31),for an
unwrapped channel of the same proportions. Using
uniform meshes of 64, 256, 1024 elements, flow rates at
open discharge ( I I p = 0), for example, agreed with the
analytic results to 4.5, 2.2 and 0.35 percent, respectively. Comparison of flow rates under other operating
conditions and of velocity profiles within the channel
gave similar results, demonstrating the convergence of
the solution as the mesh is refined.
For the deep channel, flow rates obtained with 1024
elements differed by only 0.5 percent from the results
with a 256 element mesh. At this point it would be quite
customary to assume that convergence to a correct result
is also occurring for this intrinsically more dimcult problem, where an analytic result is not available. The alternative is to test the computations against careful experimental values, and this is what we have done. This
important test of the present approach is reported subsequently, but finally here we make a comparison with
the analytically derived curvature correction due to
Booy (32).
Booy solved the equations of motion and continuity in
cylindrical polar coordinates on the basis of the following
assumptions: developed, slow, isothermal Newtonian
flow; a channel of infinite width (zero radial velocity
everywhere); zero radial pressure gradient; and zero
flight thickness. He used this solution to derive correction factors to be applied to drag flow and pressure flows
calculated from the one-dimensional unwrapped channel theory based on the barrel diameter and helix angle.
He proposed that the correction factors should each be
multiplied by the ratio of the axial channel width to the
axial screw pitch, to allow for non-zero flight thickness.
Finally, the effect of a finite width channel was to be
included using the previously available correction factors derived from the two-dimensional unwrapped
channel theory (31). Accordingly, for the present case
(HID = 0.25, = 45),see Table 1 , we read from Booys
Figs. 3 and 4 (32):

+,,

Comparison with Analytic Results

1 r,J

10

1 g4
a

106

13

poci

F i g . 6. Variation ofjlou; rate with a f o r the deep channel. Drag


jluu., n,>=0.Sc.rewspeedh7
=15!RPM.Meshszze1024elenicntr.

elements) and the highly curved channel (1024 elements) were 0.16 x lop4and 0.05 x lo-, respectively.
Inspection of the element values showed that the largest
values occurred near the flight clearance, indicating that
POLYMER ENGINEERING AND SCIENCE, MARCH, 1980, Vol. 20, No. 5

Drag-flow correction factor

0.78 (25%)

FDC

FpC

= 0.97 (t5%)

Pressure-flow correction factor


The flight thickness correction factor is 1 - 11.9/121 =
0.9. The finite width channel factors are obtained using
the channel width (orthogonal to the flight) at the mean
345

M . L. H a m i and J. F . T . P i t t m a n
depth in the channel. In the present case,

APPENDIX
The Helical System

w,,,~ = W cos 51 = 68 mm

H/wmean
= 0.14 and the factors are (31)

so

Fp

= 0.9 (25%)

FD = 0.92 ( t 5 % )
The resulting operating line is shown in Fig. 7, together
with the one-dimensional unwrapped channel results (based on barrel diameter and helix angle) and the
computed solution. In this example, the Booy correction
slightly overestimates the reduction in flow rate and
pressure-raising capacity due to curvature. The present
computer method clearly has the potential for a more
extensive investigation of curvature effects, and comparison with other results which attempt to correct for
channel curvature (2, 33-36).
None of the flow rates discussed above, either analytic
or computed, has been corrected for leakage flow over
the flights. This topic will also be taken up subsequently.
CONCLUSION
Finite element solutions for developed isothermal
Newtonian flow in both shallow and deep highly curved
screw channels have been obtained, based on a variational formulation in helical coordinates. An integral
constraint method has been used to enforce continuity,
and the dependence of the solutions on Lagrangian multiplier value has been investigated. It is found that for
sufficiently fine meshes, a clear range exists where the
solution is independent of the multiplier value. This
range is a function of geometry and mesh design, and not
of the boundary conditions or physical properties of the
problem. Solutions obtained within the region of multiplier independence are shown to converge to analytic
solutions for a wide, shallow channel. Solutions for a
deep, curved channel show a smaller reduction of
throughput due to curvature than is predicted by the
Booy correction. Computations for this case will be subsequently tested against experimental results.

We use the familiar cylindrical polar coordinates as a


reference system, and give transformations between
this and the helical system. Cylindrical polars are denoted by xl, x2, xrradial, angular and axial coordinates
respectively (we avoid the usual r , 8,z notation, since
we have used r in the dimensionless helical system). We
derive below the transformations given in Ref. (1).
Coordinates. Note that the r, s, t system uses an
anti-clockwise helix, whereas x2 is measured clockwise.
To simpllfy the coordinate transformations, we set the
origin of the helical system at XI= 0, x2 = 0, x3 =
-2?rRk4, where 2?rRk4 is the axial pitch. Refer to Fig. 1
and E 9 3 . Clearly,

xl/R -

X2
-.
2rr

271. Vx,

+ k4 R2 = Rs

(A-1)

where 2?r dxlz ki2R is the helical pitch length at


radius xl.

