casting design through multi-objective optimization
TRANSCRIPT
Casting Design Through Multi-Objective Optimization
Jean Kor*, Xiang Chen*, Zhizhong Sun**, Henry Hu**
*Department of Electrical & Computer Engineering, University of Windsor
Windsor, Ontario, �9B 3P4, Canada
e-mail: [email protected]
** Department of Mechanical, Automotive & Materials Engineering, University of Windsor
Windsor, Ontario, �9B 3P4, Canada
Abstract: The gating and riser system design plays an important role in the quality and cost of a metal casting. Due
to the lack of fixed theoretical procedures to follow, the design process is carried out on a trial-and-error basis. The
casting design optimization problem is characterized by multiple design variables, conflicting objectives and a
complex search space, making it unsuitable for sensitivity-based optimization. In this study, a formal optimization
method using Multi-Objective Evolutionary Algorithm (MOEA) was developed to overcome such complexities. A
framework for integrating the optimization procedure with numerical simulation for the design evaluation is presented.
The proposed optimization framework applied to the gating and riser system of a sand casting gave good results and
provided more flexibility in decision making.
1. INTRODUCTION
Casting design, in particular the gating and riser system
design has a direct influence on the quality of cast
components. The design of gating and riser systems is
largely based on past experience and empirical rules
(Campbell,1991) and simulation has become an
important tool for the design, analysis and optimization
of casting processes (Kor, 2006).
Coupling numerical simulation with formal optimization
methods is one way of adopting a more systematic
approach towards casting design. The casting design
optimization problem is characterized by multiple
control points and multiple conflicting objectives that
involve many parameters (Wang, 2007).
In this study, the casting design process is formalized
and a proposed optimization framework using multi-
objective evolutionary algorithm (MOEA) is presented.
Evolutionary algorithms are guided stochastic search
methods that are well known for their ability to address
such difficulties. They are global optimizers that work
with a population of solutions in parallel and have been
shown to be robust in finding optimal solutions in
complex search spaces regardless of the starting points
(Jones, 2002).
The key aspect of using MOEA for casting design
optimization is that it does not require any analytical
information for the optimization formulation. The main
goal is to improve the efficiency and quality of the
casting design process by combining simulation and
formal optimization techniques. The focus is primarily
on the gating and feeding system where numerical
simulation is employed to predict the performance of a
design.
2. MULTI-OBJECTIVE OPTIMIZATION
In a multi-objective problem, the aim is to find a set of
values for the design variables which optimizes a set of
objective functions simultaneously. When the objectives
are conflicting, finding a solution that minimizes all of
the objectives at the same time becomes impossible.
The operation of the genetic algorithm is shown in
Figure 1. The elitist Non-Dominated Sorting Genetic
Algorithm II (NSGA II) is currently one of the most
popular MOEA method used in real-world multi-
objective optimization problems (Deb, 2003). Some of
the distinguishing features of NSGA II are its fast elitist
sorting strategy that involves a combined pool of both
the parent and child populations and the elimination of
sharing parameters using an autonomous crowding
distance strategy.
NSGA II maintains a Pareto archive and introduces
elitism by comparing the current population with the
previously found best non-dominated solutions. The
selection procedure is guided towards a diverse set of
points on the Pareto front based on the concept of non-
dominated ranking and crowding distance. NSGA II is
used for this study and is interchangeably referred to as
MOEA.
3. CASTING DESIGN OPTIMIZATION USING
MOEA
3.1. Optimization frame work
In virtual product development, the computer aided
design (CAD) and analysis sequence is integrated into an
optimization strategy, whereby the product's geometry
are described by parameters for an efficient process
(Merkel, 2003).
