You are on page 1of 18

INTERNATIONAL JOURNAL FOR NUMERICAL METHODS IN ENGINEERING

Int. J. Numer. Meth. Engng. 43, 11091126 (1998)

OPTIMIZATION METHOD FOR STEADY CONDUCTION


IN SPECIAL GEOMETRY USING A
BOUNDARY ELEMENT METHOD
SEONG JIN PARK AND TAI HUN KWON
Department of Mechanical Engineering; Pohang University of Science and Technology; Pohang 790-784; Korea

ABSTRACT
In some steady heat conduction problems in special geometries which consist of a closely spaced surface
and circular holes in an in nite domain, thermal system designers may want to optimize the con guration
of circular holes in terms of their radii and locations to achieve the goal of uniform temperature distribution
over a closely spaced surface. In this paper, an ecient optimization procedure for this kind of problem is
proposed utilizing (i) the special boundary element analysis, (ii) the corresponding design sensitivity analysis
and (iii) the CONMIN algorithm. Three sample problems were solved to demonstrate the eciency and the
usefulness of the proposed optimization procedure. Some industrial engineering examples of such problems
can be found in the injection molding process, the compression molding process, and so on. ? 1998 John
Wiley & Sons, Ltd.
KEY WORDS: three-dimensional conduction heat transfer; special boundary integral formulation; Design Sensitivity
Analysis (DSA); Direct Di erentiation Approach (DDA); optimization; CONMIN algorithm

1. INTRODUCTION
Many engineering problems involve heat transfer, stress analysis, uid mechanics, electronics,
acoustics, etc. in complex three-dimensional geometries with a closely spaced surface and circular
holes. In particular, some typical examples of such geometries in the heat transfer problems are
molds of the injection molding process and compression molding process. The present study is
concerned about a model optimization problem of steady-state conduction heat transfer in these
kinds of geometries. In this thermal system, it would be of utmost importance to determine an
optimal con guration of circular holes in terms of their sizes and locations to make the temperature
distribution on closely spaced gaps (e.g. cavity surface in an injection mold) as uniform as possible.
A typical optimization process starts with a preliminary design (or initial) and searches for a
better design with the help of the numerical analysis and the design sensitivity analysis of the
current design. Based on the design sensitivity coecients, a new design is proposed during an

Correspondence to: Tai Hun Kwon, Department of Mechanical Engineering, Pohang University of Science and
Technology, San 31, Hyoja Dong, Pohang 790-784, Korea. E-mail: thkwon@vision.postech.ac.kr

Contract=grant sponsor: Korea Ministry of Education; Contract=grant number: ME95-E-19


Contract=grant sponsor: Samsung Electronics

CCC 00295981/98/06110918$17.50
? 1998 John Wiley & Sons, Ltd.

