Journal of Materials Processing Technology 77 (1998) 194 – 200

CADPATH: a complete program for the CAD-, CAE- and

CAM-winding of advanced fibre composites
B.S. Johansen a,*, Aa. Lystrup a, M.T. Jensen b
a
Materials Department, Risø National Laboratory, DK-4000 Roskilde, Denmark
b
Alfa La6al Separation A/S, Søborg, Denmark

Abstract

Filament winding is a widely-used manufacturing process in the composites industry. The efficient use of the process requires
methods of automating the design of components, including the generation of control data for filament-winding machines. This
paper gives an introduction to the geometrical basis for filament winding on surfaces of revolution, with emphasis on the
generation of covering patterns. An example of the use of the methods for an actual component is given, as implemented in
CADPATH, a system for the computer-assisted design and manufacture of filament-wound components. The main features of the
program are: (i) the calculation of possible winding paths for rotationally-symmetric components; (ii) searches for covering
patterns; (iii) updating of the geometry between successive layers; (iv) the performing of simple strength and stiffness analyses of
the resulting laminate; (v) the transfer of laminate data to FEM-programs for detailed analysis; and (vi) the transfer of data for
the generation of control files for filament-winding machines. © 1998 Elsevier Science S.A. All rights reserved.

Keywords: Components; Fibre composites; Filament winding

1. Introduction lution). This simplifies the calculations and the represen-

tation of the mandrel surface, since any surface of
Filament winding is a process that is well suited to the revolution can be approximated to any desired accuracy
manufacture of rotationally-symmetric components. In by a series of cylindrical and conical sections. The
its basic form, the principle of the filament-winding description of these sections can be stored as a list of
process is very simple: a band of fibres is wound around radial and axial positions. The approximation does not
a mandrel, thereby forming the component and the unduly limit the applicability of the program, since most
composite material. In most cases, the fibres are impreg- filament-wound components are in fact rotationally
nated with a heat-curing resin during the winding pro- symmetric.
cess. The actual arrangement of the filament-winding A formal proof for the convergence of the path on the
apparatus can take a number of different forms, a approximated surface to the path on the original surface
lathe-like configuration being common (a description of is given in Gold [2].
alternative forms is given in Peters et al. [1]). The properties of geodesic paths on surfaces of revo-
Actual experience will quickly teach the practitioner lution was described by the French mathematician,
that particular winding paths are stable, while others are Alexis Clairaut, in the 18th century. He found that the
not. Theoretical considerations show that the only paths following condition holds for the relationship between
that can be stable in the absence of friction are geodesic the radius r of the parallel circle (the circle formed by
paths, which are therefore commonly used as the basis the intersection of a plane that is perpendicular to the
for the generation of filament-winding patterns. rotational axis of the body, and passes through the
The mandrel can have any desired shape, but for the point, and the surface) and the local angle a, i.e. the
sake of simplicity, the present work is concerned only angle between the tangent to the geodesic path and the
with rotationally-symmetric mandrels (surfaces of revo- meridian (the curve formed by the intersection of the
surface and a plane which contains the axis of the body
* Corresponding author. and the point).

0924-0136/98/$19.00 © 1998 Elsevier Science S.A. All rights reserved.

PII S0924-0136(97)00417-2
 B.S. Johansen et al. / Journal of Materials Processing Technology 77 (1998) 194–200 195

Fig. 1. Developed surface of a cone.

