Professional Documents
Culture Documents
net/publication/242348667
CITATIONS READS
46 1,829
6 authors, including:
Some of the authors of this publication are also working on these related projects:
INDICATE - A Novel Instrument for Characterization of Preforms and Bottles for Injection Stretch Blow Moulding View project
All content following this page was uploaded by C. P. J. O’Connor on 25 April 2020.
C.P.J. O’Connor1*, P.J. Martin1, J. Sweeney2, G. Menary1, P. Caton-Rose2, and P.E. Spencer2
1 School of Mechanical and Aerospace Engineering, Queen’s University Belfast, Ashby Building, Stranmillis Road, Belfast, BT9 5AH,
Northern Ireland, UK.
2 School of Engineering, Design and Technology/IRC in Polymer Science and Technology, University of Bradford, Bradford BD7 1DP, UK.
Abstract
Thermoforming is widely employed in industry for the manufacture of lightweight, thin-walled products from pre-extruded
plastic sheet and its largest application is in packaging. Over many years attempts have been made to simulate the process
and thereby exploit modern computational tools for process optimisation. However, progress in this area has been greatly
hampered by insufficient knowledge of the response of polymer materials under thermoforming conditions and an inability to
measure this and other processing phenomena accurately. In recent years some address has been made to these problems
through advances in measurement technologies, and in particular, the development of high speed, high strain, biaxial testing
machines that are designed to replicate the conditions in thermoforming processes. In this work the development of an
advanced finite element-based thermoforming process simulation is presented. At its heart is a sophisticated large strain
thermally coupled (LSTC) material model for polypropylene, which has been developed after several years of research and is
founded directly on biaxial test results at elevated temperatures. This material model has been demonstrated to provide an
excellent fit to the biaxial data and to offer a very stable computational platform for the process simulation. The performance
of the working simulation was validated through comparison with matching experimental test results, and this enabled
investigation of the sensitivity of the process output (in the form of part wall thickness distribution) to changes in a range of
other processing parameters. This work confirmed that the process is most sensitive to the parameters controlling
plug/sheet contact friction. Heat transfer parameters were also shown to be significant and the requirement for the model to
Keywords: Thermoforming, simulation, polypropylene, constitutive model, plug assist, thermal coupling
Page 1
Introduction
The manufacture of packaging is a large and very important industrial sector. Within packaging a large proportion of the
market is concerned with the production of thin-walled polymer containers that are manufactured using industrial
thermoforming processes. As the name implies, thermoforming involves the shaping of a heated polymer material through
the application of pressure in an appropriate thermal environment. Most commonly the feedstock is a heated pre-extruded
polymer sheet and air pressure is the principal means of forming. However, as the required depth of draw in the product
increases it becomes necessary to employ a two stage thermoforming process, where the heated sheet is initially pre-
stretched through contact with a moving mechanical plug, before forming is completed through the application of air
pressure. This process is known as plug-assisted thermoforming and it is very widely used in the manufacture of food
packaging products such as pots, tubs and trays. In the past such products were almost exclusively manufactured from
amorphous polymers such as polystyrene (PS) and polyvinylchloride (PVC), as these can be shaped with relative ease
above their glass transition temperatures and also exhibit forgivingly wide forming temperature windows. However, in recent
times the same market has become dominated by polypropylene (PP) which has gradually replaced its amorphous rivals due
to its superior mechanical properties and lower material costs. As a semi-crystalline material, PP is more difficult to
thermoform as it must be processed within a narrow temperature range just below its crystalline melting point. This, along
with its greater tendency to sag when heated, has significantly increased the challenge facing the thermoformed packaging
industry. At the same time the packaging market has become increasingly demanding, particularly with respect to product
aesthetics, performance and costs, and the industry has responded through improvements in design and manufacturing
technologies. This has included many advances in the speed and efficiency of production machines, but despite the
widespread use of simulation elsewhere, the thermoforming industry still largely remains wedded to traditional trial and error
Over many years researchers have made significant advances in the simulation of most of the major polymer
processing techniques, for instance injection moulding, and this has increased the level of sophistication of materials
processing technology used throughout the industry. However, the equivalent body of work in thermoforming is
comparatively small, and to date its industry has failed to capitalise on many of the benefits of process simulation seen in
other areas. Nevertheless, advances in process simulation have been steadily reported, although the majority of this work
has been focussed upon simple vacuum forming processes such as the preliminary models described by Bourgin et al.
(1995), where a heated sheet was formed by air pressure alone, the material was treated as a membrane and there was no
interaction with a moving plug. Later vacuum forming studies published by Wang et al. (1999) of an acrylonitrile butadiene
styrene (ABS) square cup have reported good correspondence between experiment and simulation. In most cases
Page 2
simulations have employed finite element techniques for modelling the process components and in early work the
assumption of isothermal conditions was commonplace. As a result, these simulations generally employed membrane
elements to represent the sheet during thermoforming and tooling surfaces were modelled as rigid surfaces as in the study of
Karamanou et al. (2006). The performance of the membrane assumption compared to solid continuum elements was
presented in the study by Nam et al. (2000). It was proposed that under conditions of free inflation that the membrane
assumption corresponded well with the three-dimensional solids when the sheet width to thickness ratio is larger than 100.
Experimental work by Collins et al. (2002) has presented clear evidence of significant changes in temperature during
thermoforming and has highlighted the need to include heat transfer in thermoforming process simulations. The
development of non-isothermal simulations have been reported by Laroche and Erchiqui (1999) who used a thermally
activated material model in the finite element analysis of a vacuum formed container and Christopherson et al. (2001) in a
thermally conducting computational fluid dynamic simulation of blister packaging. Use of more sophisticated element types
such as shell (Debbaut and Homerin, 1998) or continuum (solid) elements (O’Connor et al., 2007) have been reported in
thermoforming simulations to model the sheet material. The treatment of contact slip within thermoforming simulations has
also advanced substantially over the years. In early models it was common that no slip or absolute sticking conditions were
assumed, as is the case in the perfect plug slip assumption of Nam et al. (2000) and the perfect mould stick assumption of
Warby et al. (2003). Later models have included contact slip with the use of Coulomb friction functions and this has been
aided by efforts to experimentally measure friction either by using a moving sled technique (Collins et al., 2001) or a
rotational viscometer (Beilharz et al., 2007). Researchers such as McCool et al. (2006) have also recognised the need to
make frictional behaviour temperature dependent and have used non-isothermal friction models to improve their simulations.