Preprints of the 18th IFAC World CongressMilano (Italy) August 28 - September 2, 2011
Copyright by theInternational Federation of Automatic Control (IFAC)
11642
In casting design, the two main objectives are: (1)
eliminating casting defects and (2) maximizing casting
yield. To achieve these goals, different designs can be
explored by changing the shapes and sizes of the gating
and riser components. The application of genetic
algorithms have been shown to be effective for cross-
sectional shape generators that involve a marriage of
continuous variables such as shape dimensions, and
discrete variables such as shape type and quantity
(Yoshimura, 2005). For this study, geometrical
descriptors such as length, radius, and height of the
casting components were selected as the design variables
for the optimization process as shown in Figure 2.
Fig. 1 General flowchart of the Genetic Algorithm.
Fig. 2. Binary encoding of the gating and riser design
variables.
A commercial simulation software, MAGMASOFT®
was employed for the gating and riser system design
analysis. Since the details on the complex calculations
involved were not available, the performance of a design
had to be measured from the accessible simulation
outputs.
The optimization strategy had to be connected to the
simulation environment. Here, the design analysis
software was treated like a ‘black box’ with an input
interface to accept new design suggestions and an output
interface to communicate their performance measures to
the optimization engine. These output measures were
used to evaluate the fitness of a design in MOEA.
The optimization framework and its process flow are
illustrated in Figure 3. The structure of the casting
design optimization implementation is divided into four
main parts: pre-processing, simulation, post-processing
and optimization which are described next.
Fig. 3. Optimization process flow chart of the gating
and riser system design
PRE-PROCESSING
Simulation of the cavity filling and solidification process
required the geometrical information for the casting, the
gating system and the mould in advance. To explore
different designs in an autonomous fashion, parametric
geometry functions were employed to create the gating
and riser systems with the aid of command files. By
modifying the parameter values in the geometric
functions, the design could be varied accordingly.
SIMULATION
The actual filling and solidification simulation would
take place once the meshed geometries and the necessary
casting process parameters have been established. The
type of numerical calculations employed was based on
the algorithm (Solver) type chosen. Once the simulation
is completed, the results were processed for analysis in
the next step.
POST-PROCESSING
With the 3-D post processor module in
MAGMASOFT®, the visualization of the fluid flow and
temperature field patterns in the cavity during the casting
process could be graphically analyzed. In order to
Initialize a population
Select parents for mating
Create new offspring by crossover & mutation
Delete undesirable members of population
Evaluate & insert new offspring into population
Termination criterion satisfied?
Return best chromosome
Increment population
Yes
No
Casting
Design
(STL)
Objectives, Weights,
Constraints &
Design Variables,
Initial Gating &
Riser Design
(Parametric Geometry)
Modeling
Filling &
Solidification
Simulation
MeshingExport
ResultsMAGMA API
GA
Evaluation &
Optimization
Termination
Criteria Met?
Return
Optimal
Result(s)
Yes
No
MAGMASOFT
Graphical
Results
Text
Results
Modify Design Parameters
Casting
Parameters
Preprints of the 18th IFAC World CongressMilano (Italy) August 28 - September 2, 2011
11643
formalize the casting design optimization process, these
results had to be converted to a proper format that can be
routinely handled. The Application Programming
Interface (API) of MAGMASOFT® was employed to
export the graphical results into text results.
Using the API, customized subroutines were developed
to access the MAGMASOFT® files and data structures,
extract the desired results from each control volume and
convert them into the appropriate format for further
processing. The processed results were further organized
to form the appropriate performance measures required
for the optimization performance evaluation process as
shown in Figure 4.
Fig. 4. The post-processing of simulation results for design
evaluation
OPTIMIZATION
Based on the output measures acquired from the
MAGMASOFT® simulation results, the performance
evaluation of a design was carried out against the
objective functions and constraints specified in the
problem definition. Based on the performance
evaluation process, the next direction of the MOEA
optimization process would then be determined.
In this study, the quality of a cast product was
characterized by its shrinkage porosity and liquid metal
velocity. Using the output results available from the
MAGMASOFT® simulation, the shrinkage porosity
measure was taken as the maximum porosity value
contained in the control volumes of the casting. Since a
shrinkage porosity-free casting is always desired, the
porosity requirement was P = 0%.
