You are on page 1of 10

Optimization of Injection Molding Design.

Part II: Molding Conditions Optimization


IOANNIS PANDELIDIS and 9 I N ZOU

D e p a r t m e n t of Mechanical Engineering
University of Maryland
College P a r k , M a r y l a n d 20742
The quality of a n injection molded part is affected by many factors. These
include geometric parameters associated with the mold design and the cooling
system design as well a s process parameters such as the molding conditions
during the filling phase. In the companion paper, the problem of automatic
optimization of gate location was addressed. In this paper, a methodology for
molding condition optimization is presented. The optimization problem can be
broken into three parts. A n approximate feasible molding space (AFMS) is first
determined to constrain the search space for the optimization algorithm. Quality
is quantified a s a function of flow simulation outputs and constitutes the objective
function that must be minimized. The resulting optimization is solved by iterative
search in the constrained space based on numerical optimization algorithms. The
proposed methodology is not dependent on any particular simulation package and
may be applied for any thermoplastic material and any complex mold geometry.
INTRODUCTION

he quality of a n injection molded part is affected


by many factors. These include geometric parameters associated with the mold design and the cooling
system design a s well a s process parameters such as
the molding conditions during the filling phase. In
the companion paper ( 1 ) . the problem of automatic
optimization of gate location was addressed. In this
paper, a methodology for molding conditions optimization is presented.
The optimization problem can be broken into three
parts. An approximate feasible molding space
(AFMS) is first determined to constrain the search
space for the optimization algorithm. Quality is quantified a s a function of flow simulation outputs and
constitutes the objective function that must be minimized. Quality measures were discussed in the companion paper (1).The resulting optimization is solved
by an iterative search in the constrained space based
on numerical optimization algorithms.
The selection of appropriate molding conditions in
the injection-molding process is a nontrivial task. A
quick examination of the general effect of each variable will illustrate the point. A n increase in melt
temperature causes a decrease in melt viscosity,
which results in reduced pressure requirements and
reduced stresses. On the other hand, high melt temperature may increase the possibility of material
burning and will also increase cooling time. Increasing the mold temperature reduces heat losses, and

the maximum temperature difference at the end of


the filling phase may be reduced. However, a high
mold temperature increases cooling time. Short fill
times require higher pressure because of the higher
required flow rate, resulting in higher shear rate and
shear stress. However, the temperature difference a t
the end of fill time decreases because the material at
the end of the flow will have a shorter time differential in cooling times a s compared with the gate. Then,
if the fill times are too long, pressure will increase
because the plastic temperature decreases and the
viscosity increases accordingly. It is clear that some
optimization to balance the conflicting processing
parameters is required, and in general, both quality
and cost considerations must be taken into account.
It is very difficult, however, for a molding engineer
to obtain optimum molding conditions in a complex
part strictly by trial and error, since the interrelationship of the parameters is very complex.
The proposed optimization methodology alleviates
the need for manual trial and error. In Determination
of Approximate Feasible Molding Space (AFMSI,
a n automatic determination of the feasibility
constraints and the initial molding conditions is
presented. Optimization of Molding Conditions
presents the mathematical formulation of the
problem as a constrained optimization and gives
the algorithms for the search procedure. Applications are given for a complex industrial mold
in the section titled Application, which is followed
by a summary.

POLYMER ENGINEERING AND SCIENCE, MID-AUGUST 7990, Vol. 30, No. 75

883

Ioannis Pandelidis a n d Qin Zou

The maximum flow length is estimated by

DETERMINATION OF APPROXIMATE FEASIBLE


MOLDING SPACE (AFMS)

Lj = N j * D

Determination of the approximate feasible molding


space (AFMS)will be based on a geometric approximation of the original part by a center-gated model.
The approximation is done so that the radius of the
center-gated model equals the maximum flow length
of the part. We developed a "pseudoflow" algorithm
to estimate the maximum flow path given the geometric description of the part in terms of a finite
element mesh. This algorithm can be described as

(2)

where N'is the maximum number of flow fronts and


D is the average nodal distance.
For the multigated part, each flow may fill different
volumes because of the different gate size. The flow
proportion f r for each flow can be defined as the
percentage of total flow from the sprue. Therefore,
the time interval ATr for flow front i should satisfy
the following equation:
AT,

Pseudoflow Algorithm

For a chosen injection node, all its adjacent nodes


are found and assigned a n approximate fill time.
These nodes are then considered as the first
pseudoflow front. Next, all the adjacent nodes of
this flow front are found and assigned another
approximate fill time. This is the second pseudoflow front. Pseudoflow is shown in Fig. 1. The
process continues until all nodes are filled. Notice
that no matrix solutions are required, so the algorithm is much faster t h a n a true finite element
derived flow front.

