You are on page 1of 23

Thermal Residual Stress Simulation in

Thermoplastic Filament Winding Process

MARTIN SCHLOTTERMÜLLER,* HAIBO LÜ, YORK ROTH,


NORBERT HIMMEL, RALF SCHLEDJEWSKI AND PETER MITSCHANG
Institut für Verbundwerkstoffe GmbH
University of Kaiserslautern
67663 Kaiserslautern, Germany

ABSTRACT: The present paper describes an advanced model to simulate


the residual stresses of reinforced filament wound thermoplastic composites. The
nonlinear characteristics of the problem are reduced by decoupling some
of the phenomena and the associated dependent variables into submodels. Four
submodels are used here: fiber motion, thermal, kinetic and rheological as well
as a stress–strain submodel. The models were tested on a selected GF/PP sample
chosen from a sensitivity study which is presented also. The effect of individual
process parameters on the residual stresses in filament wound parts is rated by a
sensitivity study.

KEY WORDS: residual stress, thermoplastic composite, filament winding.

INTRODUCTION

ILAMENT WINDING IS one of the predominant manufacturing processes


F for composite cylinders. The processing of fiber reinforced thermo-
plastics in filament winding offers the opportunity to specifically induce
internal stresses in components of fiber reinforced plastic structures.
Compared with thermoset matrix materials, thermoplastics have some
advantages such as the improvement of impact resistance, the reduction of
emissions during the manufacturing process, as well as the reduction of costs
by process chain reduction.

*Author to whom correspondence should be addressed. E-mail: schlottermueller@ivw.uni-kl.de

Journal of THERMOPLASTIC COMPOSITE MATERIALS, Vol. 16—November 2003 497


0892-7057/03/06 0497–23 $10.00/0 DOI: 10.1177/089270503035407
ß 2003 Sage Publications

Downloaded from jtc.sagepub.com at UNIV OF NEW ORLEANS on May 29, 2015


498 M. SCHLOTTERMÜLLER ET AL.

The residual stresses in filament wound composites have received much


attention. It has been shown that residual stresses can induce damage of
the layer and in some cases the amount of residual stress can be so high that
it promotes cracking. The residual stresses of thermoplastic filament wound
composites depend to a high degree on the processing technique and
manufacturing parameters. The assessment of residual stresses requires
an understanding of laminate behavior and the viscoelastic behavior, so that
all phenomena which play a role in the residual stresses are taken into
account.
Much research work has been done in recent years. Olofsson [1] developed
a kinetic model for the thermoset filament winding process which considers
fiber distribution, consolidation degree, lay-up and filament flow stress. This
kinetic model deals with the analysis of internal stresses of wound rings.
Carpenters and Colton [2] proposed a one-dimensional, empirical model of
the online thermoplastic filament winding consolidation process to describe
the laminate deformation and the matrix flow properties through a pressure
roller regarding the parameters winding speed, fiber distribution and contact
pressure. Drozdov and Kalamkarov [3] concentrated on the analytical
determination of optimized filament tow prestressing during the filament
winding of pressure vessels without considering the micromechanics of the
problem. Tzeng and Pipes [4] advanced this model to analyze thermal residual
stress for in situ and postconsolidated composite rings. A more intensive
mathematical description of the thermoset filament winding process
was presented by Di Vita et al. [5]. The authors here divided their simulation
model into four decoupled submodels: fiber displacement, thermal, kinetic-
rheological and stress–strain submodels.
In this paper, a model based on Di Vita’s theory is extended to analyze
internal stresses of thermoplastic composites. Again, the overall problem is
divided in four submodels with consideration in this case of thermoplastic
composites. In the experimental part, a sensitivity study was carried out to
expose the effect of the main process parameters on the residual stress state.
An example will be shown that demonstrates the capability of the above
described model.
Similar studies in the past were mostly concerned with thermoset
materials or were not studied for this case. For example Yang and Colton
[6] concentrated on the permeability of thermoplastic composites as prepreg-
tapes, commingled tow and powder coated tow. Tzeng and Pipes [7] – like
Ghashemi et al. [8] – verified their models by comparing them to existing
solutions. There is little reported in the literature about the verification of
models using measured values for residual stress. Tierney and Gillespie did
so but for the tow placement [9] process. Therefore the practical part of the
present article will show how the samples were produced and how the

Downloaded from jtc.sagepub.com at UNIV OF NEW ORLEANS on May 29, 2015


Thermal Residual Stress Simulation 499

processing values and the measured residual stress values were determined
and how well they correspond with the results of the theoretical model.
The thermoplastic winding process is commonly known and here the
method described by Funck [10] was applied using a hydrogen-oxygen flame
as heat source and in this case a consolidation roller that applies pressure at
the nip-point. The residual stress, and the strain, can be determined in
several different ways. Knight [11] used orthotropic photoelastic analysis to
determine the residual stresses. White [12] presented measurement by special
strain gauges for the hole drilling method and the layer removal procedure.
Ersoy and Vardar [13] slitted metal and composite specimens to measure the
residual stress and showed that this method also complies to composite
materials. All these methods are less suitable for ring samples whereas the
method of Funck [14] matches the requirements of filament wound test
samples and guarantees a high grade of reproducibility and accuracy. Strain
gauges measured the restoring strain when the ring was cut open. From this
strain the residual stress can be calculated.

