J . Fluid Mech. (1970), vol. 42, part 3 , p p .

609-625 609
Printed in Great Britain

The flow of a tubular film

Part 2. Interpretation of the model
and discussion of solutions
By J. R. A. PEARSON AND C. J. S. PETRIEP
Department of Chemical Engineering, University of Cambridge

(Received 3 September 1969)

The equations governing the free-surface flow of a tubular film of liquid are
derived from physical arguments, which throw some light on the formal process
described in part 1. The solutions of the equations are discussed, in particular
with reference to the film-blowing process for the manufacture of thin sheets of
thermoplastic material. The qualitative adequacy of a model based on the
dominance of viscous forces is demonstrated, and the effect of surface tension,
air drag and non-isothermal flow is discussed briefly.

1. Introduction
The work described below can be thought of either as an application of the
formal results of part 1 (Pearson & Petrie 1970a),providing in addition a physical
description of the approximations made there, or as a physically based approxi-
mate solution of a practical problem, whose formal justification can be found in
part 1. The authors hope that they have succeeded in separating the two parts of
the work sufficiently for either to be intelligible on its own.
The process studied here is one for the manufacture of a thin sheet or film of
a thermoplastic, such as polyethylene, from molten material supplied under
pressure by a screw extruder. Figure 1 illustrates the process schematically.
The liquid is forced through an annular die and the tubular film produced is
thinned by both an internal pressure and an axial tension. Thus, any element of
the film is being drawn down in two directions as it flows from the die to the take-
up rolls (which are usually vertically above the die). These are arranged to guide
the film once it has solidified (and cooled sufficiently to prevent the film sticking
to itself) from its cylindrical shape to a plane (‘layflat ’) form as it passes through
the nip rolls. The nip rolls form an airtight seal, so that between them and the
die the film forms a tubular bubble containg air at a pressure slightly above
atmospheric. The air supply led in through the centre of the die is used only t o
adjust this pressure.
The rate of cooling, and thus the distance to the freeze-line (theregion where the
molten polymer solidifies) is controlled by jets of cooling air from a ring sur-
t Present address : Department of Engineering Mathematics, University of Newcastle
upon Tyne.
39 F L M 42
610 J . R.A . Pearson and C.J . S. Petrie
rounding the bubble. The nip rolls are driven to provide the axial tension needed
to take up the film, and might be driven at either constant speed or constant
torque, usually the former. (The implications of this choice for the problem of the
control of product dimensions are discussed elsewhere-Pearson & Petrie 1970 b.)

cooling ‘air

FIGURE
1. Diagram of the am-blowing process; section in a vertical plane.

As far as the steady-state problem is concerned we can take either the speed or
the torque as prescribed, and the choice which is most convenient for our analysis
of the flow of the liquid polymer is of a given axial tension applied to the film at
the freeze-line.
What we seek to do here is to set up and use a mathematical model of the %ow
in the region between the die and the freeze-line,where we have the free-surface
 Flow of a tubular jilm. Part 2 611
flow of a highly viscous liquid. We need to prescribe at least seven parameters in
order to get a determinate problem, and these are taken to be the bubble radius
and the film thickness at the die, the freeze-line distance, the pressure difference
across the bubble, the axial tension at the freeze-line, the volumetric flow rate,
and the viscosity of the liquid. If we wish to take account of any but the essential
factors controlling the flow, more parameters will be required. We can then
predict the bubble shape, its thickness and velocity, and the forces acting in it.
In particular, the dimensionless ratios of bubble radius, film thickness and
velocity a t the freeze-line to the corresponding quantities at the die can be
predicted in terms of three numbers, which are essentially dimensionless values
of the freeze-line distance, the excess pressure inside the bubble, and the axial
tension at the freeze-line. (The velocity ratio and the axial tension can be trans-
posed between the lists of dependent and independent quantities, if the velocity,
rather than the tension, is prescribed at the freeze-line.)