Therefore,
x2
* 27Tk4R =Rt -

Also,

277

x3

t = x2k, + 2R
X

Therefore,

(-4-3)

Derivatives. Suppose a scalar (or vector/tensor component) F is expressed in the two coordinate systems:

Then

F = f l i X 1 , x2, x 3 ) = fib-,s, t )
aF - aF dr +--+-dF as
aF at
ax,
dr ax,
as ax,
a t ax,
(A-4)

Therefore,
Similarly,

aF - 1 dF
- -R ar

ax,

rs

R(k$ + r)

aF
as

aF aF + k p aF
- - V k 2 + r2 ax2

as

at

(A-5)

0.5

0.4

and

c.3

Vector Components. For the cylindrical polar system,


denote dimensional components by ui, referred to unit
magnitude base vectorski (i = 1, 2 , 3).
For the helical system, the dimensional components
u:, u:, ui are referred to unit magnitude base vectors
br,b,, bt (see Fig. 1). Express an arbitrary vector in
terms of the two systems:

0.2
ii.1

Fig. 7 . Coniparisoii of operating characteristic predictions-all


uncorrected for leakage f l ow. Cume 1 :One-dimensional shallow channel theory, Curve 2 : Finite elenierit computation, arid
Curve 3 : Curve 1 corrected according to Booy.
346

(-4-6)

u .b,, we obtain
Taking in turn the dot products POLYMER ENGlNEERING AND SCIENCE, MARCH, 1980, Vol. 20, No. 5

Finite Element Solutions f o r Flow in a Single Screw Extruder, lncluding Curvature Effects
241

(A-7)

= u:

u, = u: cos

(7d2 +

4)

rate

p_

= density

= a characteristic shear stress, equal to

= helix angle

4 b
v3

= u; cos

4 + u;

=helix angle at the internal barrel radius R


= gravity potential
Matrices

{ D ) (1~ 9 =)the three dimensionless helical velocity


NOMENCLATURE
Roman Scalars

= semi-bandwidth of the coefficient matrix

in equations representing extremalization


of the functional
c
= specific heat
D
= internal barrel diameter
H
= channel depth
F
= a characteristic rate of viscous heat generation
K
= thermal conductivity
k,, k2, k3, k, = geometric factors of the channel
= rotational speed of the screw
= Poiseuille, Thompson numbers
= axial pressure gradient
= volumetric flow rate through the extruder
= flow rate through the channel
=leakage flows, drag and pressure flows,
respectively
= dimensionless radius
= internal barrel radius
= dimensionless helical coordinate
= domain of surface integration
= dimensionless axial coordinate
= temperature
= dimensional velocity components in the
helical system
= dimensionless components in the helical
system
= a characteristic velocity magnitude
= domain of volume integration
= channel width in axial direction
= axial distance, Fig. 3
= fractional depth in the channel, R ( l - r)/H
= momentum and energy hnctionals, respectively, in vector-tensor form
= constrained
form of Yu
= Y,v expressed in helical coordinates
= element contribution to Y

Greek Scalars
= Lagrangian multiplier
= a characteristic shear rate
= a dimensionless measure of continuity en-

forcement
= viscosity
= reference viscosity
= a characteristic viscosity, applicable at

shear rate
= dimensionless extruder pressure gradient
= dimensionless extruder volumetric flow
POLYMER ENGINEERING AND SCIENCE, MARCH, 1980, Vol. 20, No. 5

components and their spatial derivatives


[GI (9x9) = a function of r and t
[ H e ] (9x9) = a function of element nodal coordinates
[ L J (9~ 9 =)afunction ofgeometrical factors andr, s, t
[L,] (9x9) = a function of r
{ M } (1~ 9=)a function of geometrical factors and r, s, t

{ ] (px1)
n1

column of nodal variables for a general


case, otherwise written {n>
[ S ] (3xp) = matrix of shape factors for a general case
{T} (1xp) = a function of spatial derivatives of shape
factors for a general case
{u,] (1~ 9 =)three helical velocity components at each
of the nodes of a linear triangle.
{u) ( 1 ~ 3 n =
) three helical velocity components at each
of the n internal nodes
0,

Vectors, Tensors

bl, bz, b3 =base vectors of a cylindrical polar system


br,
i
s
,
i
t
=base vectors of the helical system
= velocity vector
U
A
= rate of deformation tensor

= del operator

Subscript
0

=indicates values which render the functional stationary, i.e., indicates terms not
subject to variation
REFERENCES

