Best practice for benchmarking injection moulding simulation

Abstract

Injection moulding simulation benchmarking is the comparison of predicted and actual process and quality parameters. Best practice requires that the inputs and outputs from an injection moulding simulation agree to the actual conditions of a real life moulding machine, as closely as possible. The accuracy of injection moulding simulation is influenced by many factors such as: the modelling of the part geometry, runner and nozzle, model mesh type and density, mathematical finite element solution, material data, and process settings. Validation of injection moulding simulation requires good quality process measurements from a range of transducers, measuring process pressure and temperatures at different locations and including movement of the reciprocating screw to infer flow rate and accurate and repeatable measurement methods for part deflection. The objective of this paper is to give the reader an appreciation of the important issues, to ensure the inputs for an injection moulding simulation match the actual conditions of the real life moulding machine. The inputs for injection moulding simulation are reviewed in detail, firstly considering machine performance issues and secondly considering the simulations issues. This will provide the reader with knowledge and understanding to improve benchmarking procedures.

Full text

Best practice for benchmarking injection
moulding simulation
R. G. Speight, F. Costa, P. K. Kennedy*and C. Friedl
Injection moulding simulation benchmarking is the comparison of predicted and actual process
and quality parameters. Best practice requires that the inputs and outputs from an injection
moulding simulation agree to the actual conditions of a real life moulding machine, as closely as
possible. The accuracy of injection moulding simulation is influenced by many factors such as: the
modelling of the part geometry, runner and nozzle, model mesh type and density, mathematical
finite element solution, material data, and process settings. Validation of injection moulding
simulation requires good quality process measurements from a range of transducers, measuring
process pressure and temperatures at different locations and including movement of the
reciprocating screw to infer flow rate and accurate and repeatable measurement methods for part
deflection. The objective of this paper is to give the reader an appreciation of the important issues,
to ensure the inputs for an injection moulding simulation match the actual conditions of the real life
moulding machine. The inputs for injection moulding simulation are reviewed in detail, firstly
considering machine performance issues and secondly considering the simulations issues. This
will provide the reader with knowledge and understanding to improve benchmarking procedures.
Keywords: Injection moulding, Simulation, Validation, Accuracy, Error, Experiment
Introduction
Benchmarking is a technique used by engineers since the
very beginning of Computer Aided Engineering (CAE)
for injection moulding simulation. One of the earliest
documented benchmarks was by Colin Austin, founder
of Moldflow, in late 1970s during a trip to Japan. Colin
was lecturing on a seminar tour to promote Moldflow
technology, after presenting his seminar in Tokyo he
designed a simple runner balancing technique on a
mould for Toshiba. On his visit to the Toshiba plant the
next day, the engineers informed Colin, that they had
gone back to their factory the previous night, assembled
a mould with inserts in it, moulded samples and their
results agreed with the simulations. They were
impressed.
1
At that time there was a real problem with
regulating the flow of polymer melt into the runners and
cavities, with overpacking in one cavity and a short shot
in another cavity. Use of CAE for injection moulding
simulation has progressed from flow front prediction in
the early days, through to include full simulation of the
injection moulding process and its variants and asso-
ciated processes. The theory and practice for validation
of flow analysis software is described by Austin.
2
An important reason for benchmarking is to confirm
simulation results when users are first introduced to
injection moulding simulation applications. Typically an
extensive study is carried out on a representative mould,
machine and material combination, where filling pattern,
injection pressures and part warpage are reviewed. Once
results have been confirmed and practical considerations
are documented, the injection moulding simulation
applications are then implemented on a large scale, which
demands good communications between design engineers
and the shopfloor. The moulding industry is changing and
improving the link between design and manufacturing, the
flow of machine setup information and process response
continues to be critical. It is still true today that the three
key stages of the value chain: part design, mould design
and production, may actually be controlled by different
organisations, within a company (see Fig. 1). While these
organisations may be independent, it is vital that the
required inputs for the CAE simulation are validated
by part design, mould design, manufacturing and
production.
Objectives of benchmarking are to gain confidence
and experience in modelling, material and simulation
capability, and to then use simulation results to guide
future design decisions for which designers do not have
moulding data.