2. The mathematical model

The basic assumptions made are that the forces controlling the flow are the
viscous forces arising in the steady axisymmetric isothermal flow of a homo-
geneous Newtonian liquid, and that the film is thin enough for variations in the
flow field across it to be ignored, and for the velocity gradients to be approximated
locally by those of a plane film being extended bi-axially. These assumptions,
and the neglect of the effects of gravity, surface tension, air drag and the inertia
of the liquid, are justified formally to some extent in part 1. They are justified
practically, in part at least, by the fact that reasonable predictions are obtained.
Further experimental verification is required before the range of applicability
of the simple viscous model can be inferred. Certainly cases are known where other
factors cannot be neglected, in particular gravity. The present model can be ex-
tended to cover most of these cases.
Equations governing the flow have already been derived in part 1 (equations
(16) and (17 ) ) by means of a formal perturbation expansion. Here an alternative,
less formal, approach is shown to lead to the same results, and a t the same time
to help in the understanding of the essential physics of the situation. The two
relevant equations are based on a simple balance of forces, one in the axial direc-
tion and the other in the direction normal to the film surface.
We take cylindrical polar co-ordinates (p, q5, z ) as shown in figures 1 and 2,
and define the following symbols:
a is the bubble radius (measured normal to the z-axis) which takes values a,,
at z = 0 (at the die) and A a t z = 2 (at the freeze-line);correspondingdimension-
less quantities are r = a/aoand R = A/a,. (T corresponds to h,, in part 1.)
his the film thickness (measured normal to the film surface), which takes the
values h, at z = 0 and H at z = 2; since h only appears as aratio, it is not necessary
to define a dimensionless thickness. @/a, corresponds to eh,, in part 1.)
x = z/ao and X = Z/aoare dimensionless values of the axial co-ordinate and of
the freeze-line distance respectively.
0 is the angle between the bubble profile and the z-axis, so that tan 0 = da/dz.
39-2
612 J . R.A . Pearson and C. J . [Link]
p is the liquid viscosity and Q the total volumetric flow rate.
A is the pressure difference across the bubble, p (inside)- p (outside),and F, is
the axial force applied at the freeze-line.
In order to obtain the velocity gradients, we define local Cartesian co-ordinates
(tl,t2,t3)at a point P in the film, with & in the direction of flow, t2normal to
the film and t3in the transverse (circumferential) direction (see figure 2). For
definiteness we take the origin P to be on the inner surface; then the &-axis

FIGURE
2 . Co-ordinate systems; sectioned plan and elevation of a portion of the film.

meets the outer surface at t2= h. (At P , the ,& directions coincide with the xi
directions of the ‘intrinsic’ co-ordinates used in part 1.) In this co-ordinate
system, we take velocity components (vl, v2,v3),and proceed to obtain approxima-
tions to the velocity gradients av,/a&.
On the inner surface g2 = 0,v2 is zero, and on the outer surface v2 = Dh/Dt,
so that, neglecting the variation of av2/at2with t2,we obtain
av2/at2= h-lDh/Dt.
Similarly, using the axisymmetry condition and the relation t3= atan(6, we
obtain
av3/ag3= a-IDa/Dt;
 Flow of a tubularjlrn. Part 2 613
and continuity gives
&l/a& = - (h-lDh/Dt + a-lDalDt).
These quantities are all O(1); the other velocity gradients are O(h/a); they are
ignored in this analysis. Treating a and has functions of z, and using dzld[, = cos 6'
and D[,/Dt = vl, we obtain
avl/a& = - V ~ C O S B(a-lda/dz+ h-ldhldz).
8v2/at2 = v1 cos B h-l dhldx,
and 8v3/at3 = v1 cos 0 a-1 daldz.
(Itmay readily be shown that these correspond to the first-order terms obtainable
from equation (3)) part 1.)
The principal stresses are given by

pii = - p + 2p.a~,/a[~, for i = 1 , 2 , and 3,

and the condition that p,, is zero (relative to atmospheric pressure) at the free
surfaces gives, for the hydrostatic pressure p,
p = 2pvl cos 8 h-l dhldz (1)
(cf. equation (15), part 1). This imposes the condition that A < p ; i.e. A is
O(h/a) multiplied by a typical viscous stress. (There is no inconsistency in
ignoring A here while using it below, since in the equations below it balances
terms of order h multiplied by a typical viscous stress.) The stresses are functions
of t1only; and they can be integrated across the film to give the longitudinal and
transverse (hoop)forces per unit length, PL( = hpll) andP, ( = hp,,),respectively.
Using the overall equation of continuity, Q = 27rahv1, to eliminate v1 gives finally

p - _ _pQcos8
__ -1-da
+ - -2 dh
L - 7ra (adz hdz]'

and P - (3)

The balance of the total axial force between cross-sections at z and at 2 (the
freeze-line)gives, neglecting inertial forces,
SnapLcos I3 - na2A = F, - nAzA; (4)
that of the normal forces on the film gives

A = pL/RL -k pH/RH, (5)