r · sin(a)=Constant (1) starting position and an angle completely defines the

rest of the winding.
The constant is usually called Clairaut’s constant, rcl.
The relationship between radius and angle is given
The relation can be illustrated easily for the simple
and can not be altered as desired for strength or
case where the body is a cone. The surface of the cone
stiffness purposes. Instead, a compromise between man-
can be developed to a section of a circle, as shown in
ufacturing ease and the strength and stiffness require-
Fig. 1. The angle of the section depends on the apex
ments must be sought. Such a compromise may require
angle of the cone.
a combination of several layers with different values of
For the purpose of the argument, the actual angle is
rcl (and of possibly different materials).
irrelevant. For a given point on the geodesic on the
A cylindrical coordinate system, local radius r, length
cone and the corresponding point on the developed
(or height) z, and angle u, is convenient for the descrip-
surface, the radius of the parallel circle on the cone, r,
tion of the geodesic paths and surfaces of revolution.
and the radius on the circle section, R, are related
In the actual use of Clairaut’s equation for the
linearly by the relation R · sin(g/2) = r, where g is the
calculation of geodesic paths, a relation between r, u
apex angle of the cone.
and z values is needed. These relations can be deduced
For a particular point with radius r0 and angle to the
easily for developable surfaces, since the geodesic on
meridian a0, the corresponding point on the circle
the developed surface is a straight line. The relations
section is R0, and the angle of the geodesic (which is a
are listed in Bookhart and Fowler [3] and Nielsen and
straight line on the circle section) to the radius (merid-
Johansen [4].
ian), is a0. By forming the right-angled triangle ABC,
the distance from the center in the circle section, C, to
the point B, Rcl, is found to be R0 · sin(a0). By selecting 2. Patterns
another point on the geodesic with radius R, the dis-
tance in the right-angled triangle DBC from C to B is An important value for the winding is the rotation v
found to be R · sin(a). Since all bodies of revolution can of the mandrel between two points on the geodesic, at
be approximated by a series of truncated cones and a given z-position, moving in a given direction. The
since the angle does not change in the transition from geodesic between two such consecutive points is custom-
one truncated cone to the next, Eq. (1) follows directly. arily called a circuit, and the winding is performed by the
Some facts are apparent from this equation. The repetition of such circuits, each with an angular displace-
value of the radius when the angle is 90° is equal to rcl. ment of V= v mod(2p) relative to the preceding circuit.
This corresponds to the radius of the opening at the Any rational number can be expressed as the ratio
ends of the winding (the turning points), or the radius between two integers and any irrational number approx-
of the mandrel that will not be covered by the winding. imated by such a ratio to any desired precision. The
The radius of this opening will obviously be the same at angular displacement V between consecutive circuits can
both ends of the winding, and is independent of the thus be expressed as V= 2p(d/N), where 1 5 dBN. One
geometry in between. of the objects in specifying winding paths is to find
There is only one geodesic through a particular point geodesic paths with suitable values for d and N. The
with a given angle, which means that the choice of a subject is treated in detail in Lowery [5] and the
196 B.S. Johansen et al. / Journal of Materials Processing Technology 77 (1998) 194–200

procedure outlined herein is implemented in CADPATH, Pn = qnPn − 1 + Pn − 2; Qn = qnQn − 1 + Qn − 2;

a PC-program for the design and manufacture of
P − 1 = 1; P0 = 0; Q − 1 = 0; Q0 = 1
filament-wound composites (Johansen and Jensen [6]).
Basically, N in the equation specifies the number of The values for q1 to q4 in the example are 6, 1, 5, 2, and
circuits necessary to obtain coverage, and d governs the the corresponding values for P and Q are:
pattern formed during the winding (assuming that the
ratio d/N is non-reducible, i.e. d and N have no common P1 = q1P0+P−1 = 6 · 0+1 = 1
factors). Simple patterns are formed when d = 1 or Q1 = q1Q0+Q−1 = 6 · 1+0 = 6
d = N− 1 (which corresponds to d = − 1). In this case, P2 = q2P1+P0 = 1 · 1+0 = 1
each consecutive circuit is placed next to the previous Q2 = q2Q1+Q0 = 1 · 6+1 = 7
one, being displaced by 9 1/N of a revolution. An P3 = q3P2+P1 = 5 · 1+1 = 6
obvious analogy to this situation appears when d · n= Q3 = q3Q2+Q1 = 5 · 7+6 = 41
N 9 1; n= 2, 3, 4,…, where 2, 3, 4,…, circuits are wound P4 = q4P3+P2 = 2 · 6+1 = 13
on the mandrel before a new path is located adjacent to Q4 = q4Q3+Q2 = 2 · 41+7 = 89
a path that is already placed on the mandrel. This can
be generalized as d · n =N · m 91; m = 1, 2, 3,…, or: The final values P4 and Q4 recreate the original fraction,
(d/N)= (13/89). A solution to the equation d · n−
d · n− N · m= 9 1 (2)
N · m= 9 1 is found as m= P3 and n= Q3; Q3 · d−
In this form, the condition is recognized as a Diophantine P3 · N=41 · 13− 6 · 89= − 1. The algorithm is general,
equation, since d, N, n and m are all integers (whole Vinogradov [8]. It can be shown easily that it will produce
numbers). There is an infinity of solutions to the equa- the minimum values for the solutions.
tion, since it was specified that d and N have no common The algorithm is used in CADPATH, but it can also be
factors, Cohn [7], and the minimum value of n for which implemented conveniently in a spreadsheet program. As
the equation is satisfied specifies the number of circuits an example, consider the case treated above where d=13
that have to be placed on the mandrel before a circuit and N=89. The spreadsheet can then be developed as
is located adjacent (displaced by 91/N of a revolution) follows:
to the previously-placed circuit.
The solution can be found by developing the fraction A B C D E A B C D E
d/N as a continued fraction, using an algorithm similar
to Euclids for greatest common divisors, Vinogradov [8]. 1 N P−1 Q−1 1 89 1 0
This is the method described in Lowery [5], and used in 2 d P0 Q0 1 2 13 0 1 1
CADPATH. 3 r1 q1 P1 Q1 −1 3 11 6 1 6 −1
The procedure can be illustrated by an example. 4 r2 q2 P2 Q2 1 4 2 1 1 7 1
Consider the case where N = 89 and d =13. Both N and 5 r3 q3 P3 Q3 −1 5 1 5 6 41 −1
d are prime, and will consequently have no common 6 r4 q4 P4 Q4 1 6 0 2 13 89 1
factors.