MODELING OF THERMOPLASTIC FILAMENT


WINDING PROCESS

The characteristics of the analysis are highly nonlinear as the processing


technique and the manufacturing parameters can influence the internal
stress state of a thermoplastic filament wound composite. Furthermore,
research on thermoplastic filament winding shows that the fibers can move
to the interior side of the component during the winding process. As a result,
stress relaxation may occur, so that the introduced fiber stresses decrease
partially. The existence of internal stresses also depends to a high degree on
the heating and cooling conditions. Besides the melt viscosity of the matrix,
the crystallization also plays an important role.
The general filament winding process is shown in Figure 1. For the sake
of simplicity, thermal and mechanical effects along the axis of the composite
are neglected, and only transfer phenomena on the r plane is considered.
In order to establish a model of the filament winding process, mass balance,
force balance and energy balance equations need to be considered. The non-
linear characteristics of the resulting system may be reduced by decoupling
some of the phenomena and the associated dependent variables into sub-
models. Four submodels are used in the analysis:
1. Fiber motion submodel;
2. Thermal submodel;
3. Kinetic and rheological submodel;
4. Stress–strain submodel.

Downloaded from jtc.sagepub.com at UNIV OF NEW ORLEANS on May 29, 2015


500 M. SCHLOTTERMÜLLER ET AL.

Temperature measurement

Pre-heating unit Consolidation unit

VW

Measurement of tape
tension and velocity
Brake Mandrel
Flame

Figure 1. Filament winding process.

The objective is to develop the relationships between the thermo-


mechanical properties of the thermoplastic matrix composite and the
parameters describing the manufacturing process (tape angle and tension,
winding speed, heating rate, etc). In particular, the following variables will
be described as a function of time and radial position: mandrel and
composite layer temperature, fiber position, fiber tension, stress and strain,
viscosity of matrix, etc.

Fiber Motion Submodel

Due to the relatively low viscosity of the melted matrix during the winding
process, the fiber will sink under the tension force applied to the tapes. Such
a microphenomenon can be described partially by the radial displacement
of every layer. The fiber motion submodel describes the position, displace-
ment and instantaneous tension of the fibers, the thickness of each layer and
of the whole composite as well as the fiber volume fraction. Input data for
this submodel are the physical and mechanical properties of the fiber and
matrix materials, the geometrical dimensions of the mandrel and the
composite, and the processing parameters such as winding angle and tape
tension force. Other data includes the actual temperature, the degree of
solidification and the viscosity of the matrix and is provided by other
submodels.
The radial component of the fiber tension produces a radial displacement
wfðjÞ , where the index j indicates the jth layer of the composite. After reaching

Downloaded from jtc.sagepub.com at UNIV OF NEW ORLEANS on May 29, 2015


Thermal Residual Stress Simulation 501

the glass transition temperature of the matrix the fibers are locked in their
position and no more displacements occur.
The radial displacement wðjÞf depends on the balance between the radial
component of the fiber tension and the resistance that the viscous matrix
presents to fiber movement. Radial tension is a function of the fiber
modulus Ef, whereas the viscous resistance follows Darcy’s law [15] of flows
through porous media and is a function of the permeability S of the fibers.
Considering the two contributions, the displacement of a fiber layer during
a short time interval t is given by the following expression:

SfðjÞ
wfðjÞ ¼ t sin2 ðjÞ ð1Þ
ðjÞ rfðjÞ

where m(j) is the matrix viscosity, rðjÞf the radial coordinate indicating the
position of the jth layer and fðjÞ the stress in the fiber due to the force of the
tape F. The effective force F applied to the fiber is continuously changing
from its initial value F0 during the winding process. The permeability S of a
porous medium for a Newtonian fluid may be obtained by the Carman–
Kozeny equation [16] as

R2 ð1  f Þ3
S¼ ð2Þ
4kzz 2f

where R is the fiber radius, f the fiber volume fraction and kzz the Carman–
Kozeny constant in the transverse direction of the fiber bundle. The
Carman–Kozeny equation is based on the capillary model of a granular bed
as a bundle of tortuous channels and appears to hold true for isotropic
porous media. However, several authors reported less satisfactory results for
transverse flow through anisotropic porous media [17,18]. This was often
compensated by adjusting the Carman–Kozeny constant to fit the data.
Usually, the Carman–Kozeny constant kzz ¼ 50,000.
An example of fiber layer displacement of the first layer during the whole
process is shown in Figure 2. The displacement changes of other layers are
similar. The outer radius of the mandrel ¼ 0.073 m. With the temperature
decrease, the fiber movements also decrease and stop after the part is totally
consolidated.