Beilharz et al. (2007) proposed several experimentally based mathematical relations for frictional behaviour governed by
temperature, pressure and strain rate dependence. However, accurate measurement of friction during thermoforming is
recognised as an ensuing challenge for researchers and it remains as a serious impediment to the improvement of the
The most significant advance in the technology required for the simulation of thermoforming has come through the
development dedicated biaxial testing machines. For the first time these have been able to subject materials to testing
conditions precisely matching those experienced during thermoforming and have enabled the deformation response of
different polymer materials to be accurately mapped. Across the world a number of such machines are now being used by
research groups to create sophisticated and realistic models for different polymers used in thermoforming and blow moulding
applications. These include the dedicated biaxial testing machines at Oxford University (Buckley et al., 1996), the University
of Erlangen-Nuremberg (Capt et al., 2003) and Queen’s University Belfast (Martin et al., 2005). In early simulations
Newtonian or hyperelastic models were commonly employed (Nied et al., 1990), but with the combination of the need to
Page 3
introduce plug contact in many processes and the improved measurement and understanding of the materials, models are
now predominantly either viscoelastic (O’Connor et al., 2007) or viscoplastic (Wang et al., 1999). A number of very different
mathematical formulations have been proposed for different polymers as presented in the studies by O’Connor et al. (2007)
on PP, Buckley et al. (1996) on polyethylene terephthalate (PET), and Hummel and Nied (2004) on ABS. Researchers such
as Hosseini et al. (2006) have also recognised the need for these models to be thermo-mechanical and have introduced
mechanisms for thermal dependency of properties that is essential for modelling non-isothermal processes. In addition,
results from the biaxial testing of sheet polymers at high strain rates have suggested that internal or adiabatic heating during
stretching may raise the local sheet temperature during the process (Martin et al., 2005), and it has therefore been
suggested that material models need to be fully thermally coupled for accurate simulation of thermoforming (Makradi et al.,
2007). Despite the considerable research effort there is currently no approach to simulating thermoforming processes, which
has wide acceptance for its predictive capabilities and accuracy. Furthermore, thermoforming of PP presents particular
difficulties to the researcher because of the extreme temperature sensitivity of the material and the critical effect this has on
The major aim of this work was the development of an effective and fully featured finite element simulation of the
plug-assisted thermoforming process for polypropylene. It was intended to base the simulation around a newly developed
constitutive model for polypropylene which was the result of several years of dedicated research involving extensive biaxial
testing of the polymer material. As well as providing accurate mapping of the deformation response of polypropylene, the
model was specifically configured to provide a fully coupled thermo-mechanical response and thereby ensure accurate
replication of all aspects of the thermoforming process. It was also planned that the simulation should permit appropriate
levels of heat transfer from both contacting and non-contacting surfaces, and slip across contacting surfaces. In the case of
the latter it was hoped that the development of the simulation would enable closer scrutiny of the importance of friction to the
overall process. The performance of the working simulation was to be validated through comparison with matching
experimental test results, and it was planned that this would enable investigation of the sensitivity of the process output (in
the form of part wall thickness distribution) to changes in a range of other processing parameters.
The simulation presented in this work was constructed specifically to replicate the manufacture of a typical industrially
thermoformed product. The cavity used to create the yoghurt container considered here is detailed in Fig. 1.
Page 4
Figure 1: Major dimensions (in mm) of the yoghurt pot cavity, provided in section view.
At the time of this study the receptacle was being manufactured from a single-layer polypropylene sheet using an
inline plug-assisted thermoforming process. The sheet material contained a 60%:40% blend of homopolymer (Basell Moplen
HP540J) and copolymer (Basell Moplen RP210G). The initial extruded sheet thickness was 1.23mm. The processing
conditions used in the manufacturing process were determined through consultation with an industrial collaborator. This
included process timings, forming speeds and the thermoforming temperature window. The latter was established using a
combination of laboratory tests, including Differential Scanning Calorimetry (DSC) and Dynamic Mechanical Thermal
Analysis (DMTA), and industrial practice. In industry direct sheet temperature measurement is not yet widely employed for
process monitoring and instead much of the industry still relies on long-established trial and error techniques during
production. In this case an acceptable processing window of 1555°C was determined, with a target temperature of 155°C.
The thermoforming process used to manufacture the part is illustrated in Fig. 2. This shows a representation of the
main elements of the production tool taken from the thermoforming simulation, which is described in detail later.
Page 5
Figure 2: Simulation of the plug-assisted thermoforming process.
In thermoforming a pre-extruded polymer sheet is heated, or re-heated, to its previously determined forming
temperature and it is then indexed into the open area of a two part split mould, the lower half of which is illustrated in 3-
dimesional section in Fig. 2(a) and the mesh structure in 2-dimensional form in Fig. 2(b). This shows a single cavity forming
tool, which in industry would only be a small part of a much larger multi-cavity production tool. In this case the complete
industrial tool contained 28 identical cavities. The two-part mould is then closed together and the sheet is clamped above
and below with a downholder built into the upper mould. In the first stage of the thermoforming operation a moving tool,
known as a plug, descends from the top of the two part split mould and pre-stretches the heated polymer sheet into the lower
mould cavity. Typically these plugs are made of temperature resistant polymer materials and can be actuated pneumatically,
hydraulically or with electrical servo motors to control their speed and displacement. For most packaging applications the
plugs are initially unheated but quickly rise to an equilibrium production temperature during repeated production cycles. The
production plug used in this case was made from an acetal homopolymer, polyoxymethylene (POM) which was chosen for its
high stiffness, low friction and low thermal conductivity. For this application the plug completes its downward stroke of 70mm
reaching to within 5.2mm from the centre of the base of the cavity. The plug stroke takes only 0.70 seconds to complete,
moving with an average speed of 100mm/s. At this point a forming air gauge pressure rising to 5bar is rapidly applied to the
plug side of the sheet, which stretches it and propels it downwards and outwards until contact is made with the cavity walls.
These are typically manufactured from an anodised aluminium alloy, and on contact the heated polymer sheet cools rapidly.
The mould temperature is generally controlled by chilled water, which is pumped through internal cooling channels. Once
Page 6
sufficiently cooled the upper mould is retracted, the product is trimmed and then it is ejected from the lower mould and
collected by a stacking unit. The entire thermoforming cycle for this product takes less than two seconds to complete.