- AT,

where ATr and A T are the time intervals for flow i


and flow j respectively, and a n d f , are the flow
proportions for flow i and flowj.
To determine the AFMS, the recommended mold
temperature and melt temperature ranges are obtained directly from the material data base for the
specific material. According to the stress, pressure,
and temperature relations, eight boundary molding
conditions can be defined by scanning the fill time.
These eight boundary points determine eight halfspaces the intersections of which define the AFMS
for the part.
The flow rate Q for a center-gated disk mold with
constant flow rate can be defined as a function of
the molded part thickness, the flow length, the fill
time, and diameter of the runner (2).
fi

The maximum flow length is finally obtained. The


distance between two fronts can be estimated as the
average distance between the nodes, D , which can
be calculated as follows:

Q = 8.rrH/(t(R2- DF))

where m is the number of nodes that are adjacent to


the injection node and X u , Y o ,and Z, are the coordinates of the node.

(4)

where thickness = 2 H , flow length = R , fill time = t,


runner diameter = D,.
The corresponding fill pressure equation is given
as (31

+ 2Qt/(rHDF))

Po = (3Qp/8aH3)ln(l

(5)

where p is the viscosity. The viscosity can often be


adequately characterized by a first order model:
= AYBeCT

(61

where p is the viscosity and A, B, and C are experimentally determined constants. is the shear rate
and T is the melt temperature, which can be defined
as

T = T,,,

- klek2'

+ Tfrrc

(7)

where k l a n d kz are the logarithmic cooling factors


and T,r,, is a factor for heat generated by friction.
T,,(, is the initial melt temperature (2). The stress
can be calculated by the following equation:

where r is the shear stress, a n d finally the shear rate


can be calculated by ( 4 )

Note that the model presented above can only be used


a84

POLYMER ENGINEERING AND SCIENCE, MID-AUGUST 1990, Vol. 30, No. 15

Optimization of Injection Molding Design. Part I1


for approximate estimation of the boundaries of the
feasible molding space. The final optimum molding
conditions will be based on a full finite difference
model.
According to the above equations, MF (an option of
Moldflow simulation) is called iteratively for the different combinations of melt a n d mold temperatures.
For each combination, the program scans the fill time
to determine its minimum and maximum boundaries.
To use the AFMS program, a material data file must
also be created for the process. This contains material data such as the material code a n d the corresponding mold and melt temperature ranges, the
maximum melt temperature (plastic burn temperature), and maximum allowable shear stress.
Eight boundary molding conditions can be determined by applying constraints of maximum shear
stress, maximum pressure, and temperature distribution. These boundary molding conditions define
the AFMS.
The following is a n example of the program. The
material used in this sample is Nylon 6. The recommended mold and melt temperature ranges are 60 to
90C and 240 to 280C respectively. The maximum
allowable shear stress is 500,000 Pa, a n d the maximum melt temperature is 320C. The maximum flow
length of the part is 200.00 mm, and the average
thickness of the part is 6 mm.
First, the mold temperature was chosen to be 60
and 240C for the melt. Both a r e minimum temperatures. The results are shown in Table I.
As can be seen from the Table 1 , at short fill time,
the pressure is very high because of the high flow
rate required. A s the filling slows down, the pressure
drops. However, if the fill time is too long, the plastic
will get too cold at the end of flow and viscosity will
increase, which will cause the pressure to rise again.
I t is typically suggested not to mold the part slower
than this fill time (2).The limitation of the maximum
pressure is dependent on the injection machine. A
maximum pressure on a n injection machine is approximately about 138 MPa (20,000 psi). Therefore,
100 MPa is usually used as design limit.
Next note the temperature at the end of flow. If this
temperature becomes too cold, there will be problems
at the end of flow such as bad finish, weld lines, etc.
Since the freeze temperature for Nylon 6 is 230C.

temperatures below 235C a r e considered unacceptable. On the other hand, to achieve a reasonably
uniform temperature distribution throughout the
cavity, the final temperature should not be more than
5 K higher t h a n the temperature entering the cavity,
and not less t h a n 20K below the temperature entering the cavity.
A last consideration is the maximum shear stress
level, which may not exceed the maximum allowable
stress for that particular material. Two boundary
molding conditions are shown here:
mold temp. = 60"C, melt temp. = 240C.
1.5 s 5 fill time 5 2.5 s.
The results for the other combinations of minimum
and maximum mold temperature with minimum and
maximum melt temperature are shown in Tables 2,
3, and 4. By the end of the process, eight boundary
molding conditions can be obtained as shown in
Table 5.
Table 2. Mold Temperature = 90C and Melt Temperature =
240OC for Nylon 6.
Fill Time