Turbulence of the liquid metal in the casting was
assessed in terms of flow velocities obtained also from
the simulation results. Based on Campbell’s rules, the
liquid metal flow should not exceed a velocity of 0.5 m/s
to maintain the stability of the meniscus front. Since the
critical velocity Vcrit was not properly defined in either
the X, Y or Z directions(Campbell, 1998), the constraint
on the entry velocity of the liquid metal was set such that
it must not exceed 0.5 m/s in any direction of the
velocity vector, Vx, Vy and Vz for a design to be
considered feasible.
For the purpose of scoring, the magnitude of the velocity
vector 222zyxxyz VVVV ++= was calculated based on the
Vx, Vy, and Vz components in each control volume of the
casting. The maximum velocity magnitude represented
the velocity objective measure.
For maximizing casting yield, the metal yield was
calculated based on the volume ratio of the actual casting
and the total gating and riser system:
%100% ×+
=+risergatingcast
cast
VolumeVolume
VolumeYield
Since minimization of the objective function was
chosen, the yield loss, Yl = 1-Y was used for the
objective measure.
3.2 Gating and Riser Design Case Study
TEST CASTING
A cylindrical housing model was used to demonstrate
the MOEA optimization strategy. It has an outer radius
of 260 mm and 160 mm at the largest and narrowest
part, an inner radius of 120 mm and 180 mm at the upper
half and bottom part respectively and a height of 245
mm. This casting is relatively large with a total weight of
30 kg. Bottom filling of the mold was employed. The
three-dimensional CAD model of the test casting and its
gating and riser system is shown in Figure 5.
For this experiment, cylindrical shaped gating and riser
components were used for simplicity. Due to the
symmetry of the test casting, the number of design
variables was reduced to four independent parameters.
The design variables were the radii of the top risers ar,
middle risers br and their heights rh as well as the radii of
the ingates and runner gr as shown in Figure 6.
Dimensional constraints were also imposed to the design
variables to maintain realistic shapes as given in the
Table 1.
The optimization problem for this study is defined as
follows:
[ ] where
% 30)(
% 0)(
cm/s 50)(,,
)( subject to
)(
)(
)(
)( Minimize
T
l
l
hrrr
zyx
xyz
rgba
Y
P
VVV
Y
P
V
=
<
=
<
=
=
x
x
x
x
xg
x
x
x
xf
Preprints of the 18th IFAC World CongressMilano (Italy) August 28 - September 2, 2011
11644
Fig. 5. The gating and riser components of the test
casting.
Fig. 6. The gating and riser components and design
variables of the test housing casting.
COMPUTER SIMULATION
The initial gating and riser system as well as the housing
model were created using the preprocessor module of
MAGMASOFT®. The enmeshment of the geometry was
made up of approximately 110,000 elements.
Magnesium alloy AZ91 was used as the casting material
and dry silica was used for the sand mould. The
thermophysical properties of both the cast materials were
available in the database module of MAGMASOFT®.
The pouring temperature is 650°C, die initial
temperature is 20°C and the heat transfer coefficient is
800 W/m2K. The filling time was 12 seconds.
For the MOEA implementation, a string size of 20 was
used for the binary encoding. The crossover rate was 0.8
and the mutation rate was set at 0.05. Single point binary
crossover and bit-wise flip mutation operators were
used. The binary tournament size used was 2 for
selection. The initial population was randomly generated
and the number of generations, G was used as the
terminating criterion where it is set according to the
maximum number of designs allowed by the total
acceptable run time.
Table 1 Design variables and the parameter range
Parameters
(mm) ar br gr rh
Upper Bound 40 10 12 20
Lower Bound 70 40 35 100
String Size 4 4 4 4
4. RESULTS AND DISCUSSIONS
In this study, only the velocity and yield objectives were
plotted in the objective space for easier visualization
whereas the porosity objective was treated as a
constraint. Using NSGA II with N=20, a set of non-
dominated solutions was found in each generation, G.
Since constraints were explicitly handled, the solutions
found were ensured to satisfy all the design
requirements.