In the majority of industrial thermoforming processes accurate control of the product wall thickness distribution is the principal
measure of part quality. Close control ensures that parts have adequate mechanical performance but it also enables
processors to reduce material costs by minimising the starting gauge necessary in the extruded sheet. In this study the wall
thickness distributions of industrially manufactured parts were measured using a magnetic induction thickness gauge. The
thickness profile was recorded through 29 selected locations from the base centre of each part through to its upper lip and an
average taken from a total of 112 containers randomly selected from the overall output from across all 28 cavities in the
production tool. In the later results presented these measured distributions were compared with those from the simulation in
order to judge its predictive performance. A plot of the measured wall thickness distribution for the production container is
shown in Fig. 3. Like most thermoformed products of any kind the distribution is uneven with significant variations in
thickness from one location to another. In this case the thinnest part is around the bottom corner of the pot (<0.2mm), which
contrasts sharply with the much greater thickness at the base centre (>0.6mm). In much of the sidewall the material
distribution is quite well balanced at values slightly above and below 0.3mm thickness. Previous research has clearly
demonstrated the important role that the plug plays in determining these relative thicknesses (Martin and Duncan, 2007) and
in particular it has highlighted the critical role of friction during contact between the plug and the sheet (Collins et al., 2002).
With lower friction and therefore more slip, movement of material from the base towards the corner and sidewall of the pot is
promoted. However, this change must not be excessive as otherwise the imbalance can precipitate quality issues in the final
Page 7
Figure 3: Wall thickness measurements for the product, taken as a profile from the centre of the base of the
container along a vertical profile terminating at the flat lip of the receptacle.
Development of the finite element simulation
It is clear from figures Fig. 1 and Fig. 2 that the product has rotational symmetry about its centreline. This therefore allowed
a simpler two dimensional analysis to be employed in the development of the process simulation. Subsequently an
axisymmetric analysis model was constructed using the commercial finite element analysis [FEA] pre-processor software
Altair HyperMesh. The plug, sheet and mould cavity are clearly detailed in the finite element analysis of the forming process
shown in Fig. 2, as is the axis of symmetry for the process and product. An implicit FEA model was created to simulate the
plug-assisted thermoforming of the product using the industry standard analysis code Simulia ABAQUS/Standard. ABAQUS
has been widely employed for analogous applications in the past and it is regarded as one of the most effective software
tools for constructing manufacturing process simulations. The finite element mesh of the polymer sheet was constructed of
341 SAX2T axisymmetric thermo-mechanically coupled shell elements. The SAX2T is a 3-noded thin or thick axisymmetric
shell element which offers linear temperature variation in the shell surface. The simulation used five integration points
through the element thickness for the calculation of thermal conduction across the sheet. The SAX2T is part of a group of
general-purpose elements that typically provide robust and accurate solutions in many loading conditions for thin and thick
shell problems. It can deform through its thickness as a consequence of in-plane deformations, which is an obvious
requirement in this application. The SAX2T is not prone to transverse shear locking, nor does it have any unconstrained
hourglass modes. SAX2T elements are programmed to calculate large (finite) membrane strains and their sizes, locations
and densities were arrived at by assessing the mesh sensitivity to achieve a progressive convergence of analysis results.
Page 8
The mesh was constructed iteratively, such that sufficient elements were available during the process simulation to take
account of the curvature of the cavity and other form tools, particularly in transitional areas of rapid changes of geometry.
The SAX2T is also a fully thermo-mechanically coupled element type so that mechanical straining of the element can
influence the thermal state and in turn internal heat generation can influence the mechanical behaviour.
The simulation was programmed to precisely replicate the measured processing conditions and therefore the input
thickness set for the polypropylene sheet was 1.23mm. The plug and cavity materials were modelled as isothermal rigid
bodies, incapable of deformation but capable of contact and thermal interaction with the sheet. The form tooling can
therefore act as either a heat sink or heat source, but does not increase or decrease in temperature during the analysis. This
assumption significantly reduces the computational complexity with very minor consequences in terms of the predictive
accuracy of the simulation. In the actual thermoforming process there is generally a thermal gradient between the sheet and
each of the tooling surfaces, but as previously explained the most important thermal interactions are normally between the
plug and the sheet. Clamping of the sheet material between the downholder and mould was simulated by constraining the
sheet around its clamped periphery. An accurate plug traverse was created in the model to displace the plug vertically
downwards into the sheet by 70mm, as discussed previously. Upon completion of this plug displacement a thermoforming
air gauge pressure of 5bar is achieved over a total pressurisation phase of duration 0.4s. This is performed using a fluid cell
in the ABAQUS software which introduces a mass flow rate of air into the plug-side cavity of a maximum value of 9e-03kg/s.
This ramping air-pressure forms the material into the shape of the cavity, where it cools and sets rapidly upon contact.
A fully coupled thermal-stress analysis was performed since the mechanical and thermal solutions in this situation
affect each other considerably and therefore the solution must be obtained simultaneously. The analysis provided for
thermal conduction across the heated sheet and into the cooler tool surfaces, and convection from the free surfaces of the
sheet to the surrounding air. In the case of conduction this included a simple gap conductance function:
q = k (A − B )
where A and B are the temperatures of the respective surfaces in contact, q is the heat flux (W/m2) and k (W/m2K) is
the conductance coefficient. This function activates once the gap between the surfaces closed to within 0.01mm.
The thermal conduction properties of the materials were taken from literature and the values used for the gap
conductance and convection coefficients were estimated from experiments. In this case the convection coefficient chosen
was 25W/m2K and the gap conductance coefficient was 32W/m2K. Thermal radiation was not treated separately in the
analysis. However, its relatively minor influences were accounted for in the consideration and selection of the convection
coefficient chosen from the experiments. It should also be stated that at the time of writing it is not possible in
ABAQUS/Standard to perform a coupled temperature-displacement analysis that directly includes thermal radiation effects.