Received 28 April 1997


Revised 12 March 1998

1110

S. J. PARK AND T. H. KWON

iterative process by using a non-linear programming algorithm, for example, the steepest descent
algorithm and the CONMIN algorithm proposed by Haarho and Buys,1 etc. If no better design
can be found, the iterative optimization process stops. Otherwise, the iterative process is continued
until an optimal design is obtained. In order to use any rst-order method for this optimization
process, the thermal analysis and the corresponding design sensitivity analysis are essentially required for an optimal design. The design sensitivity analysis provides with Design Sensitivity
Coecients (DSCs) which are rates of changes of response variables such as temperatures or heat
uxes, in a heat transfer problem, with respect to design variables. In the literature, there are some
research works on the subject of optimization problems using nite element method and boundary
element method: for example, Saigal and Chandra2 for the two-dimensional and axisymmetric heat
di usion problems, Awa and co-workers3 for pultrusion die design, Tang and co-workers4 and
Matsumoto and co-workers5 for the two-dimensional cooling system design of injection molds,
Matsumoto and co-workers6 for the three-dimensional cooling=heating design of compression molds,
etc.
As far as the thermal analysis is concerned, Rezayat and Burton7 have proposed a special
boundary integral formulation for the steady heat conduction problem in these complex geometries.
In the literature, there are some research works on the subject of design sensitivity analysis for
several problems in mechanics such as elasticity or heat transfer: for example, Choi and co-workers8
for two-dimensional elastic structures, Saigal and Chandra2 for two-dimensional and axisymmetric
heat di usion problems, etc. Basically, there are three di erent approaches in the design sensitivity
analysis: the Finite-Di erence Approach (FDA), the Adjoint Structure Approach (ASA) and the
Direct Di erentiation Approach (DDA). These three approaches are used in conjunction with the
Finite Element Method (FEM) or the Boundary Element Method (BEM) to obtain design sensitivity
coecients. Park and Kwon9 recently proposed a design sensitivity analysis formulation for the
above steady heat conduction problem using the direct di erentiation approach based upon the
special boundary integral formulation proposed by Rezayat and Burton.7
The present paper has developed an ecient optimization procedure to achieve the goal of
uniform temperature distribution over a closely spaced surface for the above steady heat conduction problem: radii and locations of circular holes are considered as design variables; the
CONMIN algorithm has been adopted to obtain the optimal con guration of the design variables
in conjunction with the special boundary integral formulation by Rezayat and Burton7 and the
corresponding design sensitivity analysis formulation by Park and Kwon.9 Three sample problems are solved to demonstrate the eciency and the usefulness of the present optimization
procedure.
2. PROBLEM DEFINITION
The model optimal design problem of steady-state conduction heat transfer in an in nite domain with a closely spaced surface (see Figure 1) is de ned as follows: given a constant heat
ux boundary condition on the closely spaced part surface, nd optimal radius and location (design variables) of each circular hole with a constant heat transfer coecient and a constant bulk
temperature to minimize the nonuniformity of the temperature distribution on the closely spaced
part surface (or maximize the uniformity of that) with side constraints (i.e., realistic interval of
each design variable). In this model problem, the closely spaced surface and circular holes correspond to the part surface and circular cooling channels, respectively, in an injection molding
case.
? 1998 John Wiley & Sons, Ltd.

Int. J. Numer. Meth. Engng. 43, 11091126 (1998)

1111

OPTIMIZATION METHOD FOR STEADY CONDUCTION

Figure 1. A geometry of a simple model problem bounded by

, SC and SE

3. GOVERNING EQUATION
The geometry of a simple model problem consists of the closely spaced surfaces and the circular
holes in an in nite domain as shown in Figures 1 and 2. In these gures,
, SP , , SC , SE , and
n denote the in nite domain, the part surfaces (closely spaced surfaces), the mid-surfaces of the
part, the circular hole surfaces, the exterior surfaces of the domain (at in nite point), and the
outward unit normal vector in the part, respectively. We can describe the part surface by only
the mid-surface of the part instead of the whole surface (positive and negative surface as indicated
by S + and S , respectively in Figure 2) because the part surfaces are closely spaced. The governing
di erential equation for the steady-state heat conduction equation may be written as
2T = 0

in

(1)

where T is the temperature. In this model problem a constant heat ux is imposed over the part
surfaces as a boundary condition, and the boundary condition on the circular hole surfaces is
treated as a mixed boundary condition with a speci ed heat transfer coecient and a speci ed
internal mean temperature (for example, a coolant mean temperature when the circular hole is a
cooling=heating channel). We also assume that the exterior surface of domain (at in nite point)
may be treated as an in nite adiabatic sphere. Thus, the boundary conditions on the boundary
S = SP + SC + SE are given as
 
@T
@T
=
on SP
@n
@n 0
k

@T
= h(T Tm ) on SC
@n
@T
=0
@n

? 1998 John Wiley & Sons, Ltd.