The following list outlines the key parameters that are
typically compared by design and process engineers:
(i) filling
a. injection nozzle pressure
b. filling pattern, weldlines and hesitations
c. packing cavity pressure decay
(ii) warpage
a. deflections, out of plane
Moldflow Pty Ltd, Kilsyth, Australia
*Corresponding author, email Peter_kennedy@moldflow.com
124
ß2008 Institute of Materials, Minerals and Mining
Published by Maney on behalf of the Institute
Received 12 October 2007; accepted 3 December 2007
DOI 10.1179/174328908X283203 Plastics, Rubber and Composites 2008 VOL 37 NO 2/3/4

b. in plane shrinkages.
The cooperation between the design engineer, process
engineers and machine setters will improve through
information exchange. Design induced limitations will
be reduced by automated tools that pass information
transparently between the shop floor and the design
engineers, as part of a product lifecycle management
system. Such a system will provide evidence of design
problems and reduce anecdotal information, with an
important area being the comparison of predicted and
experimental data. The key to the validation of such
systems is ensuring that there is good agreement between
the simulations and the real world. Figure 2 shows a
schematic diagram of the benchmark process. This
paper will focus on the preparation of a benchmark
and the importance of appreciating and understanding
the inputs.
Value of injection moulding simulation
The value of injection moulding simulation are clear.
3
Injection moulding and its variants are the most
successful area of simulation because:
(i) the process may be represented by a relatively
simple material model, namely, the generalised
Newtonian fluid which allows the viscosity of the
fluid to be a function of the rate of deformation
(ii) the governing equations may be reduced to a
simple form that is suitable for solution on
ordinary computers
(iii) injection moulding simulation has a high return
on investment.
Injection moulding demands more of part and mould
designers, as experimentation after the mould is built is
expensive in terms of time and money. Injection
moulding simulation is relatively inexpensive in terms
of project cost and offers great benefits to those using it
early in the manufacturing process. The above factors
bring a level of complexity to injection moulding that is
not present in other plastic forming processes. All these
aspects combine to make injection moulding an ideal
focus for simulation. Simulation of injection moulding
has a higher return on investment than simulation of
other plastic forming processes.
Accuracy of injection moulding
simulation
The accuracy of injection moulding simulation is
influenced by many factors. If the aim of the simulation
is the prediction of filling patterns and the location of
any weldlines or hesitation marks, then the accurate
representation of the geometry is the most critical factor.
Errors in the modelling of wall thickness are surprisingly
common. These may be due to design changes made
towards the end of the design cycle not being present in
the CAD model used as a basis for the simulation
model, or they may be due to the tooling not exactly
matching the final CAD design. In either case, when
benchmarking fill pattern predictions against short shot
samples, it is important to check key wall thicknesses in
the simulation model against actual moulded parts.
Another class of geometry inaccuracy which can arise is
the simplification inherent in the modelling representa-
tion chosen. Injection moulding simulation has tradi-
tionally been based on a midplane shell representation of
the part geometry. More recently, Dual Domain
technology from Moldflow has allowed a solid geometry
representation, but both of these modelling methods are
reliant on the assumption of laminar Hele-Shaw flow
and are not well suited to geometries with width to
thickness ratios less than four. For such geometries, a
true solid modelling representation of the part geometry
is required because the errors in the Hele-Shaw
approximations become too great.
If the aim of simulation is the prediction of the
pressure required to fill the moulding, then inclusion of
the runner, sprue and gating design is essential. In
addition, if the simulation injection pressure will be
compared to a measured nozzle pressure, then it is also
appropriate to account for the pressure drop in the
nozzle and contraction into the nozzle tip. This can be
done either my including the nozzle body and contrac-
tion into the simulation model (typically assigning a
property similar to hot runners) or by performing an air
shot experiment. An air shot experiment is when the
injection unit of the moulding machine is retracted away
from the mould and an injection shot performed at the
typically injection speed, but with polymer extruding
freely out of the nozzle tip rather than flowing into the
top of the sprue. The recorded nozzle pressure during
the air shot experiment may typically be between 10 and
40 MPa and should be added to the predicted injection
pressure if the simulation model begins from the top of
the sprue.