Thermal Submodel

This submodel, based on the solution of the energy balance, provides the
temperature distribution in the filament wound composite as a function of

Downloaded from jtc.sagepub.com at UNIV OF NEW ORLEANS on May 29, 2015


502 M. SCHLOTTERMÜLLER ET AL.

1,00686

1,00684

1,00682

1,00680
r1/r0

1,00678
1,00676

1,00674

1,00672

1,00670
0 50 100 150 200
Time (min)
Figure 2. Radial position changes of layers during filament winding process (GF/PP, steel
mandrel).

time. Input data include the thermal properties of the raw material and the
thermal boundary conditions. The geometry and the instantaneous position
of the fibers are given by the fiber motion submodel. Unlike thermoset
matrix materials, the energy balance equation is less complicated as there is
no chemical reaction occurring and no heat generated in thermoplastic
matrix materials. For this cylindrical geometry it can be written in the
following form:
 
@T 1 @ @T
C ¼ rKr ð3Þ
@t r @r @r

where , C, and Kr represents the instantaneous density, specific heat


and thermal conductivity of the composite respectively.
The heat transfer between the composite components and the environ-
ment during filament winding process is rather complicated, as shown
in Figure 3.
It is difficult to get exact results for the temperature distribution due to
the complexity of the mechanisms and the interaction of the three heat
transfer mechanisms. As a result, an engineering method is used to evaluate
the heat transfer of every part. First, we can define a variable thermal
resistivity R to describe the heat transfer capability of every part:

T
R¼ ð4Þ
q

Downloaded from jtc.sagepub.com at UNIV OF NEW ORLEANS on May 29, 2015


Thermal Residual Stress Simulation 503

Figure 3. The heat transfer balance in filament winding.

Rradiation

Tcomponent T∞

Rconvection

Figure 4. Thermal resistivity model.

where q is the heat flux. If the total thermal resistivity of all heat transfer
types is attained, the temperature change of the composite component is
obtainable.
The heat transfer from the component to the air through convection and
radiation is considered as a parallel model (Figure 4), where Rradiation and
Rconvection are the thermal radiation and convection resistivities of the air,
Tcomponent the surface temperature of the component and T1 the ambient
temperature. The thermal resistivities that are encountered in the filament
winding process are discussed below. The thermal radiation resistivity can
be determined by the following equation [19]:
Tcomponent  T1 Tcomponent  T1
Rradiation ¼ ¼ ð5Þ
q "b ðT 4  T1
4 Þ

where " is the hemispherical emissivity of the component and  b the universal
constant for black-body radiation.
For natural convective flows and a component which is placed
horizontally, the Nusselt number can be calculated by the correlating
equation proposed by Churchill [20]:

Nu ¼ 0:36 þ 0:362 Gr1=6 þ 0:0906 Gr1=3 ð6Þ

Downloaded from jtc.sagepub.com at UNIV OF NEW ORLEANS on May 29, 2015


504 M. SCHLOTTERMÜLLER ET AL.

where Gr ¼ ðgðtw  t1 Þd 3 Þ=2 is the Grashof number,  the thermal


expansion coefficient with  ¼ T 1 for an ideal gas, g the gravitational
vector,  the kinetic viscosity, and d the diameter of the component.
For forced convective flows through a tube, the Nusselt number is given
by the Sieder–Tate equation [21]:
8
> ud
 0:14 >
< Ref ¼  1  104
1=3 f 
Nu ¼ 0:027 Re0:8
f Prf with ð7Þ
w >
: Prf ¼ Cp ¼ 0:7  16700
>


The thermal convection resistivity can be expressed as:

d
Rconvection ¼ ð8Þ
Nu

where  is the heat conductivity coefficient of the flow.


Based on the thermal resistivity model described above, the total thermal
resistivity of convection and radiation is given by:

1 1 1
0
¼ þ ð9Þ
R Rconvection Rradiation

A finite difference module was used to solve the parabolic differential


equation (3). An example is given below. GF/PP is selected as the
thermoplastic material for filament winding. The material properties
associated with the heat transfers are listed in Table 1. The room
temperature was assumed to be 25 C. In the winding process, the mandrel
was kept at a constant temperature of 112.5 C and the material was heated
up to 200 C. The time for winding was 30 min. After the winding was
completed, the component was taken aside to cool down to ambient
temperature. The parameters of air can be found from [19]. Figure 5 shows
the temperature changes in the component during the filament winding
process. In the initial stage of winding (0<t<1800 s), the temperature
of each layer increases and decreases. When the component is removed

Table 1. Material parameters for thermal submodel of


GF/PP.

Density Specific Heat Thermal Conductivity


kg/m3 J/(kg K) W/(m K)
GF/PP 1.47  103 1.1  103 0.61

Downloaded from jtc.sagepub.com at UNIV OF NEW ORLEANS on May 29, 2015


Thermal Residual Stress Simulation 505

Figure 5. Temperature changes through filament winding process.

to cool down, the temperature of each layer decreases to room temperature


gradually.