(s)

Pressure
(MPa)

Shear Stress
(MW

Temperature
("C)

0.20
0.30
0.50
0.70
1.oo
'1 5 0
'2.00
'2.50
3.00

38.60
32.50
26.20
22.70
19.60
16.80
15.30
14.50
14.70

1.239
1.019
0.798
0.679
0.574
0.474
0.415
0.374
0.344

256.0
253.0
249.0
247.0
244.0
241.O
238.0
236.0
234.0

*feasible fill time

Pressure
(MPa)

Shear Stress
(MPa)

Temperature
("C)

0.70
1.oo
*1.50
'2.00
*2.50
'3.00

21.70
18.70
15.90
14.20
13.30
12.70

0.642
0.542
0.448
0.391
0.353
0.325

247
244
241
239
237
235

'feasible fill time

Table 3. Mold Temperature = 60C and Melt Temperature =


280C for Nylon 6.
Fill Time
(s)

Pressure
(MPa)

Shear Stress
(MW

Temperature
("C)

0.30
*0.50
"0.70
*1.oo
'1.50
"2.00
*2.50
*3.00
5.00

18.20
14.40
12.30
10.50
8.80
7.80
7.10
6.60
5.40

0.509
0.399
0.340
0.287
0.237
0.207
0.187
0.172
0.136

286
284
282
279
276
273
270
268
258

* feasible fill time

Table 4. Mold Temperature = 90C and Melt Temperature =


280OC for Nylon 6.

Table 1. Mold Temperature = 6OoC and Melt Temperature =


240OC for Nylon 6.
Fill Time

(9

Fill Time

(9

Pressure
(MPa)

Shear Stress
(MW

Temperature
("C)

0.30
*0.50
*0.70
'1 .oo
'1.50
'2.00
'2.50
*3.00
'5.00
7.00

17.90
14.20
12.20
10.30
8.60
7.60
6.90
6.40
5.30
4.70

0.502
0.393
0.335
0.283
0.234
0.204
0.184
0.169
0.134
0.115

286
284
282
280
277
274
272
270
261
253

* feasible fill time

POLYMER ENGINEERING AND SCIENCE, MID-AUGUST 1990, Vol. 30, NO. 15

885

Ioannis Pandelidis and Qin Zou


the eight boundary points. It can be calculated by
following:

Table 5. Eight Boundary Molding Conditions.


Mold Temp. ("C)

Melt Temp. ("C)

Fill Time (s)

6O.(rnin)
6O.(rnin)
90.(rnax)
90 .(rnax)
6O.(rnin)
60.(rnin)
90.(rnax)
90.(rnax)

240.(rnin)
240.(rnin)
240.(rnin)
240.(rnin)
280.(rnax)
280.(rnax)
280.(rnax)
280.(rnax)

1.5O(rnin)
2.50(rnax)
1.50(min)
3.00(rnax)
0.50(rnin)
3.00(rnax)
0.50(rnin)
5 .OO(rnax)

va

These eight boundary molding conditions define


eight half-spaces, which determine a feasible molding space. The four upper points of the boundary
define two planes, and the four lower points of the
boundary also define two planes, which specify the
half-spaces at the top a n d the bottom respectively.
Eight inequalities in total define the AFMS:
Tmold miri 5 Tmold 9 Tmold max

(10)

Trnelt n u n 5 TmeIt

(11)

Tmelt mnx

+ CITmold + ~2Tmelt+ c3 5 0
t m a x + C4Tmold + CSTmelt + CgTmelt 5 0
tmm + dlTmoid + dPTmplt + d3 5 0
t m , n + d4Trriold + dsTmelt + d g 5 0
tnmx

1;;;

s'rm"I"

rnax

Tmola.mm

Determination of Initial Molding Conditions Based


on AFMS

(12)
(13)
114)
(15)

where T m o l d mand
l n Tmoldrnax
are the minimum and
maximum mold temperatures respectively, Tmelr
m,n
and TmeltmaX
are the minimum and maximum melt
temperatures respectively, and c,, d , are coefficients
determined by the eight boundary points,
The selection of appropriate initial molding conditions for the optimization program is important, since
a poor selection could lead to a local rather t h a n a
global optimum solution. I t is preferable that these
parameters are given by a molding expert. When this
information is not provided, default values for the
initial molding conditions are chosen automatically
as the center of the approximate feasible molding
space.

sTrneu r

dTmeltdTmolddt.