Good quality material data is also important when
comparing injection pressures. In this case, grade
specific viscosity data is required, including pressure
dependence if the injection pressures will be high
(.100 MPa) and including some model to represent
the entrance pressure loss, extensional viscosity or
2 Schematic diagram of benchmarking process
1 Three key stages of value chain (‘islands of automa-
tion’ operate independently)
Speightetal. Best practice for benchmarking injection moulding simulation
Plastics, Rubber and Composites 2008 VOL 37 NO 2/3/4 125

‘juncture loss’ which will occur at any strong contrac-
tions, such as entering narrow gates.
If the aim of simulation is to predict the final warped
shape of the ejected part, then the accurate reflection of
the process settings in the simulation is important. In
particular, packing time and packing pressure (or
pressure profile), cooling time and any relative difference
in coolant temperatures have a strong influences on the
amount of shrinkage and warpage. In moulding
practice, it is possible to halve the amount of warpage,
or even change the warpage direction for some parts
through changes in these process settings. Injection
speed, which has a strong influence on filling pressure, is
not usually a dominant influence on warped shape.
The availability of accurate material data will also
influence the simulation accuracy of warpage predic-
tions. For example, the use of grade specific pressure–
volume–temperature (PVT) and thermal conductivity
and specific heat data is required for accurate represen-
tation of shrinkage and warpage, as are the mechanical
properties of the solidified polymer such as modulus and
coefficient of thermal expansion. Shrinkage correction
coefficients derived from shrinkage measurements on
plaque mouldings for a range of processing conditions
and geometries can also be used to improve the accuracy
of shrinkage and warpage predictions.
4
The discretisation of the geometry (into finite difference
girds, finite elements or finite volume cells), will also play a
key role in simulation accuracy. Areas of changes in
thickness, such as the gate, should be discretised by a
minimum of three rows of elements to allow an adequate
representation of the thickness changes. Some discretisa-
tion methods allow the geometry surface to be altered to
fit the discretisation size, but this should be avoided in
injection moulding where small features such as the gate
can have critical influences on simulations. For example, a
geometry error in the sizing of a gate due to discretisation
would lead to a delayed prediction of gate freezeoff during
packing, which will have a strong influence on the
shrinkage and warpage of the part.
The mesh size must also be considered with respect to
the type of numerical solution being used. Traditionally,
for Hele-Shaw based laminar flow simulations, a mixed
finite element and finite difference method is used,
5
where a one dimensional finite difference grid is
employed in the thickness direction to capture tempera-
ture, viscosity and shear rate variations. Analogously,
when using a three-dimensional method, care must be
taken to ensure sufficient discretisation is present
through the thickness direction in all locations to
provide sufficient resolution to represent the same
property variations. Lower order or constant property
schemes such as the finite volume method require a finer
level of discretisation than finite element methods,
particularly if higher order interpolations are used in
the finite element formulation.
Verification and validation
At its most abstract, simulation involves using a computer
to solve partial differential equations with suitable
boundary conditions (mathematical model). In practice,
simulation involves the transformation of the mathema-
tical model into computer code. This involves the coding
of algorithms to solve the mathematical model and the
discretisation of the solution domain. Verification
involves testing that the intended equation or algorithm
and associated boundary conditions have been coded
faithfully. Verification is not related to the process being
simulated. It is a mathematical step that ensures no errors
are introduced in creating the simulation code.
Validation checks that the results of the coded
mathematical model have some agreement with physical
experiment. Validation therefore assesses how well a
particular model describes the physical process being
simulated.
Analysis accuracy refers to the accuracy of the results
obtained from simulation. Accuracy depends on the
fidelity of the mathematical model used in the simula-
tion, its implementation in the code, the appropriate
boundary conditions (these include processing condi-
tions), discretisation of the geometric domain in which
we seek a solution and material models that describe the
material’s physical properties. Verification is necessary
to ensure accuracy, however it is validation that
provides quantitative measures of accuracy.