(2)

on SE
Int. J. Numer. Meth. Engng. 43, 11091126 (1998)

1112

S. J. PARK AND T. H. KWON

Figure 2. Detailed diagram of a simple model problem for notations

where n;
k; h, and Tm denote the outward unit normal vector, the thermal conductivity of the material
in domain, the speci ed heat transfer coecient and the speci ed internal mean temperature,
respectively, and (@T=@n)0 is the speci ed temperature gradient.
4. BOUNDARY INTEGRAL FORMULATION FOR THERMAL ANALYSIS
A standard boundary element formulation10 for the three-dimensional Laplaces equation given by
equation (1) governing the steady conduction leads to

 
Z  
@ 1
1 @T (^)
T (x) =
T (^)
dS(^)
(3)
@n
@n r
S r
Here x and ^ are points in space, r = |^ x|, and denotes a solid angle formed by the boundary
surface. Note that = 2 at the smooth boundary surface, and = 4 or 0 at internal points or external points of the domain, respectively. Kwon11 and Forcucci and Kwon12 have used equation (3)
for mold cooling system analyses.
For any two closely spaced surfaces such as the part surfaces in the model problem, because
equation (3) leads to redundancy in the nal system of linear algebraic equations, a modi ed
procedure as described by Rezayat and Burton7 needs to be used. According to this modi cation,
the mid-surface, , is considered rather than two closely spaced surfaces, SP+ and SP as schematically depicted in Figure 2. For each mid-surface element, in order to derive the extra equation
corresponding to the additional degree of freedom, a derivative of equation (3) with respect to
the normal direction vector,  (n+ in Figure 2), at the mid-surface element, is taken. And, for
circular hole surfaces, SC , we use a special formulation based on the line-sink approximation.
This formulation avoids discretization of the circular channels along the circumference and thus
saves a large amount of computer memory and time. See the references by Rezayat and Burton,7
Himasekhar and co-workers13 and Park and Kwon9 for further details of this modi ed approach
for part surfaces, circular holes, and exterior surface.
? 1998 John Wiley & Sons, Ltd.

Int. J. Numer. Meth. Engng. 43, 11091126 (1998)

OPTIMIZATION METHOD FOR STEADY CONDUCTION

1113

The following is the nal BEM formulae for these modi cations:
for a point x on the mid-surface of the part, , we have the following pair of integral equations:
 


Z   +
@ 1
@T
1 @T
+
+
+

(T T ) dS(^)
+
T (x) + T (x) =
r @n+
@n
@n r
  #
Z 2
Z "  Z 2  
N
P
@ 1
@T
1
d T
d ak dl(^)
+
@n
r
@n r
k=1 lk
0
0
(4)
+ 4T
  


Z    +

@
@T
@T + (x)
@T
@ 1
@ 1
+ @T (x)
+


(T


=
+

T
)
dS(^)
@n+
@n
@ r
@n+
@n
@ @n r
 
   #
Z 2
Z " Z 2
N
P
@ 1
@
@ 1
@T
d T
d ak dl(^)
+
@n
@
r
@
@n
r
k=1 lk
0
0
(5)
and for a point on the axis of the cylindrical segment of the cooling channels we have the following
integral equation:
 


Z   +
@T
@ 1
1 @T
+

(T T ) dS(^)
+
0=
r @n+
@n
@n r
  #
Z "  Z 2  
Z 2
N
P
@ 1
@T
1
d T
d ak dl(^) + 4T
(6)
+
@n
r
@n r
k=1 lk
0
0
In equations (4)(6), T denotes the temperature on the exterior surface at in nite point, N is the
total number of cooling channels, ak is the radius of kth cooling channel, l is the local co-ordinate
in the axial direction of each cooling channel, and  is the local circumferential co-ordinate of
each cooling channel as shown in Figure 3.
Once the integrals on each of the elements are calculated, the discretized boundary element
formulae for the analysis can be manipulated to the following form:
( )
@T
(7)
[Hij ] {Tj } = [Gij ]
@n j
where [Hij ] and [Gij ] are functions of geometry of boundary surfaces. Some of [Hij ] and [Gij ] are
singular integrals, which can be evaluated analytically in a simple manner as suggested by Rezayat
and Burton.7 Also, the integrals over  in equations (4) (6) are evaluated in a closed form using
the complete elliptic integrals in the same manner proposed by Park and Kwon.9 Other integrals
are evaluated by Gaussian quadrature rule. Next, boundary conditions can be introduced to obtain
a system of linear algebraic equations:
[Aij ] {Tj } = {fi }

(8)

where [Aij ] and {fi } re ect boundary conditions. For this problem, temperature, T , is taken to be
an unknown on each element with @T=@n being eliminated with the help of a boundary condition.
? 1998 John Wiley & Sons, Ltd.