where R, and RH are the principal radii of curvature,
R, = a sec 0 and RL = - sec3B/(d2a/dz2).
(See e.g. Novozhilov 1959, p. 96.)
614 J . R.A . Pearson and C. J . S. Petrie
Introducing the dimensionless variables defined above and the dimensionless
parameters,
B = na$A/pQ(a dimensionless pressure difference),
T, = aoFZ/pQ(a dimensionless axial force),
and T = T,- R2B (the total dimensionless axial force a t any cross-section),
and writing ’ for dldx give, after some rearrangement,
h’/h = - &r’/r- $sec2B(T+r2B)= - &r’/r- $(l+ r ’ 2 ) (T+ r 2 B ) , (6)
and 2r2(T+ r2B)r“ = 6r’ + r sec28(T - 3r2B)= 6r’ + r( 1 + T I 2 ) (T - 3r2B). (7)
(In order to show the equivalence of these equations t o (16) and (17), part I ,
set B = P1/2$,, dx/dxl = cosd, and eliminate T by differentiation.) Thus, we
have found one integral of the equations derived in part 1, and have separated
the problem of finding the shape of the bubble from that of finding the film
thickness, ( 7 ) being an equation in r alone.
Two boundary conditions for (6) and (7) can be stated immediately. They are
h=h,, r = l at x=O. (8)
A second boundary condition for ( 7 ) could be prescribed arbitrarily as
r’ = b at x = 0,
but physical considerations suggest that it is conditions a t the freeze-line end of
the bubble that will control the process. If the material freezes (i.e. p -+ co),
then r’ must become zero beyond that line, no further deformation being possible.
It is intuitively obvious that the relation,
r ’ = 0, x = X , (9)
can be applied to the molten region also, provided PL and PH remain bounded.
To show this in the case of rapid freezing, we suppose that the viscosity changes
from its constant finite value yo to an infinite value within a region of length e
(measured in the x-direction) where E can later + 0. If this is the case, then r can
be taken as constant in (7) and we get a relation of the form,
r’’ = A p ’ + B(1+ r’2).
Here A and B are constants, fixed by the parameters defining the problem,
r’ = 0 at x = X and p varies from ,aoto infinity in the range [ X - E , X I . Ele-
mentary argument shows that for suitable p, say
P = P0(4(X - x))4

the term r“ is always O(1) and so r‘ is always O ( E ) Hence,

. by letting e + 0, we
recover (9) as the suitable boundary condition we sought. It is worth noting
that the same argument does not imply h’ = 0 a t x = X , which would otherwise
overdetermine the problem.
The consequence of these boundary conditions is that the solution of (6), ( 7 )
will not in general yield r‘ = 0 a t x = 0, although for large enough X this is very
 Flow of a tubuEarJi1m. Part 2 615
nearly true. This is not incompatible with the equations governing flow at a
die exit; though at the level of approximation we are concerned with, we cannot
investigate this matter further.

3. Discussion 3.1. The phase-plane

We now have a non-linear two-point boundary-value problem; both for the
qualitative discussion, and for the numerical solution of (7), it is convenient to
take initial conditions T = R,r' = 0 at x = X, then modify the choice of R until
we get r = 1 at x = 0. From this point of view, the parameters B (pressure),
Tz (axial tension), X (freeze-line distance) and the initial value R completely
specify the bubble shape (T = Tz -R2B),and we avoid the problem that, if we
start from x = 0, R and hence the parameter T , which appears in the equation,
are not known in advance. (It would be necessary t o guess values for r' at x = 0
and for T , two guesses instead of one.)
We rewrite (7) as
dr/dx = s,
ds/dx = (6s + r( 1 + s2) (T - 3r2B)}/{2r2(T
1
+ PB)}, (10)

and study the trajectories (solution curves) of system (10) in the phase plane
with co-ordinates ( r , s).The system
drldt = - 2r2s(T+ r2B),
I
dsldt = - 6s - r( 1 + s2)( T - 3r2B),
where d x l d t = - 2r2(T +r2B), has the same trajectories as system (lo), with x
(11)

decreasing in the direction of 6 increasing for r2(T+ r2B)> 0. Problems of in-

terpretation on r = 0 and on T + r2B = 0 will be postponed.
Since the equations are unaltered if the signs of r and s are both changed,
and since the half-lines r = 0, s > 0 and r = 0, s < 0 are solutions of system (1 l),
and so may not be crossed by any other trajectories, we confine our attention to
the half-plane r 2 0. I n order to keep the discussion manageable, we restrict
attention to B > 0 and Tz > 0, the ranges relevant to the problem which
motivates this study, and consider the three cases T > 0, T = 0 and T < 0.
These are further subdivided, according to the number and type of the singular
points, into:
l(a), T 3 > 81B/16 > 0,
l ( b ) , T3 = 81B/16 > 0,
1(c), 81BI16 > T 3 > 0,
2, T = 0,
3 ( a ) , 0 > T 3 > -9B/16,
3 ( b ) , 0 > T 3 = - 9B/l6,
3 ( ~ ) , 0 > -9B/16 > T3.
The results are summarized here and illustrated in figures 3-5. The appendix
gives more details and outlines proofs of some of the statements made here.
Case 1. There are two singular points in r 2 0, namely (0, 0),which is a saddle-
616 J . R.A . Pearson and C . J . X. Petrie

0.5

-0.5

(a) (b)
FIGURE 3. Phase plane for case 1, T > 0. Sketches of typical trajectories, arrows in the
=
direction of z decreasing: ( a ) case l(a), (B/T3)& +;
( b ) case l ( c ) , (B/T3)*= Q. v = 8,
u = r(B/T)i.

-0.5
4. Phase plane for case 2, T = 0. Sketches of typical trajectories, arrows in the
FIGURE
direction of x decreasing. v = s, u = rB$.