Accuracy is therefore not a simple comparison of
simulation results and an experiment. Errors arise for
the following reasons:
(i) software error – incorrect coding of a mathe-
matical expression and/or its associated bound-
ary conditions
(ii) geometrical error – import of geometry and the
subsequent discretisation used to define the
computational domain does not reflect the real
part
(iii) material data error – material data is not
appropriate for the materials used to produce
the part
(iv) input error – processing conditions used in the
simulation differ from those used in the
manufacturing process
(v) post processing – manipulation of calculated
data for post processing
(vi) experimental error – experimental data is in
error often due to poor experimental technique,
poor instrumentation or transducers.
Injection moulding machine
benchmarking
An understanding of the fundamental concepts of the
injection moulding machine are required, and also
specific details of the actual machine being used. It is
important to have an appreciation of the links between
the manufacturing and simulation worlds, and that the
simulation should be treated as a virtual moulding
machine. Important factors to consider are: machine
capability, accurate information about screw move-
ments, check ring valve performance, material prepara-
tion (drying), nozzle pressure or cavity pressure, not
hydraulic pressure, shot to shot variations (stability),
venting, sensor types and reliability.
Moulding: machine capability
Moulding machines are continually advancing, and
there are a wide range of machines available. The
injection stage can be pressure controlled or velocity
controlled, most modern machines are velocity con-
trolled, in open or closed control loop. There are also
different methods of actuation: open loop digital
hydraulics, closed loop proportional valve, and closed
Speight et al. Best practice for benchmarking injection moulding simulation
126 Plastics, Rubber and Composites 2008 VOL 37 NO 2/3/4

loop servovalve. The packing or compression stage is a
transition stage from velocity control to pressure
control, typically this stage is not well controlled, until
the set holding pressure is achieved. In a recent
validation study at Moldflow, a moulding machine
which had a set flow rate of 93 cm
3
s
21
, actually only
achieved an average flow rates of 52 cm
3
s
21
, based on
screw movement calculations, due to machine response
capability. There was not indication from the moulding
machine that the set flow rate was not achieved. Another
Moldflow study showed that set holding pressures
were not being reached until gate freezeoff occurred,
due to the dynamics of the control system. An impor-
tant technical requirement for the benchmarking
process is a high speed data acquisition system,
monitoring the injection moulding process. This will
provide a accurate indication of a machines perfor-
mance, and this data can then be used as the basis of the
simulation inputs.
Moulding: screw movement and check ring
valve performance
Process monitoring, as well as providing details of
machine performance can also detect machine problems,
which should always be reviewed prior to the start of a
bench mark. Moulding machines that are subjected to
continuous use will have problems, worn check ring
values can result in incorrect flow rates and packing/
holding pressures. Figure 3 demonstrates the screw
moving in packing/holding up to 50% of the filling
stroke due to a worn check ring valve. This screw
movement continued even after the cavity pressure
traces (blue) indicated that the polymer in the cavity
had frozen.
Moulding: material drying
Material drying can cause process stability issues and
also differences in melt viscosity. In research by Khanna
et al.,
6
Nylon was dried for 17 h at 80, 110, 125 and
140uC, the lower temperature resulted in a higher
moisture content which resulted in a viscosity difference
up to 500% at low shear rates and 200% at higher shear
rates.
Moulding: nozzle pressure
The ideal relationship between nozzle melt pressure and
hydraulic injection pressure is shown below
Nozzle melt pressure~(APiston=AScrew )
|hydraulic injection pressure
where A
Piston
is area of hydraulic piston (m
2
) and A
Screw
is area of screw (m
2
)
The ratio of piston to screw areas is referred to as the
screw intensification ratio or gain, and is typically
quoted as being approximately equal to 10, it may in
fact vary considerably depending on screw and piston
geometries. In practice the apparent screw intensifica-
tion ratio may vary due to:
(i) compressibility of the hydraulic oil due to
temperature changes
(ii) frictional effects between the screw and barrel
(iii) the influence of polymer melt compressibility
during the filling process.