Int. J. Numer. Meth. Engng. 43, 11091126 (1998)

1114

S. J. PARK AND T. H. KWON

Figure 3. Coordinate transformation rule in one cooling channel element

This system of equations can be solved by LU-decomposition by Gaussian elimination or an underrelaxation iterative method. It may also be mentioned that T can be easily determined during
the iteration procedure to satisfy the heat balance.

5. BOUNDARY INTEGRAL FORMULATION FOR SENSITIVITY ANALYSIS


The boundary element method is also an ecient numerical tool in design sensitivity analyses
when the problem can be treated without domain discretization and the design variables can be
de ned on the boundary. In this particular model problem, the part surface is xed, but the size
and the arrangement of the circular holes (only de ned on boundaries) can be changed to achieve
a certain design objective. Therefore, the boundary element method can be e ectively applied to
the design sensitivity analysis of this model problem. In this paper, the radii and the locations of
the circular holes are considered as design variables.
The design sensitivity analysis is essentially to determine the variation of an objective function
with respect to a variation in the design variable. In this model problem, the design sensitivity
coecients of the temperature on the mold surface with respect to all design variables provide
valuable information for the optimal design using the rst-order optimization techniques. By the
implicit di erentiation of equations (4)(6) with respect to each design variable (the direct di erentiation approach), we have derived various boundary integral formulae for each design variable:
in design sensitivity equations, all T and @T=@n in equations (4) (6) are replaced by @T=@X
and @(@T=@n)=@X , respectively, X being a design variable; in additions to this replacement, design sensitivity equations should include extra boundary integral terms because the mold geometry
itself also depends on all the design variables. Refer to Park and Kwon9 for the nal forms
of design sensitivity equations and further details of this sensitivity approach for each design
variable.
? 1998 John Wiley & Sons, Ltd.

Int. J. Numer. Meth. Engng. 43, 11091126 (1998)

OPTIMIZATION METHOD FOR STEADY CONDUCTION

1115

Once the integrals on each element are calculated, the discretized boundary element formulae
for the sensitivity analysis can be manipulated to the following form:
( )
( )
@T
@T
[Hij; X ] {Tj } + [Hij ] {Tj; X } = [Gij; X ]
+ [Gij ]
(9)

@n j
@n j; X
where ,X represents a derivative with respect to a design variable X . It may be noted that in
equation (9), {Tj }, {(@T=@n)j }, [Hij ] and [Gij ] are already obtained in the thermal analysis. The
integrals for coecients of the matrices of [Hij; X ] and [Gij; X ] can be evaluated in a manner similar
to the thermal analysis case. See Park and Kwon9 for details of evaluating these integrals. Next,
sensitivities of boundary conditions can be introduced into equation (9) to obtain a system of
linear algebraic equations as follows:
[Aij ] {Tj; X } = {fi; X } [Aij; X ] {Tj }
{fi;0 X }

(10)

where Tj; X , is taken to be an unknown on each element. It may be noted that, in equation (10),
the only forcing terms {fi;0 X } vary with the design variable X . Therefore, from equation (10),
the sensitivity analysis results can be obtained simultaneously for all design variables with multiple forcing terms, thus signi cantly saving the elapsed CPU time. The major advantage of using
LU-decomposition solver lies in the fact that the LU-decomposition of the matrix [Aij ], is formed
during the analysis, and thus can be multiply reused in the sensitivity analysis. But, an iterative
method takes less computer memory and thus becomes more useful at the cost of more computating time, than a direct solver, especially when a large number of elements are to be used for
complicated part geometries. The present study has introduced the iterative method.
6. OPTIMIZATION
In this model problem, we want to nd an optimal con guration of circular holes to make the
temperature distribution over the part surface as uniform as possible. It may be mentioned that the
injection molded part quality increases with the temperature uniformity. Towards this design goal
of the uniform part surface temperature distribution, the objective function is chosen as
Z
(T T ) 2 dA
SP
Z
(11)
F(X) =
2