The procedure stops when the remainder rn =1. The numerators and the denominators in the fractions:

can be found from the following relationships:

 B.S. Johansen et al. / Journal of Materials Processing Technology 77 (1998) 194–200 197

Fig. 2. Rotation as a function of angle, for cylindrical mandrels of different length – radius ratios.

The equations and constants are:

The number of rows (steps) necessary depends of

course on the values for N and d. The value of the radius at which coverage is desired, will be selected.
mod(A[j-2], A[j-]) will be zero after a finite number, j-2, This tolerance makes it possible to perform a search for
of steps, and the greatest common divisor of N and d suitable patterns (values of d and N) amongst the
will then be found in column A, row j-1. This will be 1 geodesic patterns. Similarly, there will be a tolerance on
if N and d are irreducible, as was assumed. the angle a.
The solution to Eq. (2) is found in cell C[j-1], m, and The procedure to find covering patterns within a
D[j-1], n. Finally, the value in cell E[j-1] gives the sign given range of band-widths wmin to wmax, and angles
of the solution. It can be shown, Vinogradov [8], that amin to amax, as automated in CADPATH, is outlined in
the sign depends on whether the number of steps in the the following.
algorithm is odd or even. For a given mandrel, there is a relationship between
the rotation during a circuit and the winding angle. An
3. Coverage example is shown in Fig. 2 where the rotation for a
circuit as a function of the angle is shown for cylindri-
The necessary width (for coverage) w of the fibre cal mandrels with plane ends, and length–radius ratios
band at a given radius r, angle a and N will be: of 0.5, 1, 2 and 5.
The search for covering patterns will proceed in the
2pr cos(a)
w= (3) following manner:
N The acceptable range of winding angles will produce
In practice, there is some tolerance on the width of the a range of rotations, vmin to vmax and a range Nmin to
band and a value of w which gives a slight overlap at Nmax by the following reformulation of Eq. (3):
198 B.S. Johansen et al. / Journal of Materials Processing Technology 77 (1998) 194–200

2pr cos(amax) 2pr cos(amin)

5Nmin Nmax 5
wmax wmin
The possible values for the fraction D/N are then
determined by the relationship:
D
vmin 5 5vmax
N
where N is allowed to take all values in the interval
Nmin to Nmax. Each fraction D/N will correspond to a
rotation v in the interval vmin to vmax and the corre-
sponding value(s) for a can be found by standard
numerical methods. The patterns generated are then
analyzed by the procedure just described, where d is
Fig. 3. Mandrel shape.
found by d =D mod N.
It is apparent from Fig. 2 that the range of v for a
An obvious requirement is that the surface should be
given range of a depends strongly on the value of a, as
covered by the fibre-reinforced material. At least one
well as on the geometry. For particular combinations, a
layer must therefore be wound with a value for Clair-
minimum for v will be found in the range, which again
nauts constant of less then the radius of the cylindrical
can produce sets of two values for a for a given
piece (r= 50.5 mm). At the same time it is essential to
combination of D and N. The relationship between v
avoid bridging, i.e. areas of the mandrel that are not in
and a can become arbitrarily complex for more compli-
contact with the filament during the winding: the prob-
cated mandrels. Note that a mandrel with a cylindrical
lem is illustrated in Figs. 4 and 5. The mathematics of
section with radius rcyl will let v go towards infinity
geodesic paths has no difficulty with concavities in the
when rcl =rcyl.
mandrel, but such shapes cannot always be manufac-
It is not uncommon that no suitable values for d and
tured by ordinary filament-winding techniques. The
N are found. Referring to Fig. 2, it is seen that there are
‘manufacturability’ of a given shape is determined in
combinations of geometry and angels that produce very
CADPATH by inspection of the tangent to the geodesic
little variation in v, and consequently little choice of
in the surface, at a given point. A plot of the distance
patterns. The possible solutions require either a depar-
from the axis to the tangent as a function of the
ture from the geodesic winding, or a change in the
z-distance from the contact point shows a hyperbola,
mandrel geometry. The latter possibility can be effected
an intersection between this curve and the mandrel
by the addition of one or two extensions with radius
being a sign of bridging.
equal to rcl for the desired winding angle. This will
It is clear from Fig. 4 that a winding with rcl =40
make it possible to obtain any desired pattern, since the
mm will cause bridging in the transition area from the
amount of rotation over a section with radius equal to
cylindrical part to the cone, while rcl around 45 mm is
rcl can be chosen arbitrarily. For the purpose of calcula-
acceptable. A search for possible patterns in this range
tion of geodesics and the generation of machine control
can produce the pattern shown in Figs. 6 and 7 (a large
data, such extensions are described conveniently with a
band-width has been selected for clarity in the illustra-
slight conicity, due to the already-mentioned property
of the geodesic that will produce an infinite rotation
over a cylindrical section with radius rcl.