1. J. Nebrensky, J. F. T. Pittman, and J. M. Smith,Polym. E n g .


Sci., 13, 209 (1973).
2. H. M. Griffith, 1. E . C. Fundumentuls, 1, 180 (1952).
3. B. Martin, J. R. A. Pearsoii, aiid B. Yates, Report KO,5,
Polyni. Proc. Res. Ceiitre, Dept. Cheni. Erig. Uiiiv. Canibridge, U.K.(1969);also, H. J. Zaniodits aiid J. R. A . Pearsoii,
Truiis. Soc. Rheol., 13, 357 (1969).
4. R. T. Fenner, Extruder Screw Design, Iliffe Books, London (1970).
5 . K. Palit, Ph.D. Thesis, Dept. Mech. Eng., Imperial College,
London (1972).
6. K. Palit and R. T. Fenner, AZChE I., 18, 628 (1972).
7. h. Palit aiitl R. T. Feiiiier, AZChE J,, 18, 1163 (1972).
8. H. J. Zaniodits, Ph.D. Thesis, Dept. Cheni. Eiig., Uiiiv.
Canibridge, U.L. (1964).
9. M. L. Booy, SPE Truns., 3, 176 (1963).
10. C. Kiparissides and J. Vlachopoulos, P o l y m . E n g . Sci., 16.,
712 (1976).
11. J. R. A. Pearsoil, Paper preseiited at E3.Ch.E. worhiiig
part) oii hoii-Newtoiiiaii Fluid Processiiig; Anisterdani,
Hollai~d(Juiie 1976);also, The CheniicuZ Eiigiireer, 317,91
(1977).
12. R. A. Worth aiitl J. Pariiab) ,PoZyni. Eiig. Sci., 17,257(1977).
347

M . L. Hami and J . F. T . Pittman


13. D. R. Carlile and R. T. FerinerJ.M.E.S., 20, 2 (1978).
14. H. A. A. Helmy and J. Parnaby, Polym. Eng. Sci., 16, 437
(1976).
15. P. Glansdorff and I. Prigogine, Physica, 30, 351 (1964).
16. B. A. Finlayson, Phys. Fluids, 15, 963 (1972).
17. M . L. Hami, Ph.D. Thesis, Dept. Chem. Eng., U.C.
Swansea, Univ. Wales (1977).
18. T. T. Tung and R. L. Laurence, Polym. Eng. Sci., 15, 401
(1975).
19. W. Mohrarid R. S. Mallouk,lnd. E n g . Chem., 51,765(1959).
20. C. Taylor arid P. Hood, Computers Fluids, 1, 73 (1973).
21. P. Hood and C. Taylor, Paper presented at the International
Symposium on Finite Elenlent Methods in Flow Problenis,
University College, Swansea (1974). Proceedings published UAH Press, University of Alabama, Huntsville
(1974).
22. B. Atkinson, C. C . H. Card, and B. M. Irons. Trans. Inst.
Chem. Engrs., 48, 276 (1970).
23. G. P. Bazeley, Y. K. Cheung, B. M. Irons, and 0. C. Zienkiewicz, Proc. 1st Cong. on Matrix Methods in Structural
Mech., Wright Patterson AFB, Ohio (1965).
24. G. Strang and G. J. Fix, An Analysis of the Finite Element
Method, Prentice Hall, Englewood Cliffs, New Jersey
(1973).

348

25. 0. C. Zienkiewicz, The Finite Element Method, (3rd


ed.), McGraw Hill, U.K. (1977).
26. 0. C. Zienkiewicz and P. N. Godbole, Paper presented at
the International Symposium on Finite Element Methods
in Flow Problems, University College, Swansea (1974);
Proceedings published, UAH Press, University of Alabama,
Huntsville (1974).
27. 0. C. Zienkiewicz, Conference on Nunierical Solution of
Differential Equations, Dundee (1973); Lecture Kotes on
Mathematics, Springer (1973).
28. I. M. Gelfand and S. V. Fomin, Calculus of Variation,
trans. ed. R. A. Silverman, Prentice-Hall, Eaglewood Cliffs,
New Jersey (1974).
29. 0. C. Zienkiewicz, The Finite Elenierit hlethod, (3rd
ed.), p. 287, McGraw Hill, U.K. (1977).
30. J. C. Nagtegaal, D. M. Parks, arid J. R. Rice, Computer
Methods Appl. Mech. Eng., 4, 153 (1974).
31. E . C. Bernhardt (Ed.), Processing of Thermoplastic Materials, p. 169-174, Rheinhold, U.S.A. (1967).
32. M. L. Booy, SPE Trans., 3, 176 (1963).
33. P. H. Squires, in Ref. 28.
34. F. Rieger and J. Sestak, Appl. Sci. Res., 28, 89 (1973).
35. D. F. Dyer, AZChE J., 15, 823 (1969).
36. Z. Tadmor, Polym. Eng. Sci., 6, 203 (1966).

POLYMER ENGINEERING AND SCIENCE, MARCH, 1980, Vol. 20, No. 5

Das könnte Ihnen auch gefallen