(16)

Trne~tmm

This may now be used as a basis for comparison with


other materials. A s a n example, poly(ethy1ene terephthalate) may be considered as alternative material to
Nylon 6 with calculated eight boundary points as
shown in Table 6. The cross sections of the AFMS at
low, medium, and high mold temperatures are shown
in Fig. 2.
Notice that at high melt temperature we typically
have a longer fill time range. This is true because the
lower viscosity of the material at that temperature
allows for a shorter fill time. By the same account,
the plastic takes a longer time to cool and, therefore,
i t i s possible to have longer fill times. It should be
remembered that this is only a first pass analysis
and other considerations may limit the indicated
AFMS.
The volume of the AFMS for polyethylene terephthalate was about 7100 and the volume of the AFMS
for Nylon 6 was about 5400. Therefore, in this example, polyethylene terephthalate was recommended. Notice the indicated results a r e true only for
that particular chosen geometry and are not a n indication of superiority of one material versus another
in general (6).

OPTIMIZATION OF MOLDING CONDITIONS


The quality measures and objective function to be
minimized are the same as we discussed in "Part I:
Gate Location Optimization" (1). To make the design
acceptable, some feasibility constraints must be satisfied. Therefore, the optimization of molding conditions can be described as a three-dimensional constrained optimization problem, which can be defined
as
minimize:
F ( X ) = aTd

+ PNmp +

(17)

subject to:

Material Selection Based on AFMS


During the selection process for a n appropriate
material for a part, the designer first considers the
chemical, physical, and mechanical properties of the
material in order to satisfy the design requirements.
However, it is often the case that several different
materials will meet the design requirements. In this
case, the volume of the AFMS can be used as one of
the considerations of material selection.
A small AFMS implies that there may be difficulties
with controlling the process on line, because of restricted latitude in the choice of molding conditions.
Therefore, if all other considerations are equal, the
material that allows for a large AFMS is preferable
to a n alternative material.
The volume of the AFMS can be calculated from
886

Table 6. Eight Boundary Molding Conditions for Poly(ethy1ene


Terephthalate).
Mold Temp. ("C)

Melt Temp. ("C)

dO.(rnin)
80.(rnin)
11O.(rnax)
11O.(rnax)
80.(min)
80.(rnin)
1 1O.(rnax)
11O.(rnax)

270.(rnin)
270.(min)
270.(rnin)
270.(rnin)
320 .(rnax)
320.(rnax)
320.(rnax)
320.(rnax)

Fill Time (s)


~~

1.OO(rnin)
3.00(rnax)
1 .OO(rnin)
5.00(rnax)
O.PO(rnin)
5 .OO(rnax)
0.20(rnin)
5.00(rnax)

POLYMER ENGINEERING AND SCIENCE, MID-AUGUST 1990, VOI. 30, NO. 15

Optimization of Injection Molding Design. Part I1

NYLON 6

'
E
L
T

300

300

280+

240..
220..

(C)
i

E
L

308..

T
E

28B,,

280,.

T 240-.
E
220..
P
(C)

T 240..
E

320..

-I

T 280+

T 280
E
260

P
(C

T 3@0

260-.
'I

320

T 380.,

220..

(C 1

(C)

280..

L
T 268..

2'60..

320..

300

T 260,.

260-,

E
P

P
(C

4-

:-+

'

Fig. 2. Automatic generation of a part AFMS f o r two materials.

where Td is the maximum temperature difference at


the end of the filling stage, Noup is the percent of
overpacked elements, and Nfh is the percent of frictional overheated elements. The coefficients a , beta,
and y are the weighting functions for Td,Noup,and
N,, respectively. These quality measures are calculated from the flow simulation temperature, time,
and pressure distributions at the end of filling, and
have been discussed in detail in the companion paper
on gate location optimization (1).X = [X,,X,, &]'are
design variables. In molding condition optimization,
the design variables are molding conditions: mold
temperature, melt temperature, and fill time respectively. g , , g2 and g3 are called inequality constraints.
P,,
is the maximum required, and PMc i s the maximum pressure the machine can apply. T,,
is the
maximum shear stress in the part, and 7 M A is the
maximum allowable shear stress of the specific material. Td is the maximum temperature difference
found in the part, and TZ is the maximum allowable
temperature difference.
Equations 15, 16, and 17 define the constraints

for the maximum injection pressure, maximum allowable shear stress, and temperature difference respectively. Equation 18 is called side constraints.
Xt(i = 1 , 2, 3 ) are the lowest allowable mold temperature, melt temperature, a n d fill time, and X p ( i = 1 ,
2, 3 ) are the highest allowable mold temperature,
melt temperature, and fill time respectively.
The augmented Lagrange multiplier (ALM) method
is used to modify the constrained problem into a n
unconstrained problem first (see Appendix A). Next
the pseudo objective function %(X)created by ALM is
minimized based on the sequential unconstrained
minimization technique (SUMT). This method will be
presented next.