Kinetic and Rheological Submodel

This model determines the viscosity of the matrix as a function of position


and time. The input data are the parameters of the kinetic and rheological
equations that are obtained during the experimental characterization
program. The composite geometry and the instantaneous position of the
fibers are given by the fiber motion submodel and the temperature
distribution is given by the thermal submodel.
Unlike thermosets, the contribution of physical effect is associated with
the molecular mobility changes, as a consequence of temperature variations,
as there is no chemical reaction occurring in the process. Empirical models
should be considered as useful tools to describe the rheological behavior of
commercial systems with unknown composition. From a process point of
view, temperature is the factor that determines, via viscosity, how readily the
polymer can melt and flow into and through the macropores of the fabric
and the minute spaces between the fibers.
For processes exhibiting very low shear rates (<100 s1) such as calendars
and double belt presses, the temperature dependent apparent viscosity of
a polymer melt can be approximated by means of the Newtonian viscosity
0 using an Arrhenius approach in the form of:
  
E0 1 1
0 ðTÞ ¼ 0 ðTn Þ exp  ð10Þ
R T Tn

Downloaded from jtc.sagepub.com at UNIV OF NEW ORLEANS on May 29, 2015


506 M. SCHLOTTERMÜLLER ET AL.

with the activation energy E0, the gas constant R (8.314  103 kg mol1
K1) and the Newtonian viscosity 0 of the polymer melt at a certain
temperature Tn.
However, the generalized Newtonian models discussed above do not
attempt to relate the polymer viscosity to its chemical and physical
properties. Williams et al. [22] conjectured that the mobility of the polymer
at temperatures above Tg should be a function of the amount of free volume
in the melt. This concept had been used to describe the viscosity of simple
liquids. The logarithm of the viscosity is taken to be proportional to the
fractional change in the free volume. Other theories of the liquid state
propose a temperature dependence of viscosity that is an Arrhenius-type
formulation. This seems to be a good physical picture of polymer liquids at
temperatures much higher than Tg. The following equation is suggested to
express the viscosity of many polymer systems if more specific values are
unknown:
 
AðT  Tg Þ 17:44ðT  Tg Þ
ln ¼ ¼ ð11Þ
0 B þ ðT  Tg Þ 51:6 þ ðT  Tg Þ
In [23] Williams–Landel–Ferry ‘‘(WLF)’’ constants for specific polymers
are listed.

Stress–Strain Submodel

The stress–strain submodel determines the distribution of stresses and


strains at the layer interphases as a function of time. Input data required for
this analysis are the mechanical properties and the thermal expansion
coefficients of the component. The submodel uses the geometrical
distribution of the layers and the temperature distribution through the
composite thickness provided by the fiber motion, thermal and kinetic
submodels, respectively. The stress–strain submodel results are the stress
state due to the initial winding tension and the stress due to thermal effects.
Assuming axial symmetry, the strain–displacement relationship for each
layer in the laminated composite ring can be expressed in cylindrical
coordinates as follows [24]:
du
"r ¼ ð12Þ
dr
and
u
" ¼ ð13Þ
r
where "r and " are the radial and circumferential strain of the component,
u is the radial displacement and r the radial coordinate.

Downloaded from jtc.sagepub.com at UNIV OF NEW ORLEANS on May 29, 2015


Thermal Residual Stress Simulation 507

Assuming plane stress conditions, the constitutive equations for a


mechanical orthotropic material with thermal anisotropy become:

1 r
"r ¼ r   þ
r T ð14Þ
Er E

r 1
" ¼  r þ  þ
 T ð15Þ
Er E

where Ei and ij are the corresponding direction Young’s modulus and
Poisson’s ratios,
i the thermal expansion coefficients,  r and   the radial
and circumferential stress and T is the temperature change. To account for
general conditions, both the mechanical and thermal properties may vary
from layer to layer according to the fiber orientation.
The equilibrium equation in the radial direction is:

dr r  
þ ¼0 ð16Þ
dr r

From Equations (12)–(16) the following equation in u results:

d 2u du
r2 þ r  k2 u ¼ C1 rT ð17Þ
dr2 dr

where k2 ¼ E =Er and C1 ¼ ð1  r Þ


r þ ðr  k2 Þ
 .
As the mandrel is assumed to be much stiffer compared to the composite
cylinder during its production, the radial displacement of the composite
cylinder vanishes at its inner boundary. Furthermore, the radial stress at
the outer boundary is zero as no external pressure is applied. Equation (17)
can be solved with these two boundary conditions.
Numerical solution of the partial differential equations is obtained
by solving the implicit finite difference equation for the radial displacement,

ujþ1, kþ1  2ujþ1, k þ ujþ1, k1 ujþ1, kþ1  ujþ1, k1


r2k þ rk
r2 2r ð18Þ
2
 k ujþ1, k ¼ C1 rk T

The numerical solution of Equation (18) results in the displacement u as a


function of r and t. Finally, the stress components  r and   can be
determined by substituting these results into Equations (14) and (15).
Mechanical and thermal properties for GF/PP are listed in Table 2.
The number of layers is ten and the winding angle is 60 . Figure 6 shows

Downloaded from jtc.sagepub.com at UNIV OF NEW ORLEANS on May 29, 2015


508 M. SCHLOTTERMÜLLER ET AL.

Table 2. Material parameters for stress-strain sub-model of GF/PP.