50.0
55.0
60.0
65.0
70.0
75.0
80.0
85.0
70.0 75.0 80.0 85.0 90.0 95.0
Yield (%)
Vxyz (cm/s)
G1
G4
G8
G12
G16
Fig.7. Shift of the Pareto front in different
generations (7SGA II, 7=20).
50.0
55.0
60.0
65.0
70.0
75.0
80.0
85.0
75.0 80.0 85.0 90.0 95.0
Yield (%)
Vxyz (cm/s)
N12 G20
N20 G16
N40 G8
Fig.8. Shift of the Pareto front in different generations and
using different population sizes.
The shift of the Pareto front found in G = 1, 4, 8, 12 and
16 is illustrated in Figure 7 and it can be seen that the
solutions improved as G progressed while better
solutions were being discovered. At the completion of
one run, 6 Pareto optimal solutions were identified and
are listed in Table 2. The solutions found had a good
spread that gave trade-off information between the
competing velocity and yield objectives as shown in
Figure 7. Based on the proposed optimization
framework, optimal gating and riser system designs
could be successfully obtained even with the absence of
auxiliary knowledge or analytical information in the
problem formulation. Similarly, an initial population
made up of all bad designs did not impede the ability of
the optimization algorithm in finding better feasible
solutions.
Table 2 Pareto optimal set obtained using 7SGAII with
7=20 and G=16
Preprints of the 18th IFAC World CongressMilano (Italy) August 28 - September 2, 2011
11645
VARIABLES OBJECTIVES
ar br gr rh Vx (cm/s)
Vy (cm/s)
Vz (cm/s)
Velocity (cm/s)
Porosity (%)
Yield (%)
68.7
10.0
23.1
32.9
46.6 41.9 42.3 75.6 0.0 87.1
58.4
10.0
23.1
63.9
35.2 42.6 43.1 70.1 0.0 86.1
62.3
13.9
26.8
63.9
33.6 39.8 39.5 65.6 0.0 84.4
59.7
12.9
28.3
61.3
39.0 41.1 30.9 64.3 0.0 82.8
54.5
13.9
32.8
63.9
39.0 37.7 20.0 57.8 0.0 79.1
70.0
12.9
34.3
61.3
39.2 35.2 22.4 57.2 0.0 76.8
70.0
12.9
34.3
61.3
39.2 35.2 22.4 57.2 0.0 76.8
4.1 Effects of Population Sizing
Various suggestions for population sizing can be found
in literature. According to the De Jong’s standard
settings (De Jong, 1975), the suggested population size
is 50. Latter research by Schaffer (Schaffer, 1989) and
Grefenstette (Grefenstette, 1986) showed that using a
moderate size of about 20-30 could solve the problem
faster without excessive computational costs. A smaller
population size of 4 to 10 was also recommended in
some recent research on micro-GA to accelerate fitness
convergence (Zhang, 2004; Ye, 2000). For this study,
three different population sizes of 12, 20 and 40 were
employed to compare their ability to find the Pareto front
as shown in Figure 8.
Based on the results, the population size of 20 was found
to be most appropriate for finding the optimal solutions
efficiently. The moderately-sized population allowed a
sufficient pool of genes to exist in the population for
effective crossover operations compared to an
inadequate size of 12. It also had a lower computational
cost compared to the population size of 40.
4.2 Effects of Requirement Change
The MOEA technique gave flexibility in decisions
making since trade-off information was given by the
Pareto optimal set. With a new tightened yield
requirement, Y > 85% imposed, another corresponding
solution that satisfies this constraint could be easily
picked from the existing set in Figure 8. This is
advantageous from the practical standpoint because
objective priorities often change according to current
conditions in real world problems.
4.3. Optimized Design
Based on the results obtained using the proposed
optimization framework, comparisons of some gating
and riser designs are shown in Figures 9 and 10. The
influence of the runner and ingates sizes on the liquid
metal velocity in the casting is shown in Figure 9. It can
be observed that using a large gr for the gating system
increased the gating ratio and allowed the liquid metal to
be slowed down before entering the casting cavity.