Page 9
The starting temperatures for the polymer sheet, plug and mould during the simulation were 155°C, 100°C and 60°C
respectively. Interfacial heat generation caused by contact was also accounted for with any heat generated from a frictional
For the treatment of contact friction, the simple Coulomb relationship was assumed for all contacting surfaces with
a single value of friction coefficient used to represent all conditions, whether static or dynamic. Initially relatively high values
for the coefficient of friction at the plug/sheet interface were applied (around =0.80) as the plug tends to have a rather
sticky initial contact with the hot polymer sheet. A lower friction coefficient of =0.40 was selected for the sheet/mould
interface, which reflects the much lower contact temperature at this interface. However, as is the case in practice, it was not
expected that this would result in any significant slip on the mould wall due to the much improved mechanical properties of
the sheet at these lower temperatures. In both cases the chosen values were consistent with practical thermoforming
experience and the published experimental measurements of Collins et al., 2001 from sled tests, and Beilharz et al., 2007
with a rotational viscometer. In the initial model the chosen coefficients of friction were independent of temperature as it was
deliberately planned to use the simulation to investigate the effects of friction in a decoupled manner.
The large strain thermally coupled [LSTC] constitutive model for p olypropylene
The key innovation in the development of the process simulation was the selection and implementation of a suitable
mathematical model to represent the deformation response of the polypropylene material at thermoforming conditions. A
review of literature performed for this study established that researchers are increasingly turning to viscoelastic models for
thermoforming applications, but there was little agreement on the best model to use for any given polymer material. For this
work the strategy chosen was to adapt an existing approach that had been developed by some of the authors over a number
of years for polycarbonate (PC) and polyethylene (PE) (Sweeney et al., 2007) and PP (Olley and Sweeney, 2011), all in cold
drawing experiments. This subsequently was demonstrated to provide excellent results in simulating the multiaxial large
strain response of PP at elevated temperature (Sweeney et al., 2009), but it had never been applied to a full industrial
process such as thermoforming before. In particular the existing model architecture needed to be enhanced to add
temperature dependence and to cope with the full non-linearity of the complete industrial operation. The modified version of
the model for this application has been entitled the Large Strain Thermally Coupled (LSTC) material model. A 1-dimensional
schematic of the LSTC model developed is shown in Fig. 4. The model structure consists of two circuits in parallel.
Page 10
Figure 4: A 1-dimensional schematic of the LSTC model for semi-crystalline polymers (ADJ = adjustable).
The top circuit (X-arm) contains a Neo-Hookean strain energy function W , which has the form of equation (1).
1
W = ( Nc kBT ) 12 + 22 + 32 − 3 (1)
2
where Nc is the number of cross-links per unit volume in this network, kB is the Boltzmann constant, T is the absolute
temperature in degrees Kelvin and i , ( i = 1, 2,3) are the principal stretch ratios. The Neo-Hookean spring represents
the change in entropy of the polymer network through the action of cross-linking behaviour in the material. There is no limit
to the chain extensibility in this X-arm of the circuit and the function makes use of Gaussian chain statistics. In series with
this is an Eyring process that provides a mathematical method to address the stress and rate dependency of the material
behaviour. In the Eyring model utilized in this paper a scalar plastic strain rate e is calculated using the expression detailed
in equation (2).
where A is a material constant which is often linked with initial induced strain at yielding, oct is the octahedral shear
stress, is the hydrostatic stress and Vp and Vs are related to the pressure and shear activation volumes of the material,
as follows:
Page 11
p s
Vp = ; Vs =
kBT kBT
where s and p are the shear and pressure activation volumes respectively. The total strain in this X-arm is a
combination of the strain in the Eyring process and the strain in the neo-Hookean process, which can be represented by
e p
deformation gradient tensors F and F (the elastic and plastic components) respectively. These are combined
F = FeF p
Thus, 1, 2, 3 in equation (1) are the eigenvalues of the pure deformation V associated with the deformation gradient
F , i.e. the left pure stretch tensor V results when rigid body motion R has been removed from the deformation gradient
tensor F . The left stretch tensor V can be similarly decomposed into elastic and plastic components such that:
F = V e V pR
e p
where V is the pure elastic stretch tensor and V is the plastic component of the left stretch tensor.
p
The scalar strain rate e of equation (2) is related to the tensor plastic strain rate D by
1 p p
e= D :D
3
p
where D is the pure plastic rate of deformation tensor calculated as
Dp = V p V p−1
where V p is the rate of change of the plastic component of the left stretch tensor. The Levy-Mises flow rule is used to
specify how the scalar plastic strain increments defined by e are separated into the components of the plastic strain
increment tensor Δε pl . Details of the incremental analysis have been given by Sweeney et al. (2009).
The lower circuit (Y-arm) contains a construct that represents the change in entropy of the polymer network
through the action of cross-linking and slip-linking behaviours in the material. It was proposed by Ball et al. (1981) that in a
rubber network there will be two types of junction points: the permanent cross-links where the chains are joined by chemical
Page 12
bonds and entanglements which are temporary junction points caused by chains becoming entangled. They proposed that
these temporary junction points are replaced by slip-links, which allow the chains to slide over one another. The contribution
to the strain energy of this slip-linking behaviour is governed by a slipperiness factor , which dictates the freedom of
motion of the chains over one another and holds a zero value in the case of no sliding, effectively making the junction a
cross-link. The Edwards-Vilgis modification of the Ball model introduces an additional parameter , which is used as a
measure of the maximum extension ( max ) of the polymer network. max corresponds to a singularity in the stressing of
the network chains with being its reciprocal ( = 1/ max ) (Edwards and Vilgis, 1988). An increase in the magnitude
of represents a reduction in the maximum chain extension. The contribution to free energy is the summation of those
contributions from the slip-link and cross-linking behaviour. The Edwards-Vilgis strain energy function used in the model is
3
(1 − ) i
2 2
1
2
3
( Nc kBT ) i =1
2 1− 2 2
3
+ ln
1 − 2
i
i =1
i
i =1
W = (3)
3 (1 + ) 1 − 2 2
+ 1 (N k T ) ( ) i + ln 1+ 2 + ln 1 − 2 3 2
2 s B ( i ) i
i =1 2
1 − i (1 − i )
3
2 2 i =1
i =1
The Neo-Hookean spring and Eyring process detailed in the 1-dimensional schematic of Fig. 4 are depicted with an
arrow crossing through them with the abbreviation “ADJ.” beside the arrowhead. This is to indicate that the behaviour of the
Neo-Hookean spring and Eyring process are now adjustable. In the model presented here, the spring and the Eyring
process have temperature dependent material parameters. These effects are accounted for in the model by introducing
temperature dependence into the cross-linking density parameter ( Nc ) in the X-arm. Additionally in the X-arm of the
constitutive model, the pre-exponentiation factor ( A ) in the Eyring process expression, detailed in equation (2), is made a
function of temperature and this changes the flow behaviour of the material. In effect the pre-exponentiation factor ( A ) is
adjusted as the deformation evolves. If there is an increase in the material temperature this will activate an increased flow
from the Eyring process. Conversely a drop in temperature will restrict the activated flow processes of the Eyring dashpot.