Young’s Moduli Thermal Expansion


(N/mm)2 Shear Modulus Poisson Coefficient (K1)
(N/mm)2 Ratio
E1 E2 G12 12
1
2
6
GF/PP 30,000 1800 5000 0.3 5  10 70  106

Figure 6. Internal stress changes during filament winding process: (a)   t; (b) r  t.

Downloaded from jtc.sagepub.com at UNIV OF NEW ORLEANS on May 29, 2015


Thermal Residual Stress Simulation 509

(a) 8

σθ (MPa) 2

-2

-4

-6

-8
0,074 0,076 0,078 0,080 0,082
Radius (m)

(b) 0,18

0,16

0,14

0,12

0,10
σr (MPa)

0,08

0,06

0,04

0,02

0,00
0,074 0,076 0,078 0,080 0,082
Radius (m)
Figure 7. Radial and circumferential residual stress distributions after mandrel removed
(a) circumferential residual stress  ; (b) radial residual stress  r .

the internal stress changes of every layer during the filament winding
process. After the mandrel is removed at t ¼ 14,000 s, the radial and
circumferential stress distributions are shown in Figure 7. The measured
experimental data are also shown in it. Larger winding angle generates
larger residual stresses.

MANUFACTURING CONDITIONS

Here, in the practical part, the same conditions are used as in the model.
From this it is clear that the process of filament winding itself has to be

Downloaded from jtc.sagepub.com at UNIV OF NEW ORLEANS on May 29, 2015


510 M. SCHLOTTERMÜLLER ET AL.

Figure 8. Head of the winding machine at IVW.

investigated and that all of its parameters shall be regarded. It is necessary


to guarantee that the assumptions made are realistic. There are some general
and important differences between thermoset and thermoplastic filament
winding. Besides the fact that a heat source is needed for the material to be
molten, the main difference is the guiding of the tape. The comprehension of
the established system machinery plays an important role in the under-
standing of the process and its modeling. Figure 8 shows the system
designed at IVW.
It can be seen in the figure that the usual guidance systems for thermoset
winding, like the rake in the fiber eye, were replaced by a self designed device
consisting of the following parts: an infrared preheating station included
in the head, a consolidation roller that can realize stepless consolid-
ation pressures and a temperature controlled roller. All temperatures are
surveyed either by pyrometers (as it can be seen in Figure 8 for the
consolidation roller) or conventional thermocouples. It is seen that the
fully consolidated tape is entering the head at the back side and is guided
by the consolidation roller that places it on the mandrel. Unlike the
resin impregnated rovings, the tapes that are already consolidated can
neither be bent nor twisted without damage. As a consequence, there is no
free tape length between the head and the mandrel and the position of the
whole winding head is a direct function of the winding pattern, for
the contact of roller and surface must not be on the edge of the roller, but
between two perpendicular parts. This is only possible when all rotational

Downloaded from jtc.sagepub.com at UNIV OF NEW ORLEANS on May 29, 2015


Thermal Residual Stress Simulation 511

axes are used. Furthermore, the consolidation roller is applying a defined


pressure at the center of the nip-point in the same location as the hydrogen-
oxygen flame is melting the tape. This is important for the modeling of
the heat transfer in this area.
The winding head as well as the mandrel is equipped with temperature
control. The temperatures recorded serve as input data for the model. All
data such as tape tension, gas volume (in norm liters) and power of the
preheating unit is acquired by the field bus system which interacts with
the SPS control of the winding machine, so that all actuators and sensors
can be accessed by the control PC and can operate synchronized to the
movements of the axis of the winding machine.

SENSITIVITY STUDY

The aim of the present sensitivity study was to determine the relevant
process parameters that should be investigated more closely and their
influence on the residual stress state. 16 process parameters were identified
in the beginning. This would mean a number of 216 experiments in a full
factorial DOE (design of experiments) multiplied by the number of samples
per experiment. Regarding the time and the material that was used it was
clearly better to effect the study in a fractional factorial concept as proposed
for example by Taguchi [25]. With this design all processing parameters were
respected but only the relevant ones were selected for the further proceeding.
The reduction of parameters is in this case from 16 to 8. In terms of
experiments this means a reduction from 128 to 8. The chosen parameters
have to be varied. Fractional factorial concepts normally operate with two
to four variations of one parameter. In this case, each parameter applies to
two different values. They are assigned to the orthogonal array (Table 3)
that gives the combinations for the tests that have to be performed. If more
values for one parameter should be investigated, the number of parameters

Table 3. Orthogonal array for residual stress test samples.