dA
T
SP

where

Z
T = ZSP

T dA

SP

dA

In equation (11), X is a design variable vector, T is the average temperature over the part surface.
Proper constraints have to be imposed on the design variables to keep the design realistic for
this optimization problem. In this particular problem, the upper and lower bounds must be placed
on the radius and position of each circular hole to keep the design practical from the manufacturing
? 1998 John Wiley & Sons, Ltd.

Int. J. Numer. Meth. Engng. 43, 11091126 (1998)

1116

S. J. PARK AND T. H. KWON

view point. Inequality constraints associated with a design variable Xi having the upper and lower
inequality bounds expressed by
Ai 6Xi 6Bi ;

i = 1; : : : ; n

(12)

can be modi ed to an equality constraint by introducing a new slack design variable Yi as follows:
Gi (Xi ; Yi ) = Xi Ai (Bi Ai ) sin 2 Yi = 0;

i = 1; : : : ; n

(13)

where n is the number of design variables.14


In order to solve the above constrained minimization problem, the present study has employed
CONMIN algorithm developed by Haaro and Buys1 since it has been successfully used in an
optimal design of heating systems in compression molds, which is quite similar to the present
model problem, by Barone and Caulk15 and Forcucci and Kwon.12 CONMIN algorithm employs
the Augmented Lagrangian Multiplier (ALM) method to deal with the equality constraints, and
DavidonFletcherPowell method16 for the unconstrained minimization during the successive unconstrained minimization procedure.
7. RESULTS AND DISCUSSION
Before the discussion of optimization results, it might be noted that the accuracy of the analysis
formulation and that of the sensitivity analysis formulation were demonstrated by Rezayat and
Burton7 and by Park and Kwon,9 respectively. The optimization method proposed in this paper
has been applied to three representative examples. We have used triangular elements for the part
surface and line elements for the circular hole surface as boundary elements. It might be mentioned
that each element has a constant temperature and a constant heat ux in the boundary element
analysis in the present study. We used the following the relative norm convergent criteria for
stopping the iteration:
new

F
F old
2


610
F old
where F is the objective function. The computer used in this study is SUN SPARC 10 (229
MFLOPS=102 mips).
Plane plate with two circular holes
The part geometry with an initial design of two circular holes considered in this example is
shown in Figure 4. The radii and the locations of circular hole #1 and circular hole #2 are made
intentionally unsymmetric to see if the optimized result leads to a symmetric con guration as a
way of validation of the proposed method. The discretization for the boundary element analysis is
shown in Figure 5 where each asterisk represents a node. The number of elements in the part and
two circular holes are 588 and 30, respectively. The thermal conductivity of the domain material
used in this example is 100 W=m K. For this example, we used the following boundary conditions:
(1) for part surface: uniform temperature gradient (@T=@n)0 = 103 C=m;
(2) for both cooling channels: Tm = 20 C, h = 105 W=m 2 K:
The design variables of this example are the radius, the y-co-ordinate and z-co-ordinate of each
circular hole. (The e ect of x-co-ordinate of eachcircular hole is negligible.) The constraints of
? 1998 John Wiley & Sons, Ltd.