Larger values of pre-exponentiation factor will precipitate yielding at lower values of stress and strain, which is observable
Page 13
with many polymers. The material model is provided with a range of values for the pre-exponentiation factor ( A ) and the
number of crosslinks per unit volume ( Nc ) for a range of temperatures covering the thermoforming window of the polymer.
A kinematic approach to accommodate the potential for internal heat generation in the material caused by
mechanical deformation is included in the model. A significant amount of energy is dissipated during the stretching process.
However, a significant proportion of this energy can lead to adiabatic heating of the material itself. In the operation of the
pl
model a scalar volumetric heat generation per unit volume, r , caused by the mechanical working of the material is
calculated using a backward Euler scheme to integrate an incremental plastic strain expression (4)
1
r pl = Δε pl : σtX + σtX+t (4)
2t
where the stress in the X-arm only at the previous configuration, σ t , and the current configuration, σt +t , are summed
X X
and the double dot product of the resultant stress tensor and a tensor of true plastic strain increments, Δε pl , in the Eyring
process is formed. The product of the two represents the total energy dissipated in the Eyring process alone over the
increment. This is divided by the time increment between the previous and current configuration, t . The relevant stress
is assumed to be the average of σt and σt +t , hence the divisor of 2. The parameter represents the fraction of the
energy dissipated by the Eyring process, which is returned to the material as heat. It can hold values between 0 and 1, but
in the current model its value was set to 0.30 after a range of values were checked for consistency with experimental
pl
observation of temperature rises during biaxial testing. The heat generation per unit volume, which is calculated as r in
In order to determine the values of the material parameters for the LSTC model, samples of polypropylene sheet were
extensively tested using the custom-built flexible biaxial stretching machine (FBSM) at Queen’s University. Details of the
features and operation of the test machine are described in Martin et al. (2005). The experiments were conducted over a
wide range of conditions representing the thermoforming process. This included temperatures from 145-159°C, strains from
0.00 to 3.16, and nominal strain rates from 0.5 to 31.5/s. In addition, data was obtained for different modes of deformation
including constant width, equi-biaxial and sequential in order for it to be fully representative of the different stages of
thermoforming processes. An example of the model constants obtained by fitting to biaxial test data for the polypropylene
blend described earlier is shown in Table 1. The data given was obtained at a test temperature of 155°C, which matched the
Page 14
chosen forming temperature for the product. The data presented was tested at a nominal strain rate of enom =1.0/s and a
stretch ratio of =3.0. The temperature dependency of the material response was provided by piecewise fitting of the
model constants from the available biaxial test data as presented in Table 2.
Table 1: LSTC model material parameters for the polypropylene blend at 155°C, enom =1.0/s, stretch ratio of =3.0.
Table 2: Temperature dependent materials parameters for the polypropylene blend at given temperatures, enom
=1.0/s, stretch ratio of =3.0.
In general the fit obtained is quite good for the equi-biaxial results presented in Fig. 5. Fig. 5 compares true stress-strain
data for the PP blend in equi-biaxial stretch conditions at three different test temperatures. The model predictions for each of
the three conditions are displayed in Fig. 5. The simulation results presented in Fig. 5 are a deliverable of using the LSTC
model in conjunction with the data set presented in tables 1 and 2. The initial stiffness response is predicted well by the
LSTC model in the curves detailed in Fig. 5. The yielding behaviour is similarly accurately simulated by the mathematical
model as is the strain hardening evident in the PP blend at strains greater than 60%. However the LSTC construct found it
difficult to replicate the significant post-yield strain softening response of this polypropylene blend which is very prevalent at
lower test temperatures in the 10-50% strain region. This strain softening is attributed by the authors to a reduction in the
strength of the material caused by internal heating of the material, which is itself a deliverable of the considerable mechanical
working of the polypropylene as it is progressively stretched and deformed during tests. This is an area of the model
Page 15
7.00
Exp. T=149°C Sim. T=149°C
6.00 Exp. T=155°C Sim. T=155°C
True stress, N/mm2
4.00
3.00
2.00
1.00
0.00
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 1.1
True strain
Fig. 5: Model [Sim.] vs. experimental [Exp.] data (equi-biaxial tests on the PP blend at 155°C, enom =1.0/s, stretch
ratio of =3.0.
Simulation results
The performance of the process simulation employing the LSTC model for polypropylene in modelling the thermoforming of
the small yoghurt pot detailed in figures Fig. 1 and Fig. 2 is demonstrated in figures Fig. 6 to Fig. 12. In all of these figures
the output from the process is assessed in terms of the final wall thickness distribution in the formed product. All the data
presented in figures Fig. 6 to Fig. 12 are drawn from fully thermo-mechanically coupled simulations. The evolution of the part
wall thickness (STH) with time is shown pictorially in stages in Fig. 6. The progressive changes in sheet material shell
thickness (STH) during the plugging stage and the subsequent blowing stage respectively, at six different junctures in the
process, is illustrated in Fig. 6. The thermoforming process is illustrated from the start to the end of the plug’s displacement
into the clamped sheet, which covers a total period of 0.70s, and then the final application of air pressure, which takes a
further 0.40s. In the images shown, reductions in the local sheet thickness are indicated by increases in the relative
brightness of bands in the walls of the partially formed parts. The PP material is seen to effectively wrap around the plug
shape as it progressively deforms and presses into the sheet during its traverse into the lower recesses of the cavity up until
time t=0.70s. Once the pressure is ramped during the pressure cycle at t>0.70s the sheet is seen to billow and peel away
from the plug and ultimately comes into contact with the surrounding cavity, whereupon it cools and sets forming the final PP
product shape.
Page 16
Figure 6: Evolution of the wall thickness (STH) distribution in millimetres during the plugging step (t=0.00 to 0.70s)
and blowing stage (t=0.70 to 1.10s), illustrated with a 180 slice of container.