4. Example

The principles outlined in the previous sections are

illustrated in the following, by an example (Johansen
and Jensen [9]). The mandrel section is shown in Fig. 3,
the component being a small centrifuge bowl. The
properties of fibre-reinforced-plastics, high specific
strength and good corrosion resistance, make it rele-
vant to investigate the manufacturing possibilities for
this item by filament winding. A collapsible steel man-
drel is used in order to make it possible to separate the
part and the mandrel after the winding and curing. Fig. 4. Bridging, rcl =40 mm.
 B.S. Johansen et al. / Journal of Materials Processing Technology 77 (1998) 194–200 199

Fig. 5. Bridging control, rcl =45 mm.

Fig. 6. Example of a geodesic path.

Fig. 8. Mandrel during the winding.

tion). The values for N and d are 25 and 8. The position

on the different circuits on the circumference at a given the mandrel, machined with the proper diameter for the
z-position is shown in Fig. 7. rcl selected. A very stable path is obtained by letting the
Winding with patterns with rcl in the range around 45 fibre wrap around this section 1/4 to 1/2 turn, which
mm is possible, but somewhat difficult. The geodesic is makes it possible to choose any pattern desired. The
stable in principle, but sensitive to the build-up of cost is the use of a little extra material, which is
material during the winding. There is a tendency for the insignificant in this case (the lower end of the part is cut
band to slip off the corner at the end of the cylindrical off before use). As already mentioned, this extension is
part. This can be avoided by the use of an extension to modeled with a slight conicity in the calculation of the
geodesics.
The same principle can be used for the windings with
larger values of rcl. Fig. 8 shows the mandrel during the
winding of the first layer. At the right is an extension
with r=75 mm, which is used during the winding of
one of the subsequent layers. Any pattern with rcl =75
mm can thus be achieved. The use of an extension is
also convenient during the transition from one layer to
the next. This must normally be done by one or more
non-geodesic circuits in order to change the winding
angle gradually, so that the friction is sufficient to
prevent slippage. A disk with a groove for the fibre
makes it possible to change the winding angle in one
Fig. 7. Section of the winding pattern. step (going from a lower value of rcl to a larger).
200 B.S. Johansen et al. / Journal of Materials Processing Technology 77 (1998) 194–200

Acknowledgements tory, Livermore, California, 1966.

[3] T.W. Bookhart, A.H. Fowler, Geodesic Paths on Surfaces of
Revolution: A Computer-Aided Filament Winding Program,
This work has been carried out within the Centre for
Technical Report Y-1660, Union Carbide Corporation, Nuclear
Polymer Composites under the Danish Development Division, Y-12 Plant, Oak Ridge, Tennesee, 1966.
Programme for Materials Technology. The Centre for [4] F.E. Nielsen, B.S. Johansen, Geometry of Filament Winding and
Polymer Composites is a research and development Computer-Aided Design, Technical Report ISBN 87-7756-180-5,
cooperation on materials and technology between Dan- Centre for Polymer Composites, Taastrup, Denmark, 1992.
ish industry and Danish research and development [5] P.A. Lowery, Continued fractions and the derivations of uniform-
coverage filament winding patterns, Sampe J. 26 (5) (1990) 57–64.
centres.
[6] B.S. Johansen, M.T. Jensen, Cadpath 2.0 — Users Manual, Tech-
nical Report ISBN 87-7756-340-9, Centre for Polymer Com-
posites, Taastrup, Denmark, 1994.
References [7] P.M. Cohn, Algebra, vol. 1, Wiley, New York, 1982.
[8] I.M. Vinogradov, Elements of Number Theory, Dover Publica-
[1] W.D. Humphrey, S.T. Peters, R.F. Foral, Filament Winding tions, 1954.
Composite Structure Fabrication, Sampe, 1991. [9] B.S. Johansen, M.T. Jensen, Demonstration of the Application of
[2] S.P. Gold, Filament Winding Surface of Revolution, Technical CADPATH (in Danish), Technical Report ISBN 87-7756-377-8,
Report SCL-DR-65-105, Sandia Corporation, Livermore Labora- Centre for Polymer Composites, Taastrup, Denmark, 1994.