Int. J. Numer. Meth. Engng. 43, 11091126 (1998)

OPTIMIZATION METHOD FOR STEADY CONDUCTION

1117

Figure 4. Plane plate with two circular holes: initial design

each design variable are


016a1 610;
106y1 6100;
1006z1 6100;

016a2 610
1006y2 610
1006z2 6100

where subscripts 1 and 2 denote the circular hole ID number. The optimization procedure required
4 iterations of unconstrained minimizations to yield the optimal con guration and the total elapsed
CPU time was 17 : 25 : 42 (h : min : s). Figure 6(a) shows the con guration path from the initial
con guration to the optimal one and Figure 6(b) indicates the corresponding values of normalized
objective function (with respect to the value of objective function of the initial design) during the
iteration. It is interesting to note that the con guration changes toward a symmetric con guration
as it should. The objective function decreases upto 247 per cent. From the above results, we have
found the following characteristics of each design variable:
(a) As the radius decreases (resulting in less cooling e ect), both |T T | (the numerator of the
objective function) and T (the denominator of the objective function) increase. Therefore,
there exists at least an optimal radius of each circular hole, which was con rmed by extensive
computations.
? 1998 John Wiley & Sons, Ltd.

Int. J. Numer. Meth. Engng. 43, 11091126 (1998)

1118

S. J. PARK AND T. H. KWON

Figure 5. Plane plate with two circular holes: boundary element mesh

(b) At a given radius (about 4 mm), when the distance between the part surface and the circular
hole changes, |T T | is found to be smallest at the distance of 50 mm. On the other hand,
T increases with the distance. Therefore, there exists an optimal distance of 475 mm as a
local minimum. In fact, when the distance is over 200 mm, the objective function begins to
decrease as the distance increases. Since the distance over 200 mm is too large to be realistic
in a model, the local minimum is what we wanted to nd.
(c) The optimal z-co-ordinate coincides with the symmetrical location, that is, the centre of the
part.
Figure 7 shows the distribution of 1 (T= T ) (referred to the uniformity distribution, hereafter)
over the part surface in the initial design (a) and in the optimal con guration (b), which clearly
indicates that the optimal con guration has more uniform temperature distribution than the initial
one. The results of this example show that the proposed optimization procedure works quite
successfully. Furthermore, we have con rmed that the optimization method ended up with the
same optimal con guration from several di erent initial designs.
Plane plate with six circular holes
Figure 8 shows the part geometry with an initial design of six circular holes for the second
example. The number of boundary elements in the part and six circular holes used in the analysis
are 588 and 60, respectively. The thermal conductivity of the domain material used in this example
is 10 W=m K. For this example, we used the following boundary conditions:
(i) for part surface: uniform temperature gradient (@T=@n)0 = 103 C=m
(ii) for both cooling channels: Tm = 20 C, h = 5 104 W=m 2 K
The design variables of this example are the radius, the x-co-ordinate and z-co-ordinate of each
circular hole. (The e ect of y-co-ordinate of each circular hole is negligible.) The constraints of
each design variable are
016a1 610;
016a4 610;
006x1 69:0;
006x4 69:0;
106z1 6100;
1006z4 601;
? 1998 John Wiley & Sons, Ltd.

016a2 610;
016a5 610;
956x2 6105;
956x5 6105;
106z2 6100;
1006z5 601;

016a3 610
016a6 610
1106x3 6200
1106x6 6200
106z3 6100
1006z6 601

Int. J. Numer. Meth. Engng. 43, 11091126 (1998)

OPTIMIZATION METHOD FOR STEADY CONDUCTION

1119

Figure 6. Results of optimization: (a) path from initial design to optimal con guration; (b) normalized objective function
versus iteration

where subscripts 16 denote the circular hole ID number. The optimization procedure required
9 iterations of unconstrained minimizations to yield the optimal con guration and the elapsed
CPU time was 31 : 19 : 42. Figure 9 shows the initial design (dotted lines) and the nal optimal con guration (solid lines). Figure 10 shows the distribution of 1 (T= T ) over the part
surface both in the initial design and in the optimal con guration. And the normalized objective function reduced to about 592 per cent at the optimal con guration. As expected, the
circular holes become symmetrical and the distance of circular holes at the centre from the
? 1998 John Wiley & Sons, Ltd.