9
L

-I

(a) (b)
FIGURE 5. Phase plane for case 3, T < 0. Sketches of typical trajectories, arrows in the
+.
direction of x decreasing. ( a )Case 3 ( a ) ,( - B/T3)*= 8; ( 6 ) case 3 (c), ( - B/T3)4= 2) = S ,
u = r( - B/T)*.
 Flow of a tubular Jilm. Part 2 617
point, and ((T/3B)t, 0)) which is a focus in case 1( a ) and a node in cases 1 (b)
and 1 ( c ) .This latter singular point is stable as 6 increases (x decreases, from the
freeze-line towards the die). It can further be shown that there are no closed
trajectories; hence, that every trajectory starting in r > 0 tends to ((T/3B)4, 0)
a s x + -CQ.
Case 2. The only singular point is the origin, which is a node, stable as 5 in-
creases.
Case 3. The origin is a saddle-point; in case 3 (c), this is the only singular point.
In case 3 ( b ) , there is a saddle-node at (( - T/B)3,1).I n case 3 ( a ) , the point
(( - T / B ) t q, ) is a node (stable as 5 increases);the point (( - T/B)t,l / q )is a saddle-
point, where q is the smaller root of 4q2 - 6( - BIT74 q + 4 = 0.
In case 3, r < ( - T/B)acorresponds to T + r2B < 0 ; so in figure 5 the arrows
on the trajectories (in the direction of x decreasing) show 6 decreasing for
r < ( - T / B ) tand increasing for r > ( - T/B)*.For system (lo),moreover, the
points (( - T/B)t,q ) and (( - T/B)t,l / q )are not strictly speaking singular points,
since they are not themselves solutions of the equations, and solutions tending
to these points do in fact reach them in a finite distance (x). They are rather
points of bifurcation of these solutions, where dsldx ( = d2r/dx2)is indeterminate.
The physical interpretation of this non-uniqueness is discussed below. (See also
appendix.)

3.2. Results of the qualitative analysis

First we discuss case 3, where T, is so small that there is a real, positive value of r ,
r = (R2-T,/B)t, for which T +r2B vanishes, and ( 7 )becomes singular. From (6)
(which is (4)in dimensionless form), we see that this means that the longitudinal
tension in the bubble PLvanishes at this value ofr, so that the radius of curvature
RLin ( 5 )is indeterminate. With this interpretation (that the film becomes slack),
it is not surprising that our model fails to predict a unique shape for the bubble;
to keep in touch with the practical process, we insist that the axial tension applied
is sufficient to keep the film taut between the die and the freeze-line. It is sufficient
for this to require that T, > B(R2- 1 ) .
Numerical solution of the equations (see below) shows that cases 2 and 3 give
rise to large blow ratios R and very small freeze-line distances X and thickness
reductions h,/H compared with the values observed in practice, so subsequent
discussion is based on case 1. As was mentioned above the qualitative analysis
provides additional reasons for the choice of boundary condition that was made
(r‘ = 0 at x = X). As x decreases (proceeding towards the die), the trajectories
approach the singular point and, for large enough freeze-line heights, r’ must be
small at the die. (Computation suggests that r’ will fall below 0-1,in a distance of
about 3 die diameters, measured from the freeze-line.) Thus, the observed be-
haviour of the solutions is predicted without the necessity of imposing any
condition at x = 0. A similar argument does not apply for x increasing, as we
approach the freeze-line; moreover, d2r/dx2is large far from the singular point,
so that a small change in X would cause a large change in drlda: at x = X (i.e. the
bubble shape would be critically dependent on X , and similarly on the other
parameters, unless the condition on drldx is imposed at the freeze-line).
618 J . R. A. Pearson and C. J . S. Petrie
We can also make some numerical predictions for long bubbles, since all
trajectories tend to the singular point r = (T/3B)t,r' = 0, as x + - 00. Hence, as
X - + O ~the
, die radius tends to (T/3B)*;since r ( 0 ) = 1, we have in the limit
T = 3B, so that the blow ratio R tends to the value (T,/B- 3 ) t . I n case 1(c),
provided R is not too large, r decreases monotonically from R to (TI3B)B;so
the die radius 1 must be greater than the latter value. Hence,
R2> Tz/B- 3,
and we have a minimum blow ratio attained for large freeze-line distances. The
numerical work confirms that this behaviour is relevant for values of the para-
meters in the practical range of interest; it also shows that the limiting value is
nearly attained in many cases for freeze-line distances of about 10 times the die
radius (X M 10).
Again, this limiting value is independent of liquid flow rate and viscosity,
since T,/B = FZ/raiA;it depends only on the applied forces and the die radius.
For a long bubble, the blow ratio increases with increased axial tension, and
with decreased die radius and internal pressure. This last result is less surprising
when one recalls the behaviour of a spherical bubble acted on by an internal
pressure and surface tension forces. (The excess pressure required to sustain the
bubble is inversely proportional to its radius.)
We can use the foregoing to estimate the effect of increasing the freeze-line
distance on the thickness reduction h,/H. Once the bubble is long enough for the
limiting value of R to be substantially attained, any increase in X corresponds
to a lengthening of the neck of the bubble, where r is close to 1,and r' is close to 0.
We consider freeze-linedistances X , and X,, with corresponding film thicknesses
HI and H,; if R is the same in both cases, we have

(ho/Hl)/(h,o/Hz)
= expIx'
X,
a( 1+ rI2)(T + r2B)dz.
Between X , and X, (measuring from the freeze-line),r w 1 and TI M 0, so that
exp P ( X , - X,t>
( h O / ~ l ) l ( ~ O / H A%5

(using T/3B w 1).Estimates obtained in this way are compared with computed
values of the ratio H,/H, in table 1.