The equivalent wall thickness distribution from the centre of the base to the clamped lip of each of the partially formed parts
is plotted in Fig. 7 for the plugging step and in Fig. 8 for the blowing step. The final wall thickness distribution of the fully
formed part from experiments is included for comparison. It is clear from these illustrations that the majority of the thinning in
the material initially takes place in the unsupported sidewall. They also show that the material progressively wraps around
the plug as the stroke continues. The material that comes into physical contact with the plug around the base of the part is
Page 17
1.2 Experimental
t=0.23s
t=0.47s
1.0
t=0.70s
Wall thickness, mm
0.8
0.6
0.4
0.2
0.0
0.0 10.0 20.0 30.0 40.0 50.0 60.0 70.0 80.0 90.0 100.0
Distance from the centre of the cup base, mm
Figure 7: Evolution of the wall thickness distribution during the plugging step.
1.2
Experimental
1.0
t=0.83s
Wall thickness, mm
t=0.89s
0.8
t=1.10s
0.6
0.4
0.2
0.0
0.0 10.0 20.0 30.0 40.0 50.0 60.0 70.0 80.0 90.0 100.0
Distance from the centre of the cup base, mm
Figure 8: Evolution of the wall thickness distribution during the blowing step.
This material is clearly restricted from thinning by the high frictional forces preventing slip of the material in this area. This
precipitates the thinning of the material progressively up the sidewall in the direction of the lip of the container. Overall the
unsupported material in the gap between the plug face and the clamp area at the mould interface has performed the
overwhelming majority of the deformation. At the end of the plugging stroke (0.70s in Fig. 7) some material elements in the
Page 18
sidewall have lost nearly 75% of their initial thickness. As time progresses further, the pressure ramps and progressively
starts to peel the material off the plug. The sheet noticeably billows at 0.83s in Fig. 6, and a significant alteration to the
deformation occurs. At this point the sidewall material starts to reach the outer cavity surface and due to a combination of
rapid cooling and frictional contact the thinning of this material ceases as it becomes locked to the surface. The forming
pressure then continues to stretch the remaining free material around the bottom of the part until there is contact of the sheet
across the entire internal surface of the cavity (1.10s in Fig. 6 and Fig. 8). The switching of the major deformation of the
material from the sidewall to the base of the part is very apparent from 0.70s onwards in Fig. 6. Beyond this point the
pressure effectively acts like a physical plug and forces the material down into the lower reaches of the cavity, which has a
major effect on the thickness distribution of the material. This can be verified by observing the very substantial difference in
the wall thickness distribution from 0.83 to 0.89s in Fig. 8. It is clear from this figure that the forming of much of the upper
sidewall of the part is almost complete after the plug step. The blowing step is therefore largely concentrated on defining the
final wall thicknesses in the part base and lower sidewall. It is evident that this step is responsible for taking the relatively
thick regions of the sheet that were initially largely captive through contact with the base of the plug and redistributing this
material towards the furthest extremities of the lower part of the cavity.
The overall performance of the simulation may be seen by comparing the final predicted part wall thickness
distribution at 1.10s in Fig. 8 with the measured distribution obtained from experiments. The correlation between the two is
generally very good, particularly in the base region of the part from 0-25mm and around the top lip (90-100mm), where the
predictions are within the scatter of the physical experimental measurements as indicated by the error bars. The largest
discrepancies are in the sidewall where the simulation underpredicts the thickness in the lower sidewall (25-50mm) and
balances this with a modest overprediction of the thickness in the upper sidewall (55-95mm). However, the average sidewall
thickness is almost the same. It is thought that these differences in the prediction of the sidewall thickness are not due to
any fundamental weakness in the LSTC material model, but instead are largely attributable to the adoption of a single, fixed
coefficient of friction for the plug/sheet contact in the model. This appears to permit too much sliding across the face of the
plug in the early stages of plugging (Fig. 7) which is unresponsive to thermal interactions in that area caused by plug-sheet
contact. As previously introduced, experimental studies on plug friction have already confirmed that the measured
coefficients of friction are very temperature sensitive and have concluded that friction models for thermoforming need to be
temperature dependent. Despite this weakness, the predictive performance of the LSTC model-based thermoforming
simulation shown here is much improved over results previously published for the same product when using material models
based on standard linear viscoelasticity theories (O’Connor et al., 2007). Additionally, throughout these simulation
Page 19
processes the LSTC model proved to be very robust computationally and, providing close control was maintained over the
solution time step, it proved to be considerably more stable than analyses utilising standard linear viscoelastic models.
As part of this investigation, the sensitivity of the simulation to changes in a number of the key processing parameters was
investigated as the research evolved. The availability for the first time of a proven material model within the simulation
provided a unique opportunity for the effects of other parameters to be examined. These results are shown in Fig. 9 through
to Fig. 12. The influence that contact friction between the plug and sheet has on the part wall thickness distribution is
provided in Fig. 9. Here the coefficient of friction between the plug and the polymer sheet has been varied between =0.40
and =3.20. In these examples the fully coupled model was employed along with the contact conditions and process
variables outlined earlier. The results show that the lower the coefficient of friction used, the greater the underestimation of
the wall thickness in the base region of the part. The best results are clearly obtained for a friction coefficient of =0.8.
This value of friction coefficient is consistent with published data for PP in contact in the work of Beilharz et al. (2007). These
results confirm the previously reported observations of the importance of friction to the process and confirm the need for the
1.2
1.0
Experimental
Wall thickness, mm
0.2
0.0
0.0 10.0 20.0 30.0 40.0 50.0 60.0 70.0 80.0 90.0 100.0
Distance from the centre of the cup base, mm
Figure 9: The influence of friction coefficient of the plug sheet interface upon the wall thickness distribution.