Int. J. Numer. Meth. Engng. 43, 11091126 (1998)

1120

S. J. PARK AND T. H. KWON

Figure 7. Uniformity distribution on the part surface in (a) initial design (T = 1180) and (b) optimal con guration
(T = 1311) [levelvalue = (11 level number)=20]

part surface becomes larger than those at both ends to make the temperature as uniform as
possible.
Wedged shape with ve circular holes
The wedged-shape part geometry with an initial design of ve circular holes considered in this
third example is shown in Figure 11. The number of boundary elements in the part and ve
circular holes are 564 and 30, respectively. The thermal conductivity of the domain material used
in this example is 10 W=m K. For this example, we used the following boundary conditions:
(1) for part surface: uniform temperature gradient (@T=@n)0 = 103 C=m
(2) for both cooling channels: Tm = 20 C, h = 5 104 W=m 2 K
? 1998 John Wiley & Sons, Ltd.

Int. J. Numer. Meth. Engng. 43, 11091126 (1998)

OPTIMIZATION METHOD FOR STEADY CONDUCTION

1121

Figure 8. Plane plate with six circular holes: initial design

Figure 9. Initial design and optimal con guration


? 1998 John Wiley & Sons, Ltd.

Int. J. Numer. Meth. Engng. 43, 11091126 (1998)

1122

S. J. PARK AND T. H. KWON

Figure 10. Uniformity distribution on the part surface in (a) initial design (T = 754) and (b) optimal con guration
(T = 992) [level value = (11 level number)=20]

The design variables of this example are the radius, the x-co-ordinate and z-co-ordinate of each
circular hole. (The e ect of y-co-ordinate of each circular hole is negligible.) The constraints of
each design variable are
016a1 610;
016a4 610;
1006x1 600;
1006x4 610;
106z1 6100;
1006z4 610;
? 1998 John Wiley & Sons, Ltd.

016a2 610;
016a3 610
016a5 610;
006x2 6100;
106x3 6100
1006x5 610;
006z2 6100; 1006z3 600
1006z5 610
Int. J. Numer. Meth. Engng. 43, 11091126 (1998)

OPTIMIZATION METHOD FOR STEADY CONDUCTION

1123

Figure 11. Wedged shape with ve circular holes: initial design

Figure 12. Initial design and optimal con guration


? 1998 John Wiley & Sons, Ltd.

Int. J. Numer. Meth. Engng. 43, 11091126 (1998)

1124

S. J. PARK AND T. H. KWON

Figure 13. Uniformity distribution on the part surface in (a) initial design (T = 913) and (b) optimal con guration
(T = 923) [level value = (11 level number)=20]

where subscripts 15 denote the circular hole ID number. The optimization procedure required 5
iterations of unconstrained minimizations to yield the optimal con guration and the elapsed CPU
time was 25 : 46 : 24. Figure 12 shows the initial design (dotted lines) and the nal optimal con guration (solid lines). The result indicates that the optimization method changes the con guration
in the direction of increasing the cooling e ect on the plus plane and decreasing it on the minus
plane. Figure 13 shows the distribution of 1 (T= T ) over the part surface both in the initial design and in the optimal one. The normalized objective function reduced to about 240 per cent at
the optimal con guration. This optimization method quite signi cantly improves the temperature
uniformity.
? 1998 John Wiley & Sons, Ltd.

Int. J. Numer. Meth. Engng. 43, 11091126 (1998)

OPTIMIZATION METHOD FOR STEADY CONDUCTION

1125

8. CONCLUDING REMARKS
This paper is concerned with the overall numerical procedure for the optimal design problem of
some steady heat conductions in special geometries which consist of a closely spaced surface and
circular holes. Towards the goal of uniform temperature distribution over a closely spaced surface,
we could optimize the con guration of circular holes in terms of their radii and locations using
the proposed optimization procedure which consists of the followings:
1. the boundary element analysis using the special boundary integral formulation;
2. the design sensitivity analysis using the direct di erentiation approach based upon the above
special boundary integral formulation;
3. the CONMIN algorithm.
Optimizations for three representative example problems were performed to show the eciency
and the usefulness of the present optimization procedure. It may be noted that the procedures and
techniques presented in this paper can be practically applied to mold thermal system designs of
manufacturing industries such as the injection molding process, the compression molding process,
and so on.
ACKNOWLEDGEMENTS

This work is supported by Korea Ministry of Education through Mechanical Engineering Research
Fund (ME95-E-19) and partially by grants from R & D Center in Samsung Electronics (Consumer
Electronics Business). The authors gratefully acknowledge these nancial supports.