Sequential Unconstrained Minimization


Technique
In minimizing a n unconstrained F ( X ) , the sufficient conditions for F ( X * ) to be a local minimum are
that

POLYMER ENGINEERING AND SCIENCE, MID-AUGUST 1990, Vol. 30, No. 15

VF(X*) = 0

(22)
887

Ioannis Pandelidis and Qin Zou

where

defined as
S" = -V@(X")

+ PnS"-l

(26)

where p" is a scalar number and is defined as


V@(X").V*(X")
Pn

I 8x3 J
and that the Hessian matrix, H ( X ) at X" is positive
definite. X" is a global optimum design, if and only if
F@*) = 0 and H ( X ) is positive definite for all X. In
other words, X* ensures the design to be a global
optimum only when the F ( X )is convex. In most cases
in molding condition optimization, F ( X )is not convex
and, therefore, the identified minimum is not guaranteed to be the global optimum.
A s discussed before, to solve unconstrained optimization problems requires finding a vector X" such
that the gradient of the objective function F(X*) becomes zero. However, it is almost impossible to infer
X* directly by solving the equation VF(X) = 0 since
VF(X) may be a highly nonlinear function. In the case
of molding condition optimization, the objective function F ( X ) is a function of a flow simulation outputs,
so we do not even have an explicit form for F ( X ) .
Therefore, instead of solving VF(X) directly, F ( X ) is
minimized by an iteration procedure. Beginning with
an initial design vector Xo, a new design vector X' is
chosen in some way, which will give a lower value of
(X').The search iterations for this type of unconstrained problem can be defined as follows:

X""

= 2("

+ (YS"

(24)

where n is the iteration number, S is a vector representing the search direction in the design space, and
(Y is a scalar number that represents the distance to
move in the S direction.
It can be seen that this optimization algorithm can
be divided into two basic parts. One is finding a
search direction, S , among which the objective function will be minimized. Another is finding the scalar
number 01" that defines a proper distance of moving
in direction S.

= V@(X"-') -V@(X"-l).

Therefore, in our molding condition optimization, we


define V@(X)as

V@(X)=

Iax3 J
where

-a@(X)- @(XI.Xz, X3 + Ax31 - @(XI,XZ.X3)


ax3

where Pn is a scalar number defined as

(33)

There are several methods available for defining a


search direction for F ( X ) (7-9). In this paper, a conjugate direction method is employed to perform the
search. This method, also called Fletcher-Reeves
method, is a first-order method that uses the gradient
information of the objective function to determine a
search direction S (10).Relatively fast convergence
can then be achieved, which is important because of
the time-consuming functional evaluations involved
with using flow simulation software. The initial
search direction So uses the steepest descent direction, which is defined a s
(25)

The conjugate direction on subsequent iterations is


888

(31)

and X1 is the mold temperature, X, is the melt temperature, and Xs is the fill time.
Theoretically, AX should be very small. However,
the flow simulation package is not sensitive to small
changes in mold temperature and melt temperature.
AX, and AXz are chosen to equal ZK, and AX3 is equal
to 0.015 (s)after running several cases to test the
sensitivity of mold temperature, melt temperature,
and fill time. Because the flow simulation package is
much more sensitive to fill time than to mold temperature and melt temperature, it is necessary to
make a small modification in the search direction.
The search direction is defined as

Defining a Search Direction S With Conjugate


Direction Method

s o = -O*(XO).

AX3

V@'(x)is defined as

where Fa,Fbrand F , are experimentally determined


constants about their respective sensitivities. To
keep the search direction in a descent direction, we
have to restart the process periodically during the
optimization because of the nonquadratic nature of
the objective function or numerical imprecision. The

POLYMER ENGINEERING AND SCIENCE, MID-AUGUST 1990, Vol. 30, No. 15

Optimization of Injection Molding Design. Part 11

three unknowns:

condition for restarting the process is defined as


S".V@'(X") 2 0.

a.

(35)

+ al + a2 = $*(1)
+ 4al + 16a2 = @*(4).
a.

If S".V@'(Xn)
2 0, it implies that the search direction
S" can not improve the objective function any more.
Therefore, it should be set equal to -V+(X") again.
The conjugate direction method algorithm can be
defined a s follows:

a.

F(X + as).