The influence that the speed of the plug has upon the thickness distribution is illustrated in Fig. 10, which shows
the wall thickness distribution obtained for speeds of 0.01, 0.10, and 1.00m/s. For these results the fully coupled model and
Page 20
a friction coefficient of =0.8 were employed, all other model variables remaining unchanged. It is apparent that the overall
effect of speed is very small for small changes in plugging rate, which confirms the findings from previously published
experimental observations of the thermoforming process (McCool and Martin, 2010). However if the speed is increased by
larger magnitudes, i.e. factors of 10 as in Fig. 10 more pronounced changes in the behaviour of the material can be
observed. In this instance a significant change in plug speed effects the thickness distribution at the plug-sheet interface
and near to the entrance of the cavity at the lip of the container. A considerable increase in plug speed is shown to cause
the sheet to maintain significantly more material in the base of the pot and precipitate increased thinning near the lip of the
receptacle. Outside of these two areas, a change in plug speed has relatively minor effects on the distribution of the material
1.2
1.0
Wall thickness, mm
Experimental
0.8 Plug speed=1.00m/s
Plug speed=0.10m/s
0.6 Plug speed=0.01m/s
0.4
0.2
0.0
0.0 10.0 20.0 30.0 40.0 50.0 60.0 70.0 80.0 90.0 100.0
Distance from the centre of the cup base, mm
Figure 10: The influence of average plug speed upon the wall thickness distribution.
The effect of thermal convection from the sheet, in the form of the convective heat transfer coefficient, or film
coefficient ( h ), chosen for the simulation is shown in Fig. 11. In this case the results for the chosen initial value of 25W/m2K
for the baseline simulation are compared to those with convection film coefficients of 5W/m2K and 100W/m2K. This lower
value of h =5W/m2K would be representative of a still air condition whereas the higher value would be appropriate for a very
energetic convective airflow. For thermoforming the actual situation is more complex as the sheet is also moving rapidly
(through stretching), so it is expected that the higher values for h would be more representative of actual convection
conditions. Changing the film coefficient is seen to have a sizeable effect on the overall wall thickness distribution in the part,
Page 21
although its influence is again most pronounced in the base and lip regions. At the lower convection values the thickness in
the base is slightly reduced and further away from the experimental measurements, thus confirming that the higher value is
more appropriate for the actual thermoforming process. The higher value convection film coefficient influences the material
to sustain the thickness of the material underneath the plug face. Since the LSTC material model is thermo-mechanically
coupled then changes caused by the more rapid cooling of the material are to be anticipated and indeed this is consistent
1.2
1.0 Experimental
Convection film coefficient, h=100W/(m^2K)
Wall thickness, mm
0.6
0.4
0.2
0.0
0.0 10.0 20.0 30.0 40.0 50.0 60.0 70.0 80.0 90.0 100.0
Distance from the centre of the cup base, mm
Figure 11: The effect of convective heat transfer film coefficient upon the wall thickness distribution.
The final parameter investigated was the rate of application of the forming air pressure, which effectively controls
the straining rate experienced by the sheet material during the blowing stage. In this case this was adjusted by changing the
mass flow rate of air into the upper plug-side fluid cell. Varying the mass flow of air into the upper plug-side cell controls the
length of time required to reach the full forming gauge pressure of 5bar. A higher mass flow rate into the fluid cell will see the
forming pressure achieved sooner, whilst a lower mass flow rate will increase the length of time required to reach the full
5bar. A higher air mass flow rate leads to an increased rate of straining of the material and has the effect of maintaining a
thicker distribution of material in the base of the PP blend container. PP as a material is rate dependent and the fact that the
wall thickness distribution changes when part of the thermoforming process is slowed or accelerated is consistent. This
result is clearly demonstrated in Fig.12 where it is seen that the lowering of the mass flow rate into the fluid cell causes the
receptacle base to thin as compared to the two higher rate simulations presented.
Page 22
1.2
Experimental
1.0
Air mass flow rate=3.60e-02kg/s
Wall thickness, mm
0.4
0.2
0.0
0.0 10.0 20.0 30.0 40.0 50.0 60.0 70.0 80.0 90.0 100.0
Distance from the centre of the cup base, mm
Figure 12: The influence of the pressure cycle air mass flow rate upon the wall thickness distribution.
A comparison of the results from the parameters study shows that all of the key processing parameters
investigated have clearly discernible effects on the wall thickness distribution. In most cases the greatest effects are seen
around the base and lip regions of the formed part, and it is clear that these could not be ignored without significantly
compromising the overall performance of the simulation. Of the parameters investigated, variations in the coefficient of
friction at the sheet/plug interface have clearly the largest capacity to change the performance of the simulation. The
establishment of firm temperature-dependent friction relationships for use in thermoforming and the development of reliable
test methods therefore remain as clear priorities for future research. The importance of predicting the thickness distribution
in the sidewall of the container is underlined as this is a direct deliverable of the plug-sheet interaction and the frictional
conditions therein. Improved modelling approaches must be developed to predict this interaction with greater accuracy.
However, this work has clearly demonstrated the significant effects that the various heat transfer parameters can have on the
simulation and have confirmed the need for a fully thermo-mechanically coupled approach to the process model.
Conclusions
A fully featured process simulation has been successfully constructed for the plug-assisted thermoforming process and it has
been shown to provide good correlation with experimental measurements. It contains a sophisticated large strain thermally
coupled (LSTC) material model for polypropylene that has been carefully constructed using the results of extensive biaxial
stretching tests at thermoforming conditions. This material model has been demonstrated to provide a superior fit to biaxial
data and to offer a very stable computational platform for the process simulation. It has enabled a range of other processing
Page 23
parameters to be investigated, which has confirmed that the process is most sensitive to the parameters controlling
plug/sheet contact friction. Heat transfer parameters have also been shown to be significant and the requirement for the
References
Ball, R.C., Doi, M., Edwards, S.F. and Warner, M., 1981. Elasticity of entangled networks. Polymer, 22, 9.
Beilharz, F., Kouba, K., Blokland, H. and Eyerer, P., 2007. Experimental investigation of friction, thermal conductivity and
heat transfer and their mathematical description for the thermoforming process simulation. In: Proceedings of Polymer
Processing Society Regional Meeting, Gothenburg, Sweden.
Bourgin, P., Cormeau, I. and Saint-Matin, T., 1995. A first step towards the modelling of the thermoforming of plastic sheets.
Journal of Materials Processing Technology, 54, 1-11.
Buckley, C.P., Jones, D.C. and Jones, D.P., 1996. Hot-drawing of poly(ethylene terephthalate) under biaxial stress:
application of a three-dimensional glass--rubber constitutive model. Polymer, 37(12), 2403-2414.
Capt, L., Rettenberger, S., Münstedt, H. and Kamal, M.R., 2003. Simultaneous biaxial deformation behavior of isotactic
polypropylene films. Polymer Engineering & Science, 43(7), 1428-1441.