Pressure difference ( B ) 0.1 0.1 0.1 0.2 0.2 0.3

Axial tension ( T z ) 0.5 0.5 2.0 1.0 2.5 2.0
Limiting blow ratio 1.41 1.41 4.12 1.41 3.08 1.91
( R = ( T z / B - 3)t)
Lower freeze-line distance (X,) 20 15 10 10 4 6
Upper freeze-line distance (XI) 30 20 15 15 10 10
Estimate of H,/H,, 2.72 1.65 1.65 2.72 3.32 3.32
exp ( B W ,- X , ) )
Computed valuo of H , / H , 2.84 1.74 1.86 2.89 3.61 3.46
TABLE1. Comparison of estimated and computed values of the change in thickness
reduction due to a change in freeze-line distance
 Flow of a tubular Jilm. Part 2 619

3.3. Numerical results

Numerical estimates of the bubble shape and thickness were obtained by a
Runge-Kutta integration procedure, no special precautions being necessary.
Values of B, T,, and X were fixed, R was guessed, and (7) was integrated from
x = X (with r = R and &/ax = 0 ) to x = 0. This process was repeated with
improved guesses for R,until the condition r = 1 at x = 0 was satisfied.i Then (6)
was integrated to give h,/H (and h,/h(x) if desired). Some typical bubble shapes
are shown in figure 6 for values of the parameters corresponding to cases 1 ( a ) ,
1 (c), 2 and 3 (a).The shape for case 1 (c) is similar to those observed in practice.

IR

X X
-+
FIGURE
6. Sketches of typical bubble shapes: ( a ) case 1(a),
(6) case 1(c), (c) case 2, ( d ) case 3(a).

For the film-blowingprocess two of the important parameters are the product
dimensions, which are determined by A and H , so the dimensionless ratios
R ( = Afa,) and h,/H are the quantities we wish to predict as functions of the
dimensionless parameters B, T, and X (i.e. of the physical variables A, F,, 8,a,,
p and Q ) .For results of practical interest, we may restrict attention to the ranges

t If no other information was available (e.g. from calculations with similar values of
the parameters), R was chosen by linear interpolation. The f i s t two values used in that case
) 3$, the limiting value of R as X + CO, and (Tz/B)*,the limiting value of R
were ( ( T z / B-
as the film tension was allowed to fall to zero at some point in a long film. I n practice,
the final value was often quite near the f i s t of these values, as can be seen in figure 7.
620 J . R.A . Pearson and C. J . S. Petrie
1.5 < R < 3, 10 < h,/H < 30 and 8 < X < 20; hence, we have the restrictions
0.075 < B 6 0.4 and 0.5 d T, d 2.5. It is not easy to estimate p (since in practice
it could vary from 104 t o lo6 poise along the film, on account of the variations
of temperature, and, t o a lesser extent, of shear rate); there are no data from which
Fz can be obtained (so far as we know). Thus, the above is probably the most
reliable way of estimating the relevant values of the parameters. If we take the
values (appropriate t o a small-scale experimental arrangement) a, = 3.75 cm,
Q = 4cm3/sec, A = 70 N/m2 ( z 7 x lo-* atmospheres), p = 3 x lo5 poise, and
F, = 5 N ( z1lb. wt.), we obtain B = 0.097 and T, = 1.56.

3
R1.

I I ! I I
10 20 30 40 50
JLOlH
--3

FIGURE 7. Typical results: Blow ratio R against thickness reduction ho/H. Curves of
constant B and T z ,B and X, and T z and X.(a)B = 0.1,T z = 2 ; (b) B = 0.2, T z = 2;
(c) B = 0.1, Tz = 1 ; ( d ) B = 0.2,T z = 1 ; ( e ) B = 0.1, T z = 0.5;(f)B = 0.1, Y , = 10;
( 9 ) B = 0.2, X = 5; ( h ) B zz 0.1, X = 20; ( i )B ~ 0 . 2 X , = 10; ( j ) T z = 2, X = 10;
(k)T z = 1, X = 20; ( I ) T z = 1, X = 10.