The effect of riser size on the shrinkage porosity formed
in the casting is shown in Figure 10. It can be seen that
using top risers with a low height rh and wide radius ar
could not eliminate the porosity in the casting. Similar
infeasible results were also obtained when a tall riser
with a small radius was employed. Porosity can only be
eliminated from the casting with an optimally sized riser.
The identification of the proper gating and riser sizes
was autonomously examined by the optimization
algorithm and did not require any human intervention.
Fig. 9. Filling velocity results in the casting with
gating size of (a) gr = 20 mm, (b) gr = 25 mm, (c) gr =
30 mm.
Fig. 10. Shrinkage porosity result of the casting with
(a) short, wide riser, (b) tall, thin riser, (c) medium
sized riser, and (d) an optimal riser.
(a)
(b)
(c)
(d)
(a) (b)
(c) (d)
Preprints of the 18th IFAC World CongressMilano (Italy) August 28 - September 2, 2011
11646
5. CONCLUSIONS
In this study, the proposed optimization framework
managed to autonomously obtain 6 optimal design
choices after 320 design evaluations in a relatively short
period of time. Using the trial-and-error approach to
manually design, iterate and redesign could possibly take
up much more man hours and yet may not necessarily
obtain optimal designs. A population size of 20 was
found to be most efficient for finding the optimal
solutions. Regardless of the infeasible initial population,
optimal gating and riser system designs were
successfully obtained even with the absence of analytical
information. The main advantage was the ability to
obtain trade-off information between the competing
yield and quality objectives. This provided an insight
into the system characteristics and allowed flexibility in
decision making.
6. ACKNOWLEDGEMENTS
The authors would like to acknowledge the support of
NSERC and the University of Windsor for this work.
REFERENCES
Antonsson, E.K., Cagan, J 2001, “Formal Engineering Design
Synthesis”, Cambridge University Press, Cambridge, UK.
Campbell, J., 1991, “Castings”, Butterworth-Heinemann Ltd., Oxford, England.
Campbell, J., 1998, “The Ten Castings Rules Guidelines For The Reliable Production of Reliable Castings: A Draft Process Specification”, Proceedings of the Materials Solutions Conference on Aluminum Casting Technology, Chicago, pp. 3-19.
Carroll, D.L., 1997, “FORTRAN Genetic Algorithm Driver”, Website Article, University of Illinois, http://www.staff. uiuc.edu/carroll/ga.html.
Chung, J. S., 2002, "Process Optimal Design in Forging by Genetic Algorithm", Journal of Manufacturing Science and Engineering, 124(2), pp. 397-408.
Ciampoli, F., and Chew, J.W., 2007, "Automatic Optimization of Preswirl Nozzle Design", Journal of Engineering for Gas Turbines and Power, 129(2), pp. 387-393.
Coello,C.A., and G.T., Pulido, 2001, “A Micro-Genetic Algorithms For Multiobjective Optimization”, Proceedings of The 1st International Conference of Evolutionary Multi-Criterion Optimization, Springer Berlin Heidelberg, Zurich, Switzerland, 1993/2001, pp. 126-141.
De Jong, K., 1975, “An Analysis of The Behavior of A Class of Genetic Adaptive Systems”, PhD Thesis, Dissertation Abstracts International, 36(10):5140b, University Microfilms No. 76-9381, University of Michigan, Michigan.
Deb, K., and Jain, S., 2003, “Multi-Speed Gearbox Design Using Multi-Objective Evolutionary Algorithms”, Journal of Mechanical Design, 125(3), pp. 609-619.
Fuoco, R., and Correa, E.R., 2004, “The Effect of Gating System Design on The Quality of Aluminum Gravity Castings”, Website Article, http://www.moderncasting.com.
Grefenstette, J., 1986, “Optimization of Control Parameters for Genetic Algorithms”, IEEE Transactions on Systems, Man and Cybernetics, 16(1), pp. 122 – 128.