11. If convergence is reached, stop. Else go to 7.


Polynomial Approximationfor Finding Optimum
Distance (I* in Direction S"

Since X" = X"-' + asn,minimizing the objective


function @(X)is reduced to a one-dimensional problem for finding the optimum a in the given search
direction S", which has the form

The optimum a

a* satisfies

(37)

where S" is the given search direction.


Many methods are available for solving this type of
problem such as the golden section method, polynomial approximation method, and Newton's method.
For our problem, the polynomial approximation
method is the most effective technique for finding
the optimum a. The basic idea of polynomial approximation is to calculate the function F(X) at several
different points and then fit a polynomial F ( X ) to
these known points. The minimum of the polynomial
F(X)is considered a good approximated minimum of
the true function F(X).
Because each objective function evaluation involves flow simulation, which is very time consuming, a three-point polynomial approximation is used
to approximate the function +(a) as a quadratic
polynomial that has the form
+*(a)= a.

+ a l a + a2a2.

a2

(40)
(41)

(42)

- 4@*(1)
@*(4)
12

(43)
(44)

The optimum a can be solved as


=

--a1
2a2

(45)

and is considered a good approximation of the optimum a* for +(a).

Convergence
It is very important to set appropriate criteria for
terminating the search, since the efficiency and reliability of the optimization process are greatly affected by these considerations. To ensure that the
optimization process is stable, several termination
criteria are used in our optimization process.
Maximum Number of Iterations

The maximum number of iterations will ensure


that the search process will not continue to iterate
indefinitely by stopping a search process when the
number of iterations, P , exceeds a large prespecified
number, P,. This may happen when the process is
extremely slow because of numerical or algorithmic
difficulties, nonconvergence, or simply because of
some programming errors. In our program for optimizing molding conditions, P, is chosen as 100.
Absolute Change of Objective F u n c t i o n

The second termination criterion is the absolute


difference of the objective function F(X) on successive iterations. In this case, the convergence can be
defined as follows:
IF(Xp;)- F(XP-')I 5

(46)

where t, is a specified tolerance. In our program, this


criterion is chosen a s tA = 0.051F(X0)I.
Relative Change of Objective Function

The third termination criterion is the relative


change of the objective function F(X) between successive iterations. It can be defined as
(47)

(38)

The @(a)can be calculated at three points: a = 0, N =


1 , and a = 4. We have three linear equations with

= 9*(0)

4
1
al = -@*(l)
- @*(O) + -aJ*(4)
3
12

x+

(39)

Solving for coefficients,

Algorithm for Conjugate Direction Method

1 . Choose Xo.
2. go+ X,V F = VF(X). a = VF.VF.
3. -VF + SF.
4. Call subroutine to find p* which minimizes
F(X + a*S).
5. If p = 0 then stop, else go to 6.
a*s-+ 15.
6.
a
7. VF(X) + VF, V F . V F + b, y = -.
b
(ISn-' + S", b + a, S " . V F + Slope.
8. -Vf
9. If Slope 2 0 then go to 3, else go to 10.
10. Call subroutine to find a to minimize

= cP*(O)

where t, is a specified relative change. Instead of


dividing by the magnitude of F(XP),dividing by the

POLYMER ENGiNEERlNG AND SCIENCE, MID-AUGUST 1990, Vol. 30, No. 15

889

Ioannis Pandelidis and Qin Zou

maximum of the magnitude of F ( X P )and 10'' avoids


dividing by zero when the objective function F ( X )
reaches zero. Both absolute and relative change convergence criteria are used to ensure that the convergence is identified whether the magnitude of F ( X ) is
very small or very large. The flow chart for molding
condition optimization is shown in Fig.3.
APPLICATION
A complex industrial drill housing is presented as
a n example. This is the same drill housing used in
the companion paper (1) to illustrate gate location
optimization. The results of the gate location optimization had shown that even though significant
improvement was achieved, the design constraints
were not satisfied after gate location optimization
under the molding conditions suggested by the manufacturer. The reason was that the temperature difference between the highest and the lowest temperature a t the end of fill was still unacceptably high. I t
was thus necessary to optimize molding conditions
by holding the gate location at the calculated optimum.
Initial molding conditions were shown as suggested
by the manufacturer:
0

mold temperature, 90C


melt temperature, 290C
fill time, 1.50 s.

IX- INITIkL MOLDING CONDITIONS

The optimum gate location obtained from gate location optimization was used as the injection gate. The
material used was the same as in gate location optimization (Nylon 6). The results under these initial
molding conditions are shown in Table 7.
The constraints were set as follows:
7OoC 5 mold temperature 5 100.OC,
270,O"C Imelt temperature I310.0"C
1.00 s Ifill time 5 3.00 s,
temperature difference 5 20K,
maximum pressure 5 100 MPa,
maximum shear stress I5.00 X lo5 Pa.