Pressure difference ( B ) 0.2 0.175 0.165 0.1 0.09

Axial tension (Tz) 2.3 2.0 1.85 1.15 1.0
Freeze-line distance (X) 8 9 10 20 23
TABLE2. Typical values of the dimensionless parameters for blow ratio ( R ) = 3 and
thickness reduction (ho/H)= 20

The effect of the parameters B, Tz and X on the product dimensions is shown in

figure 7, where R is plotted against h,/H for fixed values of pairs of these para-
meters. Table 2 shows how these parameters are interrelated by giving typical
values of the three of them for R = 3 and h,/H = 20. (One of B, T, and X can
be chosen arbitrarily.)
As mentioned in $ 1 , the take-up of the film, which has here been assumed to
imply a prescribed axial tension T, a t the freeze-line, could be a t constant velocity,
 Plow of a tubularfilm. Part 2 62 1
equivalent to fixing the axial velocity a t the freeze-line. From continuity
(Q = 27rahv,), we deduce that R = v,(O) h,/v,(X) H, giving a straight line of slope
v,(O)/v,(X) on figure 7. Thus, fixing values of B and X (giving another line on
figure 7), as well as this ratio, suffices to determine the solution of the problem.
To compute a solution from these conditions, a value of T, would be guessed,
and the value of T, needed to give the prescribed value of the ratio v,(O)/v,(X)
would be found by iteration.
[Link] factors
The effect on the feasibility of this approach to the analysis of the film-blowing
process of some of the many neglected factors has been discussed from the point
of view of the asymptotic analysis in part 1 (Pearson & Petrie 1 9 7 0 ~ )Here .
remarks will be confined to four topics where the less formal approach can be
expected to be helpful. I n particular, no mention is made here of gravity, inertia
or effects due to a thick film. (In practice the ratio of film thickness to bubble
radius will lie between 0.05 and 0.005 at the die, and will be smaller downstream.)
The details of the flow at the die exit, where the flow changes from a constrained
to a free-surface flow, have been ignored, despite the quite large 'die-swell'
effects observed in the flow of molten polymers. (See e.g. Pearson 1966, p. 48.) The
assumption, that the effects of this transition are confined to a region near the
die exit, allows the crude approach of 'correcting' the initial values a, and h,
from the die dimensions to the values the die dimensions would have to take in
the absence of any such effects, so as to give the same downstream flow. With
the present state of knowledge of the transition flow, this is an empirical correction.
Air drag can perhaps be dealt with (iteratively if necessary), by taking the
results of the above analysis in its absence, and calculating the air drag on a
bubble of that fixed shape and velocity. Taylor (1959) leads one to hope that the
effect will be small. (Taylor estimates a 7 yo velocity reduction due to air drag
on a water bell.)
The effect of surface tension can easily be allowed for in this approach, with the
proviso that, if the surface tension forces are very much greater than the viscous
forces, the film thickness is not found in the first approximation, since the equa-
tions replacing (6) and ( 7 ) become equivalent. We write PL+ 2r and PH+ 2I' for
PLand'P in (4) and (5)' where I? is the surface tension a t the liquid-air interface,
and then terms 4Grsece and 2Gr2sece are added to the right-hand sides of (6)
and (7), respectively, where G = 271.4 I'/,uQ, the ratio of surface tension to viscous
forces. The modified equations have not been studied in detail, but the limiting
value of R as X + co is readily obtained from
1 = (&)"([l+&]'+[m] G2 4 ), where R 2 = (T,-T)/B,

and the phase plane is not altered in any major way for T > 0 and G not too
large. The non-uniqueness in case 3 is not avoided by taking surface tension into
consideration.
Temperature affects the mechanics of the flow through the dependence of
viscosity on temperature; an attempt was made to estimate this effect by allowing
/I to vary with position along the film. The viscosity was taken to be ,uo at the
622 J . R. A . Pearson and C. J . S. Petrie
freeze-line, where it changes discontinuously as in all the models considered,
and to decrease towards the die (withincreasingtemperature) in a predetermined
way. Functions,u/,u, = 1- O.O5(X - x),exp { - O.O5(X - x)>and exp { - 0.5(X - x)}
were used for X up to 8, 8 and 10 respectively, giving viscosity reductions of
40 %, 33 yo and 99 % over the length of the bubble. The bubble shape was not
significantly altered in any of these cases, the major effect being a considerable
increase in the thickness reduction h,/H over the value it took in the constant
viscosity case. This is in the main due to more rapid thinning of the film in the
long neck of the bubble, where the liquid is hottest and least viscous. Obviously,
this will have an important effect on the quantitative predictions, but it leaves
the qualitative results substantially unaffected. A similar conclusion probably
holds for the effect of the variation of viscosity with rate of shear.

4. Conclusions
We can, with reasonable confidence, deduce from the results of this work that
the dominant factor controlling the flow is the balance between the viscous
forces and the externally applied forces. The major shortcoming of the quan-
titative predictions (for the practical process of making thermoplastic film) is
likely to arise from the neglect of the temperature variation and its effect on the
liquid viscosity. The effects of surface tension and air drag are certainly worth
investigating, but seem unlikely to affect the main features of the flow. In large
bubbles of thick film being slowly drawn, gravity becomes a limiting factor.