Guleyupoglu, S., 1997, “Casting Process Design Guidelines”, AFS Transactions, 105, pp. 869-876.
Jones, D.F., Mirrazavi, S.K., and Tamiz, M., 2002, “Multi-Objective Meta-Heuristics: An Overview of The Current State-of-The-Art”, European Journal of Operational Research, Elsevier, 137(1), pp.1-9.
Kor, J., Li, Z., Hu, H., Chen, X., Wang, Q., and Yang, W., 2006, “Numerical Understanding of John Campbell's Casting Design Rules”, Proceedings of Simulation of Aluminum Shape Casting Processing: From Alloy Design to Mechanical Properties, Q.G. Wang et al., eds., Warrendale, PA, pp. 177-187.
Manzari, M.T, Gethin, DT., and Lewis, R.W., 2000, “Optimum Design of Chills in Sand Casting Process”, Proceedings of the International Mechanical Engineering Congress and Exposition, ASME, Orlando, FL.
McDavid, R.M., and Dantzig, J., 1998, “Design Sensitivity and Finite Element Analysis of Free Surface Flows with Application to Optimal Design of Casting Rigging Systems”, International Journal for Numerical Methods in Fluids, Wiley Interscience, 28(3), pp. 419-442.
Merkel, M., and Schumacher, A., 2003, “An Automated Optimization Process for a CAE Driven Product Development”, Journal of Mechanical Design, 125(4), pp. 694-700.
Otsuka, Y., Trapaga, G., and Szekely, J., 1994, “Optimal Casting Design by Application of Mold Filling and Solidification Simulation”, Light Metals, Minerals, Metals and Materials Society, USA, pp. 897-905.
Rezvani, M., Yang, X., and Campbell, J. 1999, “The Effect of Ingate Design on The Strength and Reliability of Al. Castings”, AFS Transactions, 107, pp. 181-188.
Schaffer, J.D., Caruana, R.A., Eshelman, L J., and Das, R., 1989, “A Study of Control Parameters Affecting Online Performance of Genetic Algorithms For Function Optimization”, Proceedings of The 3rd International Conference on Genetic Algorithms, pp. 51-60.
Scott-Murphy , A., Kalyanasundaram, S., and Cardew-Hall,M., 2004, "A Hybrid Draw Die Optimization Technique for Sheet Metal Forming", Journal of Manufacturing Science and Engineering, 126(3), pp. 582-590.
Tai,C.C., and Lin, J.C., 1998, “A Runner-Optimization Design Study of A Die-Casting Die”, Journal of Materials Processing Technology, 84(1), pp. 1-12.
Valtierra-Gallardo, J., S., Talamantes-Silva, and Gonzalez–Villarreal, J.A., 2005, “Gating Design After Campbell - A Practical Experience with The Daily Struggle with Porosity”, Proceedings of Shape Casting: The John Campbell Symposium, M. Tiryakioglu and P. N. Crepeau, eds., pp. 455-462.
Wang, G.G., and Shan, S., 2007, “Review of Metamodeling Techniques in Support of Engineering Design Optimization”, Journal of Mechanical Design, 129(4), pp. 370-380.
Ye, K.Q., Li, W., and Sudjianto, A., 2000, “Algorithmic Construction of Optimal Symmetric Latin Hypercube Designs”, Journal of Statistical Planning and Inference, pp. 145-159.
Yoshimura, M., Shinji Nishiwaki, S., and Izui, K., 2005, “A Multiple Cross-Sectional Shape Optimization Method for Automotive Body Frames”, Journal of Mechanical Design, 127(1), pp. 49-57.
Zhang, F., Chen, S., and Khalid, M., 2004, “Multi-Point Optimization of Transonic Wing by Real-Coded Genetic Algorithm”, Proceedings of the 10th AIAA/ISSMO Multidisciplinary Analysis and Optimization Conference, Albany, NY.
Preprints of the 18th IFAC World CongressMilano (Italy) August 28 - September 2, 2011
11647