The optimization process converged in 20 iterations. The optimum molding conditions were:
mold temperature, 98C.
melt temperature, 290.35"C,
fill time, 1.33 s.
Under the optimum molding conditions, the results
are shown in Table 8.
A comparison of the results is shown in Table 9
where it can be seen that temperature h a s been
reduced by 19.0%,the number of overpacked elements h a s been reduced by 5.9%. the number of
frictional overheated elements h a s been increased by
22.9%, maximum pressure h a s been increased by
3.5%.and maximum shear stress h a s been increased
by 1.9%. The most critical factor. namely the temperature difference (Td!B)r
was reduced by 19.0%.The
other factors were not very important in this case, so
their variation did not contribute significantly to the
objective function.

MODIFY CONSTRAINED PROBLEM


INTO UNCONSTRAINEDPROBLEM

Table 7. Results Under the Initial Molding Conditions.


Gate location

F(X)

POLYNOMIAL APPROXIMATION

Temperature difference
Overpack elements
Friction heating elements
Maximum pressure
Maximum shear stress

node 460
100
22.6 K
1.7%
7.4%
22.3 MPa
0.46 MPa

Table 8. The Results Under the Optimum Molding Conditions.


Gate

x-x+B's

CALCULATE CONJUGATE DIRECTION

S- - V O + BS

F(X)
Temperature difference
Overpack elements
Friction heating elements
Maximum pressure
Maximum shear stress

node 460
37.80
18.3 K
1.6%
9.6%
23.1 MPa
0.48 MPa

Table 9. Comparison of Results Before and After


Optimization.
Before Opt. After Opt. Improvement

F(X)
100
Temperature difference 22.6 K
Overpack elements
1.7%
Friction heating elements 7.4%
Maximum shear stress
0.53 MPa
890

37.80
18.3 K
1.6%
9.6%
0.54 MPa

62.2%
19.0%
5.9%
-22.9%
-1.9%

POLYMER ENGINEERING AND SCIENCE, MID-AUGUST 1990, Vol. 30, NO. 15

Optimization of Injection Molding Design. Part I1


The temperature difference was reduced by decreasing fill time and increasing mold temperature
since high mold temperature a n d short fill time reduce heat loss during the filling. Shorter fill time
needs higher injection pressure simply because of
high flow rate. The high flow rate caused a n increase
in shear rate and shear stresses, which explains the
small degradation in the last three quality measures.
SUMMARY

We successfully developed a system that can automatically optimize molding conditions by using the
results of flow simulation. The results of our optimization system have been successfully applied to
complex industrial molds. The conjugate direction
method and ALM penalty function method are suited
to our optimization of molding conditions. Since the
objective function to be minimized is highly nonlinear, molding conditions obtained from our optimization program are not guaranteed to be the global
optimum. The molding conditions therefore should
be used under supervision of a n injection molding
expert to ensure that the results are reasonable.

a r e referred to as the Kuhn-Tucker conditions (10)


and can be stated as:
1. X* is feasible.
2. a,g,(X*)=O,j= l , . . . ,r n a , r O .
3 . V F ( X * ) C_;"=I ajVgj(X*) E L I a k + m V h k ( X * ) ,

If a penalty function is created by augmented Lagrange multiplier (10). the pseudo objective function
h a s the following term:
rn

a. r,) =

F(X)+ ,=IC I ajdj + rpd?l


I

+ k=c [ a k + m h k ( X ) + rph2(X)1
1

W3
rp) = F(X)+ rpP(X)

(48)

where F ( X ) is the original objective function. P ( & ) is


a n imposed penalty function and r, is a scalar number that determines the penalty magnitude.
There are several methods available to create a
pseudo objective function for the original objective
F ( X ) (10). The method presented here is called
the augmented Lagrange multiplier or ALM method.
Because this method includes the conditions of
optimality into the optimization algorithm, it is more
reliable and efficient t h a n other methods such as the
exterior penalty method or the interior penalty
method (11). Let u s consider a general unconstrained
problem.
minimize:

F(X)

(49)

subject to:

g,(X) 5 0 j
hk(X)=

., rn

(50)

k = 1, ., l

(51)

= 1,

X + s x , ~ x , Ui =

1, . , n

(52)

The necessary conditions of constrained optimality

(53)

The stationary conditions for the pseudo objective


function @(X,r,,, a ) can be defined as:

V@(X r,, a ) = VF(X)

+ &El

aJVd(X)

+ ck=l ak+171Vhk(X) + c$l 2rPdJ(X)vdJ(x) (55)

APPENDIX A: THE AUGMENTED LAGRANGIAN


MULTIPLIER METHOD

To use the sequential unconstrained minimization


technique to solve a constrained optimization problem, the constrained problem h a s to be modified into
a n unconstrained problem first. The basic idea is to
transform the constrained objective function F(X)
into a n unconstrained objective function @(X,r,) by
adding some penalty when the constraints are violated or nearly violated (10).The modified objective
function @(X,r,). called the pseudo objective function, can be written in this form:

aj L

0.