Figure 4 shows nozzle melt pressure measured using a
Dynisco nozzle melt pressure sensor and nozzle melt
pressure derived from multiplication of the theoretical
screw intensification ratio of hydraulic pressure, show-
ing a 10 mPa difference.
7
In practice, nozzle melt
pressure is often derived from hydraulic pressure, due
to difficulties of installing nozzle melt pressure sensors.
Moulding: shot to shot
Typical injection machine controllers typically have very
limited capabilities for data overlaying, making it
difficult to determine shot to shot variations. Figure 4
demonstrates differences in nozzle melt pressure, it is
3 Screw position in packing/holding stage, for worn
check ring valve 4 Screw nozzle melt pressure and hydraulic pressure
comparisons
5 Overlaid screw position and injection pressure profiles
Speightetal. Best practice for benchmarking injection moulding simulation
Plastics, Rubber and Composites 2008 VOL 37 NO 2/3/4 127

important to note that the curves are the mean of 25
cycles, and the coefficient of variation is displayed, so
that the quality of the data is clearly visible. Figure 5
shows successive pressure sensor and screw position
traces (short glass fibre reinforced polyethylene ter-
ephthalate (PET)/polybutylene terephthalate (PBT)
blend, fibre filled) immediately following a change in
processing settings (injection speed). The screw position
trace appears stable, but the pressure traces are not
stable, this is a material effect, not a machine effect and
is probably a result of an unoptimised melt preparation
stage.
Figure 6 shows the pressure traces once the process
has stabilized, the successive pressure traces typically
stabilise after 5 to 10 shots.
Moulding: sensors
A machine independent process monitoring system for
data acquisition is essential for injection moulding
validation. The system has to be capable of capturing
the true machine and process dynamics. Coates and
Speight
8
describe the full range of sensors that are
available for the injection moulding process. Sensor
selection is determined by considering multiple factors,
such as cost, robustness/reliability, dynamics, repeat-
ability, linearity, influence of external indirect factors
(such as temperature on a pressure sensor), size of
measuring head and ultimately suitability. Figure 7
shows cavity pressure readings for three equidistant
pressure sensors in a 1 mm rectangular cavity, filled with
polypropylene at constant flow rate. The flow front is
expected to reach cavp3 one second after cavp2, but it is
clearly visible that the pressure rises earlier. The pressure
sensor is performing correctly, but is measuring the air
pressure build up due to poor mould venting.
Moulding practice: warpage
Flow front position, injection pressure and warpage are
common benchmarking parameters. Warpage bench-
marking can be influenced by coolant temperatures,
packing pressure settings, shot to shot variations
(stability) and a reliable and repeatable way of measur-
ing warpage. Warpages measured within the tool makers
mould tolerance, i.e. ¡10 mm, are really outside the
bounds of the simulation world.
Injection moulding simulation
benchmarking
Injection moulding simulation benchmarking of CAE
requires an appreciation of simulation technology and
knowledge of the assumptions made in the simulation.
Simulation: filling inputs
The most important input that influences filling pattern
is mould geometry, so an accurate representation of the
mould geometry is essential. The most important inputs
that influence injection pressure are:
(i) geometry,
(ii) switchover from velocity to pressure control stages
(iii) material viscosity
(iv) injection speed (profile or constant).
Simulation: geometry
Often time is spent analysing simulation results of an
intricate feature, only to find that the feature was not
modelled correctly. Models are often assumed to be
correct. It is always advisable to have a moulded part to
review, to check for any obvious errors, it is also
recommended to consider the life of the tool, as several
modifications may have been performed on the tool. The
model used for the injection moulding simulation may be
from the part design, so the tool maker’s shrinkage
allowance will not be included in any dimensions. Nozzle,
runner and gate geometries are not always included in a
simulation model, these features can have a considerable
effect on simulation results, particularly pressure to fill.
Simulation: switchover
Switchover from velocity to pressure control is set by:
screw position or time, in this case the full geometry must
be modelled accurately, or volume percentage filled or
automatic. Moulding simulation often uses the latter and
it is essential to check that this is a reasonable approxi-
mation of the settings used on the moulding machine.