Part of the work reported here was carried out while one of us (C.J.S. P.) held a
Science Research Council Fellowship in the Department of Chemical Engineering
a t Cambridge. We are grateful to the Science Research Council, and to the Head
of the Department, for enabling us to carry out this work. Some of the computa-
tional work was done in the computing laboratories of Cambridge and of New-
castle upon Tyne Universities; we are also grateful to the directors of these
laboratories for the use of their facilities.

Appendix. The phase plane for system (11)

dr/d[ = - 2r2s(T+r2B),
ds/d[ = - 6s - r( 1 + s2) (T - 3r2B).
(i) The origin is a non-elementary singular point to which the theorems of
Keil (Sansone & Conti 1964, pp. 256-267) may be applied. For T $. 0 (cases 1 and
3), we write u = Tr, v = 6s + Tr, t = 66, and A = BIT3, to obtain
du/dt = g(u, v),
dv/dt = w +f(u, v ) ,
where
and g(u, v) = +&(V - u )(1+ A d ) .
 P l m of a tubular Jilm. Part 2 623
We identify (v,a )with (x,y) of Keil’stheorems, and see that system (A 1)satisfies
the hypotheses of the theorems; namely, that f and g are dominated by linear
terms near (0, O ) , and that in a neighbourhood of (0, 0), excluding (0,O) itself,
duldt and dvldt do not both vanish.
From the first theorem, there are two and only two trajectories tangent to the
v-axis at the origin, and for system (Al) these are clearly the half-lines u = 0,
v > 0 and u = 0, v < 0. The two regions, into which this pair of trajectories
divides the plane, are considered separately; from the second theorem, the
trajectories in each region fall into one of two classes: either (1) all trajectories
are parabolic (i.e. they tend to the origin) and tangent to the u-axis a t the origin,
or (2) one trajectory is parabolic and tangent to the u-axis at the origin, while
all the other trajectories are hyperbolic (as are the trajectories near a saddle-
point). Thus, the origin is either a node ((1) in both regions), a saddle-point
((2)in both regions),or a saddle-node (( 1) in one region and (2) in the other region).
We distinguish between these alternatives by means of the third theorem,
by studying the slope dvldu of trajectories on either side of the isocline Jo (where
dvldu = 0).We consider first the half-plane u > 0, where, if dvldu increases with v
increasing across J,, we have (l),and, if dv/du decreases,we have (2).The converse
is true in the half-plane u < 0.
Here we approximate Jo near the origin by
v = &( 1 + 108A) u3 + &AU5,
so that for A > - 1/108 J, lies in the first and third quadrants (uv > 0). On v = 0,
dv/du is given by
av/au = (-u3(1+ i o s ~ ) - 5 ~ u 5 ) / { - 2 ~ 3 ( 1 + ~ U 2 ) } ,
which is positive near the origin for A > -IjlOS, so that, from continuity
arguments, dvldu in this case decreases as J, is crossed in the direction of v
increasing. (And, in u < 0, it increases.) For A < - l/lOS, J, lies in the second
and fourth quadrants, and dvldu is negative near the origin, leading to the same
conclusion. Thus, for all values of A , the origin is a saddle-point with separatrices
tangent to the u- and v-axes (i.e. the separatrices are the lines r = 0 and
6s + Tr = 0 in the (r, s)-plane).(See figures 3 and 5.)
For T = 0, we set u = rB), v = s and t = - 6g to obtain
du/dt = QUCV,
dvldt = v - 4 ~ 3 1( + $),
and apply the same methods. (See figure 4.)
(ii) At the singular point ((T/3B)i,0) of case 1 the equations are, writing
W = r - (T/3B)t, dw/d< = ( 8T2/9B)s + 0(w2+ s2),
ds/d< = - ~ T +
w6s + 0 ( w 2 s2). +
The standard methods (see e.g. Sansone & Conti 1964, pp. 44-47) lead to the
results that the point ((T/SB)#,0) is a stable focus for 16T3/9B > 9, and a stable
node for 16T3/9B < 9. I n the latter case, the critical directions are given by
S/W = Y(B/T3)4[l f (1 - 16T3/81B)$].
624 J . R. A . Pearson and C . J . [Link]
(iii)Similarly, the results stated in the main text for the singular points of case
3 ( a )can be obtained from
+ O ( ( r- ( - T/B)4)2+(s- q ) 2 ) ,
d ( r - ( - T/B)*)/d(= - 4q( - T3/B)4( r - ( - T/B)*)
+
d ( s - q)/df; = - 10(q2 1) T ( r- ( - T/B)B)+ (8q(- T3/B)4- 6 ) ( s - q )
+ O(( r - ( - + ( s- q ) 2 ) ,
for the point (1,q ) , and
d ( r - ( - T/B)i)/d =
f; - ( 4 / q )( - T3/B)*
(r - ( - T / B ) & )
+ O ( ( r - ( -T/B)i)2+(s-~ / q ) ~ ) ,
; -(10T(q2+ l)/q2)( r - ( - T / B ) ~ ) + ( 8 ( - T 3 / B ) h / q - 6 ) ( sl/q)
d(s- I / q ) / d f = -
+ O ( ( r- ( - T / B ) i )+2 ( s- l/q)2)
where q = $( - B / T 3 ) t - (( - 9 B / 1 6 T 3 )- 1);) and 0 > T 3 > - 9 B / 1 6 .
(iv) In case 3 ( b ) ,we write u = r - ( - T/B)4,v = s - 1 to obtain
du/df; = - 3~ + O(u2+ v2),
dv/dc = ~ ~ ( - B / T ) ~ u + O ( U ~ + V ~ ) ,
so that u = v = 0 is a non-elementary singular point, which may be shown to be
a saddle-node. (Sansone I%Conti 1964, pp. 256-267.)
(v) I n case 3, solutions with dr1d.z: tending to co are possible and we can get
more information by considering the ( r , @-plane, where t a n 8 = d r / d x ; i.e. we
have as phase space the surface of a cylinder rather than ‘a plane. System (1 1)
becomes