2rphk+rn(X)Vhk+m(X)

where d(X) = rnax g ( X ) ,- .

At the optimum values a: of the multipliers


have
1

VF(X) -C

(Yk,

we

rn

C
,=I

aJVdj(X) +

C
k=

azvhk+rn(X)

(56)

The precise constraint satisfaction can be achieved


with some positive and finite rp. In other words, the
r,, a*) can be the true minimum of
minimum of
F ( X ) with some positive a n d finite r,. However, it is
impossible to find out the optimum a beforehand, so
we have to provide a reasonable initial r, and a. If
the search h a s not converged, the new o(k and rp can
be updated as follows:
rp

= Yrp

a, = aJ + 2r,rnax[g,(X),- a,/r,l
ak+m

ak+m

+ 2rphk(X)

(57)
(58)
(59)

After testing the pseudo objective function @(X,r,,


(Yk) under several different cases, the initial value for
r,, a l . a2,and a3 are chosen as 0.25, 0.3, 0.3, 0.4
respectively. The different r,, a l r a2, a n d a3 will
change the results of the optimization since (Yk represents the relative importance of the kth constraint
so they should be chosen carefully. A large a should
be selected for a n important constraint so that the
more penalty will be added to the pseudo objective
function when the constraint is violated or near violated.
ACKNOWLEDGMENTS

The generous financial and technical support of


Black & Decker Co., Monsanto Co., and G.E. Co. is

POLYMER ENGINEERING AND SCIENCE, MID-AUGUST 1990, Vol. 30, NO. 15

891

Ioannis Pandelidis and Qin Zou

gratefully acknowledged. We also appreciate the financial support of the Engineering Research Center
and the use of the CAD/CAM computing facility of
the University of Maryland.
NOMENCLATURE
= average nodal distance.
= maximum pseudoflow length.
= maximum number of flow fronts.
= flow proportion for flow i.
= volume filled by flow i.
= time interval for flow i.
= runner diameter.
= shear stress.
= fill pressure.

= temperature difference.
=

melt and mold temperature difference.

= percent of overpacked elements.

percent of frictional overheated elements.


= weighting coefficients for Td,
Noup,and
Nfh respectively.
= maximum allowable pressure.
= maximum allowable shear stress.
= search direction.
= pseudo objective function.
=

892

REFERENCES
1. 0. Pandelidis a n d 9. Zou, Part I, Polym. Eng. Sci., this
issue.
2. C. Austin, Moldflow-Thermoplastics and Thermoset
Flow Analysis, pp. 1.8. 3.8, Moldflow Pty. Ltd., Australia (1987).
3. S. Middleman, Fundamentals of Polymer Processing,
p. 270, McGraw-Hill Book Company, New York City
(1977).
4. C. Austin, in Developments in Injection Molding-3, A.
Whelan a n d J . P. Goff, eds., Elsevier Applied Science
Publishers (1985).
5. I. I. Rubin, Injection Molding: Theory and Practice, p.
136, J o h n Wiley a n d Sons, Inc., New York City (1972).
6. I. 0. Pandelidis a n d 9. Zou, SPE ANTEC Tech. Papers,
3 4 , 2 3 3 (1988).
7. David Luenberger, in Linear and Nonlinear Programming, Addison Wesley Publishing Company, Inc., New
York (1984).
8. A. V. Fiacco a n d G. P. McCormick, in Nonlinear Programming: Sequential Unconstrained Minimization
Techniques, J o h n Wiley a n d Sons, New York (1968).
9. David Wisner a n d R. Chattergy, in Introduction to
Nonlinear Programming-A Problem Solving Approach, North Holland, New York (1978).
10. J. Vanderplaats, in Numerical Optimization: Techniques f o r Engineering Design. with Applications,
pp. 89, 140, 122, 123. 17, 127, McGraw-Hill Book Co..
New York City (1984).
11. J. H. Cassis a n d L. A . Schrnit, Int. J. Num., Meth.
Engin., 10(1),3 (1976).

POLYMER ENGINEERING AND SCIENCE, MID-AUGUST 7990, Yo/. 30, No. 75