Simulation: viscosity
In the situation when a material is not present in the
material database it is common practice to choose
‘similar’ material based on melt flow index (MFI) or
viscosity, using the same polymer family, similar levels of
6 Overlaid screw position and injection pressure profiles
7 Cavity pressure readings for three equidistant pressure
sensors in 1 mm rectangular cavity, filled at constant
flow rate, polypropylene
Speight et al. Best practice for benchmarking injection moulding simulation
128 Plastics, Rubber and Composites 2008 VOL 37 NO 2/3/4

filler and perhaps the same manufacturer. It may be
expected that a 10% difference in filler weight percentage
gives y10% difference in pressure prediction. If using
material data from a different manufacturer expect up to
40% difference in pressure. This is because viscosity is also
indirectly affected by the material’s thermal properties.
Simulation: injection speed
It is important to set the correct injection velocity (flow
rate profile). Figure 8 shows the experimental and
simulation results compared, where the flow rate used in
simulation was determined from readings on two cavity
pressure sensors. The difference in position between the
sensors and the time difference for the flow to reach the
sensors was used to determine an injection speed for
the simulation. However injection speed calculated this
way shows poor agreement with measured pressures.
Figure 9 shows the results of simulation using the actual
screw velocity profile of the machine used in the ex-
periment. The prediction of pressure is greatly improved.
Simulation: packing pressure
Injection moulding simulation has many built-in fea-
tures to make running a simulation easy. One feature is
the automatic packing/holding pressure profile, which
uses by default 80% of the maximum injection pressure
for 10 s. Often the actual pressure profile is overlooked.
Figure 10 illustrates the effect of using an arbitrary
pressure profile on warpages results. This example serves
to highlight the need for users to appreciate any settings
that are determined by the simulation program for the
purpose of making the program easier to use.
Conclusions
Benchmarking of injection moulding simulation requires
a systematic approach to problem elimination when
comparing simulation and moulding practice.
Many different factors may be the cause of error.
1. Machine capability and response time.
2. Material preparation, characterisation and stability.
3. Measurement methods (pressure or deflection).
4. Geometry inaccuracies.
5. Variation in process settings used in experiment
and inputs to simulation software.
Awareness of these points is essential when compar-
ing simulation results to those obtained on moulding
machines.
Fortunately, even without perfect agreement, simula-
tion can provide great insight into performance
sensitivities to process, geometry and material. These
sensitivities may be used by engineers to improve pro-
duct design and process settings for actual production.
8 Experimental and predicted injection pressures 9 Experimental and predicted injection pressures
10 Default and actual packing/holding pressure profiles
Speightetal. Best practice for benchmarking injection moulding simulation
Plastics, Rubber and Composites 2008 VOL 37 NO 2/3/4 129

Acknowledgement
This paper is based on a presentation at the Polymer
Process Engineering Conference held in Bradford, UK,
in July 2007.
References
1. B. Whitford: ‘Success through excellence, a professional innova-
tor’, 389; 1992, Perth, Beaumont Publishing House.
2. C. Austin: Proc. ANTEC 90, Dallas, TX, USA, May 1990, SPE,
208–211.
3. A. R. Thomas, R. G. Speight and P. K. Kennedy: J. Plast. Rubber
Compos., 2005, 34, (5/6), 211–221.
4. P. Kennedy and R. Zheng: Proc. ANTEC 2002, San Francisco,
CA, USA, May 2002, SPE, 593–599.
5. C. A. Heiber and S. F. Shen: J. Non-Newt. Fluid Mech,1980,7,1–32.
6. Y. P. Khanna, P. K. Han and E. D. Day: Polym. Eng. Sci., 1996,
36, (13), 1745–1754.
7. R. G. Speight, P. D. Coates, J. B. Hull and C. Peters: Proc. I Mech
E Part E, 1997, 211E, (2), 115–128.
8. P. D. Coates and R. G. Speight: Proc.I Mech E Part B,1995,209B,357.
Speight et al. Best practice for benchmarking injection moulding simulation
130 Plastics, Rubber and Composites 2008 VOL 37 NO 2/3/4