Variant Mandrel Tape Winding Tape Number of


No. Heating Annealing Material Liner Angle( ) Force (N) Layers
1 No No CF/PEEK No 90 5 5
2 No No CF/PEEK Yes 60 70 10
3 No Yes GF/PP No 90 70 10
4 No Yes GF/PP Yes 60 5 5
5 Yes No GF/PP No 60 5 10
6 Yes No GF/PP Yes 90 70 5
7 Yes Yes CF/PEEK No 60 70 5
8 Yes Yes CF/PEEK Yes 90 5 10

Downloaded from jtc.sagepub.com at UNIV OF NEW ORLEANS on May 29, 2015


512 M. SCHLOTTERMÜLLER ET AL.

has to be reduced or the number of combinations is increased. From


the orthogonal array one can determine with the aid of the variance analysis
the significance and influence of the parameters. The main influence on the
process was attributed to the following parameters in accordance to
the experience from former projects and from what has been investigated in
corresponding research works: Mandrel temperature, cooling conditions,
tape material, application of a liner, winding angle, applied tape force, and
number of layers. For these parameters, respectively, two values were
determined. The combination of the values is given by the DOE-concept
according to the number of parameters that are investigated.
Eight different tests were established (Table 3) to combine the process
parameters.
Every variant was manufactured with three parts that were tested on
residual stress, resp. strain. The results were subjected to a variance analysis.
A detailed description for the required analysis of variance can be found in
[26–28]. The analysis was put forward to show the influence of the process
parameters. The complete procedure of Robust Product Design developed
by Taguchi is not realizable at this stage and was therefore not continued. It
will be executed with the model proposed in this paper after the verification
of the model.

MANUFACTURING AND TEST DEVICES

The test samples were produced on the described filament winding


machine as so it is much easier to get the simulation and the test results onto a
common data base because the process parameters were recorded over time
during the whole process. Figure 9 shows a process parameter plot of a trial
to investigate the influence of the roller temperature control. The
temperature of the mandrel surface is obtained by a thermocouple in the
mandrel material just below the surface. The control algorithm is an on/off
condition within a range of 5 C. The pyrometer that measured the surface
temperature of the substrate is in a fixed position in the middle of the
mandrel with the purpose of showing the time the substrate takes to cool
between its lay down and the lay down of the next layer for this hoop wound
sample. The temperature of the consolidation roller has a great influence on
the temperature of the substrate which is one of the reasons that IVW
developed its control of all parameters by one central processing unit. After
the last peak of the substrate temperature is passed, it takes another 70 s until
the winding of this 19 layer test sample is finished. The remaining time shows
the cooling conditions of the sample remaining on the steel mandrel. All heat
sources were switched off after the last layer had been placed but the mandrel
kept on turning until the end of the winding process and cooling phase.

Downloaded from jtc.sagepub.com at UNIV OF NEW ORLEANS on May 29, 2015


Thermal Residual Stress Simulation 513
Figure 9. Plot of process parameters.
Downloaded from jtc.sagepub.com at UNIV OF NEW ORLEANS on May 29, 2015
514 M. SCHLOTTERMÜLLER ET AL.

The samples produced with the winding machine have a geometry derived
from the NOL tensile strength ring test samples. This means that the
diameter is the same as for NOL rings, but the width has to be different to
avoid the influence of boundary conditions in the area of residual stress
measurement. Gould [29], for example, showed how to determine the
minimum axial length of the test samples that is required to avoid boundary
influence. From the equation

3ð1  2 Þ Eh
k4 ¼ 2 2
¼ ð19Þ
a h 4Da2

it is easy to determine the k and with the condition that the length L of the
sample must be greater than 6/k. For two open boundaries of a tube, the
noncritical length of the samples can be calculated. As there are strain
gauges being applied, the length of the samples must be increased by the
width of the gauges. The strain was measured simultaneously on the inner
and outer diameter.
The residual stress measuring device consists of a support and diamond
sawing blade for cutting the test rings. The support is designed to prevent
the cut samples from snapping in either an opening or closing direction. The
strain release is affected by slowly opening the support. The strain
measurement is continued during the cutting process, so that any potential
influence of the cutting on the strain measurement is immediately revealed.
Figure 10 shows a picture of the test device designed at IVW.

Figure 10. Strain measuring device.