Christopherson, R., Debbaut, B. and Rubin, Y., 2001. Simulation of pharmaceutical blister pack thermoforming using a non-
isothermal integral model. Journal of Plastic Film and Sheeting, 17(3), 239-251.
Collins, P., Martin, P.J., Harkin-Jones, E.M.A., 2001. Experimental investigation of slip in plug-assisted thermoforming. In:
Proceedings of SPE ANTEC Conference, Dallas, TX, USA.
Collins, P., Harkin-Jones, E.M.A. and Martin, P.J., 2002. The Role of Tool/Sheet Contact in Plug-assisted Thermoforming.
International Polymer Processing, 17(7), 361-369.
Debbaut, B. and Homerin, O., 1998. 3D numerical simulation of thermoforming: prediction of thickness and extensions. In
Proceedings of SPE ANTEC Conference, Atlanta, GA, USA.
Edwards, S.F. and Vilgis, T.A., 1988. The tube model theory of rubber elasticity. Reports on Progress in Physics, 51, 55.
Hosseini, H., Vasilivuch, B.B and Mehrabani-Zeinabad, A., 2006. Rheological modeling of plug-assist thermoforming. Journal
of Applied Polymer Science, 101, 5.
Hummel, S.R. and Nied, H.F., 2004. A procedure for measuring biaxial viscoelastic behaviour of thermoplastics.
Experimental Mechanics, 44(4), 381-386.
Karamanou, M., Warby, M.K. and Whiteman, J.R. , 2006. Computational modelling of thermoforming processes in the case
of finite viscoelastic materials. Computer Methods in Applied Mechanics and Engineering, 195, 5220-5238.
Laroche, D. and Erchiqui, F., 1999. Experimental and theoretical study of the thermoforming of industrial polymers. Journal
of Plastic Film and Sheeting, 15(4), 7.
Martin, P.J., Tan, C.W., Tshai, K.Y., McCool, R., Menary, G., Armstrong, C.G. and Harkin-Jones, E.M.A., 2005. Biaxial
characterisation of materials for thermoforming and blow moulding. Plastics, Rubber and Composites, 34, 276-282.
Martin, P.J. and Duncan, P., 2007. The role of plug design in determining wall thickness distribution in thermoforming.
Polymer Engineering & Science, 47(6), 804-813.
Makradi, A. Belouettar, S., Ahzi, S. and Puissant, S., 2007. Thermoforming process of amorphous polymer sheets: modeling
and finite element simulations. Journal of Applied Polymer Science, 106, 7.
McCool, R., Martin, P.J. and Harkin-Jones, E., 2006. Process modelling for control of product wall thickness in
thermoforming. Plastics, Rubber and Composites, 35, 340-347.
Page 24
McCool, R. and Martin, P.J., 2010. The role of process parameters in determining wall thickness distribution in plug-assisted
thermoforming. Polymer Engineering and Science, 50(10), 1923-1934.
Nam, G.J., Ahn, K.H. and Lee, J.W., 2000. Three-dimensional simulation of thermoforming process and its comparison with
experiments. Polymer Engineering & Science, 40(10), 2232-2240.
Nied, H.F., Taylor, C.A. and Delorenzi, H.G., 1990. Three-dimensional finite element simulation of thermoforming. Polymer
Engineering & Science, 30(20), 1314-1322.
O’Connor, C.P.J., Martin, P.J., Menary, G.H., McConville, E., 2007. Modelling the thermoforming of polypropylene using
Abaqus. In: Proceedings of Abaqus User’s Conference, Daventry, UK.
Sweeney, J., Spares, R., Caton-Rose, P. and Coates, P.D., 2007. A unified model of necking and shearbanding in
amorphous and semicrystalline polymers, Journal of Applied Polymer Science, 106, 1095 – 1105.
Olley, P. and Sweeney, J., 2011. A multi-process Eyring model for large strain plastic deformation. Journal of Applied
Polymer Science, 119, 2246-2260.
Sweeney, J., Spares, R. and Woodhead, M., 2009. A constitutive model for large multiaxial deformations of solid
polypropylene at high temperature. Polymer Engineering and Science, 49(10), 1902-1908.
Wang, S., Makinouchi, A., Tosa, T., Kidokoro, K., Okamoto, M., Kotaka, T. And Nakagawa, T., 1999. Numerical simulation of
Acrylonitrile-butadiene-Styrene material’s vacuum forming process. Journal of Materials Processing Technology, 91, 219-
225.
Warby, M.K., Whiteman, J.R., Jiang, W.G., Warwick, P. and Wright, T., 2003. Finite element simulation of thermoforming
processes for polymer sheets. Mathematics and Computers in Simulation, 61(3-6), 209-218.
Figure captions
Figure 1: Major dimensions (in mm) of the yoghurt pot cavity, provided in section view.
Figure 2: Simulation of the plug-assisted thermoforming process.
Figure 3: Wall thickness measurements for the product, taken as a profile from the centre of the base of the
container along a vertical profile terminating at the flat lip of the receptacle.
Figure 4: A 1-dimensional schematic of the LSTC model for semi-crystalline polymers (ADJ = adjustable).
Fig. 5: Model [Sim.] vs. experimental [Exp.] data (equi-biaxial tests on the PP blend at 155C, enom =1.0/s, stretch
ratio of =3.0.
Figure 6: Evolution of the wall thickness (STH) distribution in millimetres during the plugging step (t=0.00 to 0.70s)
and blowing stage (t=0.70 to 1.10s), illustrated with a 180 slice of container.
Figure 7: Evolution of the wall thickness distribution during the plugging step.
Figure 8: Evolution of the wall thickness distribution during the blowing step.
Figure 9: The influence of friction coefficient of the plug sheet interface upon the wall thickness distribution
Figure 10: The influence of average plug speed upon the wall thickness distribution.
Figure 11: The effect of convective heat transfer film coefficient upon the wall thickness distribution.
Figure 12: The influence of the pressure cycle air mass flow rate upon the wall thickness distribution.
Table captions
Table 1: LSTC model material parameters for the polypropylene blend at 155C, enom =1.0/s, stretch ratio of =3.0.
Table 2: Temperature dependent materials parameters for the polypropylene blend at given temperatures, enom
=1.0/s, stretch ratio of =3.0.
Page 25