where
du/dt = - 2u2(u2- 1 ) sin 8,
d0/dt = cos 8(u(1+ 3u2)- m sin 2 8 ) ,
u = r( -BIT);, m = 6 ( -BIT3):
I (A 3)

and dt/dx = 2 ( -T 3 / B ) i / 2 r 2 (t
T r2B)cos 0 .
In case 3 ( a ) ,system (A 3) has six singularities on u = 1and a further four on u = 0
in - n < 8 d n. Writing a for the smallest positive root of
sin28 = 4/m (0 < a < an),
theseare ( l , a ) ,(l,fin-a), (l,Qn), (1, -n+ct), (1, -&--a), (1, -in-)and ( O , O ) ,
(0,in), (0,T),(0, -in). Solutions relevant to the physical problem start on
8 = 0 with u > 1 (at the freeze-line)and such solutions, and in fact any solutions
starting in u > 1, reach either the singularity (1, a ) or (1, in),apart from the
separatrices approaching (1, QT - a ) and Ieaving (1, - in).At these singularities,
the solution of system (10)is indeterminate, but there are only certain possibilities
open to it. For example, solutions leaving (1, a),apart from the ingoing separatrix
to the origin, must approach either (1, - in)or (0, in),and, by such arguments,
certain types of solution can be predicted. The investigation of the solutions of
the equations in this case are not discussed in more detail here, because they
are not relevant to the particular physical problem motivating the analysis.
(vi) The proof that in case 1 all trajectories in r > 0 tend to the singular point
(( T/3B)B,0 ) depends on showing that there is a family of closed curves in this half-
 Flow of a tubular jilm. Part 2 625
plane, which are always crossed from their exterior to their interior by trajectories
as x decreases. System (11) may be written
- - m v + u ( l - 3u2) (1 +v2)
dv -
au 2u2(1+ u2)v
(where u = r(T/B)3,t = [(T3/B)4,m = 6(B/T3)9and v = s), which can be in-
tegrated to give
v2 = Azc/(l + u 2 ) 2 - 1+E(u),

where A = +
(1 U ) 2(1 + V2)/Uforthe trajectory passing through the point ( U , V ) ,
and

Writing v; = A u / ( l+ u ~ ) ~1, -we see that the curves v2 = v: are closed, sym-
metrical about v = 0, and cut v = 0 once between u = 0 and u = 1/43, and once
for u > 1/43. ( A as defined above is never less than 5 6J3, that minimum value
giving a real value ( 0 )for v only at u = 1/43.)
We treat vl as an approximation to v with error E , and show that E is always
such that 1v1 < lull as we proceed in the direction of x decreasing. Now m > 0,
u > 0, and for v > 0 u decreases with x decreasing along a trajectory, so we
take u < U , and see immediately that E < 0. Similarly, for v < 0, we take u > U ,
and again E < 0, so that v2 < v:, which is the desired result. Since the family of
curves v2 = v: fills the half-plane u > 0,this completes the proof.

REFERENCES
NOVOZHILOV,
V. V. 1959 The Theory of Thin Shells. Groningen: Noordhoff.
PEARSON, J. R. A. 1966 Mechanical Principles of Polymer Melt Processing. Oxford:
Pergamon.
PEARSON,J. R. A. & PETRIE,C. J. S. 19700 J . Fluid Mech. 40, 1.
PEARSON,J. R. A. & PETRIE,
C. J. S. 1970b Plastica & Polymers, 38, 85.
SANSONE,G. & CONTI. R. 1964 Non-linear Differential Equations. Oxford: Pergamon.
TAYLOR,0. I. 1959 Proc. Roy. Soc. A 253, 289-295.

F L M 42