Downloaded from jtc.sagepub.com at UNIV OF NEW ORLEANS on May 29, 2015


Thermal Residual Stress Simulation 515

TEST RESULTS

Three samples were manufactured under constant conditions. Three parts


are enough to show the influence of the process parameters as it is not the
aim to determine a scaled value for the residual stress. The variance analysis
led to the result that the parameters winding angle, annealing, number of
layers and mandrel heating have a significant effect – here the measured
strain as a direct measure for the residual stress – to a level of 99%. The
significance of the parameters allows one to determine the percentage values
of influences. In this case the influences were  28% for the winding angle,
 20% for the annealing,  9% for the number of layers and  8% for the
mandrel heating. Compared to simulated studies the influence of the
mandrel temperature seems very low but in this case it should be noted that
the mandrel used had forced heating but no cooling, the latter correspond-
ing to mandrels in the industry that have no temperature control at all.
Therefore the temperature of the heated mandrel was constant at the chosen
temperature whereas the temperature of the unheated mandrel rose with
increasing production time caused by the applied flame. This is not a
problem for the simulation. Real processing conditions were always
respected but comparing the present results to other simulations one
should be aware that the unheated mandrel is not the same as a cold
mandrel, which is also reflected in the significance of the parameter mandrel
temperature. As other simulations often use a mandrel at constant cold
temperature one should keep this in mind when comparing them to the
present data.
The high influence of the winding angle is not simply a result of geometry,
but the mandrel temperature data shows that the steady forth and back
movement of the flame over the mandrel surface leads to much faster
heating of the mandrel than the radial and restricted heating resulting from
hoop winding. Generally, the dependence of residual stress on the process
parameters is as follows: winding angle and number of layers influence the
residual stress state directly. This is evident because the winding angle and
therefore the fiber direction define the properties of the part in terms of
modulus and strain depending on the direction of the measurement. The
smaller the angle gets, the smaller the residual stress in the circumferential
direction becomes. This is a result of the decreasing effect from the different
volume changes of fiber (in the direction of the winding angle) and the
matrix (behaving isotropically). The number of layers plays a major role in
the thermodynamics of the part. Very thick parts remain warm longer
because of their ratio between volume and surface with the result that the
residual stresses are higher than in thin shells due to the slower cooling rates
in combination with larger temperature gradients. The same effects are

Downloaded from jtc.sagepub.com at UNIV OF NEW ORLEANS on May 29, 2015


516 M. SCHLOTTERMÜLLER ET AL.

responsible for the influence of annealing. Slower cooling rates through


annealing result in less residual stress. Finally, the mandrel heating exerts
the same influence because of its role in the temperature gradient during and
after production.
Generally the dependence is as follows: A hot mandrel and/or an
additional annealing lead to a stress state with negative stress at the mandrel
interface and positive stress at the surface. As a direct sign it can be
remarked that such ring samples have the tendency to snap when a piece is
cut out in the axial direction. On the other hand a cooled mandrel results in
an opposite stress distribution and in rings that open up.

CORRESPONDENCE OF MODEL AND TEST RESULTS

To show the correspondence of model and production the test results


from the variant number 5 of the orthogonal array in Table 3 and the results
from one additional test described below were chosen. The calculation was
carried out under the same conditions as the production of the real parts.
Figure 11 gives an overview of the values that were found for both
model and test samples. The samples were manufactured under the
following conditions: 60 winding angle in the case of variant number
5 ( in Figure 11) resp. 90 for the additional test ( in Figure 11). All

30

20

90°
10
σθ (MPa)

60°
0

-10

-20

1 2 3 4 5 6 7 8 9 10
Layer
60° Experimental results
90° Experimental results
Theoretical results
Figure 11. Residual stress of test samples and prediction model.

Downloaded from jtc.sagepub.com at UNIV OF NEW ORLEANS on May 29, 2015


Thermal Residual Stress Simulation 517

other parameters were the same for both samples: GF/PP material, no
liner, 110 C mandrel temperature, free cooling, 10 layers, and a winding
force of 5 N.
The results are agreeable for the sample with 60 winding angle. They are
within the range of 8–15%. The difference between model and test values,
especially for the 90 version, is most likely caused by the creep occurring
during the production process itself. It is to be underlined that in the present
paper not only is a new model of residual stress presented, but also the
authors showed its application towards a production process that runs on
an industrial scale.

CONCLUSION

A model for the thermoplastic matrix filament winding process and its
implementation in a simulation of the internal residual stresses was
presented. The utilization of the model requires the solution of a system
of coupled differential equations describing the different phenomena. This
solution can be obtained by applying a number of numerical techniques.
Some relevant results are reported in this paper. The model is validated
through experiments. Furthermore, a sensitivity study under production
conditions has shown that the potential of residual stresses can only be
exploited when the temperatures during production are sufficiently
controlled. The necessary equipment has been identified and the imple-
mentation of theoretical knowledge was successfully demonstrated.

ACKNOWLEDGMENT

The authors would like to acknowledge the financial support of this


research project provided by the German Research Foundation (DFG)
through grant NE 546/7-1.

REFERENCES

1. Olofsson, S.K. (1993). Analysis of the Residual Material State in Wet Filament Wound
Thick Pipes, Ninth Annual Meeting, pp. 201–202, Polymer Processing Society, Manchester.
2. Carpenter, C.E. and Colton, J.S. (1993). On-line Consolidation Mechanisms in
Thermoplastic Filament Winding. Composite Modeling and Processing Science,
In: Proceedings of the Ninth International Conference on Composite Materials (ICCM/9),
Madrid, pp. 430–437.
3. Drozdov, A.D. and Kalamkarov, A.L. (1995). Optimization of Winding Process for
Composite Pressure Vessel, International Journal of Pressure Vessels and Piping, 62: 69–81.
4. Tzeng, J.T. and Pipes, R.B. (1992). Thermal Residual Stress Analysis for in situ and
Post-Consolidated Composite Rings, Composites Manufacturing, 3(4): 273–279.