REFERENCES
1. P. C. Haarho and J. D. Buys, A new method for the optimization of a non-linear function subject to non-linear
constraint, Comput. J., 13, 178 184 (1970).
2. S. Saigal and A. Chandra, Shape sensitivities and optimal con gurations for heat di usion problems: a BEM approach,
ASME J.=Heat Transfer, 113, 287 295 (1991).
3. Awa, West and Price, Heater con guration Design of a Pultrusion Die using Design Optimization Techniques,
in Gabriele (ed.), Proc. 1992 ASME Intl Computers in Engng. Conf.; Vol. 1; Computers in Engineering 1992,
San Francisco, 1992.
4. L. Q. Tang, C. Chassapis and S. Manoochehri, Optimal cooling system design for multi-cavity injection molding,
Finite Elements Anal. Des., 26, 229 251 (1997).
5. T. Matsumoto, M. Tanaka, M. Muyagama and N. Ishii, Optimal design of cooling lines in injection moulds by using
boundary element design sensitivity analysis, Finite Elements Anal. Des., 14, 177 185 (1993).
6. T. Matsumoto, M. Tanaka and M. Miyagawa, Boundary element system for mold cooling=heating design, in
C. A. Brebbia and J. J. Rencis (eds.), Boundary Element XV; Vol. 2; Stress Analysis, Computational Mechanics
Publications, Boston, 1993, pp. 461 475.
7. M. Rezayat and T. Burton, A boundary-integral formulation for complex three-dimensional geometries, Int. J. Numer.
Meth. Engng., 29, 263 273 (1990).
8. J. H. Choi, B. M. Kwak and D. H., Lee, A uni ed approach in shape design sensitivity analysis of elastic structures
in the boundary integral equation formulation, in M. Tanaka et al. (eds), Boundary Element XIIProc. 12th Int.
Conf., Vol. 2, Computational Mechanics Publications, Boston, 1990, pp. 225 236.
9. S. J. Park and T. H. Kwon, Sensitivity analysis formulation for three-dimensional conduction heat transfer with complex
geometries using a boundary element method, Int. J. Numer. Meth. Engng., 39, 2837 2862 (1996).
10. C. A. Brebbia, J. C. F. Telles and L. C. Wrobel, Boundary Element TechniquesTheory and Application in
Engineering, Springer, Berlin, 1984.
11. T. H. Kwon, Mold cooling system design using boundary element method, ASME J. Engng. Ind., 110, 384 394
(1989).
? 1998 John Wiley & Sons, Ltd.

Int. J. Numer. Meth. Engng. 43, 11091126 (1998)

1126

S. J. PARK AND T. H. KWON

12. S. J. Forcucci and T. H. Kwon, A Computer Aided Design System for Three-Dimensional Compression Mold Heating,
ASME J. Engng. Ind., 111, 361 368 (1989).
13. K. Himasekhar, J. Lottey and K. K. Wang, CAD of mold cooling in injection molding using a three-dimensional
numerical simulation, ASME J. Engng. Ind., 144, 213 221 (1992).
14. R. L. Fox, Optimization Methods for Engineering Design, Addison-Wesley, Menlo Park, 1971.
15. M. R. Barone and C. A. Caulk, Optimal arrangement of holes in a two-dimensional heat conductor by a special
boundary integral method, Int. J. Numer. Meth. Engng., 18, 675 685 (1990).
16. R. Fletcher and M. J. Powell, A rapidly convergent descent method for minimization, Comput. J., 6, 163 168 (1963).

? 1998 John Wiley & Sons, Ltd.

Int. J. Numer. Meth. Engng. 43, 11091126 (1998)