Downloaded from jtc.sagepub.com at UNIV OF NEW ORLEANS on May 29, 2015


518 M. SCHLOTTERMÜLLER ET AL.

5. Di Vita, G., Perugini, P., Kenny, J. and Marchetti, M. (1993). Process Simulation Models
for Filament Winding Technology. Composite Modeling and Processing Science,
In: Proceedings of the Ninth International Conference on Composite Materials (ICCM/9),
pp. 438–446.
6. Yang, H. and Colton, J.S. (February 1994). Microstructure-based Processing Parameters
of Thermoplastic Composite Materials. Part II: Experimental Results, Polymer Composites,
15(1).
7. Tzeng, J.T. and Pipes, R.B. (1992). Thermal Residual Stress Analysis for in situ and Post-
Consolidated Composite Rings, Composites Manufacturing, 3(4): 273–279.
8. Ghashemi, M.N., Gillespie, Jr., J.W. and Cope, R.D. (1992). Prediction of Process-Induced
Stresses for In-Situ Thermoplastic Filament Winding of Cylinders, Third International
Conference CAD-COMP, Computer Aided Design in Composite Material Technology,
pp. 225–253.
9. Tierney, J. and Gillespie, Jr., J.W. (1998). Control of Warpage and Residual Stresses during
the Automated Tow Placement Process, ANTEC ’98, Plastic on My Mind, Vol. 2, pp. 2356–
2360, Atlanta Georgia, 26–30 April 1998.
10. Funck, R. and Neitzel, M. (1995). Improved Thermoplastic Tape Winding using Laser or
Direct Flame Heating, Composites Manufacturing, 6(3–4): 189–192.
11. Knight Jr., C.E. (February 1972). Orthotropic Photoelastic Analysis of Residual Stresses in
Filament-Wound Rings, Experimental Mechanics, pp. 107–112.
12. White, J.R. (1984). Origins and Measurement of Internal Stress in Plastics, Polymer
Testing, 4: 165–191.
13. Ersoy, N. and Vardar, Ö. (2000). Measurement of Residual Stresses in Layered Composites
by Compliance Method, Journal of Composite Material, 34(07).
14. Funck, R. Entwicklung innovativer Fertigungstechniken zur Verarbeitung kontinuierlicher
faserverstärkter Thermoplaste im Wickelverfahren, Fortschrittsberichte VDI Reihe 2,
Nr. 393.
15. Lee, S.Y. and Springer, G.S. (1990). Filament Winding Cylinders: I. Process Model,
Journal of Composite Materials, 24: 1270–1285.
16. Mayer, C., Wang, X. and Neitzel, M. (1998). Macro- and Micro-Impregnation Phenomena
in Continuous Manufacturing of Fabric Reinforced Thermoplastic Composites.
Composites, Part A, 29A: 783–793.
17. Chmielewski, C., Petty, C.A. and Jayaraman, K. (1990). Crossflow of Elastic Liquids
through Arrays of Cylinders, Journal of Non-Newtonian Fluid Mechanics, 35: 309–325.
18. Aström, B.T., Pipes, R.B. and Advani, S.G. (1992). On Flow through Aligned Fibers
Beds and Its Application to Composite Processing, Journal of Composite Materials, 26(9):
1351–1373.
19. Holman, J.P. (1981). Heat Transfer, McGraw-Hill Book Company, New York.
20. Churchill, S.W. and Chu, H.H.S. (1975). Correlating Equations for Laminar and Turbulent
Free Convection from a Horizontal Cylinders, International Journal of Heat and Mass
Transfer, 18(9).
21. Sieder, E.N. and Tate, C. E. (1936). Heat Transfer and Pressure Drop of Liquids in Tubes,
Industrial Engineering Chemistry, 28: 1429.
22. Grulke, E.A. (1994). Polymer Process Engineering, pp. 379–380, PTR Prentice Hall,
Englewood Cliffs, NJ.
23. Ferry, J.D. (1980). Viscoelastic Properties of Polymers, 3rd edn, Wiley, New York.
24. Nowinski, J.L. (1978). Theory of Thermoelasticity with Applications, Sijthoff & Noordhoff
International Publishers, New York.
25. Taguchi, G. (1983). Taguchi on Robust Technology Development, ASME Press, New York.
26. Dean, A. and Voss, D. (1999). Design and Analysis of Experiments, Springer-Verlag,
New York.

Downloaded from jtc.sagepub.com at UNIV OF NEW ORLEANS on May 29, 2015


Thermal Residual Stress Simulation 519

27. Rasch, D., Verdooren, L.R. and Gowers, J.I. (1999). Fundamentals in the Design and
Analysis of Experiments and Surveys, Oldenbourg.
28. Krottmaier, J. (1991). Versuchsplanung, Der Weg zur Qualität des Jahres 2000, Verlag
TÜV Rheinland.
29. Gould, P.H. (1977). Satic Analysis of Shells, Lexington Books D. C. Heath and Company,
Lexington.

Downloaded from jtc.sagepub.com at UNIV OF NEW ORLEANS on May 29, 2015

You might also like